#1537: merge — restore the closing brace of the range Paste Formatting test

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FypgCX8G9hbs6XYZFsgjbL

.github/dependabot.yml

@@ -7,11 +7,30 @@ updates:
       interval: "daily"
       time: "06:00"
       timezone: "America/New_York"
-    # Consolidate updates into a single PR rather than one PR per crate
+    # Minor and patch updates share one PR. A breaking bump gets its own PR so one API change
+    # doesn't hold the safe updates back (#2973). For cargo, Dependabot counts a 0.y bump of a
+    # 0.x crate as major, as Cargo does. A dependency goes to the first group it matches, so the
+    # crate families that must move together come before the catch-all.
     groups:
+      egui:
+        patterns:
+          - "egui"
+          - "egui_*"
+          - "eframe"
+          - "winit"
+      fonts:
+        patterns:
+          - "skrifa"
+          - "harfrust"
+      hayro:
+        patterns:
+          - "hayro-*"
       cargo-dependencies:
         patterns:
           - "*"
+        update-types:
+          - "minor"
+          - "patch"
     commit-message:
       prefix: "chore(deps)"
 

.github/workflows/house-rules.yml

@@ -0,0 +1,28 @@
+name: House rules
+
+on:
+  push:
+    branches: [main]
+    paths:
+      - AGENTS.md
+      - CLAUDE.md
+      - .github/workflows/house-rules.yml
+  pull_request:
+    paths:
+      - AGENTS.md
+      - CLAUDE.md
+      - .github/workflows/house-rules.yml
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: house-rules-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  house-rules:
+    uses: storytold/craft-common-actions/.github/workflows/house-rules.yml@main
+    with:
+      rust: true

.github/workflows/release.yml

@@ -18,7 +18,14 @@ concurrency:
   cancel-in-progress: false
 
 jobs:
+  # Regenerates contributors/contributors.json for the release commit; every build bundles it.
+  credits:
+    permissions:
+      contents: read
+      pull-requests: read
+    uses: storytold/craft-common-actions/.github/workflows/refresh-contributors.yml@main
   release:
+    needs: credits
     # Only this job publishes the GitHub release, so only it gets write access.
     permissions:
       contents: write
@@ -31,6 +38,7 @@ jobs:
       craft-fonts-ref: b076417b5880bd4154bd22f00695b1d28c6bea83
       web-craft-fonts: true
       version: ${{ inputs.version }}
+      contributors-artifact: ${{ needs.credits.outputs.artifact-name }}
     secrets:
       APPLE_CERTIFICATE: ${{ secrets.APPLE_CERTIFICATE }}
       APPLE_CERTIFICATE_PASSWORD: ${{ secrets.APPLE_CERTIFICATE_PASSWORD }}
@@ -46,3 +54,4 @@ jobs:
       AZURE_SIGNING_ENDPOINT: ${{ secrets.AZURE_SIGNING_ENDPOINT }}
       AZURE_SIGNING_ACCOUNT: ${{ secrets.AZURE_SIGNING_ACCOUNT }}
       AZURE_CERT_PROFILE: ${{ secrets.AZURE_CERT_PROFILE }}
+      LINUX_SIGNING_KEY: ${{ secrets.LINUX_SIGNING_KEY }}

AGENTS.md

@@ -1,10 +1,62 @@
+<!-- craft-house-rules:start core source=dde09ea31e2e457f5b4653a6e76f7cdf55944bdf sha256=540e6cea75d12d0710c3dca89658328e81e508ecef102446ca5bb95b7c0b92bb -->
+## House rules
+
+These apply in every Craft repository. A later rule may be stricter, not weaker. The source is `standards/house-rules-core.md` in `storytold/craft-contrib`. Propose changes there, not in a copy.
+
+1. **A person is accountable.** Whoever runs an agent answers for what it opens and for its review comments.
+2. **Check, then claim.** Before writing code, read the issue comments, search pull requests for `#N` and the keywords, and check recent commits on `main`. Comment on the issue first. Release a claim you will not finish.
+3. **Verify, don't guess.** Open a definition before using it. Confirm the cause in the code, even when someone names a location.
+4. **Keep fixes narrow and tested.** Fix what the issue describes, where the bug is. Add a test that fails without the fix.
+5. **Review the diff before opening a pull request.** Identifiers exist as used, the test fails without the fix, nothing unrelated changed, and the description is true.
+6. **Look for a better pull request.** If another fixes the same thing, close yours when it is better or equal, and say why.
+7. **Report only what ran.** Say what you tested and what you could not run. Without a local build or test, say so and rely on CI before marking ready.
+8. **`main` changes through a pull request.** Do not push to it. Branch protection enforces this. There is no exception.
+9. **Stop and ask only about** licensing outside the allowlist, publishing, releases, a new remote, secrets, spending money, destructive git, and anything else hard to undo. Otherwise keep working. `plan/` is a gitignored notebook, not rules. Without it, do not ask for it and do not stop.
+10. **Keep secrets and personal details out** of the repository, logs and requests.
+11. **CI stays shared.** Workflows call `storytold/craft-common-actions`. Do not change CI in a feature or fix pull request. If CI cannot cover a change, open an issue there.
+12. **Guards say why.** A check that looks removable says what it prevents. Read that before changing it.
+13. **Instructions must be readable by everyone.** No notes about one machine, and no links to files a reader cannot open. `craftrules` is retired. Do not link to it.
+14. **Keep pull requests moving.** Concurrent pull requests are fine; unanswered actionable feedback is not. Respond in 7 days, a reminder at 14, closure eligible at 30, from when an actionable request waits on the author. Suggestions and unconfirmed bot findings do not start the clock. A fix, explanation or agreed plan resets it; unrelated activity does not. Waiting on a reviewer or maintainer pauses it and does not count against the author. A maintainer decides closure; it is never automatic. Reuse search results, do not poll, and stop at a rate limit.
+15. **No agent attribution.** A pull request does not say an agent wrote it. A repository rule may not require that.
+<!-- craft-house-rules:end -->
+<!-- craft-house-rules:start rust source=dde09ea31e2e457f5b4653a6e76f7cdf55944bdf sha256=387e0ccb4b49e567f60951d7d6fff7305054c89f43ad812b9993dc7c38528013 -->
+## Rust Craft repositories
+
+These apply to native Craft apps and to shared Rust libraries. Select them by repository type. A repository that merely contains some Rust does not take this block. Rules later in a repository's file may be stricter, but not weaker. The source is `standards/house-rules-rust.md` in [storytold/craft-contrib](https://github.com/storytold/craft-contrib).
+
+1. **Never crash.** Non-test code does not panic and has no `unsafe`, except where the repository file names an exception. Return errors, treat input as hostile, and bound recursion and allocations. New production crate roots copy the `deny` attribute from an existing crate: `clippy::unwrap_used`, `clippy::expect_used`, `clippy::panic`, `clippy::unimplemented`, `clippy::todo`, `clippy::unreachable`. Every crash fix comes with a test that panicked before the fix.
+2. **Clean room.** Learn from observed behaviour and public specifications. Do not copy another product's code, assets, strings or artwork. Check a dependency's licence before adding it.
+3. **Assets and fonts.** Every asset is original or suitably licensed, and is recorded in `ATTRIBUTION.md` in the same change. Font files live in [storytold/craft-fonts](https://github.com/storytold/craft-fonts), not in the app repository.
+4. **Rust only for native Craft apps.** No handwritten JavaScript or TypeScript, and no Tauri, Electron or webview shell, in those apps. Shared tooling follows its documented technical scope and maintainer-approved exceptions. Craft PR Review's existing Tauri and TypeScript UI is one such exception. It does not relax this rule for native apps. Preserve any existing web build.
+5. **Time goes through `chrono`.** Do not hand-roll calendar conversion. For dates beyond what `chrono` can represent, follow [standards/time.md](https://github.com/storytold/craft-contrib/blob/main/standards/time.md).
+6. **Contributor credits.** Add or edit only your own person's entry in `contributors/people.toml`, and only after asking them. Never guess a name or hand-edit `contributors.json`.
+
+## Licence allowlist
+
+Dependencies and assets use SPDX identifiers. An expression is allowed when every identifier is in this list. A repository may name a shorter list. It may not add an identifier here by editing its own copy of these rules.
+
+- `ISC`
+- `MIT`
+- `BSD-2-Clause`
+- `BSD-3-Clause`
+- `Apache-2.0`
+- `OFL-1.1`
+- `Ubuntu-font-1.0`
+- `Bitstream-Vera`
+- `CC0-1.0`
+- `CC-BY-4.0`
+- `CC-BY-SA-4.0`
+- `LicenseRef-Public-Domain`
+
+`LicenseRef-Storyteller-Trademark` is not in this list. It covers Storyteller brand marks only, where a repository already says so.
+<!-- craft-house-rules:end -->
+
 # WordCraft — instructions for agents
 
 WordCraft is a clean-room, open-source, Rust-native word processor targeting Microsoft Word parity — and going further on speed, openness and agent control. It runs natively on macOS, Windows, Linux and BSD, and on the web via WASM. Siblings with the same conventions: `../photocraft` (Photoshop), `../vectorcraft` (Illustrator), `../filmcraft` (Premiere), `../lightcraft` (Lightroom), `../pdfcraft` (Acrobat), `../effectcraft` (After Effects), `../designcraft` (InDesign).
 
 ## Start every session here
-1. Read `plan/STATUS.md` (current milestone, next task), then the task in `plan/execution-plan.md` and the relevant `plan/architecture.md` section. `plan/` is gitignored (local only); if it's missing, start from `ROADMAP.md`, `docs/gaps.md` and `docs/roadmap.md` (Current focus).
-2. Follow the autonomous operation protocol (`plan/execution-plan.md` §5). Don't stop to ask unless a decision is genuinely the owner's (licensing, publishing, pushing to new remotes, secrets).
+Read [`ROADMAP.md`](ROADMAP.md), [`docs/gaps.md`](docs/gaps.md) and [`docs/roadmap.md`](docs/roadmap.md). `plan/` is an optional gitignored notebook. Do not ask for it and do not stop when it is absent.
 
 ## Never crash
 People trust WordCraft with their writing; a crash loses their work. **This outranks feature work**: never ship a feature through a panic path, and fix a crash before building on top of it.
@@ -28,7 +80,7 @@ People trust WordCraft with their writing; a crash loses their work. **This outr
 - Never read, disassemble or copy anything inside the Word bundle (file names/listings only). Never commit files produced by Word. Never copy Microsoft wording beyond feature names.
 - File formats come from public specs: ECMA-376 / ISO 29500 (OOXML), the published RTF spec, OASIS ODF 1.3. Never copy GPL/LGPL/AGPL code (LibreOffice, AbiWord, pandoc…).
 
-## Architecture (see `plan/architecture.md`)
+## Architecture
 | Layer | Crate | Job |
 |---|---|---|
 | L0 | `geom` | units, rects, measurement parsing |
@@ -44,7 +96,6 @@ People trust WordCraft with their writing; a crash loses their work. **This outr
 - **Everything is a command.** User-visible behaviour = a `CommandSpec` in `crates/engine/src/cmd/*.rs` (id, label, ribbon location, shortcut, params doc, `enabled`, `run`) + tests. UI-only commands (`ui.*`) live in `crates/ui-egui/src/lib.rs`. The ribbon, shortcuts, command search, CLI, control channel and MCP all dispatch by id. Programmatic calls never open dialogs.
 - **Parity is measured:** `crates/engine/src/catalog.rs` lists Word's ribbon/menu features with command ids. `cargo xtask parity` writes `docs/parity-checklist.md`; a test enforces a floor that only rises. Implement missing ids to raise it.
 - **Layering** is enforced by `cargo xtask layers`. Nothing below L6 depends on egui/eframe/winit/rfd. The UI is thin and reads `Session` state; colours come from `theme::Tokens`.
-- **Time goes through `chrono`** (workspace dependency). Never hand-roll UTC/calendar conversion (days-to-date algorithms, leap-year math). A crate owns only its output format string. The one exception is range extension: past the years chrono can represent, shift the date by whole 400-year Gregorian cycles (146,097 days, after which the calendar repeats), let chrono compute the month and day, and add the cycles back to the year (splitting seconds into hours, minutes and seconds is plain arithmetic).
 - **Rust only** (no handwritten JS/TS). **Never break wasm** (`cargo xtask wasm`).
 
 ## Quality gates
@@ -85,3 +136,5 @@ Progress docs: `ROADMAP.md` (stage, headline numbers, progress log), `docs/targe
 - Workflows in `.github/workflows/` call the shared workflows in [`storytold/craft-common-actions`](https://github.com/storytold/craft-common-actions) at `@main`, with their inputs and secrets. No other jobs: if no shared workflow does what's needed, it doesn't go here.
 - Don't add or change CI in a feature or fix PR. If the existing workflows can't build or test a change, link an existing craft-common-actions issue or PR, or open a new issue, or a PR only for a small fix that is safe for every app and does not increase CI time.
 - Some workflow files here still run their own jobs. Don't add to them. Moving a job into craft-common-actions is its own task, never part of another PR, and the job stays here until a shared workflow runs it.
+
+New workspace crates include `workspace = true` under `[lints]` in `Cargo.toml`. Production crate roots declare the crate-level lint attributes before modules; crate-level `//!` documentation may precede them.

CLAUDE.md

@@ -1,71 +1 @@
-# WordCraft — instructions for agents
-
-WordCraft is a clean-room, open-source, Rust-native word processor targeting Microsoft Word parity — and going further on speed, openness and agent control. It runs natively on macOS, Windows, Linux and BSD, and on the web via WASM. Siblings with the same conventions: `../photocraft` (Photoshop), `../vectorcraft` (Illustrator), `../filmcraft` (Premiere), `../lightcraft` (Lightroom), `../pdfcraft` (Acrobat), `../effectcraft` (After Effects), `../designcraft` (InDesign).
-
-## Start every session here
-1. Read `plan/STATUS.md` (current milestone, next task), then the task in `plan/execution-plan.md` and the relevant `plan/architecture.md` section. `plan/` is gitignored (local only); if it's missing, start from `ROADMAP.md`, `docs/gaps.md` and `docs/roadmap.md` (Current focus).
-2. Follow the autonomous operation protocol (`plan/execution-plan.md` §5). Don't stop to ask unless a decision is genuinely the owner's (licensing, publishing, pushing to new remotes, secrets).
-
-## Never crash
-People trust WordCraft with their writing; a crash loses their work. **This outranks feature work**: never ship a feature through a panic path, and fix a crash before building on top of it.
-- **No panics in non-test code:** no `unwrap()`, `expect()`, `panic!`, `unreachable!`, `todo!`, `unimplemented!`; no `unsafe` (`unsafe_code = "forbid"`).
-- **Errors are `Result<T, E>`** through the crate's error type and `?`. An unfinished feature returns an error or is disabled; it never panics.
-- **Input-derived numbers are hostile** (files, commands, MCP/control params): `get()` instead of `[i]`, slice strings only at char boundaries, checked/saturating arithmetic, clamp sizes, cap allocations, bound recursion.
-- **Last-resort guard:** `Session::run` turns an escaped panic into an error and restores the document. It's a safety net, not a licence.
-- **Prove it:** every crash fix lands with a regression test. `crates/engine/src/tests.rs::hostile_params_never_panic` runs every command with junk params — keep it green.
-- Production crate roots carry `#![deny(clippy::unwrap_used, clippy::expect_used, clippy::panic, clippy::unimplemented, clippy::todo, clippy::unreachable)]`; new crates start with it and add `[lints]` with `workspace = true` to their `Cargo.toml`. Tests may `unwrap` (`clippy.toml`).
-
-## Assets: absolutely no Microsoft (or Adobe/Avid/Autodesk) iconography or images
-**This rule is non-negotiable.** Breaking it is the worst mistake you can make in this repo.
-- Never copy, trace, screenshot-crop or recreate icons, images, templates, themes, clip art, fonts, cursors, sounds or any other asset from Microsoft Word/Office or any Microsoft, Adobe, Avid or Autodesk product. Not even "temporarily".
-- Every asset must be **original** (drawn in code — prefer this: `crates/ui-egui/src/icons.rs`), **public domain / CC0**, **OSI open source**, or **Creative Commons that allows redistribution**, or licensed open source by a contributor who made it.
-- **Every asset file has a row in [`ATTRIBUTION.md`](ATTRIBUTION.md)** (author, source, licence). `cargo xtask assets` (part of `cargo xtask ci`) fails otherwise. Add the row in the same commit as the file.
-- Fonts: new font files go to [`storytold/craft-fonts`](https://github.com/storytold/craft-fonts), never this repo; it's an optional build input (`CRAFT_FONTS_DIR`). Spelling word lists and hyphenation patterns live in [`storytold/craft-i18n`](https://github.com/storytold/craft-i18n) `language-data/`, never this repo; they are embedded from `CRAFT_I18N_DIR` (pinned by `craft-i18n.rev`, which CI checks out). Without it the app builds, but English spelling falls back to an installed Hunspell English or the built-in Moby word list (with no American/British variant table) and pattern hyphenation is off (Bulgarian, Turkish and Russian still hyphenate by rule), so proofing tests need it. Documents asking for proprietary fonts (Calibri, Aptos, Cambria…) are rendered with installed system fonts or metric-compatible open substitutes (`crates/fonts/src/word.rs`) — never bundle proprietary fonts.
-- Templates, sample documents and themes are our own text and colours (`crates/engine/src/sample.rs`, `cmd/design.rs`). Demo images are public domain, with sources recorded.
-
-## Clean room
-- Microsoft Word is installed on the dev machine and may be **observed** black-box: run it, use its UI with synthetic documents, take screenshots **of its window only** (by window id — the machine runs other work; never capture the whole desktop) stored only under `plan/word/screenshots/` (gitignored, never committed).
-- Never read, disassemble or copy anything inside the Word bundle (file names/listings only). Never commit files produced by Word. Never copy Microsoft wording beyond feature names.
-- File formats come from public specs: ECMA-376 / ISO 29500 (OOXML), the published RTF spec, OASIS ODF 1.3. Never copy GPL/LGPL/AGPL code (LibreOffice, AbiWord, pandoc…).
-
-## Architecture (see `plan/architecture.md`)
-| Layer | Crate | Job |
-|---|---|---|
-| L0 | `geom` | units, rects, measurement parsing |
-| L1 | `doc`, `fonts` | document model + edit primitives; font DB, shaping, Word font substitution |
-| L2 | `layout`, `docx`, `formats` | line breaking, pagination, tables, headers/footers, hit testing; OOXML I/O; TXT/MD/HTML/RTF/ODT |
-| L3 | `render`, `pdf` | vello_cpu rasteriser; krilla PDF |
-| L4 | `engine` | Session, undo, **command registry**, Word feature catalog, samples, I/O dispatch |
-| L5 | `mcp` | MCP server (headless or bridged to the app) |
-| L6 | `ui-egui` | Word-style egui front end (swappable) |
-| apps | `apps/wordcraft`, `apps/wordcraft-cli`, `apps/wordcraft-web` | binaries |
-| external | `craft-proof` (craft-libs, used as `wordcraft-proof`, sits at L1) | proofing: spelling, grammar, hyphenation; its word lists and patterns are craft-i18n's `language-data/` (`CRAFT_I18N_DIR`) |
-
-- **Everything is a command.** User-visible behaviour = a `CommandSpec` in `crates/engine/src/cmd/*.rs` (id, label, ribbon location, shortcut, params doc, `enabled`, `run`) + tests. UI-only commands (`ui.*`) live in `crates/ui-egui/src/lib.rs`. The ribbon, shortcuts, command search, CLI, control channel and MCP all dispatch by id. Programmatic calls never open dialogs.
-- **Parity is measured:** `crates/engine/src/catalog.rs` lists Word's ribbon/menu features with command ids. `cargo xtask parity` writes `docs/parity-checklist.md`; a test enforces a floor that only rises. Implement missing ids to raise it.
-- **Layering** is enforced by `cargo xtask layers`. Nothing below L6 depends on egui/eframe/winit/rfd. The UI is thin and reads `Session` state; colours come from `theme::Tokens`.
-- **Rust only** (no handwritten JS/TS). **Never break wasm** (`cargo xtask wasm`).
-
-## Quality gates
-Before every commit: `cargo xtask ci` (fmt, clippy -D warnings, tests, assets, layers, wasm). Commit after every arc of work that builds, with a task id in the message (`M2.1: line spacing dialog`).
-
-## Typing behaves like Word
-Typing, Enter, Backspace, Tab, lists, AutoCorrect and AutoFormat follow Word's observed behaviour in [`docs/typing-parity.md`](docs/typing-parity.md), pinned by `crates/engine/src/tests/typing.rs` (keys go through the same commands as the keyboard). Change behaviour, doc and tests together; a typing fix lands with a test there. Compare headlessly (`ui_shot`); don't drive GUIs with keystrokes on the shared machine.
-
-## Running and looking at the app
-- `cargo run --release -p wordcraft -- --sample --control 7981` (sample document + control channel).
-- Drive it with JSON lines on `127.0.0.1:7981`, each with the window's key from `~/.config/wordcraft/control-key.local-7981` (macOS: `~/Library/Application Support/WordCraft/`), e.g. `{"id":1,"key":"…","method":"engine.execute","params":{"command":"text.insert","params":{"text":"Hello"}}}` then `{"id":2,"key":"…","method":"ui.screenshot","params":{"path":"/tmp/shot.png"}}`. Methods: `crates/ui-egui/src/control.rs`; docs: `docs/control-protocol.md` (Keys).
-- **Headless window screenshots** (no focus stealing, works with a locked screen): `cargo run --release -p wordcraft-ui-egui --example ui_shot -- script.jsonl` (see the example's header).
-- **For UI work, look at the result** (screenshot, read the PNG) and compare with Word side by side.
-- CLI: `wordcraft-cli convert in.docx out.pdf`, `wordcraft-cli run --template sample --cmd 'select.text={"text":"Studio"}' --cmd format.bold --save out.docx`, `wordcraft-cli commands`, `wordcraft-cli parity`.
-- MCP: `wordcraft-cli mcp` (headless) or `wordcraft-cli mcp --connect 127.0.0.1:7981` (drives the running app). See `docs/mcp.md`.
-- Shell gotcha: `mv`/`cp` may be aliased interactive — use `/bin/mv -f` / `/bin/cp -f`.
-- Parallel agents: separate `CARGO_TARGET_DIR` per agent; edit only the crates you own; delete your target dir when done (disk).
-
-## Roadmap
-Progress docs: `ROADMAP.md` (stage, headline numbers, progress log), `docs/target-app-parity.md` (the assessment), `docs/gaps.md` (ranked work list), `docs/roadmap.md` (milestones, Current focus), `docs/architecture.md`, and the parity checklists `docs/file-format-parity.md`, `docs/layout-parity.md`, `docs/typing-parity.md`, `docs/ui-parity.md`, `docs/hardware-parity.md`, `docs/localization-parity.md`. When work lands, update the gap, the parity doc and the progress log, and bump each touched doc's status line and revision history.
-
-## CI workflows
-- Workflows in `.github/workflows/` call the shared workflows in [`storytold/craft-common-actions`](https://github.com/storytold/craft-common-actions) at `@main`, with their inputs and secrets. No other jobs: if no shared workflow does what's needed, it doesn't go here.
-- Don't add or change CI in a feature or fix PR. If the existing workflows can't build or test a change, link an existing craft-common-actions issue or PR, or open a new issue, or a PR only for a small fix that is safe for every app and does not increase CI time.
-- Some workflow files here still run their own jobs. Don't add to them. Moving a job into craft-common-actions is its own task, never part of another PR, and the job stays here until a shared workflow runs it.
+@AGENTS.md

Cargo.lock

@@ -1081,7 +1081,7 @@ dependencies = [
 [[package]]
 name = "craft-pdf"
 version = "0.1.0"
-source = "git+https://github.com/storytold/craft-libs?rev=92d392cff132eaf53e13afff77f38b5c11b73aa0#92d392cff132eaf53e13afff77f38b5c11b73aa0"
+source = "git+https://github.com/storytold/craft-libs?rev=c5a1cbfb5f9049f9270955caffb5bee5997aebf7#c5a1cbfb5f9049f9270955caffb5bee5997aebf7"
 dependencies = [
  "krilla",
  "thiserror 2.0.21",
@@ -1090,9 +1090,10 @@ dependencies = [
 [[package]]
 name = "craft-print"
 version = "0.1.0"
-source = "git+https://github.com/storytold/craft-libs?rev=92d392cff132eaf53e13afff77f38b5c11b73aa0#92d392cff132eaf53e13afff77f38b5c11b73aa0"
+source = "git+https://github.com/storytold/craft-libs?rev=a3e0b5a9440d18f19422a9dca13516747b7d3181#a3e0b5a9440d18f19422a9dca13516747b7d3181"
 dependencies = [
  "craft-winprint-gdi",
+ "log",
  "miniz_oxide 0.9.1",
  "roxmltree",
  "tempfile",
@@ -1157,6 +1158,26 @@ dependencies = [
  "zip 9.1.0",
 ]
 
+[[package]]
+name = "craft-web-api"
+version = "0.1.0"
+source = "git+https://github.com/storytold/craft-libs?rev=fd5e2feff6ea25bed371a99cc732b3968d621505#fd5e2feff6ea25bed371a99cc732b3968d621505"
+dependencies = [
+ "js-sys",
+ "serde_json",
+ "wasm-bindgen",
+ "wasm-bindgen-futures",
+]
+
+[[package]]
+name = "craft-web-host"
+version = "0.1.0"
+source = "git+https://github.com/storytold/craft-libs?rev=fd5e2feff6ea25bed371a99cc732b3968d621505#fd5e2feff6ea25bed371a99cc732b3968d621505"
+dependencies = [
+ "wasm-bindgen",
+ "web-sys",
+]
+
 [[package]]
 name = "craft-window"
 version = "0.1.0"
@@ -2960,7 +2981,7 @@ dependencies = [
 [[package]]
 name = "krilla"
 version = "0.8.2"
-source = "git+https://github.com/storytold/craft-libs?rev=92d392cff132eaf53e13afff77f38b5c11b73aa0#92d392cff132eaf53e13afff77f38b5c11b73aa0"
+source = "git+https://github.com/storytold/craft-libs?rev=c5a1cbfb5f9049f9270955caffb5bee5997aebf7#c5a1cbfb5f9049f9270955caffb5bee5997aebf7"
 dependencies = [
  "base64 0.22.1",
  "bumpalo",
@@ -6785,7 +6806,7 @@ dependencies = [
 
 [[package]]
 name = "wordcraft"
-version = "0.15.0"
+version = "0.15.1"
 dependencies = [
  "arboard",
  "chrono",
@@ -6805,7 +6826,9 @@ dependencies = [
  "open",
  "pollster 1.0.1",
  "rfd",
+ "serde",
  "serde_json",
+ "tempfile",
  "windows",
  "winit",
  "winresource",
@@ -6815,12 +6838,13 @@ dependencies = [
  "wordcraft-engine",
  "wordcraft-fonts",
  "wordcraft-ui-egui",
+ "x11rb",
  "zbus",
 ]
 
 [[package]]
 name = "wordcraft-cli"
-version = "0.15.0"
+version = "0.15.1"
 dependencies = [
  "serde_json",
  "sys-locale",
@@ -6833,11 +6857,11 @@ dependencies = [
 
 [[package]]
 name = "wordcraft-control-key"
-version = "0.15.0"
+version = "0.15.1"
 
 [[package]]
 name = "wordcraft-doc"
-version = "0.15.0"
+version = "0.15.1"
 dependencies = [
  "base64 0.23.1",
  "log",
@@ -6851,7 +6875,7 @@ dependencies = [
 
 [[package]]
 name = "wordcraft-docbin"
-version = "0.15.0"
+version = "0.15.1"
 dependencies = [
  "cfb",
  "chrono",
@@ -6868,7 +6892,7 @@ dependencies = [
 
 [[package]]
 name = "wordcraft-docx"
-version = "0.15.0"
+version = "0.15.1"
 dependencies = [
  "aes",
  "base64 0.23.1",
@@ -6880,6 +6904,7 @@ dependencies = [
  "log",
  "proptest",
  "quick-xml 0.41.0",
+ "serde_json",
  "sha1 0.11.0",
  "sha2 0.11.1",
  "thiserror 2.0.21",
@@ -6892,7 +6917,7 @@ dependencies = [
 
 [[package]]
 name = "wordcraft-engine"
-version = "0.15.0"
+version = "0.15.1"
 dependencies = [
  "cfb",
  "chrono",
@@ -6925,7 +6950,7 @@ dependencies = [
 
 [[package]]
 name = "wordcraft-fonts"
-version = "0.15.0"
+version = "0.15.1"
 dependencies = [
  "brotli-decompressor",
  "harfrust 0.13.3",
@@ -6939,26 +6964,34 @@ dependencies = [
 
 [[package]]
 name = "wordcraft-formats"
-version = "0.15.0"
+version = "0.15.1"
 dependencies = [
+ "aes",
+ "base64 0.23.1",
  "cfb",
  "chardetng",
  "chrono",
  "encoding_rs",
+ "getrandom 0.3.4",
+ "hmac",
  "image",
  "js-sys",
  "log",
+ "miniz_oxide 0.9.1",
  "proptest",
  "quick-xml 0.41.0",
+ "sha1 0.11.0",
+ "sha2 0.11.1",
  "wordcraft-doc",
  "wordcraft-geom",
  "wordcraft-metafile",
+ "zeroize",
  "zip 3.0.0",
 ]
 
 [[package]]
 name = "wordcraft-geom"
-version = "0.15.0"
+version = "0.15.1"
 dependencies = [
  "image",
  "serde",
@@ -6966,7 +6999,7 @@ dependencies = [
 
 [[package]]
 name = "wordcraft-layout"
-version = "0.15.0"
+version = "0.15.1"
 dependencies = [
  "craft-proof",
  "image",
@@ -6986,7 +7019,7 @@ dependencies = [
 
 [[package]]
 name = "wordcraft-mcp"
-version = "0.15.0"
+version = "0.15.1"
 dependencies = [
  "log",
  "serde_json",
@@ -6998,7 +7031,7 @@ dependencies = [
 
 [[package]]
 name = "wordcraft-metafile"
-version = "0.15.0"
+version = "0.15.1"
 dependencies = [
  "kurbo",
  "thiserror 2.0.21",
@@ -7006,7 +7039,7 @@ dependencies = [
 
 [[package]]
 name = "wordcraft-pdf"
-version = "0.15.0"
+version = "0.15.1"
 dependencies = [
  "chrono",
  "craft-pdf",
@@ -7027,7 +7060,7 @@ dependencies = [
 
 [[package]]
 name = "wordcraft-render"
-version = "0.15.0"
+version = "0.15.1"
 dependencies = [
  "image",
  "log",
@@ -7041,8 +7074,9 @@ dependencies = [
 
 [[package]]
 name = "wordcraft-ui-egui"
-version = "0.15.0"
+version = "0.15.1"
 dependencies = [
+ "accesskit_consumer",
  "cfb",
  "chrono",
  "craft-i18n",
@@ -7073,8 +7107,10 @@ dependencies = [
 
 [[package]]
 name = "wordcraft-web"
-version = "0.15.0"
+version = "0.15.1"
 dependencies = [
+ "craft-web-api",
+ "craft-web-host",
  "eframe",
  "egui",
  "js-sys",
@@ -7092,7 +7128,7 @@ dependencies = [
 
 [[package]]
 name = "wordcraft-zotero"
-version = "0.15.0"
+version = "0.15.1"
 dependencies = [
  "log",
  "proptest",
@@ -7255,7 +7291,7 @@ checksum = "9440ea3e5aeabb0ac63af70daf835274065238cdd0cec83418f417eae38bacee"
 
 [[package]]
 name = "xtask"
-version = "0.15.0"
+version = "0.15.1"
 dependencies = [
  "serde_json",
 ]

Cargo.toml

@@ -4,7 +4,7 @@ members = ["crates/*", "apps/*", "xtask"]
 default-members = ["crates/*", "apps/wordcraft", "apps/wordcraft-cli", "xtask"]
 
 [workspace.package]
-version = "0.15.0"
+version = "0.15.1"
 edition = "2024"
 license = "MIT OR Apache-2.0"
 rust-version = "1.95"
@@ -67,7 +67,7 @@ wuff = "0.2.9"
 brotli-decompressor = "5"
 vello_cpu = "0.2"
 image = { version = "0.25", default-features = false, features = ["png", "jpeg", "webp", "gif", "bmp", "tiff"] }
-craft-pdf = { git = "https://github.com/storytold/craft-libs", rev = "92d392cff132eaf53e13afff77f38b5c11b73aa0", version = "=0.1.0", default-features = false, features = ["write"] }
+craft-pdf = { git = "https://github.com/storytold/craft-libs", rev = "c5a1cbfb5f9049f9270955caffb5bee5997aebf7", version = "=0.1.0", default-features = false, features = ["write"] }
 # craft-update: app-specific pretend versions apply to startup and manual checks.
 craft-update = { git = "https://github.com/storytold/craft-libs", version = "=0.1.0", rev = "c5a1cbfb5f9049f9270955caffb5bee5997aebf7", features = ["egui"] }
 quick-xml = "0.41"
@@ -92,6 +92,8 @@ zeroize = "1"
 # dependency of other crates.
 libc = "0.2"
 
+# time
+
 # ui
 egui = "0.37"
 egui_extras = { version = "0.37", default-features = false }
@@ -132,7 +134,7 @@ debug = 0
 
 # Required by craft-pdf: the backend rejects writers that failed during unwinding.
 [patch.crates-io]
-krilla = { git = "https://github.com/storytold/craft-libs", rev = "92d392cff132eaf53e13afff77f38b5c11b73aa0" }
+krilla = { git = "https://github.com/storytold/craft-libs", rev = "c5a1cbfb5f9049f9270955caffb5bee5997aebf7" }
 ecolor = { git = "https://github.com/storytold/egui.git", rev = "6d518fc799681571963842f95b55a6b68431b15a" }
 emath = { git = "https://github.com/storytold/egui.git", rev = "6d518fc799681571963842f95b55a6b68431b15a" }
 epaint = { git = "https://github.com/storytold/egui.git", rev = "6d518fc799681571963842f95b55a6b68431b15a" }

GLOSSARY.md

@@ -24,3 +24,27 @@ The proofing languages offered in Set Proofing Language, Review › Language and
 menu: the languages WordCraft names, plus those of every installed (and, in Set Proofing Language,
 every available) Language pack, each once.
 _Avoid_: Language menu (that is only one of the places it is shown)
+
+**Your languages**:
+The proofing languages Options › Language lists: each English variant in use (the default proofing
+language and any in the open document), every language with an installed
+Language pack, and every proofing language used in the open document.
+_Avoid_: Authoring languages, preferred languages
+
+**Default proofing language**:
+The proofing language new blank documents get, set with Set as default; without one, the interface
+language.
+_Avoid_: Preferred language
+
+**Proofing status**:
+Where a language's proofing comes from: built in, an installed Language pack, your own dictionary,
+a system dictionary, available to install, or not available.
+
+**Custom dictionary**:
+The words you added with Add to Dictionary, accepted in every language.
+_Avoid_: User dictionary (in the interface)
+
+**Dictionaries folder**:
+A folder of Hunspell dictionaries you supply, used before the system's (an installed Language pack
+comes after it).
+_Avoid_: Custom dictionaries

README.md

@@ -299,10 +299,21 @@ adapter. The log file records the choice (lines starting with
 RISC-V (riscv64): `wordcraft-<ver>-linux-riscv64.tar.gz`, a tarball cross-compiled for glibc 2.39+
 (Ubuntu 24.04, Debian 13 and newer).
 
+**Read Aloud** (Review › Read Aloud) speaks through a speech program installed on the system: it
+tries `espeak-ng`, then `espeak`, and when neither is installed it talks to Speech Dispatcher.
+The `.rpm` package recommends `espeak-ng`; the `.deb` recommends `espeak-ng` or
+`speech-dispatcher`, so the package manager installs one of the two by default. With the
+AppImage or the tarball, install one yourself if neither is there (`sudo apt install espeak-ng`,
+`sudo dnf install espeak-ng`, `sudo pacman -S espeak-ng`). The Flatpak doesn't speak yet: it has
+no audio access and doesn't include `espeak-ng`, `espeak` or Speech Dispatcher.
+
 **Wayland:** on a Wayland desktop the app runs through XWayland when it is available, because
 winit (the windowing library) supports dragging files into the window only on X11. To use the
 native Wayland backend instead (for example for sharper text with fractional scaling, at the cost
-of file drag-and-drop), start it with `WORDCRAFT_WAYLAND=1`:
+of file drag-and-drop), choose **File › Options › Advanced › Linux display server › Wayland**; it
+applies the next time WordCraft starts, and **X11 (XWayland)** or **Automatic** switches back. The
+`WORDCRAFT_WAYLAND` environment variable overrides the setting: `1` for native Wayland, `0` for
+XWayland:
 
 ```sh
 WORDCRAFT_WAYLAND=1 wordcraft

apps/wordcraft-cli/src/main.rs

@@ -98,8 +98,8 @@ fn open(path: &Path, password: Option<&str>) -> Result<Session, String> {
     Ok(s)
 }
 
-/// The commands that open a file, and which of their arguments not starting with `--` it is (as
-/// `pos` counts). `run` opens the value of `--file`.
+/// The commands that open a file, and which positional argument it is, excluding option values.
+/// `run` opens the value of `--file`.
 const OPENS_FILE: &[(&str, usize)] = &[("convert", 0), ("info", 0), ("text", 0), ("inspect", 0), ("render", 0), ("zotero", 1)];
 
 /// The arguments as text. Only the file the command opens may have a name that isn't valid Unicode,
@@ -108,8 +108,14 @@ const OPENS_FILE: &[(&str, usize)] = &[("convert", 0), ("info", 0), ("text", 0),
 fn text_args<S: AsRef<OsStr>>(raw: &[S]) -> Result<(Vec<String>, Option<PathBuf>), String> {
     let args: Vec<String> = raw.iter().map(|a| a.as_ref().to_string_lossy().into_owned()).collect();
     let cmd = args.first().map(String::as_str).unwrap_or("");
+    let specs = match cmd {
+        "convert" => CONVERT_OPTIONS,
+        "render" => RENDER_OPTIONS,
+        "zotero" => ZOTERO_OPTIONS,
+        _ => PASSWORD_OPTIONS,
+    };
     let file_at = match OPENS_FILE.iter().find(|(c, _)| *c == cmd) {
-        Some((_, n)) => args.iter().enumerate().skip(1).filter(|(_, a)| !a.starts_with("--")).nth(*n).map(|(i, _)| i),
+        Some((_, n)) => positional_indices(args.get(1..).unwrap_or_default(), specs).get(*n).map(|i| i + 1),
         None if cmd == "run" => args.iter().position(|a| a == "--file").map(|i| i + 1),
         None => None,
     };
@@ -145,8 +151,13 @@ fn password_from(args: &[String]) -> Result<Option<String>, String> {
 }
 
 /// The positional arguments, skipping every option's value so an option value is never read as a
-/// path. Mirrors [`validate_options`], which has already checked the options are known.
+/// path. Uses the same option specifications as [`validate_options`].
 fn positionals<'a>(args: &'a [String], specs: &[OptionSpec]) -> Vec<&'a String> {
+    positional_indices(args, specs).into_iter().filter_map(|i| args.get(i)).collect()
+}
+
+/// Keep the original indices too, so [`text_args`] can open a positional path by its native name.
+fn positional_indices(args: &[String], specs: &[OptionSpec]) -> Vec<usize> {
     let mut out = Vec::new();
     let mut i = 0;
     while let Some(arg) = args.get(i) {
@@ -156,7 +167,7 @@ fn positionals<'a>(args: &'a [String], specs: &[OptionSpec]) -> Vec<&'a String>
             i += 2;
         } else {
             if !arg.starts_with("--") {
-                out.push(arg);
+                out.push(i);
             }
             i += 1;
         }
@@ -430,7 +441,12 @@ fn run<S: AsRef<OsStr>>(args: &[S]) -> Result<(), Failure> {
         "convert" => {
             validate_options("convert", &rest, CONVERT_OPTIONS)?;
             // `--format wordml2003` (or `flatopc`) picks the kind of `.xml` file; its value isn't a path.
-            let format = arg_value(&rest, "--format");
+            let format = match find_option_arg(&rest, "--format") {
+                OptionArg::Absent => None,
+                OptionArg::Duplicate => return Err("convert: duplicate --format option".into()),
+                OptionArg::MissingValue => return Err("convert: --format requires a value".into()),
+                OptionArg::Value(value) => Some(value),
+            };
             let paths = positionals(&rest, CONVERT_OPTIONS);
             let (i, o) = (paths.first().ok_or_else(missing_argument)?.to_string(), paths.get(1).ok_or_else(missing_argument)?.to_string());
             let markup = rest.iter().any(|a| a == "--markup");

apps/wordcraft-cli/src/tests.rs

@@ -75,6 +75,22 @@ fn unknown_options_are_rejected_before_file_access() {
     assert_eq!(run(&args(&["zotero", "refresh", "missing.docx", "--prot", "1"])), Err("zotero: unknown option --prot".into()));
 }
 
+#[test]
+fn convert_rejects_malformed_format_options_before_file_access() {
+    for (options, error) in [
+        (vec!["--format"], "convert: --format requires a value"),
+        (vec!["--format", "--markup"], "convert: --format requires a value"),
+        (vec!["--format", "flatopc", "--format", "wordml2003"], "convert: duplicate --format option"),
+    ] {
+        let mut argv = args(&["convert", "missing.docx", "out.xml"]);
+        argv.extend(args(&options));
+        assert_eq!(run(&argv), Err(error.into()), "{options:?}");
+    }
+    // A valid single value reaches file access, rather than a format-option error.
+    let error = run(&args(&["convert", "missing.docx", "out.xml", "--format", "flatopc"])).unwrap_err().to_string();
+    assert!(error.starts_with("missing.docx:"), "{error}");
+}
+
 #[test]
 fn help_or_h_after_a_subcommand_asks_for_the_usage() {
     for cmd in SUBCOMMANDS {
@@ -195,6 +211,10 @@ fn only_the_opened_file_may_be_other_than_unicode() {
     let ok = |raw: Vec<OsString>| text_args(&raw).unwrap().1;
     assert_eq!(ok(vec!["zotero".into(), "refresh".into(), latin1()]), Some(PathBuf::from(latin1())));
     assert_eq!(ok(vec!["render".into(), latin1(), "out.png".into(), "--page".into(), "2".into()]), Some(PathBuf::from(latin1())));
+    assert_eq!(ok(vec!["render".into(), "--page".into(), "2".into(), latin1(), "out.png".into()]), Some(PathBuf::from(latin1())));
+    assert_eq!(ok(vec!["convert".into(), "--format".into(), "flatopc".into(), latin1(), "out.xml".into()]), Some(PathBuf::from(latin1())));
+    assert_eq!(ok(vec!["info".into(), "--password".into(), "secret".into(), latin1()]), Some(PathBuf::from(latin1())));
+    assert_eq!(ok(vec!["zotero".into(), "--port".into(), "23119".into(), "refresh".into(), latin1()]), Some(PathBuf::from(latin1())));
     assert_eq!(ok(vec!["run".into(), "--print".into(), "--file".into(), latin1()]), Some(PathBuf::from(latin1())));
     assert_eq!(ok(vec!["info".into(), "in.docx".into()]), None);
     // Every command that opens a file reaches the open with it (no such file here), so
@@ -212,6 +232,7 @@ fn only_the_opened_file_may_be_other_than_unicode() {
     for (raw, n) in [
         (vec!["zotero".into(), latin1(), "in.docx".into()], 2),
         (vec!["render".into(), "in.docx".into(), latin1()], 3),
+        (vec!["render".into(), "--page".into(), latin1(), "in.docx".into(), "out.png".into()], 3),
         (vec!["run".into(), "--save".into(), latin1()], 3),
         (vec!["commands".into(), latin1()], 2),
     ] {

apps/wordcraft-cli/tests/it/non_utf8_args.rs

@@ -53,6 +53,15 @@ fn the_file_a_command_opens_may_have_any_name() {
     let _ = std::fs::remove_dir_all(dir);
 }
 
+#[test]
+fn an_option_before_the_input_keeps_its_native_name() {
+    let (dir, file) = folder("option-first");
+    let saved = dir.join("out.xml");
+    ok(&cli(&["convert".into(), "--format".into(), "flatopc".into(), file, saved.clone().into()]));
+    assert!(std::fs::read_to_string(saved).unwrap().contains("Bonjour"));
+    let _ = std::fs::remove_dir_all(dir);
+}
+
 #[test]
 fn any_other_such_argument_is_an_error_naming_it() {
     let (dir, file) = folder("error");

apps/wordcraft-web/Cargo.toml

@@ -16,6 +16,10 @@ path = "src/main.rs"
 [dependencies]
 wordcraft-engine = { workspace = true }
 wordcraft-ui-egui = { workspace = true }
+# The host messages (storytold/craft-libs): one contract and one bridge for the web builds.
+# The page command API (storytold/craft-libs#413): window.wordcraft over the control channel.
+craft-web-api = { git = "https://github.com/storytold/craft-libs", rev = "fd5e2feff6ea25bed371a99cc732b3968d621505", version = "0.1.0" }
+craft-web-host = { git = "https://github.com/storytold/craft-libs", rev = "fd5e2feff6ea25bed371a99cc732b3968d621505", version = "0.1.0" }
 egui = { workspace = true }
 log = { workspace = true }
 serde_json = { workspace = true }
@@ -29,4 +33,4 @@ rfd = { workspace = true }
 wasm-bindgen = "0.2"
 wasm-bindgen-futures = "0.4"
 js-sys = "0.3"
-web-sys = { version = "0.3", features = ["Window", "Document", "Element", "HtmlElement", "HtmlCanvasElement", "HtmlAnchorElement", "HtmlIFrameElement", "Blob", "BlobPropertyBag", "Url", "Location", "CssStyleDeclaration", "Event", "EventTarget", "BeforeUnloadEvent", "ClipboardEvent", "DataTransfer", "FileList", "File", "DomStringList", "IdbFactory", "IdbOpenDbRequest", "IdbRequest", "IdbDatabase", "IdbObjectStore", "IdbTransaction", "IdbTransactionMode", "MediaQueryList", "Storage", "MessageEvent", "UrlSearchParams"] }
+web-sys = { version = "0.3", features = ["Window", "Document", "Element", "HtmlElement", "HtmlCanvasElement", "HtmlAnchorElement", "HtmlIFrameElement", "Blob", "BlobPropertyBag", "Url", "Location", "CssStyleDeclaration", "Event", "EventTarget", "BeforeUnloadEvent", "ClipboardEvent", "DataTransfer", "FileList", "File", "DomStringList", "IdbFactory", "IdbOpenDbRequest", "IdbRequest", "IdbDatabase", "IdbObjectStore", "IdbTransaction", "IdbTransactionMode", "MediaQueryList", "Storage", "MessageEvent", "UrlSearchParams", "CustomEvent", "CustomEventInit"] }

apps/wordcraft-web/src/api.rs

@@ -0,0 +1,61 @@
+//! `window.wordcraft`: the page command API (`craft-web-api`, storytold/craft-libs#413) over the
+//! app's own control channel, the one the desktop control server drives (`docs/control-protocol.md`).
+//!
+//! The crate owns the JavaScript side (`request`, `execute`, `commands`, `docInspect`, `uiInspect`,
+//! `screenshot`, as promises with a timeout). This module is WordCraft's part: a [`Request`] that
+//! queues a [`ControlRequest`] for the app, which answers it on its next frame, and waits for the
+//! reply. WordCraft's control request has no abandon state, so a call the page gave up on still runs
+//! when the app reaches it, as it does for a desktop client that timed out.
+
+use std::rc::Rc;
+use std::sync::mpsc::{Receiver, TryRecvError, channel};
+
+use craft_web_api::methods::Methods;
+use craft_web_api::page::{PageApi, Request};
+use serde_json::{Value, json};
+use wasm_bindgen_futures::JsFuture;
+use wordcraft_ui_egui::ControlRequest;
+
+/// The app id in `window.<app>`.
+const APP: &str = "wordcraft";
+
+/// How often a waiting call looks for the app's reply, in milliseconds (about a frame).
+const POLL_MS: i32 = 16;
+
+/// Put `window.wordcraft` on the page and return the queue the app drains each frame (hand it to
+/// `WordApp::with_control`). An existing `window.wordcraft` keeps the functions it already has.
+pub fn install(ctx: egui::Context) -> Receiver<ControlRequest> {
+    let (queue, requests) = channel::<ControlRequest>();
+    let request: Request = Rc::new(move |method, params| {
+        let (control, reply) = ControlRequest::new(method, params);
+        if queue.send(control).is_err() {
+            return Box::pin(async { json!({"ok": false, "error": "the app is not running"}) });
+        }
+        ctx.request_repaint();
+        Box::pin(wait(reply))
+    });
+    if let Err(e) = Rc::new(PageApi::new(APP, Methods::default(), request)).install() {
+        log::warn!("couldn't install window.{APP}: {e:?}");
+    }
+    requests
+}
+
+/// The app's reply, polled about once a frame.
+async fn wait(reply: Receiver<Value>) -> Value {
+    loop {
+        match reply.try_recv() {
+            Ok(reply) => return reply,
+            Err(TryRecvError::Empty) => sleep(POLL_MS).await,
+            Err(TryRecvError::Disconnected) => return json!({"ok": false, "error": "the app dropped the request"}),
+        }
+    }
+}
+
+async fn sleep(ms: i32) {
+    let promise = js_sys::Promise::new(&mut |resolve, _| {
+        if let Some(window) = web_sys::window() {
+            let _ = window.set_timeout_with_callback_and_timeout_and_arguments_0(&resolve, ms);
+        }
+    });
+    let _ = JsFuture::from(promise).await;
+}

apps/wordcraft-web/src/host.rs

@@ -1,14 +1,20 @@
-//! The `postMessage` bridge to a same-origin parent page (`?host=parent`), for pages that embed
-//! WordCraft in an `<iframe>` and keep its files (a Nextcloud app, a personal file server). The
-//! same messages as GridCraft's and PdfCraft's bridges; each is a plain object with a `type`:
+//! Host modes: a page that runs WordCraft (as an `<iframe>`, or in its own document) opens
+//! documents in it and stores what it saves instead of the browser downloading. The messages and
+//! the two transports are `craft-web-host`'s (storytold/craft-libs), the same contract as the
+//! other Craft apps' bridges, with the `wordcraft:` prefix. This module is WordCraft's part of it.
 //!
-//! - `wordcraft:ready` (to the parent): `{ version }`, sent once the app is running and
-//!   listening; send a file now.
-//! - `wordcraft:open` (from the parent): `{ name, bytes }`, bytes an `ArrayBuffer` or
-//!   `Uint8Array`. It opens like a document picked with File › Open or dropped on the window.
-//! - `wordcraft:save` (to the parent): `{ name, saveAs, bytes }` (a transferred `ArrayBuffer`) for
-//!   every file WordCraft would otherwise download. The parent stores it and reports failures
-//!   itself. `saveAs` says whether the file becomes the document's file:
+//! - `?host=parent` (an `<iframe>`, such as a Nextcloud app) talks to the same-origin parent over
+//!   `postMessage`; `data-host="page"` on the canvas (the page's own document, no iframe) uses
+//!   DOM `CustomEvent`s on the canvas instead. `?host=parent` wins when both are present. With no
+//!   parent, or a parent on another origin, there is no bridge: the normal app with downloads.
+//! - `wordcraft:ready` once the app is running and listening; send a file now. WordCraft reads
+//!   bytes, so it says `file: false` and a host sends `bytes`.
+//! - `wordcraft:open` (from the host): `{ name, bytes }`. It opens like a document picked with
+//!   File › Open or dropped on the window; a file over [`MAX_OPEN_BYTES`] is refused before it is
+//!   copied.
+//! - `wordcraft:save` (to the host): `{ name, saveAs, bytes }` for every file WordCraft would
+//!   otherwise download. The host stores it and reports failures itself. `saveAs` says whether
+//!   the file becomes the document's file:
 //!   - Save: `false`. It writes the opened file under its own name and format, as the desktop
 //!     app does. A name the parent never sent (an untitled document, an imported .doc saved as
 //!     .docx) is a new file: `true`, as for Save As.
@@ -17,70 +23,57 @@
 //!     is now that file; the next Save sends its name with `false`.
 //!   - Save As or Export to a format that doesn't (PDF, Markdown, plain text), Export as PDF and
 //!     Export Page as PNG: `false`. They are copies; the document's file stays the same.
-//! - `wordcraft:dirty` (to the parent): `{ dirty }` whenever unsaved changes appear or go away.
-//! - `wordcraft:failed` (to the parent): `{ error }` when the app couldn't start.
+//! - `wordcraft:dirty` whenever unsaved changes appear or go away, `wordcraft:failed` when the
+//!   app couldn't start.
 //!
-//! Only the parent window on the page's own origin is heard or answered. In host mode the parent
-//! guards leaving the page, and `?author=` names the user for comments and tracked changes.
+//! In both host modes the host sends the document to edit, so no sample opens, saves are handed
+//! to the host rather than downloaded, the host guards leaving the page, and `?author=` names the
+//! user for comments and tracked changes.
 
-use std::cell::Cell;
 use std::rc::Rc;
 
-use wasm_bindgen::JsCast as _;
-use wasm_bindgen::JsValue;
-use wasm_bindgen::closure::Closure;
+use craft_web_host::web::{FileData, Host, Message};
 use wordcraft_ui_egui::{Inbox, Services, WordApp};
 
+/// The app id in the host messages (`wordcraft:ready`).
+const APP: &str = "wordcraft";
+
+/// The name a document gets when the host sent none (the crate says `document` for a blank name).
+const DEFAULT_NAME: &str = "Document.docx";
+
+/// The largest document taken from the host page, refused before it is copied.
+const MAX_OPEN_BYTES: u64 = 512 << 20;
+
 /// The longest user name taken from `?author=`.
 const MAX_AUTHOR_CHARS: usize = 200;
 
 #[derive(Clone)]
 pub struct Bridge {
-    window: web_sys::Window,
-    parent: web_sys::Window,
-    origin: String,
+    host: Rc<Host>,
 }
 
 impl Bridge {
-    /// `None` when the page isn't framed (there is no parent to talk to), or when the parent is
-    /// on another origin: messages only go to this origin, so the browser would drop every save
-    /// without an error while the document showed as saved.
-    pub fn new() -> Option<Self> {
-        let window = web_sys::window()?;
-        let parent = window.parent().ok().flatten()?;
-        if js_sys::Object::is(&parent, &window) {
-            return None;
-        }
-        let origin = window.location().origin().ok()?;
-        // An opaque origin (a `file://` page in some browsers) can't be a `postMessage` target.
-        if origin == "null" {
-            return None;
-        }
-        // Reading another origin's location throws, which is `None` too.
-        if parent.location().origin().ok()? != origin {
-            return None;
-        }
-        Some(Self { window, parent, origin })
+    /// The bridge for this page, or `None` when WordCraft isn't hosted: no `?host=parent` and no
+    /// `data-host="page"` on the canvas, an unframed `?host=parent` page (it stays the normal app
+    /// with downloads), or a parent on another origin.
+    pub fn detect(canvas: &web_sys::HtmlCanvasElement) -> Option<Self> {
+        Some(Self { host: Rc::new(Host::connect(APP, canvas)?) })
     }
 
-    /// Hand saved and exported files to the parent instead of downloading them, report unsaved
+    /// Hand saved and exported files to the host instead of downloading them, report unsaved
     /// changes to it, and hear the files it sends.
     pub fn connect(&self, services: &mut Services, inbox: Inbox, ctx: egui::Context) {
         self.listen(inbox, ctx);
         let bridge = self.clone();
         services.host_save = Some(Box::new(move |name: &str, bytes: &[u8], save_as: bool| bridge.post_save(name, bytes, save_as)));
-        // Anything else that would download goes to the parent too.
+        // Anything else that would download goes to the host too.
         let bridge = self.clone();
         services.download = Some(Box::new(move |name: &str, bytes: &[u8]| bridge.post_save(name, bytes, false)));
         // The browser's save picker is never used, so no save finishes later.
         services.async_saves = None;
-        let bridge = self.clone();
-        let reported: Rc<Cell<Option<bool>>> = Rc::default();
-        services.on_dirty = Some(Box::new(move |dirty: bool| {
-            if reported.get() != Some(dirty) && bridge.send(&[("type", "wordcraft:dirty".into()), ("dirty", dirty.into())]) {
-                reported.set(Some(dirty));
-            }
-        }));
+        // The crate does not repeat a state it already reported.
+        let host = self.host.clone();
+        services.on_dirty = Some(Box::new(move |dirty: bool| host.post_dirty(dirty)));
     }
 
     /// `?author=`: the host's name for the user signs comments and tracked changes.
@@ -92,78 +85,43 @@ impl Bridge {
     }
 
     pub fn post_ready(&self) {
-        self.send(&[("type", "wordcraft:ready".into()), ("version", env!("CARGO_PKG_VERSION").into())]);
+        let _ = self.host.post_ready(env!("CARGO_PKG_VERSION"), false);
     }
 
-    /// The app couldn't start (no WebGPU or WebGL2, for example): let the parent say so.
+    /// The app couldn't start (no WebGPU or WebGL2, for example): let the host say so.
     pub fn post_failed(&self, error: &str) {
-        self.send(&[("type", "wordcraft:failed".into()), ("error", error.into())]);
+        let _ = self.host.post_failed(error, None);
     }
 
     fn post_save(&self, path: &str, bytes: &[u8], save_as: bool) -> Result<(), String> {
         let name = std::path::Path::new(path).file_name().map(|n| n.to_string_lossy().to_string()).unwrap_or_else(|| path.to_string());
-        let buffer = js_sys::Uint8Array::from(bytes).buffer();
-        let message = message(&[
-            ("type", "wordcraft:save".into()),
-            ("name", name.as_str().into()),
-            ("saveAs", save_as.into()),
-            ("bytes", buffer.clone().into()),
-        ])
-        .ok_or_else(|| "couldn't build the message".to_string())?;
-        self.parent
-            .post_message_with_transfer(&message, &self.origin, &js_sys::Array::of1(&buffer))
-            .map_err(|e| e.as_string().unwrap_or_else(|| format!("{e:?}")))
+        self.host.post_save(&name, bytes, None, save_as)
     }
 
-    /// Hear `wordcraft:open` from the parent and queue the file like a dropped one.
+    /// Hear `wordcraft:open` and queue the file like a dropped one.
     fn listen(&self, inbox: Inbox, ctx: egui::Context) {
-        let parent = self.parent.clone();
-        let origin = self.origin.clone();
-        let on_message = Closure::<dyn FnMut(web_sys::MessageEvent)>::new(move |event: web_sys::MessageEvent| {
-            if event.origin() != origin || !event.source().is_some_and(|s| js_sys::Object::is(&s, &parent)) {
-                return;
+        self.host.listen(move |message| {
+            if let Message::Open { name, data } = message {
+                deliver_open(&inbox, &ctx, &name, &data);
             }
-            let data = event.data();
-            if string(&data, "type").as_deref() != Some("wordcraft:open") {
-                return;
-            }
-            let name = string(&data, "name").filter(|n| !n.trim().is_empty()).unwrap_or_else(|| "Document.docx".to_string());
-            let Some(bytes) = bytes(&data) else { return };
-            inbox.lock().unwrap_or_else(|e| e.into_inner()).push((name, bytes));
-            ctx.request_repaint();
         });
-        if self.window.add_event_listener_with_callback("message", on_message.as_ref().unchecked_ref()).is_ok() {
-            // The listener lives as long as the page.
-            on_message.forget();
-        }
-    }
-
-    /// Post a message to the parent; false if it couldn't be sent.
-    fn send(&self, fields: &[(&str, JsValue)]) -> bool {
-        message(fields).is_some_and(|m| self.parent.post_message(&m, &self.origin).is_ok())
     }
 }
 
-fn message(fields: &[(&str, JsValue)]) -> Option<js_sys::Object> {
-    let m = js_sys::Object::new();
-    for (key, value) in fields {
-        js_sys::Reflect::set(&m, &JsValue::from_str(key), value).ok()?;
-    }
-    Some(m)
-}
-
-fn string(data: &JsValue, key: &str) -> Option<String> {
-    js_sys::Reflect::get(data, &JsValue::from_str(key)).ok()?.as_string()
-}
-
-/// `bytes` as an `ArrayBuffer` or `Uint8Array`.
-fn bytes(data: &JsValue) -> Option<Vec<u8>> {
-    let bytes = js_sys::Reflect::get(data, &JsValue::from_str("bytes")).ok()?;
-    if let Some(array) = bytes.dyn_ref::<js_sys::Uint8Array>() {
-        Some(array.to_vec())
-    } else if bytes.is_instance_of::<js_sys::ArrayBuffer>() {
-        Some(js_sys::Uint8Array::new(&bytes).to_vec())
-    } else {
-        None
+/// The file an open message carries, pushed onto the inbox for the next frame, like a dropped
+/// file. A host's file is untrusted input: an oversized one is refused before it is copied into
+/// the app's memory (the data's own length, not the host's claim). The crate has cleaned the name
+/// (and cut it to 255 characters); a blank one comes as `document` and becomes `Document.docx`.
+fn deliver_open(inbox: &Inbox, ctx: &egui::Context, name: &str, data: &FileData) {
+    let name = if name == "document" { DEFAULT_NAME } else { name }.to_string();
+    let FileData::Bytes(array) = data else {
+        log::warn!("host file {name} came as a File; WordCraft reads bytes only; not opened");
+        return;
+    };
+    if data.size() > MAX_OPEN_BYTES as f64 {
+        log::warn!("host file is {} bytes, over the {} MiB host open limit; not opened", data.size(), MAX_OPEN_BYTES >> 20);
+        return;
     }
+    inbox.lock().unwrap_or_else(|e| e.into_inner()).push((name, array.to_vec()));
+    ctx.request_repaint();
 }

apps/wordcraft-web/src/main.rs

@@ -9,8 +9,10 @@
 //! Preferences (theme, interface size, editing options…) are kept in `localStorage` (`prefs`).
 //! URL flags: `?webgl` forces WebGL2; `?sample` opens the sample document.
 //! `?host=parent` (inside an `<iframe>`) talks to a same-origin parent page instead: it sends the
-//! document to open and stores what WordCraft saves (`host`), and `&author=` names the user for
-//! comments and tracked changes.
+//! document to open and stores what WordCraft saves (`host`). A page can also host WordCraft in
+//! its own document (no iframe) with `data-host="page"` on the canvas, talking over DOM
+//! `CustomEvent`s on the canvas. `&author=` names the user for comments and tracked changes in
+//! either host mode.
 #![deny(clippy::unwrap_used, clippy::expect_used, clippy::panic, clippy::unimplemented, clippy::todo, clippy::unreachable)]
 
 #[cfg(any(target_arch = "wasm32", test))]
@@ -18,6 +20,8 @@ mod prefs;
 #[cfg(any(target_arch = "wasm32", test))]
 mod save_fallback;
 
+#[cfg(target_arch = "wasm32")]
+mod api;
 #[cfg(target_arch = "wasm32")]
 mod host;
 #[cfg(target_arch = "wasm32")]

apps/wordcraft-web/src/web.rs

@@ -10,7 +10,7 @@ use wordcraft_ui_egui::async_saves::{AsyncSaves, SaveOutcome};
 use wordcraft_ui_egui::{Inbox, Services, WordApp};
 
 const DOC_EXTS: &[&str] =
-    &["docx", "docm", "dotx", "dotm", "doc", "dot", "odt", "rtf", "txt", "md", "html", "htm", "mht", "mhtml", "tex", "json", "xml"];
+    &["docx", "docm", "dotx", "dotm", "doc", "dot", "odt", "ott", "fodt", "rtf", "txt", "md", "html", "htm", "mht", "mhtml", "tex", "json", "xml"];
 const IMAGE_EXTS: &[&str] = &["png", "jpg", "jpeg", "gif", "webp", "bmp", "wmf", "emf"];
 const RECIPIENT_EXTS: &[&str] = wordcraft_engine::cmd::mailings::RECIPIENT_EXTENSIONS;
 mod font_store;
@@ -28,9 +28,12 @@ pub fn start() {
             log::error!("missing <canvas id=\"{CANVAS_ID}\">");
             return;
         };
-        // `?host=parent`: a same-origin page framing WordCraft opens documents in it and stores
-        // what it saves (`host`). That page guards leaving it, so this one doesn't.
-        let bridge = if query_param("host").as_deref() == Some("parent") { crate::host::Bridge::new() } else { None };
+        // Host modes (`craft-web-host`): `?host=parent`, a same-origin page framing WordCraft, opens documents in
+        // it and stores what it saves (`host`); unframed there is no parent, so that URL on its
+        // own stays the normal app with downloads. `data-host="page"` on the canvas opts a page
+        // in instead: it runs WordCraft in its own document and talks over DOM `CustomEvent`s
+        // on the canvas (`host`). The host guards leaving the page, so this one doesn't.
+        let bridge = crate::host::Bridge::detect(&canvas);
         let app_bridge = bridge.clone();
         let dirty = Rc::new(Cell::new(false));
         let saves = AsyncSaves::default();
@@ -73,7 +76,8 @@ pub fn start() {
                     if let Some(bridge) = &app_bridge {
                         bridge.connect(&mut services, inbox.clone(), cc.egui_ctx.clone());
                     }
-                    let mut app = WordApp::new(session, services);
+                    let control = crate::api::install(cc.egui_ctx.clone());
+                    let mut app = WordApp::new(session, services).with_control(control);
                     // Pasted pictures wait apart from dropped files: the app inserts one only when the page is the edit target (#1349).
                     app.services.pasted = Some(pasted);
                     app.autosave = false;
@@ -655,6 +659,8 @@ fn mime_for(name: &str) -> &'static str {
         Some("pdf") => "application/pdf",
         Some("docx") => "application/vnd.openxmlformats-officedocument.wordprocessingml.document",
         Some("odt") => "application/vnd.oasis.opendocument.text",
+        Some("ott") => "application/vnd.oasis.opendocument.text-template",
+        Some("fodt") => "application/vnd.oasis.opendocument.text-flat-xml",
         Some("rtf") => "application/rtf",
         Some("html") => "text/html",
         Some("md" | "txt") => "text/plain",

apps/wordcraft/Cargo.toml

@@ -22,6 +22,8 @@ wordcraft-control-key = { workspace = true }
 eframe = { workspace = true, features = ["default_fonts", "wgpu", "accesskit", "clipboard", "wayland", "x11", "links"] }
 egui = { workspace = true }
 serde_json = { workspace = true }
+serde = { workspace = true }
+tempfile = "3"
 log = { workspace = true }
 chrono = { workspace = true }
 
@@ -38,13 +40,14 @@ fmv-macos-events = "=0.1.0"
 # The system's Reduce motion setting (System Settings › Accessibility › Display), read through
 # NSWorkspace: a safe call in these bindings. Already in the build through eframe, rfd and arboard,
 # at this version; this adds the `NSAccessibility` bindings and, to fit the title bar around the
-# traffic lights, the window, button and view ones.
+# traffic lights, the window, button and view ones; and `NSPasteboard` for the clipboard's RTF.
 objc2-app-kit = { version = "0.3", default-features = false, features = ["std", "NSWorkspace", "NSAccessibility", "NSApplication", "NSWindow", "NSButton", "NSControl", "NSView", "NSResponder", "NSPasteboard"] }
 objc2 = "0.6"
 # A copied equation's MathML goes on the pasteboard as `public.mathml.presentation` beside
 # arboard's HTML and text (#1297), through NSPasteboard's safe `setString:forType:`. Already in
-# the build through arboard, at this version; this adds only the NSString bindings.
-objc2-foundation = { version = "0.3", default-features = false, features = ["std", "NSString"] }
+# the build through arboard, at this version; this adds only the NSString bindings, and NSData
+# for the clipboard's RTF (`public.rtf`).
+objc2-foundation = { version = "0.3", default-features = false, features = ["std", "NSString", "NSData"] }
 # Closing the window must quit through NSApplication rather than by dropping the NSWindow
 # inside the AppKit run loop (see `schedule_macos_terminate`).
 dispatch2 = "0.3"
@@ -58,6 +61,9 @@ zbus = { version = "5", default-features = false, features = ["async-io", "block
 # types that arboard doesn't read, and the formats offered. Already in the build through arboard's
 # `wayland-data-control` feature, at this version.
 wl-clipboard-rs = "0.9"
+# Another program's RTF on X11 (`text/rtf`), which arboard doesn't read. Already in the build
+# through arboard, at this version.
+x11rb = "0.13"
 
 [target.'cfg(target_os = "windows")'.dependencies]
 # The upstream default-apps button opens Windows Settings; printing uses craft-print instead.
@@ -73,7 +79,9 @@ windows = { version = "0.62", features = ["UI_ViewManagement", "System_UserProfi
 clipboard-win = { version = "5.4", features = ["std"] }
 
 [target.'cfg(not(target_arch = "wasm32"))'.dependencies]
-craft-print = { git = "https://github.com/storytold/craft-libs", rev = "92d392cff132eaf53e13afff77f38b5c11b73aa0" }
+# Pinned apart from craft-ui, which doesn't depend on it: `printer_queues`, the printer list that
+# asks no driver (#5710, storytold/craft-libs#618), and `submit_cancellable` (storytold/craft-libs#622).
+craft-print = { git = "https://github.com/storytold/craft-libs", rev = "a3e0b5a9440d18f19422a9dca13516747b7d3181" }
 rfd = { workspace = true }
 # Waits for an `rfd::AsyncFileDialog` on a background thread (the dialogs don't block the UI, #94);
 # rfd itself already depends on it.

apps/wordcraft/src/appimage_env.rs

@@ -0,0 +1,139 @@
+//! An AppImage's environment inherited from another program (#5653).
+//!
+//! The AppImage runtime gives the program it starts `APPIMAGE` (the image's file), `APPDIR` (where
+//! it is mounted), `ARGV0` and `OWD`, and whatever that program starts inherits them. So an AppImage
+//! app that opens a document through `xdg-open` starts an installed WordCraft with *its* `APPIMAGE`,
+//! and the self-update, which trusts that variable, replaces the other app's file with WordCraft.
+//!
+//! This copy is in the image only when its executable is under `APPDIR`. Otherwise the variables
+//! are another program's, and the process starts over without them, before anything reads them.
+//! (Removing them in place is `unsafe` under this edition, and `exec` keeps the process id, the
+//! arguments and the standard streams.)
+
+use std::ffi::OsStr;
+use std::io::Write;
+use std::os::unix::process::CommandExt;
+use std::path::Path;
+use std::process::Command;
+
+/// What the AppImage runtime sets; the same list as the documented workaround on the issue.
+const RUNTIME_VARS: [&str; 4] = ["APPIMAGE", "APPDIR", "ARGV0", "OWD"];
+
+/// Whether the executable runs from the image mounted at `appdir`. Anything unclear counts as not:
+/// the update then goes through the package or the plain file this copy really is, and at worst
+/// says it can't, where the other answer would overwrite a file that isn't WordCraft's.
+fn runs_from_image(exe: Option<&Path>, appdir: Option<&OsStr>) -> bool {
+    let (Some(exe), Some(appdir)) = (exe, appdir) else { return false };
+    let appdir = Path::new(appdir);
+    if !appdir.is_absolute() {
+        return false;
+    }
+    let Ok(appdir) = appdir.canonicalize() else { return false };
+    // No image mounts at the root, where every program on the machine would count as inside it.
+    appdir.parent().is_some() && exe.starts_with(&appdir)
+}
+
+/// Starts this process over without another program's AppImage variables; returns only when there
+/// are none to drop (`None`) or starting over failed. Call it first, before any other thread starts.
+pub(crate) fn drop_inherited() -> Option<std::io::Error> {
+    let appimage = std::env::var_os("APPIMAGE")?;
+    let exe = std::env::current_exe().ok();
+    if runs_from_image(exe.as_deref(), std::env::var_os("APPDIR").as_deref()) {
+        return None;
+    }
+    // The logger isn't up yet, and wouldn't survive `exec`. Unlike `eprintln!`, never panics.
+    let _ = writeln!(std::io::stderr(), "wordcraft: ignoring the AppImage {:?} inherited from another program", Path::new(&appimage));
+    let mut args = std::env::args_os();
+    // This file, whatever has happened to its path since it started.
+    let mut again = Command::new("/proc/self/exe");
+    if let Some(arg0) = args.next() {
+        again.arg0(arg0);
+    }
+    again.args(args);
+    for var in RUNTIME_VARS {
+        again.env_remove(var);
+    }
+    Some(again.exec())
+}
+
+#[cfg(test)]
+mod tests {
+    use super::runs_from_image;
+    use std::path::{Path, PathBuf};
+
+    /// A folder of its own for one test, removed when it ends.
+    struct Tmp(PathBuf);
+
+    impl Tmp {
+        /// An empty folder named after `test` and this process, made afresh.
+        fn new(test: &str) -> Tmp {
+            let dir = std::env::temp_dir().join(format!("wordcraft-appimage-{test}-{}", std::process::id()));
+            let _ = std::fs::remove_dir_all(&dir);
+            std::fs::create_dir_all(&dir).unwrap();
+            Tmp(dir)
+        }
+    }
+
+    impl Drop for Tmp {
+        /// Removes the folder and what the test left in it.
+        fn drop(&mut self) {
+            let _ = std::fs::remove_dir_all(&self.0);
+        }
+    }
+
+    /// A mount with the executable inside it, by real path (`/tmp` may be a link).
+    fn mount(root: &Path, name: &str) -> (PathBuf, PathBuf) {
+        let dir = root.join(name);
+        std::fs::create_dir_all(dir.join("usr/bin")).unwrap();
+        let dir = dir.canonicalize().unwrap();
+        let exe = dir.join("usr/bin/wordcraft");
+        (dir, exe)
+    }
+
+    #[test]
+    fn an_executable_in_the_mount_is_the_image() {
+        let tmp = Tmp::new("own");
+        let (dir, exe) = mount(&tmp.0, ".mount_WordCr");
+        assert!(runs_from_image(Some(&exe), Some(dir.as_os_str())));
+    }
+
+    #[test]
+    fn a_link_to_the_mount_is_the_same_mount() {
+        let tmp = Tmp::new("link");
+        let (dir, exe) = mount(&tmp.0, ".mount_WordCr");
+        let link = tmp.0.join("link");
+        std::os::unix::fs::symlink(&dir, &link).unwrap();
+        assert!(runs_from_image(Some(&exe), Some(link.as_os_str())));
+    }
+
+    #[test]
+    fn an_installed_executable_is_not_in_another_programs_mount() {
+        let tmp = Tmp::new("other");
+        let (other, _) = mount(&tmp.0, ".mount_other");
+        assert!(!runs_from_image(Some(Path::new("/usr/bin/wordcraft")), Some(other.as_os_str())));
+    }
+
+    /// Paths compare by component: one mount's name can start with another's.
+    #[test]
+    fn a_mount_whose_name_starts_the_same_is_another_mount() {
+        let tmp = Tmp::new("prefix");
+        let (other, _) = mount(&tmp.0, ".mount_ab");
+        let (_, exe) = mount(&tmp.0, ".mount_abc");
+        assert!(!runs_from_image(Some(&exe), Some(other.as_os_str())));
+    }
+
+    #[test]
+    fn anything_unclear_is_not_the_image() {
+        let tmp = Tmp::new("unclear");
+        let (dir, exe) = mount(&tmp.0, ".mount_WordCr");
+        // No `APPDIR`, one that's gone (the other program has exited) or empty.
+        assert!(!runs_from_image(Some(&exe), None));
+        assert!(!runs_from_image(Some(&exe), Some(tmp.0.join("gone").as_os_str())));
+        assert!(!runs_from_image(Some(&exe), Some("".as_ref())));
+        // The root, and a relative path that happens to lead here.
+        assert!(!runs_from_image(Some(&exe), Some("/".as_ref())));
+        assert!(!runs_from_image(Some(&exe), Some(".".as_ref())));
+        // No executable path at all.
+        assert!(!runs_from_image(None, Some(dir.as_os_str())));
+    }
+}

apps/wordcraft/src/apple_events.rs

@@ -36,14 +36,15 @@ impl AppleEvents {
 
 /// Open the documents and act on the quit requests that arrived since the last frame. A document
 /// opens like File › Open: beside the open ones, in place of an untouched blank one, or, when it
-/// is already open, by showing it. Quitting closes the window, which asks about every document
+/// is already open, by showing it; a Word template starts a new document based on it, as Word's
+/// default verb for one does (#3433). Quitting closes the window, which asks about every document
 /// with unsaved changes.
 pub fn poll(inbox: &Inbox, app: &mut WordApp, ctx: &egui::Context) {
     for e in inbox.drain() {
         match e {
             Event::Open(paths) => {
                 for p in paths.iter().filter_map(|p| p.to_str()) {
-                    match app.open_file(p, None) {
+                    match app.open_from_platform(p, None) {
                         // Leave Backstage, as opening from the Open dialog does.
                         Ok(_) => app.ui.backstage = false,
                         Err(e) => {

apps/wordcraft/src/clipboard.rs

@@ -18,6 +18,11 @@
 //! they paste math (learn.microsoft.com/office/math/mathml, "Clipboard support"). Pasting into
 //! Word and LibreOffice is not yet verified. Linux and the BSDs have no Word, and arboard can't
 //! add MIME types of other names, so there the HTML carries the MathML alone.
+//!
+//! RTF goes beside them where we can add it: Windows ("Rich Text Format") and macOS
+//! (`public.rtf`) after arboard's HTML, Wayland (`text/rtf`) in one offer with the text and HTML.
+//! On X11 arboard owns the selection and serves only its own targets, so a copy there carries text
+//! and HTML only; reading another program's RTF works on all four.
 
 use wordcraft_engine::cmd::paste::MAX_PASTE_BYTES;
 use wordcraft_ui_egui::paste_picture::ClipboardPicture;
@@ -31,17 +36,24 @@ thread_local! {
     static WRITER: std::cell::RefCell<Option<arboard::Clipboard>> = const { std::cell::RefCell::new(None) };
 }
 
-/// Put `html` on the clipboard with `text` as its plain-text alternative, and `mathml` (a copy of
-/// one equation alone; empty for any other copy) in the platform's MathML formats. `false` when
-/// the HTML and text couldn't be set (the caller copies the text alone), including a Wayland
-/// session without the data-control protocol, where arboard would fall back to X11 and native
-/// Wayland programs might not see it. MathML that can't be added is logged: the HTML still
-/// carries it.
-pub fn set_html(html: &str, text: &str, mathml: &str) -> bool {
+/// Put a copy on the clipboard: `text`, with `html` and `rtf` beside it where the platform takes
+/// them (an empty one isn't offered), and `mathml` (a copy of one equation alone; empty for any
+/// other copy) in the platform's MathML formats. `false` when the text and HTML couldn't be set
+/// (the caller copies the text alone), including a Wayland session without the data-control
+/// protocol, where arboard would fall back to X11 and native Wayland programs might not see it.
+/// RTF or MathML that can't be added is logged: the HTML still carries the copy.
+pub fn set_rich(text: &str, html: &str, rtf: &str, mathml: &str) -> bool {
+    // Wayland: one offer with the text, HTML and RTF (no MathML formats there; see above).
+    if platform::set_rich(text, html, rtf) {
+        return true;
+    }
     if !platform::can_set() {
         return false;
     }
     let set = set_html_text(html, text);
+    if set && !rtf.is_empty() {
+        platform::add_rtf(rtf);
+    }
     if set && !mathml.is_empty() && mathml.len() <= MAX_PASTE_BYTES {
         platform::add_mathml(mathml);
     }
@@ -54,7 +66,8 @@ fn set_html_text(html: &str, text: &str) -> bool {
         if slot.is_none() {
             *slot = open();
         }
-        match slot.as_mut().map(|c| c.set().html(html, Some(text))) {
+        let set = slot.as_mut().map(|c| if html.is_empty() { c.set_text(text) } else { c.set().html(html, Some(text)) });
+        match set {
             Some(Ok(())) => true,
             Some(Err(e)) => {
                 log::warn!("clipboard: {e}");
@@ -76,8 +89,8 @@ pub fn html() -> Option<String> {
     open()?.get().html().ok().filter(|h| h.len() <= MAX_PASTE_BYTES)
 }
 
-/// The clipboard's RTF: Windows ("Rich Text Format") and Wayland (`text/rtf`); not offered by
-/// arboard, so read here.
+/// The clipboard's RTF: Windows ("Rich Text Format"), macOS (`public.rtf`), Wayland and X11
+/// (`text/rtf`); not offered by arboard, so read here.
 pub fn rtf() -> Option<String> {
     let bytes = platform::rtf()?;
     if bytes.len() > MAX_PASTE_BYTES {
@@ -176,15 +189,51 @@ mod platform {
             )
     }
 
+    /// Put the copy on a Wayland clipboard in one offer: the text (with the usual plain-text
+    /// aliases), HTML and RTF, served from a background thread until something else is copied.
+    /// `false` outside Wayland (X11 goes through arboard) or without the data-control protocol.
+    pub fn set_rich(text: &str, html: &str, rtf: &str) -> bool {
+        use wl_clipboard_rs::copy::{MimeSource, MimeType, Options, Source};
+
+        if !wayland() {
+            return false;
+        }
+        let offer = |mime: &str, data: &str| MimeSource { source: Source::Bytes(data.as_bytes().into()), mime_type: MimeType::Specific(mime.into()) };
+        // The text first: the plain-text aliases serve the first text type offered.
+        let mut sources = vec![offer("text/plain;charset=utf-8", text)];
+        for (mime, data) in [("text/html", html), ("text/rtf", rtf)] {
+            if !data.is_empty() {
+                sources.push(offer(mime, data));
+            }
+        }
+        match Options::new().copy_multi(sources) {
+            Ok(()) => true,
+            Err(e) => {
+                log::warn!("clipboard: {e}");
+                false
+            }
+        }
+    }
+
+    /// Wayland copies carry their RTF already; X11 ones can't (see the module docs).
+    pub fn add_rtf(_rtf: &str) {}
+
+    /// RTF types a clipboard may offer, best first.
+    const RTF_MIMES: [&str; 3] = ["text/rtf", "application/rtf", "text/richtext"];
+
+    /// The RTF type to ask for among those `offered`.
+    fn rtf_target(offered: &[String]) -> Option<&'static str> {
+        RTF_MIMES.into_iter().find(|m| offered.iter().any(|o| o == m))
+    }
+
     /// arboard serves its own MIME types only; Word, which reads the MathML formats, isn't here.
     pub fn add_mathml(_mathml: &str) {}
 
     pub fn rtf() -> Option<Vec<u8>> {
         if !wayland() {
-            return None;
+            return std::env::var_os("DISPLAY").is_some().then(|| x11::rtf(super::MAX_PASTE_BYTES)).flatten();
         }
-        let offered = formats();
-        let mime = ["text/rtf", "application/rtf", "text/richtext"].into_iter().find(|m| offered.iter().any(|o| o == m))?;
+        let mime = rtf_target(&formats())?;
         read(mime, super::MAX_PASTE_BYTES as u64)
     }
 
@@ -196,6 +245,125 @@ mod platform {
         let mime = super::PICTURE_MIMES.into_iter().find(|m| offered.iter().any(|o| o == m))?;
         read(mime, wordcraft_ui_egui::paste_picture::MAX_FILE_BYTES)
     }
+
+    /// Whether a property read of `len` bytes, with `bytes_after` left unread, is the whole value
+    /// and at most `cap` bytes.
+    fn complete(len: usize, bytes_after: u32, cap: usize) -> bool {
+        bytes_after == 0 && len <= cap
+    }
+
+    /// Reading another program's RTF on X11, which arboard doesn't offer: ask the CLIPBOARD
+    /// selection's owner for its targets, then for its RTF, into a property of a hidden window.
+    mod x11 {
+        use std::time::{Duration, Instant};
+
+        use x11rb::connection::Connection;
+        use x11rb::protocol::Event;
+        use x11rb::protocol::xproto::{Atom, AtomEnum, ConnectionExt as _, CreateWindowAux, GetPropertyReply, Window, WindowClass};
+        use x11rb::rust_connection::RustConnection;
+
+        /// How long a paste waits for the selection's owner, in all.
+        const TIMEOUT: Duration = Duration::from_secs(1);
+        /// Most targets looked at.
+        const MAX_TARGETS: usize = 128;
+
+        struct Reader {
+            conn: RustConnection,
+            win: Window,
+            clipboard: Atom,
+            prop: Atom,
+            deadline: Instant,
+        }
+
+        fn atom(conn: &RustConnection, name: &str) -> Option<Atom> {
+            conn.intern_atom(false, name.as_bytes()).ok()?.reply().ok().map(|r| r.atom)
+        }
+
+        impl Reader {
+            fn new() -> Option<Self> {
+                let (conn, screen) = x11rb::connect(None).ok()?;
+                let root = conn.setup().roots.get(screen)?.root;
+                let win = conn.generate_id().ok()?;
+                conn.create_window(0, win, root, 0, 0, 1, 1, 0, WindowClass::INPUT_ONLY, x11rb::COPY_FROM_PARENT, &CreateWindowAux::new()).ok()?;
+                let clipboard = atom(&conn, "CLIPBOARD")?;
+                let prop = atom(&conn, "WORDCRAFT_PASTE")?;
+                Some(Self { conn, win, clipboard, prop, deadline: Instant::now() + TIMEOUT })
+            }
+
+            /// The selection converted to `target`, at most `cap` bytes; `None` when the owner
+            /// refuses or doesn't answer in time. A longer value is `None` when `whole`, else
+            /// its first `cap` bytes.
+            fn convert(&self, target: Atom, cap: usize, whole: bool) -> Option<GetPropertyReply> {
+                self.conn.convert_selection(self.win, self.clipboard, target, self.prop, x11rb::CURRENT_TIME).ok()?;
+                self.conn.flush().ok()?;
+                loop {
+                    // Checked every time round, so a stream of other events can't keep us here.
+                    if Instant::now() >= self.deadline {
+                        return None;
+                    }
+                    match self.conn.poll_for_event().ok()? {
+                        Some(Event::SelectionNotify(e)) if e.requestor == self.win => {
+                            if e.property == x11rb::NONE {
+                                return None;
+                            }
+                            break;
+                        }
+                        Some(_) => {}
+                        None => std::thread::sleep(Duration::from_millis(5)),
+                    }
+                }
+                // In 32-bit units; for a whole value one more than the cap, so a larger one shows
+                // as `bytes_after`.
+                let units = u32::try_from(cap / 4 + usize::from(whole)).unwrap_or(u32::MAX);
+                let r = self.conn.get_property(true, self.win, self.prop, AtomEnum::ANY, 0, units).ok()?.reply().ok()?;
+                (!whole || super::complete(r.value.len(), r.bytes_after, cap)).then_some(r)
+            }
+        }
+
+        impl Drop for Reader {
+            fn drop(&mut self) {
+                let _ = self.conn.destroy_window(self.win);
+                let _ = self.conn.flush();
+            }
+        }
+
+        /// The CLIPBOARD selection's RTF, at most `cap` bytes. Copies too large to send at once
+        /// arrive in pieces (INCR), which aren't read: `None` then.
+        pub fn rtf(cap: usize) -> Option<Vec<u8>> {
+            let r = Reader::new()?;
+            // The first targets only: a longer list is cut short, not refused.
+            let targets = r.convert(atom(&r.conn, "TARGETS")?, MAX_TARGETS.saturating_mul(4), false)?;
+            let cookies: Vec<_> = targets.value32()?.take(MAX_TARGETS).filter_map(|a| r.conn.get_atom_name(a).ok()).collect();
+            let names: Vec<String> =
+                cookies.into_iter().filter_map(|c| c.reply().ok()).map(|n| String::from_utf8_lossy(&n.name).into_owned()).collect();
+            let target = atom(&r.conn, super::rtf_target(&names)?)?;
+            let incr = atom(&r.conn, "INCR")?;
+            let reply = r.convert(target, cap, true)?;
+            (reply.type_ != incr).then_some(reply.value)
+        }
+    }
+
+    #[cfg(test)]
+    mod tests {
+        use super::*;
+
+        #[test]
+        fn rtf_target_prefers_text_rtf() {
+            let names = |n: &[&str]| n.iter().map(|s| s.to_string()).collect::<Vec<_>>();
+            assert_eq!(rtf_target(&names(&["TARGETS", "text/richtext", "application/rtf", "text/rtf"])), Some("text/rtf"));
+            assert_eq!(rtf_target(&names(&["text/richtext", "UTF8_STRING"])), Some("text/richtext"));
+            assert_eq!(rtf_target(&names(&["text/html", "text/rtf;charset=x"])), None);
+            assert_eq!(rtf_target(&[]), None);
+        }
+
+        #[test]
+        fn property_reads_must_be_whole_and_capped() {
+            assert!(complete(10, 0, 10));
+            assert!(!complete(11, 0, 10), "over the cap");
+            assert!(!complete(8, 4, 10), "more left to read");
+            assert!(complete(0, 0, 0));
+        }
+    }
 }
 
 #[cfg(target_os = "windows")]
@@ -251,6 +419,23 @@ mod platform {
         true
     }
 
+    /// arboard sets the text and HTML.
+    pub fn set_rich(_text: &str, _html: &str, _rtf: &str) -> bool {
+        false
+    }
+
+    /// Add "Rich Text Format" to what arboard just set, NUL-terminated as is usual.
+    pub fn add_rtf(rtf: &str) {
+        let Some(format) = raw::register_format("Rich Text Format") else { return };
+        let Ok(_open) = Clipboard::new_attempts(10) else { return };
+        let mut bytes = Vec::with_capacity(rtf.len().saturating_add(1));
+        bytes.extend_from_slice(rtf.as_bytes());
+        bytes.push(0);
+        if let Err(e) = raw::set_without_clear(format.get(), &bytes) {
+            log::warn!("clipboard RTF: {e}");
+        }
+    }
+
     /// Add the MathML to what arboard just set, as "MathML" and "MathML Presentation": UTF-16LE
     /// text (with its terminating NUL, as Windows text formats have).
     pub fn add_mathml(mathml: &str) {
@@ -276,14 +461,18 @@ const MATHML_FORMATS: [&str; 2] = ["MathML", "MathML Presentation"];
 #[cfg(target_os = "macos")]
 mod platform {
     use objc2_app_kit::NSPasteboard;
-    use objc2_foundation::NSString;
+    use objc2_foundation::{NSData, NSString};
+
+    /// RTF's pasteboard type (`NSPasteboardTypeRTF`, whose extern static needs unsafe to read).
+    const RTF: &str = "public.rtf";
 
     pub fn formats() -> Vec<String> {
         Vec::new()
     }
 
     pub fn rtf() -> Option<Vec<u8>> {
-        None
+        let data = NSPasteboard::generalPasteboard().dataForType(&NSString::from_str(RTF))?;
+        (data.len() <= super::MAX_PASTE_BYTES).then(|| data.to_vec())
     }
 
     pub fn encoded_picture() -> Option<Vec<u8>> {
@@ -294,6 +483,18 @@ mod platform {
         true
     }
 
+    /// arboard sets the text and HTML.
+    pub fn set_rich(_text: &str, _html: &str, _rtf: &str) -> bool {
+        false
+    }
+
+    /// Add RTF to the pasteboard item arboard just wrote (it cleared the pasteboard first).
+    pub fn add_rtf(rtf: &str) {
+        if !NSPasteboard::generalPasteboard().setData_forType(Some(&NSData::with_bytes(rtf.as_bytes())), &NSString::from_str(RTF)) {
+            log::warn!("clipboard: the RTF couldn't be added");
+        }
+    }
+
     /// Add the MathML to the pasteboard item arboard just wrote, as the MathML types (W3C MathML 3
     /// §6.2.4): presentation markup, and the generic type for readers that only know that one.
     pub fn add_mathml(mathml: &str) {
@@ -329,5 +530,11 @@ mod platform {
         true
     }
 
+    pub fn set_rich(_text: &str, _html: &str, _rtf: &str) -> bool {
+        false
+    }
+
+    pub fn add_rtf(_rtf: &str) {}
+
     pub fn add_mathml(_mathml: &str) {}
 }

apps/wordcraft/src/file_dialogs.rs

@@ -239,6 +239,8 @@ fn format_name(ext: &str) -> String {
         "html" => "Web Page".into(),
         "rtf" => "Rich Text Format".into(),
         "odt" => "OpenDocument Text".into(),
+        "ott" => "OpenDocument Text Template".into(),
+        "fodt" => "Flat OpenDocument Text".into(),
         "tex" => "LaTeX".into(),
         "png" => "PNG Image".into(),
         "json" => "JSON".into(),

apps/wordcraft/src/graphics.rs

@@ -15,9 +15,25 @@ use eframe::{
 /// compiler aborts on them, the device is lost and eframe exits before the window opens. WordCraft
 /// issues no indirect draws or dispatches, so Intel Macs skip them; every other GPU keeps wgpu's
 /// default. Same fix as PhotoCraft's storytold/PhotoCraft#3171.
+///
+/// `x86_64` is the build's architecture. An x86_64 build also runs on Apple silicon under Rosetta
+/// (#5199), so it skips only when the machine itself isn't Apple silicon.
 #[cfg_attr(not(target_os = "macos"), allow(dead_code))]
-pub fn skips_indirect_validation(macos: bool, intel: bool) -> bool {
-    macos && intel
+pub fn skips_indirect_validation(macos: bool, x86_64: bool) -> bool {
+    // Short-circuits: only x86_64 macOS builds ask the system.
+    macos && x86_64 && !runs_on_apple_silicon()
+}
+
+/// Whether the machine is Apple silicon, also when this process is translated by Rosetta:
+/// `hw.optional.arm64` describes the hardware, not the process. Assumes not (an Intel Mac) when
+/// the system can't tell.
+#[cfg_attr(not(target_os = "macos"), allow(dead_code))]
+fn runs_on_apple_silicon() -> bool {
+    cfg!(target_os = "macos")
+        && std::process::Command::new("/usr/sbin/sysctl")
+            .args(["-n", "hw.optional.arm64"])
+            .output()
+            .is_ok_and(|out| out.status.success() && out.stdout.trim_ascii() == b"1")
 }
 
 /// Leaves wgpu's indirect-call validation out of the instance eframe creates when `skip`.
@@ -181,7 +197,8 @@ mod tests {
     #[test]
     fn intel_macs_skip_the_indirect_validation_shaders() {
         // #2652: Metal's Intel compiler aborted on them, losing the device before the window opened.
-        assert!(skips_indirect_validation(true, true));
+        // An x86_64 build on Apple silicon (Rosetta) isn't an Intel Mac (#5199).
+        assert_eq!(skips_indirect_validation(true, true), !host_is_apple_silicon());
         // Apple silicon, Windows and Linux keep wgpu's validation.
         for (macos, intel) in [(true, false), (false, true), (false, false)] {
             assert!(!skips_indirect_validation(macos, intel), "macos={macos} intel={intel}");
@@ -205,6 +222,36 @@ mod tests {
         }
     }
 
+    /// Whether the machine running the tests is Apple silicon, whatever this build's architecture.
+    fn host_is_apple_silicon() -> bool {
+        cfg!(target_os = "macos")
+            && std::process::Command::new("/usr/sbin/sysctl")
+                .args(["-n", "hw.optional.arm64"])
+                .output()
+                .is_ok_and(|out| out.status.success() && out.stdout.trim_ascii() == b"1")
+    }
+
+    /// #5199: an x86_64 build running under Rosetta on Apple silicon isn't an Intel Mac and keeps
+    /// wgpu's indirect-call validation.
+    #[test]
+    fn x86_64_builds_on_apple_silicon_keep_indirect_validation() {
+        if !host_is_apple_silicon() {
+            return;
+        }
+        // What main.rs passes from an x86_64 build.
+        assert!(!skips_indirect_validation(true, true), "Rosetta build on Apple silicon must keep validation");
+        let mut options = NativeOptions::default();
+        let WgpuSetup::CreateNew(before) = &options.wgpu_options.wgpu_setup else {
+            panic!("default native options must create a wgpu instance");
+        };
+        let before = before.instance_descriptor.flags;
+        configure_indirect_validation(&mut options, skips_indirect_validation(true, true));
+        let WgpuSetup::CreateNew(create) = &options.wgpu_options.wgpu_setup else {
+            panic!("default native options must create a wgpu instance");
+        };
+        assert_eq!(create.instance_descriptor.flags, before, "indirect-call validation kept");
+    }
+
     fn configured_backends(backends: Backends) -> Backends {
         let mut options = NativeOptions::default();
         configure_backends(&mut options, backends);

apps/wordcraft/src/main.rs

@@ -10,6 +10,8 @@
 #![deny(clippy::unwrap_used, clippy::expect_used, clippy::panic, clippy::unimplemented, clippy::todo, clippy::unreachable)]
 
 mod animations;
+#[cfg(target_os = "linux")]
+mod appimage_env;
 #[cfg(target_os = "macos")]
 mod apple_events;
 mod clipboard;
@@ -27,7 +29,7 @@ mod traffic_lights;
 use wordcraft_engine::Session;
 use wordcraft_ui_egui::{
     Services, UiState, WordApp,
-    window_geometry::{RawWindowState, WindowGeometry, take_rescue},
+    window_geometry::{RawWindowState, WindowGeometry, reachable, take_rescue},
 };
 
 /// The app, the restored window geometry (or, with none, the default: no position) until the
@@ -54,13 +56,21 @@ impl eframe::App for App {
         apple_events::poll(&self.5, &mut self.0, ctx);
         self.0.logic(ctx);
         self.3.show(frame);
-        let window = self.4;
+        let mut window = self.4;
+        // Only until the off-screen check below has run: it is the one reader.
+        if self.1.is_some() {
+            let (on_a_monitor, primary_monitor_rect) = monitor_geometry(frame, ctx.pixels_per_point());
+            window.on_a_monitor = on_a_monitor;
+            window.primary_monitor_rect = primary_monitor_rect;
+        }
         // A window that doesn't fit on its screen (the default size on a laptop) shrinks to fit
         // there, so all of the scroll bar can be dragged (#1215). Checked on the frame that first
         // knows the monitor: the off-screen check below takes the pending geometry then.
         if let Some((pos, size)) = self.1.and_then(|g| g.fit_on_monitor(&window)) {
             ctx.send_viewport_cmd(egui::ViewportCommand::InnerSize(size));
             ctx.send_viewport_cmd(egui::ViewportCommand::OuterPosition(pos));
+            let b = self.0.window_builder().clone().with_inner_size(size);
+            self.0.set_window_builder(b);
         }
         // The geometry saved for the next start is the active window's.
         let prev = self.0.ui.window;
@@ -126,6 +136,37 @@ impl eframe::App for App {
     }
 }
 
+/// Whether a grippable part of the window is on a monitor, and the primary monitor bounds.
+/// Winit reports physical pixels; egui positions use points. `None` where the platform cannot
+/// report the window or primary monitor (Wayland, headless).
+fn monitor_geometry(frame: &eframe::Frame, pixels_per_point: f32) -> (Option<bool>, Option<egui::Rect>) {
+    let Some(w) = frame.winit_window() else {
+        return (None, None);
+    };
+    if !pixels_per_point.is_finite() || pixels_per_point <= 0.0 {
+        return (None, None);
+    }
+    let Some(pos) = w.outer_position().ok() else {
+        return (None, None);
+    };
+    let size = w.outer_size();
+    let physical_rect = |(x, y): (i32, i32), (width, height): (u32, u32)| {
+        egui::Rect::from_min_size(egui::pos2(x as f32, y as f32), egui::vec2(width as f32, height as f32))
+    };
+    let monitors: Vec<egui::Rect> =
+        w.available_monitors().map(|m| physical_rect((m.position().x, m.position().y), (m.size().width, m.size().height))).collect();
+    let on_a_monitor = (!monitors.is_empty()).then(|| reachable(physical_rect((pos.x, pos.y), (size.width, size.height)), &monitors));
+    let primary_monitor_rect = w.primary_monitor().map(|m| {
+        let pos = m.position();
+        let size = m.size();
+        egui::Rect::from_min_size(
+            egui::pos2(pos.x as f32, pos.y as f32) / pixels_per_point,
+            egui::vec2(size.width as f32, size.height as f32) / pixels_per_point,
+        )
+    });
+    (on_a_monitor, primary_monitor_rect)
+}
+
 /// Quit the way Command-Q does, on the next turn of the main queue.
 ///
 /// Called when the window is closing. Doing it on the next turn lets the close
@@ -216,13 +257,45 @@ fn load_user_fonts() {
     }
 }
 
-/// The saved window geometry, read before the window opens (`load_prefs` runs after).
-fn saved_window() -> Option<WindowGeometry> {
+/// The saved preferences, read before the window opens for its geometry and, on Linux and the
+/// BSDs, the display server (`load_prefs` runs after).
+fn saved_ui() -> Option<UiState> {
     if !prefs_enabled() {
         return None;
     }
     let bytes = std::fs::read(prefs_path()?).ok()?;
-    UiState::from_json(&bytes)?.window?.sanitized()
+    UiState::from_json(&bytes)
+}
+
+/// Whether the window runs through XWayland (X11) rather than native Wayland, and the log line
+/// saying why. `WORDCRAFT_WAYLAND` wins (`0`, `false`, `no` or `off` asks for X11, any other value
+/// for Wayland), then File › Options › Advanced › Linux display server (#2184). Automatic is X11
+/// whenever there's an X server: winit has no file drag-and-drop on Wayland, so dropping files from
+/// the file manager showed a "no" cursor. X11 is only forced with an X server (`DISPLAY`), and
+/// unforced winit picks Wayland when it's there and X11 otherwise, so a choice that isn't
+/// available falls back to the other instead of failing to start.
+#[cfg(all(unix, not(target_os = "macos")))]
+fn uses_x11(setting: wordcraft_ui_egui::DisplayServer, env: impl Fn(&str) -> Option<std::ffi::OsString>) -> (bool, String) {
+    use wordcraft_ui_egui::DisplayServer;
+    let set = |name: &str| env(name).is_some_and(|v| !v.is_empty());
+    let (wanted, by) = match env("WORDCRAFT_WAYLAND") {
+        Some(v) if matches!(v.to_string_lossy().trim().to_ascii_lowercase().as_str(), "0" | "false" | "no" | "off") => {
+            (DisplayServer::X11, "WORDCRAFT_WAYLAND")
+        }
+        Some(_) => (DisplayServer::Wayland, "WORDCRAFT_WAYLAND"),
+        None => (setting, "the setting"),
+    };
+    let x11 = wanted != DisplayServer::Wayland && set("DISPLAY");
+    let runs = if x11 {
+        "X11"
+    } else if set("WAYLAND_DISPLAY") || set("WAYLAND_SOCKET") {
+        "native Wayland"
+    } else if set("DISPLAY") {
+        "X11 (no Wayland)"
+    } else {
+        "neither (no Wayland or X server found)"
+    };
+    (x11, format!("display: {runs}; {} asks for {}", by, wanted.code()))
 }
 
 fn save_prefs(app: &WordApp) {
@@ -268,6 +341,7 @@ fn services(file_dialog: file_dialogs::Hook, ctx: &egui::Context) -> Services {
         file_dialog: Some(file_dialog),
         printers: Some(print::discover()),
         print_native: Some(print::submit()),
+        cancel_print: Some(print::cancel()),
         clipboard_picture: Some(Box::new(clipboard::picture)),
         make_default_app: default_apps::hook(),
         // Personal templates live in `Templates` in the settings folder (#2852), made on first save.
@@ -275,7 +349,7 @@ fn services(file_dialog: file_dialogs::Hook, ctx: &egui::Context) -> Services {
         animations_enabled: Some(animations::hook(ctx)),
         clipboard_text: Some(Box::new(clipboard::text)),
         clipboard_html: Some(Box::new(clipboard::html)),
-        set_clipboard_html: Some(Box::new(clipboard::set_html)),
+        set_clipboard_rich: Some(Box::new(clipboard::set_rich)),
         clipboard_rtf: Some(Box::new(clipboard::rtf)),
         clipboard_formats: Some(Box::new(clipboard::formats)),
         ..Default::default()
@@ -351,14 +425,15 @@ fn without_launch_args(args: Vec<String>) -> Vec<String> {
 
 /// Open a file whose name isn't valid Unicode (#1440). `file.open` takes its path as text, so it
 /// gets the file's bytes under the lossy name, and the document then keeps the real path to save to.
-/// The bytes are read as `file.open` reads a path: only a regular file, and at most 2 GB.
+/// The bytes are read as `file.open` reads a path: only a regular file, and at most 2 GB. A Word
+/// template starts a new document instead ([`WordApp::open_from_platform`]), which has no path.
 fn open_non_unicode(app: &mut WordApp, path: &std::path::Path) {
     let name = path.to_string_lossy();
     let opened = wordcraft_engine::io::read_document(path)
-        .and_then(|bytes| app.run("file.open", serde_json::json!({"path": name, "data": wordcraft_engine::cmd::insert::base64_encode(&bytes)})));
+        .and_then(|bytes| app.open_from_platform(&name, Some(&wordcraft_engine::cmd::insert::base64_encode(&bytes))));
     match opened {
         // Not when it waits for a password: that open names the document itself.
-        Ok(r) if r.get("pending").is_none() => app.session.path = Some(path.into()),
+        Ok(r) if r.get("pending").is_none() && !wordcraft_ui_egui::documents::is_word_template(&name) => app.session.path = Some(path.into()),
         Ok(_) => {}
         Err(e) => {
             app.report_open_failure(&name, &e);
@@ -370,14 +445,31 @@ fn open_non_unicode(app: &mut WordApp, path: &std::path::Path) {
 /// geometry, opens requested documents or the Start screen, starts the control server when
 /// asked, and runs the eframe event loop.
 fn main() -> eframe::Result {
+    #[cfg(windows)]
+    if let Some(code) = print::isolated::helper_mode() {
+        std::process::exit(code);
+    }
     // Windows: the update helper is a copy of this program started with `--finish-update`.
     #[cfg(windows)]
     if let Some(result) = craft_update::self_update::finish(&wordcraft_ui_egui::UPDATE_APP, &std::env::args_os().skip(1).collect::<Vec<_>>()) {
         std::process::exit(i32::from(result.is_err()));
     }
+    // Linux: before anything reads another program's AppImage variables, or a thread starts (#5653).
+    #[cfg(target_os = "linux")]
+    let start_over_failed = appimage_env::drop_inherited();
+    // Still carrying them, this process must never update itself: the update would replace the
+    // other program's file.
+    #[cfg(target_os = "linux")]
+    let updates_unsafe = start_over_failed.is_some();
+    #[cfg(not(target_os = "linux"))]
+    let updates_unsafe = false;
     // First, so the panic hook and every start-up warning are recorded (`logging`).
     let logger = logging::install();
     logging::install_panic_hook();
+    #[cfg(target_os = "linux")]
+    if let Some(e) = start_over_failed {
+        log::error!("couldn't start again without another program's AppImage variables, so the self-update is off in this window: {e}");
+    }
     let mut control_port: Option<u16> = std::env::var("WORDCRAFT_CONTROL_PORT").ok().and_then(|p| p.parse().ok());
     let mut files = Vec::new();
     let mut sample = false;
@@ -449,25 +541,28 @@ fn main() -> eframe::Result {
     // FXC, which needs no DLL of ours (#603; see graphics.rs).
     #[cfg(target_os = "windows")]
     graphics::configure(&mut options, eframe::wgpu::Backends::from_env());
-    // Intel Macs: Metal's compiler aborts on wgpu's indirect-validation shaders (#2652).
+    // Intel Macs: Metal's compiler aborts on wgpu's indirect-validation shaders (#2652). An x86_64
+    // build under Rosetta on Apple silicon keeps them (#5199).
     #[cfg(target_os = "macos")]
     graphics::configure_indirect_validation(&mut options, graphics::skips_indirect_validation(true, cfg!(target_arch = "x86_64")));
     load_user_fonts();
     // Windows opened for further documents start like the first one, before its saved place.
     let window_builder = options.viewport.clone();
-    let restored = saved_window();
+    let saved = saved_ui();
+    let restored = saved.as_ref().and_then(|ui| ui.window?.sanitized());
     if let Some(window) = restored {
         options.viewport = window.apply(options.viewport);
     }
     #[cfg(all(unix, not(target_os = "macos")))]
-    // winit has no file drag-and-drop on Wayland (only on X11), so dropping files from the file
-    // manager showed a "no" cursor. Run through XWayland when it's there; WORDCRAFT_WAYLAND=1
-    // keeps the native Wayland backend.
-    if std::env::var_os("DISPLAY").is_some() && std::env::var_os("WORDCRAFT_WAYLAND").is_none() {
-        options.event_loop_builder = Some(Box::new(|b| {
-            use winit::platform::x11::EventLoopBuilderExtX11;
-            b.with_x11();
-        }));
+    {
+        let (x11, why) = uses_x11(saved.as_ref().map(|ui| ui.display_server).unwrap_or_default(), |name| std::env::var_os(name));
+        log::info!("{why}");
+        if x11 {
+            options.event_loop_builder = Some(Box::new(|b| {
+                use winit::platform::x11::EventLoopBuilderExtX11;
+                b.with_x11();
+            }));
+        }
     }
     // Registered before the event loop starts, so it catches the Finder event that launched us as
     // well as later ones. Lives until the event loop returns; the app creator only borrows it.
@@ -489,6 +584,10 @@ fn main() -> eframe::Result {
             let mut app = WordApp::new(session, services(file_dialog, &cc.egui_ctx));
             app.set_startup_update_check(true);
             load_prefs(&mut app);
+            // After the preferences, so the reason it gives is in the interface language.
+            if updates_unsafe {
+                app.disable_self_update();
+            }
             load_user_dictionary(&mut app);
             // Language packs from the suite's font store; an Install click is the person's own.
             wordcraft_engine::packs::use_font_store(true);
@@ -528,8 +627,9 @@ fn main() -> eframe::Result {
             if files.is_empty() && other_files.is_empty() && !sample && control_port.is_none() {
                 app.start_screen();
             }
+            // A Word template named here starts a new document based on it (#3433).
             for f in files {
-                if let Err(e) = app.run("file.open", serde_json::json!({"path": f})) {
+                if let Err(e) = app.open_from_platform(&f, None) {
                     app.report_open_failure(&f, &e);
                 }
             }
@@ -581,6 +681,32 @@ mod tests {
         let _ = std::fs::remove_dir_all(dir);
     }
 
+    /// #2184: `WORDCRAFT_WAYLAND` beats the setting, Automatic keeps XWayland where there's an X
+    /// server, and a choice that isn't available falls back to the other.
+    #[test]
+    fn the_display_server_follows_the_variable_then_the_setting() {
+        use wordcraft_ui_egui::DisplayServer::{Auto, Wayland, X11};
+        let x11 = |setting, vars: &[(&str, &str)]| {
+            uses_x11(setting, |name| vars.iter().find(|(k, _)| *k == name).map(|(_, v)| std::ffi::OsString::from(v))).0
+        };
+        let both = [("DISPLAY", ":0"), ("WAYLAND_DISPLAY", "wayland-0")];
+        assert!(x11(Auto, &both), "Automatic keeps XWayland, for file drag-and-drop");
+        assert!(x11(X11, &both));
+        assert!(!x11(Wayland, &both));
+        for (value, wants_x11) in [("1", false), ("", false), ("yes", false), ("0", true), (" Off ", true), ("FALSE", true), ("no", true)] {
+            let vars = [both[0], both[1], ("WORDCRAFT_WAYLAND", value)];
+            for setting in [Auto, Wayland, X11] {
+                assert_eq!(x11(setting, &vars), wants_x11, "WORDCRAFT_WAYLAND={value:?} beats {setting:?}");
+            }
+        }
+        // Without an X server X11 isn't forced: winit connects to Wayland instead of failing.
+        let wayland_only = [("WAYLAND_DISPLAY", "wayland-0"), ("DISPLAY", "")];
+        assert!(!x11(X11, &wayland_only) && !x11(Auto, &wayland_only));
+        // Without Wayland, Wayland isn't forced either: unforced winit picks X11.
+        let (forced, why) = uses_x11(Wayland, |name| (name == "DISPLAY").then(|| ":0".into()));
+        assert!(!forced && why.contains("X11 (no Wayland)"), "{why}");
+    }
+
     /// A named pipe or a device under such a name is a warning, not a read: the pipe blocked
     /// start-up until something wrote to it, and `/dev/zero` read until memory ran out.
     #[test]

apps/wordcraft/src/print.rs

@@ -1,13 +1,21 @@
 //! Desktop spooler adapter. Discovery and submission never run on the UI thread.
+#[cfg(any(windows, test))]
+pub(crate) mod isolated;
+
 use wordcraft_ui_egui::printing::{Discover, Duplex, Printer, Settings, Submit};
 
 /// Returns a discovery callback that enumerates spooler printers on a named worker thread and
 /// sends the list through a channel, keeping spooler calls off the UI thread.
+///
+/// The Print panel needs only names and the default, so the list comes from
+/// `craft_print::printer_queues`, which asks no printer driver anything: on Windows a networked
+/// printer in power save can hold a driver's capability query for a long time, and the list must
+/// not wait on it (#5710). Submission still checks the job's options with the driver.
 pub fn discover() -> Discover {
     Box::new(|| {
         let (sender, receiver) = std::sync::mpsc::channel();
         let _ = std::thread::Builder::new().name("print-discovery".into()).spawn(move || {
-            let printers = craft_print::printers()
+            let printers = craft_print::printer_queues()
                 .into_iter()
                 .map(|printer| Printer { name: printer.name, label: printer.description, default: printer.default })
                 .collect();
@@ -17,22 +25,59 @@ pub fn discover() -> Discover {
     })
 }
 
+/// A submission worker and how to stop its job: `true` when it will never be sent, `false` when
+/// it was already being sent.
+struct Job {
+    worker: std::thread::JoinHandle<()>,
+    cancel: Box<dyn Fn() -> bool + Send>,
+    // Acknowledged helper cancellation permits joining it to reap the process and its input PDF.
+    isolated: bool,
+}
+
 // Keep submission workers reachable until normal app shutdown; dropping a JoinHandle would
 // let process exit end an in-process Windows job or the CUPS hand-off before the spooler has it.
-static JOBS: std::sync::Mutex<Vec<std::thread::JoinHandle<()>>> = std::sync::Mutex::new(Vec::new());
+static JOBS: std::sync::Mutex<Vec<Job>> = std::sync::Mutex::new(Vec::new());
 
+fn jobs() -> std::sync::MutexGuard<'static, Vec<Job>> {
+    JOBS.lock().unwrap_or_else(std::sync::PoisonError::into_inner)
+}
+
+/// Waits for the jobs already being sent, so the spooler has them before the process exits. A
+/// job still waiting (on a sleeping printer's driver, say: #5710) is cancelled instead, so it is
+/// never sent and quitting doesn't wait on the driver.
 pub fn finish_pending_jobs() {
-    let workers = std::mem::take(&mut *JOBS.lock().unwrap_or_else(std::sync::PoisonError::into_inner));
-    for worker in workers {
-        if worker.join().is_err() {
+    let pending = std::mem::take(&mut *jobs());
+    // Cancel them all first: a job left waiting while another is joined could start sending.
+    let finishing: Vec<_> = pending
+        .into_iter()
+        .filter(|job| {
+            let cancelled = (job.cancel)();
+            job.isolated || !cancelled
+        })
+        .collect();
+    for job in finishing {
+        if job.worker.join().is_err() {
             log::error!("the print worker stopped unexpectedly during shutdown");
         }
     }
 }
 
+/// The Print page's Cancel: stops the job being submitted (the Print page submits one at a
+/// time, the latest) unless it has started reaching the spooler. `true` when it will never be
+/// sent. A job that has finished sending answers `false`, so its receipt is still shown.
+pub fn cancel() -> Box<dyn Fn() -> bool> {
+    Box::new(|| jobs().last().is_none_or(|job| (job.cancel)()))
+}
+
+/// Submit off the UI thread, isolating Windows drivers while retaining CUPS token cancellation.
 pub fn submit() -> Submit {
     Box::new(|bytes, settings: Settings, collate, options| {
         let (sender, receiver) = std::sync::mpsc::channel();
+        #[cfg(not(windows))]
+        let cancel = craft_print::Cancel::new();
+        #[cfg(windows)]
+        let cancel = isolated::Submission::new();
+        let token = cancel.clone();
         let worker = std::thread::Builder::new().name("print-submit".into()).spawn(move || {
             let job = craft_print::Job {
                 printer: Some(settings.printer),
@@ -48,12 +93,16 @@ pub fn submit() -> Submit {
                 title: "WordCraft document".into(),
                 options,
             };
-            let _ = sender.send(craft_print::submit(&bytes, &job).map_err(|error| error.to_string()));
+            #[cfg(not(windows))]
+            let sent = craft_print::submit_cancellable(&bytes, &job, &mut |_, _| {}, &token).map_err(|error| error.to_string());
+            #[cfg(windows)]
+            let sent = token.submit(&bytes, &job);
+            let _ = sender.send(sent);
         });
         if let Ok(worker) = worker {
-            let mut jobs = JOBS.lock().unwrap_or_else(std::sync::PoisonError::into_inner);
-            jobs.retain(|job| !job.is_finished());
-            jobs.push(worker);
+            let mut jobs = jobs();
+            jobs.retain(|job| !job.worker.is_finished());
+            jobs.push(Job { worker, cancel: Box::new(move || cancel.cancel()), isolated: cfg!(windows) });
         }
         receiver
     })
@@ -62,20 +111,67 @@ pub fn submit() -> Submit {
 #[cfg(test)]
 mod tests {
     use super::*;
-    #[test]
-    fn shutdown_waits_for_pending_submission_workers() {
+    /// A worker that waits until released, then reports that it finished.
+    fn worker() -> (std::thread::JoinHandle<()>, std::sync::mpsc::Sender<()>, std::sync::mpsc::Receiver<()>) {
         let (release, wait) = std::sync::mpsc::channel();
         let (completed, completion) = std::sync::mpsc::channel();
         let worker = std::thread::spawn(move || {
-            wait.recv().unwrap();
-            completed.send(()).unwrap();
+            let _ = wait.recv();
+            let _ = completed.send(());
         });
-        JOBS.lock().unwrap().push(worker);
-        let shutdown = std::thread::spawn(finish_pending_jobs);
-        assert!(completion.try_recv().is_err(), "the print worker is still pending");
-        release.send(()).unwrap();
-        shutdown.join().unwrap();
-        completion.recv().unwrap();
-        assert!(JOBS.lock().unwrap().is_empty(), "workers were joined before shutdown returned");
+        (worker, release, completion)
+    }
+
+    /// A cancel action that records when it was asked and answers `stops`.
+    fn asked(stops: bool) -> (Box<dyn Fn() -> bool + Send>, std::sync::Arc<std::sync::atomic::AtomicBool>) {
+        let flag = std::sync::Arc::new(std::sync::atomic::AtomicBool::new(false));
+        let set = flag.clone();
+        let cancel = Box::new(move || {
+            set.store(true, std::sync::atomic::Ordering::SeqCst);
+            stops
+        });
+        (cancel, flag)
+    }
+
+    /// Shutdown waits for a job already being sent, so the spooler has it before the process
+    /// exits, but not for one still waiting on a driver (a sleeping printer, #5710): that one is
+    /// cancelled, so it is never sent, and quitting doesn't hang on it. Every job is cancelled
+    /// before any is waited for, so a waiting job can't start sending meanwhile. The Print page's
+    /// Cancel asks only the latest job. One test, because JOBS is global.
+    #[test]
+    fn shutdown_waits_for_jobs_being_sent_and_cancels_the_rest() {
+        let (sending, release_sending, sent) = worker();
+        let (stuck, release_stuck, _) = worker();
+        let (stop_sending, sending_asked) = asked(false);
+        let (stop_stuck, cancelled) = asked(true);
+        // The job being sent comes first, so a one-at-a-time shutdown would join it before
+        // cancelling the waiting one.
+        *JOBS.lock().unwrap() =
+            vec![Job { worker: sending, cancel: stop_sending, isolated: false }, Job { worker: stuck, cancel: stop_stuck, isolated: false }];
+        // Cancel from the Print page asks the latest job only.
+        assert!(cancel()(), "the latest job was still waiting: never sent");
+        assert!(!sending_asked.load(std::sync::atomic::Ordering::SeqCst), "an earlier job isn't the one being submitted");
+        cancelled.store(false, std::sync::atomic::Ordering::SeqCst);
+        let (finished, shutdown) = std::sync::mpsc::channel();
+        std::thread::spawn(move || {
+            finish_pending_jobs();
+            let _ = finished.send(());
+        });
+        let wait = std::time::Duration::from_millis(200);
+        assert!(shutdown.recv_timeout(wait).is_err(), "shutdown waits for the job being sent");
+        assert!(cancelled.load(std::sync::atomic::Ordering::SeqCst), "the waiting job was cancelled before the wait");
+        release_sending.send(()).unwrap();
+        assert!(shutdown.recv_timeout(std::time::Duration::from_secs(10)).is_ok(), "shutdown doesn't wait for the stuck job");
+        sent.recv().unwrap();
+        assert!(JOBS.lock().unwrap().is_empty(), "shutdown took every job");
+        // A job that has already been sent answers false, so the Print page keeps its receipt.
+        let (done, release_done, finished) = worker();
+        release_done.send(()).unwrap();
+        finished.recv().unwrap();
+        *JOBS.lock().unwrap() = vec![Job { worker: done, cancel: Box::new(|| false), isolated: false }];
+        assert!(!cancel()(), "a sent job isn't reported as cancelled");
+        JOBS.lock().unwrap().clear();
+        // The stuck worker was left behind without blocking shutdown; let it end.
+        release_stuck.send(()).unwrap();
     }
 }

apps/wordcraft/src/print/isolated.rs

@@ -0,0 +1,592 @@
+//! One Windows driver submission per process: a cancelled driver cannot retain another
+//! job's provider lock. The child acknowledges its Cancel token before it exits.
+
+use std::{
+    io::{BufRead, BufReader, Read, Write},
+    path::Path,
+    process::{Child, ChildStdin, Command, Stdio},
+    sync::{Arc, Condvar, Mutex},
+    time::Duration,
+};
+
+use serde::{Deserialize, Serialize};
+
+#[cfg(windows)]
+const FLAG: &str = "--wordcraft-print-worker";
+const PREFIX: &str = "WORDCRAFT_PRINT:";
+const MAX_MESSAGE: u64 = 64 * 1024;
+const MAX_PDF: u64 = 512 * 1024 * 1024;
+const CANCELLED: &str = "the print job was cancelled before anything was sent to the printer";
+
+#[derive(Serialize, Deserialize)]
+enum Duplex {
+    Off,
+    LongEdge,
+    ShortEdge,
+}
+
+#[derive(Serialize, Deserialize)]
+struct Request {
+    printer: Option<String>,
+    copies: u32,
+    collate: bool,
+    duplex: Option<Duplex>,
+    grayscale: bool,
+    shrink: bool,
+    title: String,
+    options: Vec<(String, String)>,
+}
+
+impl Request {
+    /// Preserve all spooler options in a platform-independent wire representation.
+    fn from_job(job: &craft_print::Job) -> Self {
+        Self {
+            printer: job.printer.clone(),
+            copies: job.copies,
+            collate: job.collate,
+            duplex: job.duplex.map(|d| match d {
+                craft_print::Duplex::Off => Duplex::Off,
+                craft_print::Duplex::LongEdge => Duplex::LongEdge,
+                craft_print::Duplex::ShortEdge => Duplex::ShortEdge,
+            }),
+            grayscale: job.grayscale,
+            shrink: job.scaling == craft_print::Scaling::ShrinkToFit,
+            title: job.title.clone(),
+            options: job.options.clone(),
+        }
+    }
+
+    /// Restore the job without substituting printer defaults or changing copy semantics.
+    fn job(self) -> craft_print::Job {
+        craft_print::Job {
+            printer: self.printer,
+            copies: self.copies,
+            collate: self.collate,
+            duplex: self.duplex.map(|d| match d {
+                Duplex::Off => craft_print::Duplex::Off,
+                Duplex::LongEdge => craft_print::Duplex::LongEdge,
+                Duplex::ShortEdge => craft_print::Duplex::ShortEdge,
+            }),
+            grayscale: self.grayscale,
+            scaling: if self.shrink { craft_print::Scaling::ShrinkToFit } else { craft_print::Scaling::Exact },
+            title: self.title,
+            options: self.options,
+        }
+    }
+}
+
+#[derive(Serialize, Deserialize)]
+enum Event {
+    Ready,
+    Cancelled,
+    Sending,
+    Done(Result<String, String>),
+}
+
+enum State {
+    Starting,
+    Waking(Arc<Mutex<Child>>),
+    Running(ChildStdin),
+    Sending(ChildStdin),
+    Cancelled,
+    Finished,
+}
+
+#[derive(Clone)]
+pub(crate) struct Submission(Arc<(Mutex<State>, Condvar)>);
+
+impl Submission {
+    /// Create one submission whose driver cannot run until the parent grants startup.
+    pub(crate) fn new() -> Self {
+        Self(Arc::new((Mutex::new(State::Starting), Condvar::new())))
+    }
+
+    /// Recover the lifecycle state after poisoning without invoking another driver call.
+    fn state(&self) -> std::sync::MutexGuard<'_, State> {
+        self.0.0.lock().unwrap_or_else(std::sync::PoisonError::into_inner)
+    }
+
+    /// Only an acknowledgement of the child's token can cancel a running helper. A timeout
+    /// is a refusal, never permission to kill a process that may already be sending.
+    pub(crate) fn cancel(&self) -> bool {
+        let mut state = self.state();
+        match &mut *state {
+            State::Starting => {
+                *state = State::Cancelled;
+                return true;
+            }
+            State::Waking(child) => {
+                let _ = child.lock().unwrap_or_else(std::sync::PoisonError::into_inner).kill();
+                *state = State::Cancelled;
+                return true;
+            }
+            State::Cancelled => return true,
+            State::Finished | State::Sending(_) => return false,
+            State::Running(input) => {
+                if input.write_all(b"cancel\n").and_then(|_| input.flush()).is_err() {
+                    return false;
+                }
+            }
+        }
+        let (state, _) = self
+            .0
+            .1
+            .wait_timeout_while(state, Duration::from_secs(1), |s| matches!(s, State::Running(_)))
+            .unwrap_or_else(std::sync::PoisonError::into_inner);
+        matches!(*state, State::Cancelled)
+    }
+
+    #[cfg(windows)]
+    /// Submit through this executable's private helper mode and await its final receipt.
+    pub(crate) fn submit(&self, pdf: &[u8], job: &craft_print::Job) -> Result<String, String> {
+        self.submit_with(pdf, job, |directory| {
+            let mut command = Command::new(std::env::current_exe()?);
+            command.arg(FLAG).arg(directory);
+            Ok(command)
+        })
+    }
+
+    /// Publish a terminal state on every result, waking a cancellation caller awaiting an ack.
+    fn submit_with(&self, pdf: &[u8], job: &craft_print::Job, command: impl FnOnce(&Path) -> std::io::Result<Command>) -> Result<String, String> {
+        let result = self.run(pdf, job, command);
+        let mut state = self.state();
+        if !matches!(*state, State::Cancelled) {
+            *state = State::Finished;
+        }
+        self.0.1.notify_all();
+        if matches!(*state, State::Cancelled) { Err(CANCELLED.into()) } else { result }
+    }
+
+    /// Own the private inputs and child until receipt, acknowledged cancellation or failure.
+    fn run(&self, pdf: &[u8], job: &craft_print::Job, command: impl FnOnce(&Path) -> std::io::Result<Command>) -> Result<String, String> {
+        if matches!(*self.state(), State::Cancelled) {
+            return Err(CANCELLED.into());
+        }
+        if pdf.len() as u64 > MAX_PDF {
+            return Err("the print PDF exceeds the 512 MiB helper limit".into());
+        }
+        let request = serde_json::to_vec(&Request::from_job(job)).map_err(error)?;
+        if request.len() as u64 > MAX_MESSAGE {
+            return Err("the print settings exceed the helper message limit".into());
+        }
+        let directory = tempfile::tempdir().map_err(error)?;
+        std::fs::write(directory.path().join("job.json"), request).map_err(error)?;
+        std::fs::write(directory.path().join("input.pdf"), pdf).map_err(error)?;
+        let child =
+            command(directory.path()).map_err(error)?.stdin(Stdio::piped()).stdout(Stdio::piped()).stderr(Stdio::null()).spawn().map_err(error)?;
+        let child = Arc::new(Mutex::new(child));
+        let outcome = (|| {
+            let (input, output) = {
+                let mut process = child.lock().unwrap_or_else(std::sync::PoisonError::into_inner);
+                (process.stdin.take().ok_or("print helper has no input pipe")?, process.stdout.take().ok_or("print helper has no output pipe")?)
+            };
+            {
+                let mut state = self.state();
+                if matches!(*state, State::Cancelled) {
+                    return Err(CANCELLED.into());
+                }
+                *state = State::Waking(child.clone());
+            }
+            let mut output = BufReader::new(output);
+            loop {
+                let message = line(&mut output)?.ok_or("the print helper stopped before it was ready")?;
+                let Some((_, json)) = message.split_once(PREFIX) else { continue };
+                if !matches!(serde_json::from_str::<Event>(json).map_err(error)?, Event::Ready) {
+                    return Err("the print helper did not acknowledge startup".into());
+                }
+                break;
+            }
+            {
+                let mut state = self.state();
+                if matches!(*state, State::Cancelled) {
+                    return Err(CANCELLED.into());
+                }
+                *state = State::Running(input);
+                if let State::Running(input) = &mut *state {
+                    input.write_all(b"start\n").and_then(|_| input.flush()).map_err(error)?;
+                }
+            }
+            while let Some(line) = line(&mut output)? {
+                let Some((_, json)) = line.split_once(PREFIX) else { continue };
+                match serde_json::from_str::<Event>(json).map_err(error)? {
+                    Event::Ready => return Err("the print helper repeated its startup acknowledgement".into()),
+                    Event::Cancelled => {
+                        *self.state() = State::Cancelled;
+                        self.0.1.notify_all();
+                        return Err(CANCELLED.into());
+                    }
+                    Event::Sending => {
+                        let mut state = self.state();
+                        let previous = std::mem::replace(&mut *state, State::Finished);
+                        *state = match previous {
+                            State::Running(input) | State::Sending(input) => State::Sending(input),
+                            other => other,
+                        };
+                        self.0.1.notify_all();
+                    }
+                    Event::Done(result) => return result,
+                }
+            }
+            Err("the print helper stopped without a receipt".into())
+        })();
+        // A cancelled child has acknowledged that it cannot send. Other outcomes are terminal
+        // receipts or a broken protocol/pipe; reap the process and its private input files.
+        if outcome.is_err() {
+            let _ = child.lock().unwrap_or_else(std::sync::PoisonError::into_inner).kill();
+        }
+        let status = child.lock().unwrap_or_else(std::sync::PoisonError::into_inner).wait().map_err(error)?;
+        if outcome.is_ok() && !status.success() {
+            return Err("the print helper failed after its receipt".into());
+        }
+        outcome
+    }
+}
+
+/// Keep helper transport failures distinct from a printer's own error message.
+fn error(error: impl std::fmt::Display) -> String {
+    format!("print helper: {error}")
+}
+
+/// Read at most one byte beyond the cap so oversized files fail without unbounded allocation.
+fn bounded(mut input: impl Read, limit: u64) -> Result<Vec<u8>, String> {
+    let mut bytes = Vec::new();
+    input.by_ref().take(limit.saturating_add(1)).read_to_end(&mut bytes).map_err(error)?;
+    if bytes.len() as u64 > limit { Err("print helper input exceeds its limit".into()) } else { Ok(bytes) }
+}
+
+/// Read one bounded UTF-8 protocol record, distinguishing EOF from an oversized message.
+fn line(input: &mut impl BufRead) -> Result<Option<String>, String> {
+    let mut bytes = Vec::new();
+    input.take(MAX_MESSAGE + 1).read_until(b'\n', &mut bytes).map_err(error)?;
+    if bytes.len() as u64 > MAX_MESSAGE {
+        return Err("print helper message exceeds its limit".into());
+    }
+    if bytes.is_empty() { Ok(None) } else { String::from_utf8(bytes).map(|s| Some(s.trim_end().to_owned())).map_err(error) }
+}
+
+/// Serialize complete events under stdout's lock so the driver and listener cannot interleave.
+fn emit(event: &Event) -> Result<(), String> {
+    let mut output = std::io::stdout().lock();
+    write!(output, "{PREFIX}").map_err(error)?;
+    serde_json::to_writer(&mut output, event).map_err(error)?;
+    writeln!(output).and_then(|_| output.flush()).map_err(error)
+}
+
+/// Remove both plaintext inputs even when one fails; already-removed inputs count as success.
+fn remove_inputs(directory: &Path) -> Result<(), String> {
+    let mut first_error = None;
+    for name in ["job.json", "input.pdf"] {
+        if let Err(failure) = std::fs::remove_file(directory.join(name))
+            && failure.kind() != std::io::ErrorKind::NotFound
+        {
+            first_error.get_or_insert_with(|| error(format!("could not remove {name}: {failure}")));
+        }
+    }
+    first_error.map_or(Ok(()), Err)
+}
+
+/// Load and erase the private inputs before driver work, while an independent listener can cancel.
+fn helper(
+    directory: &Path,
+    cancel_request: impl Fn(&craft_print::Cancel) -> bool + Send + 'static,
+    submit: impl FnOnce(&[u8], &craft_print::Job, &craft_print::Cancel) -> Result<String, String>,
+) -> Result<(), String> {
+    let mut input = BufReader::new(std::io::stdin());
+    emit(&Event::Ready)?;
+    if line(&mut input)?.as_deref() != Some("start") {
+        return Err("print helper requires a start acknowledgement".into());
+    }
+    let cancel = craft_print::Cancel::new();
+    let listener = cancel.clone();
+    std::thread::Builder::new()
+        .name("print-cancel".into())
+        .spawn(move || {
+            loop {
+                match line(&mut input) {
+                    Ok(Some(command)) if command == "cancel" => {
+                        if cancel_request(&listener) {
+                            let _ = emit(&Event::Cancelled);
+                            std::process::exit(0);
+                        }
+                        if emit(&Event::Sending).is_err() {
+                            break;
+                        }
+                    }
+                    _ => break,
+                }
+            }
+            // If the parent disappears, never leave a waiting driver process behind. Sending
+            // jobs finish normally so the OS gets their pages even if the UI has gone away.
+            if cancel_request(&listener) {
+                std::process::exit(0);
+            }
+        })
+        .map_err(error)?;
+    let request: Request =
+        serde_json::from_slice(&bounded(std::fs::File::open(directory.join("job.json")).map_err(error)?, MAX_MESSAGE)?).map_err(error)?;
+    let pdf = bounded(std::fs::File::open(directory.join("input.pdf")).map_err(error)?, MAX_PDF)?;
+    remove_inputs(directory)?;
+    let result = submit(&pdf, &request.job(), &cancel);
+    emit(&Event::Done(result))
+}
+
+#[cfg(windows)]
+/// Handle only the private first-argument flag, before logging or GUI initialization.
+pub(crate) fn helper_mode() -> Option<i32> {
+    let mut args = std::env::args_os().skip(1);
+    if args.next().as_deref() != Some(std::ffi::OsStr::new(FLAG)) {
+        return None;
+    }
+    let result = args.next().ok_or_else(|| "print helper directory missing".to_owned()).and_then(|dir| {
+        helper(Path::new(&dir), craft_print::Cancel::cancel, |pdf, job, cancel| {
+            craft_print::submit_cancellable(pdf, job, &mut |_, _| {}, cancel).map_err(error)
+        })
+    });
+    if let Err(message) = &result {
+        let _ = emit(&Event::Done(Err(message.clone())));
+    }
+    Some(i32::from(result.is_err()))
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    /// A real test child exercises production pipes, token handling and process cleanup.
+    #[test]
+    #[ignore = "launched by the process protocol tests"]
+    fn helper_process() {
+        let Some(directory) = std::env::var_os("WORDCRAFT_TEST_PRINT_DIR") else { return };
+        if std::env::var("WORDCRAFT_TEST_PRINT_MODE").as_deref() == Ok("waking") {
+            std::fs::write(Path::new(&directory).join("waking"), b"starting").unwrap();
+            loop {
+                std::thread::park();
+            }
+        }
+        let result = helper(
+            Path::new(&directory),
+            |token| {
+                if std::env::var("WORDCRAFT_TEST_PRINT_MODE").as_deref() == Ok("sending") { false } else { token.cancel() }
+            },
+            |pdf, job, _token| {
+                assert_eq!(pdf, b"synthetic PDF");
+                assert_eq!(job.printer.as_deref(), Some("test queue"));
+                match std::env::var("WORDCRAFT_TEST_PRINT_MODE").as_deref() {
+                    Ok("held") => {
+                        let _provider = Mutex::new(());
+                        let _held = _provider.lock().unwrap();
+                        std::fs::write(Path::new(&directory).join("held"), b"driver entered").unwrap();
+                        loop {
+                            std::thread::park();
+                        }
+                    }
+                    Ok("crash") => std::process::exit(7),
+                    Ok("sending") => {
+                        std::fs::write(Path::new(&directory).join("sending"), b"sending").unwrap();
+                        while !Path::new(&directory).join("release").exists() {
+                            std::thread::sleep(Duration::from_millis(5));
+                        }
+                        Ok("spooled synthetic job".into())
+                    }
+                    _ => Ok("spooled synthetic job".into()),
+                }
+            },
+        );
+        std::process::exit(i32::from(result.is_err()));
+    }
+
+    /// Launch just the ignored helper entry, without invoking printer hardware.
+    fn command(directory: &Path, mode: &str) -> std::io::Result<Command> {
+        let mut command = Command::new(std::env::current_exe()?);
+        let name = module_path!().split_once("::").map_or(module_path!(), |(_, name)| name);
+        command.args(["--ignored", "--exact", &format!("{name}::helper_process"), "--nocapture"]);
+        command.env("WORDCRAFT_TEST_PRINT_DIR", directory).env("WORDCRAFT_TEST_PRINT_MODE", mode);
+        Ok(command)
+    }
+
+    /// Give the child a literal printer identity to verify transport.
+    fn job() -> craft_print::Job {
+        craft_print::Job { printer: Some("test queue".into()), ..Default::default() }
+    }
+
+    #[test]
+    /// Cancelling before submission cannot launch a driver process.
+    fn cancelled_start_never_launches_a_helper() {
+        let submission = Submission::new();
+        assert!(submission.cancel());
+        let result = submission.submit_with(b"synthetic PDF", &job(), |_| Err(std::io::Error::other("must not launch")));
+        assert_eq!(result, Err(CANCELLED.into()));
+    }
+
+    #[test]
+    /// A cancellation during creation withholds the start grant and removes private inputs.
+    fn cancellation_during_startup_never_starts_the_driver() {
+        let submission = Submission::new();
+        let worker = submission.clone();
+        let (started, starting) = std::sync::mpsc::channel();
+        let (release, wait) = std::sync::mpsc::channel();
+        let running = std::thread::spawn(move || {
+            worker.submit_with(b"synthetic PDF", &job(), |directory| {
+                started.send(directory.to_path_buf()).unwrap();
+                wait.recv().unwrap();
+                command(directory, "held")
+            })
+        });
+        let directory = starting.recv_timeout(Duration::from_secs(10)).unwrap();
+        assert!(submission.cancel());
+        release.send(()).unwrap();
+        assert_eq!(running.join().unwrap(), Err(CANCELLED.into()));
+        assert!(!directory.exists(), "startup cancellation reaps the child and removes its inputs");
+    }
+
+    #[test]
+    /// A hung startup has no permission to send and can be killed without a token ack.
+    fn an_unresponsive_starting_process_can_be_stopped_without_an_acknowledgement() {
+        let submission = Submission::new();
+        let worker = submission.clone();
+        let (path_tx, path_rx) = std::sync::mpsc::channel();
+        let running = std::thread::spawn(move || {
+            worker.submit_with(b"synthetic PDF", &job(), |dir| {
+                path_tx.send(dir.to_path_buf()).unwrap();
+                command(dir, "waking")
+            })
+        });
+        let directory = path_rx.recv_timeout(Duration::from_secs(10)).unwrap();
+        let deadline = std::time::Instant::now() + Duration::from_secs(10);
+        while !directory.join("waking").exists() && std::time::Instant::now() < deadline {
+            std::thread::sleep(Duration::from_millis(5));
+        }
+        assert!(directory.join("waking").exists());
+        assert!(submission.cancel(), "a process without a start grant cannot send anything");
+        assert_eq!(running.join().unwrap(), Err(CANCELLED.into()));
+        assert!(!directory.exists());
+    }
+
+    #[test]
+    /// Job transport preserves every explicit setting, including Unicode and duplex direction.
+    fn job_options_survive_the_process_boundary() {
+        let expected = craft_print::Job {
+            printer: Some("Office λ".into()),
+            copies: 3,
+            collate: false,
+            duplex: Some(craft_print::Duplex::ShortEdge),
+            grayscale: true,
+            scaling: craft_print::Scaling::ShrinkToFit,
+            title: "A document".into(),
+            options: vec![("media".into(), "iso_a4_210x297mm".into())],
+        };
+        let bytes = serde_json::to_vec(&Request::from_job(&expected)).unwrap();
+        let received: Request = serde_json::from_slice(&bytes).unwrap();
+        assert_eq!(received.job(), expected);
+    }
+
+    #[test]
+    /// Retry succeeds immediately after cancellation, before joining the old worker.
+    fn cancelled_driver_does_not_block_the_next_process() {
+        let submission = Submission::new();
+        let worker = submission.clone();
+        let (path_tx, path_rx) = std::sync::mpsc::channel();
+        let running = std::thread::spawn(move || {
+            worker.submit_with(b"synthetic PDF", &job(), |dir| {
+                path_tx.send(dir.to_path_buf()).unwrap();
+                command(dir, "held")
+            })
+        });
+        let directory = path_rx.recv_timeout(Duration::from_secs(10)).unwrap();
+        let deadline = std::time::Instant::now() + Duration::from_secs(10);
+        while !directory.join("held").exists() && std::time::Instant::now() < deadline {
+            std::thread::sleep(Duration::from_millis(5));
+        }
+        assert!(directory.join("held").exists(), "the child is held inside its driver");
+        assert!(!directory.join("input.pdf").exists(), "the stalled driver retains no plaintext PDF");
+        assert!(!directory.join("job.json").exists(), "loaded settings were erased before driver work");
+        assert!(submission.cancel(), "a stalled driver still acknowledges cancellation");
+        let next = Submission::new();
+        assert_eq!(next.submit_with(b"synthetic PDF", &job(), |dir| command(dir, "ready")), Ok("spooled synthetic job".into()));
+        assert_eq!(running.join().unwrap(), Err(CANCELLED.into()));
+        assert!(!directory.exists(), "the reaped helper's PDF and settings were removed");
+        assert!(!next.cancel(), "a completed job keeps its receipt");
+    }
+
+    #[test]
+    /// A refused cancellation keeps the sending process alive until its receipt.
+    fn cancellation_refusal_preserves_the_sending_job() {
+        let submission = Submission::new();
+        let worker = submission.clone();
+        let (path_tx, path_rx) = std::sync::mpsc::channel();
+        let running = std::thread::spawn(move || {
+            worker.submit_with(b"synthetic PDF", &job(), |dir| {
+                path_tx.send(dir.to_path_buf()).unwrap();
+                command(dir, "sending")
+            })
+        });
+        let directory = path_rx.recv_timeout(Duration::from_secs(10)).unwrap();
+        let deadline = std::time::Instant::now() + Duration::from_secs(10);
+        while !directory.join("sending").exists() && std::time::Instant::now() < deadline {
+            std::thread::sleep(Duration::from_millis(5));
+        }
+        assert!(directory.join("sending").exists());
+        assert!(!submission.cancel(), "a sending job refuses cancellation");
+        assert!(matches!(*submission.state(), State::Sending(_)));
+        assert!(!running.is_finished(), "the child is retained until it finishes sending");
+        std::fs::write(directory.join("release"), b"continue").unwrap();
+        assert_eq!(running.join().unwrap(), Ok("spooled synthetic job".into()));
+        assert!(!directory.exists());
+    }
+
+    #[test]
+    /// Closing the parent's pipe stops a waiting driver and permits parent-side cleanup.
+    fn parent_disconnect_stops_a_waiting_driver() {
+        let submission = Submission::new();
+        let worker = submission.clone();
+        let (path_tx, path_rx) = std::sync::mpsc::channel();
+        let running = std::thread::spawn(move || {
+            worker.submit_with(b"synthetic PDF", &job(), |dir| {
+                path_tx.send(dir.to_path_buf()).unwrap();
+                command(dir, "held")
+            })
+        });
+        let directory = path_rx.recv_timeout(Duration::from_secs(10)).unwrap();
+        let deadline = std::time::Instant::now() + Duration::from_secs(10);
+        while !directory.join("held").exists() && std::time::Instant::now() < deadline {
+            std::thread::sleep(Duration::from_millis(5));
+        }
+        assert!(directory.join("held").exists());
+        *submission.state() = State::Finished; // Drops the parent's pipe as on process exit.
+        assert_eq!(running.join().unwrap(), Err("the print helper stopped without a receipt".into()));
+        assert!(!directory.exists());
+    }
+
+    #[test]
+    /// Helper failure removes its input directory; input limits reject oversize data.
+    fn helper_crashes_are_errors_and_inputs_are_bounded() {
+        let submission = Submission::new();
+        let mut inputs = None;
+        assert_eq!(
+            submission.submit_with(b"synthetic PDF", &job(), |dir| {
+                inputs = Some(dir.to_path_buf());
+                command(dir, "crash")
+            }),
+            Err("the print helper stopped without a receipt".into())
+        );
+        assert!(!inputs.unwrap().exists(), "a helper crash removes the PDF and settings");
+        assert_eq!(bounded(&b"abcd"[..], 3), Err("print helper input exceeds its limit".into()));
+        assert_eq!(bounded(&b"abc"[..], 3), Ok(b"abc".to_vec()));
+    }
+
+    /// Cleanup is repeatable and reports real failures while still attempting both inputs.
+    #[test]
+    fn input_cleanup_tolerates_missing_files_and_reports_failures() {
+        let directory = tempfile::tempdir().unwrap();
+        std::fs::write(directory.path().join("job.json"), b"settings").unwrap();
+        std::fs::write(directory.path().join("input.pdf"), b"private PDF").unwrap();
+        assert_eq!(remove_inputs(directory.path()), Ok(()));
+        assert_eq!(remove_inputs(directory.path()), Ok(()));
+        assert!(!directory.path().join("input.pdf").exists());
+        std::fs::create_dir(directory.path().join("job.json")).unwrap();
+        std::fs::write(directory.path().join("input.pdf"), b"private PDF").unwrap();
+        let error = remove_inputs(directory.path()).unwrap_err();
+        assert!(error.contains("could not remove job.json"), "{error}");
+        assert!(!directory.path().join("input.pdf").exists(), "a settings cleanup failure does not retain the PDF");
+    }
+}

apps/wordcraft/tests/it/appimage_env.rs

@@ -0,0 +1,99 @@
+//! A WordCraft started by an AppImage app inherits that app's `APPIMAGE`, which the self-update
+//! would take for this copy's file and replace (#5653). The real program, started both ways.
+
+use std::ffi::OsString;
+use std::os::unix::ffi::OsStringExt;
+use std::path::{Path, PathBuf};
+use std::process::{Command, Output, Stdio};
+use std::time::{Duration, Instant};
+
+const NOTICE: &str = "ignoring the AppImage";
+
+/// A folder of its own for one test, removed when it ends.
+struct Tmp(PathBuf);
+
+impl Tmp {
+    /// An empty folder named after `test` and this process, made afresh.
+    fn new(test: &str) -> Tmp {
+        let dir = std::env::temp_dir().join(format!("wordcraft-appimage-it-{test}-{}", std::process::id()));
+        let _ = std::fs::remove_dir_all(&dir);
+        std::fs::create_dir_all(&dir).unwrap();
+        Tmp(dir)
+    }
+}
+
+impl Drop for Tmp {
+    /// Removes the folder and what the test left in it.
+    fn drop(&mut self) {
+        let _ = std::fs::remove_dir_all(&self.0);
+    }
+}
+
+/// Runs the real app with `args`, as an AppImage app with `appimage` mounted at `appdir` would
+/// start it, and waits up to 30 s for it to exit (with no display it can't open a window).
+fn app(args: &[OsString], appimage: &Path, appdir: &Path) -> Output {
+    let mut child = Command::new(env!("CARGO_BIN_EXE_wordcraft"))
+        .args(args)
+        .env("APPIMAGE", appimage)
+        .env("APPDIR", appdir)
+        .env("ARGV0", "other-app")
+        .env("OWD", "/")
+        // With no display, an app that tries to open a window fails at once instead.
+        .env_remove("DISPLAY")
+        .env_remove("WAYLAND_DISPLAY")
+        .env_remove("WAYLAND_SOCKET")
+        .env_remove("WORDCRAFT_CONTROL_PORT")
+        .env("WORDCRAFT_NO_PREFS", "1")
+        .stdin(Stdio::null())
+        .stdout(Stdio::piped())
+        .stderr(Stdio::piped())
+        .spawn()
+        .unwrap();
+    // A start-over that kept the variables would start over for ever, and end here.
+    let deadline = Instant::now() + Duration::from_secs(30);
+    while child.try_wait().unwrap().is_none() {
+        if Instant::now() > deadline {
+            let _ = child.kill();
+            panic!("wordcraft {args:?} was still running after 30 s");
+        }
+        std::thread::sleep(Duration::from_millis(50));
+    }
+    child.wait_with_output().unwrap()
+}
+
+/// The notice comes once: the program it started over as no longer sees the variables.
+#[test]
+fn another_programs_variables_are_dropped_once_and_the_program_carries_on() {
+    let tmp = Tmp::new("dropped");
+    let other = tmp.0.join("other-app.AppImage");
+    std::fs::write(&other, b"another program").unwrap();
+    let out = app(&["--version".into()], &other, &tmp.0);
+    let (stdout, stderr) = (String::from_utf8_lossy(&out.stdout), String::from_utf8_lossy(&out.stderr));
+    assert!(out.status.success(), "status {:?}, stderr: {stderr}", out.status);
+    assert!(stdout.starts_with("wordcraft "), "stdout: {stdout}");
+    assert_eq!(stderr.matches(NOTICE).count(), 1, "stderr: {stderr}");
+    assert!(stderr.contains("other-app.AppImage"), "stderr: {stderr}");
+    assert_eq!(std::fs::read(&other).unwrap(), b"another program");
+}
+
+/// Arguments reach the program it starts over as byte for byte, even one that isn't Unicode.
+#[test]
+fn starting_over_keeps_the_arguments() {
+    let tmp = Tmp::new("args");
+    let out = app(&["--sample".into(), "--control".into(), OsString::from_vec(b"79\xe981".to_vec())], &tmp.0.join("gone"), &tmp.0);
+    let stderr = String::from_utf8_lossy(&out.stderr);
+    assert_eq!(out.status.code(), Some(2), "stderr: {stderr}");
+    assert!(stderr.contains("argument 3 is not valid Unicode: 79\u{FFFD}81"), "stderr: {stderr}");
+}
+
+/// Its own image's variables stay, or the self-update couldn't find the file to replace.
+#[test]
+fn its_own_images_variables_stay() {
+    let exe = Path::new(env!("CARGO_BIN_EXE_wordcraft")).canonicalize().unwrap();
+    let mount = exe.parent().unwrap();
+    let out = app(&["--version".into()], &mount.join("WordCraft.AppImage"), mount);
+    let (stdout, stderr) = (String::from_utf8_lossy(&out.stdout), String::from_utf8_lossy(&out.stderr));
+    assert!(out.status.success(), "status {:?}, stderr: {stderr}", out.status);
+    assert!(stdout.starts_with("wordcraft "), "stdout: {stdout}");
+    assert!(!stderr.contains(NOTICE), "stderr: {stderr}");
+}

apps/wordcraft/tests/it/main.rs

@@ -1,3 +1,5 @@
+#[cfg(target_os = "linux")]
+mod appimage_env;
 mod control_arg;
 mod help;
 #[cfg(unix)]

contributors/people.toml

@@ -14,6 +14,10 @@
 # To remove your names, delete your entry (or set fields empty) in a PR. The next
 # contributors.py run, and the next build, drop them.
 
+[people.SenuraMalaka]
+real_name = "Senura Dissanayake"
+sync_github_name = true
+
 [people.divyanshu2074]
 real_name = "Divyanshu Gupta"
 sync_github_name = true
@@ -35,7 +39,9 @@ sync_github_name = false
 real_name = "Saeed Azarehgar"
 
 [people.vangel76]
-sync_github_name = true
+real_name = "Valentin Klein"
+display_name = "vangel76"
+sync_github_name = false
 
 [people.dkhlapov]
 real_name = "Dmitrii Khlapov"
@@ -160,6 +166,8 @@ sync_github_name = true
 
 [people.maxm11]
 real_name = "Max Murphy"
+display_name = "Max Murphy"
+sync_github_name = false
 
 [people.SlavenDj]
 sync_github_name = true
@@ -187,3 +195,12 @@ sync_github_name = true
 
 [people.dylanmtaylor]
 sync_github_name = true
+
+[people.entropybeater]
+display_name = "entropybeater"
+
+[people.helenkwok]
+real_name = "Helen Kwok"
+
+[people.ashokram-si]
+real_name = "Ashok Ram"

craft-i18n.rev

@@ -1 +1 @@
-9d6d54d6d9ff603cc41e12e5f6ba51af3554ab66
+1e515422afdda2f317af430c1f9d68a5f83935b4

crates/control-key/src/lib.rs

@@ -1,5 +1,6 @@
 //! The control channel's key file, one rule for the app (which writes it) and the MCP bridge
 //! (which reads it): `<settings>/control-key.<instance>`. See `docs/control-protocol.md`, Keys.
+#![deny(clippy::unwrap_used, clippy::expect_used, clippy::panic, clippy::unimplemented, clippy::todo, clippy::unreachable)]
 
 use std::path::{Path, PathBuf};
 

crates/doc/Cargo.toml

@@ -10,8 +10,9 @@ workspace = true
 
 [features]
 # Test inputs other crates' tests share (`cfb_guard::repeating_difat_sample`,
-# `zip_guard::zip64_entry_count_sample`); turned on only as a
-# dev-dependency, so never built into the app.
+# `zip_guard::zip64_entry_count_sample`) and the step count of
+# `work::count`; turned on only as a dev-dependency, so never built into
+# the app.
 test-support = []
 
 [dependencies]

crates/doc/src/cfb_guard.rs

@@ -7,8 +7,9 @@
 /// makes it allocate gigabytes. Here the DIFAT is walked the way `cfb` walks it (header entries,
 /// then the DIFAT chain, at most one step per sector; trailing zero and free entries trimmed
 /// the same way) and the file is refused when it lists one FAT sector twice, or more FAT sectors
-/// than its sectors need (one FAT sector covers `sector_len / 4` of them, plus one spare), so the
-/// FAT `cfb` builds is about the size the file needs. Anything else that is wrong is left for
+/// than the file has sectors — every FAT sector is a sector of the file, so a longer list is
+/// duplicates or names `cfb` refuses anyway, and the FAT `cfb` builds (one sector of entries per
+/// listed name) stays within about the file's own size. Anything else that is wrong is left for
 /// `cfb` to report.
 pub fn check(bytes: &[u8]) -> Result<(), String> {
     const MAGIC: [u8; 8] = [0xD0, 0xCF, 0x11, 0xE0, 0xA1, 0xB1, 0x1A, 0xE1];
@@ -52,12 +53,14 @@ pub fn check(bytes: &[u8]) -> Result<(), String> {
     while difat.last() == Some(&FREE) {
         difat.pop();
     }
-    // Each listed FAT sector adds a whole sector of entries. A FAT sector holds the entries of
-    // `sector_len / 4` sectors, so a file needs no more FAT sectors than that covers (one spare),
-    // and none of them twice. A free entry in the middle is left for `cfb` to refuse.
+    // Each listed FAT sector adds a whole sector of entries. A real file lists no more FAT
+    // sectors than the file has — every FAT sector is a sector of the file, and none of them
+    // twice (checked below) — so the FAT `cfb` builds stays within about the file's own size.
+    // The #3642 blowup listed 2048 × 127 names in a ~1 MB file. A tighter bound (one FAT sector
+    // per `sector_len / 4` sectors plus a spare) refused real files, whose writers reserve FAT
+    // for growth: an encrypted package of 78 sectors lists 4 and didn't open (#5109).
     let listed: Vec<u32> = difat.into_iter().filter(|&s| s != FREE).collect();
-    let needed = num_sectors.div_ceil(sector_len / 4).saturating_add(1);
-    if listed.len() > needed {
+    if listed.len() > num_sectors {
         return Err(format!("the DIFAT lists {} FAT sectors in a file of {num_sectors} sectors", listed.len()));
     }
     let mut seen = std::collections::HashSet::with_capacity(listed.len());
@@ -123,10 +126,16 @@ mod tests {
     }
 
     #[test]
-    fn a_fat_sector_listed_twice_or_more_than_the_file_needs_is_refused() {
-        // Each list is shorter than the file has sectors, which the guard used to accept.
+    fn a_fat_sector_listed_twice_or_more_than_the_file_has_is_refused() {
+        // A duplicate, and a list longer than the file has sectors: every FAT sector is a
+        // sector of the file, so a longer list is one of the two.
         assert!(super::check(&header_difat(10, &[0, 0])).is_err_and(|e| e.contains("FAT sector 0 twice")));
-        assert!(super::check(&header_difat(10, &[0, 1, 2])).is_err_and(|e| e.contains("3 FAT sectors")));
+        assert!(
+            super::check(&header_difat(10, &[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10])).is_err_and(|e| e.contains("11 FAT sectors in a file of 10 sectors"))
+        );
+        // A writer may reserve FAT for growth: a real encrypted package of 78 sectors lists
+        // 4 FAT sectors and didn't open since the #3642 guard landed (#5109).
+        assert!(super::check(&header_difat(78, &[0, 1, 2, 3])).is_ok());
         assert!(super::check(&header_difat(10, &[0])).is_ok());
         assert!(super::check(&header_difat(200, &[0, 1])).is_ok());
     }

crates/doc/src/connector.rs

@@ -213,6 +213,12 @@ pub struct ShapeExtra {
     /// main colour (the shape's `stroke`).
     #[serde(skip_serializing_if = "Option::is_none")]
     pub outline_fill: Option<crate::fill::Fill>,
+    /// An ODF custom shape's geometry as read (its `draw:enhanced-geometry`, equations and
+    /// handles), whose path the shape's freeform was worked out from: saved back as it was, so
+    /// LibreOffice draws its own shape (or a preset WordCraft draws as a rectangle: kept with
+    /// it). Dropped when the shape is made another kind.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub odf_geometry: Option<std::sync::Arc<String>>,
     /// A horizontal rule (VML `o:hr`: a web page's `<hr>` pasted into Word): it takes its whole
     /// line, and is drawn across [`HorizontalRule::pct`] of it or at the shape's own width.
     #[serde(skip_serializing_if = "Option::is_none")]
@@ -384,7 +390,6 @@ pub fn end_direction(kind: ShapeKind, w: f32, h: f32, at_end: bool) -> (f32, f32
                 (0.0, 1.0)
             }
         }
-        ShapeKind::Line => (w, -h),
         _ => (w, h),
     };
     let len = (dx * dx + dy * dy).sqrt();

crates/doc/src/control.rs

@@ -254,6 +254,18 @@ pub struct ContentControl {
     /// `w:sdtEndPr`, kept verbatim (XML).
     #[serde(skip_serializing_if = "String::is_empty")]
     pub end_pr_xml: String,
+    /// The size of the form control it came from (an ODF `draw:control`'s `svg:width` and
+    /// `svg:height`), points: kept for saving back as one.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub form_size: Option<[f32; 2]>,
+    /// How far that form control stands raised from the baseline (points, negative lowered:
+    /// an ODF control hanging from the baseline is lowered by its height), kept for saving back.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub form_raise: Option<f32>,
+    /// A form control WordCraft has no kind for (an ODF button, image button, label…): its form
+    /// XML, kept verbatim and saved back as it was. It shows its label, locked.
+    #[serde(skip_serializing_if = "String::is_empty")]
+    pub form_xml: String,
 }
 
 impl ContentControl {

crates/doc/src/document.rs

@@ -7,7 +7,7 @@ use super::settings::{CoreProps, CustomProp, Settings};
 use super::text_walk::{TextMode, notes_and_boxes, story_word_count_of, text_of, walk};
 use std::sync::Arc;
 
-use super::{BTreeMap, CharProps, Document, InlineObject, Numbering, Paragraph, SectionProps, StyleSheet, Table, edit, numbering, para, section};
+use super::{BTreeMap, CharProps, Document, InlineObject, Paragraph, SectionProps, StyleSheet, Table, edit, numbering, para, section};
 
 /// [`Document::media`] by content, so adding media is linear in the bytes added rather than a
 /// scan of every earlier image (CWE-407). Rebuilt whenever `media` changed size behind its back;
@@ -62,8 +62,8 @@ impl Document {
             body: vec![para_block(Paragraph::new())],
             last_section: SectionProps::default(),
             parts: BTreeMap::new(),
-            styles: StyleSheet::builtin(),
-            numbering: Numbering::default(),
+            styles: Arc::new(StyleSheet::builtin()),
+            numbering: Arc::default(),
             comments: BTreeMap::new(),
             revisions: Vec::new(),
             settings: Settings::default(),
@@ -76,6 +76,7 @@ impl Document {
             passthrough: BTreeMap::new(),
             glossary: Vec::new(),
             media_index: MediaIndex::default(),
+            kept: Default::default(),
         }
     }
 
@@ -152,6 +153,7 @@ impl Document {
             passthrough,
             glossary,
             media_index: _,
+            kept: _,
         } = self;
         let Document {
             body: other_body,
@@ -171,6 +173,7 @@ impl Document {
             passthrough: other_passthrough,
             glossary: other_glossary,
             media_index: _,
+            kept: _,
         } = other;
         same_blocks(body, other_body)
             && same_parts(parts, other_parts)
@@ -210,7 +213,9 @@ impl Document {
         if !parent.len().is_multiple_of(3) {
             return None;
         }
+        crate::work::steps(1);
         for ch in parent.chunks(3) {
+            crate::work::steps(1);
             let [b, r, c] = ch else { return None };
             let Block::Table(t) = &**bl.get(*b as usize)? else { return None };
             bl = &t.rows.get(*r as usize)?.cells.get(*c as usize)?.blocks;
@@ -734,6 +739,52 @@ impl Document {
         }
     }
 
+    /// [`Document::objects_mut`] for filling in what an object shows from what it is (a chart's
+    /// drawing from its data), which doesn't change the document: the markup kept as read for
+    /// the paragraphs and tables holding them (see [`crate::kept`]) follows them.
+    pub fn fill_objects(&mut self, wants: &dyn Fn(&InlineObject) -> bool, f: &mut dyn FnMut(&mut InlineObject)) {
+        // The stories (text boxes', notes') as read before, when filling changes one: they stay
+        // as read after it, so the paragraphs showing them keep their markup.
+        let holds = |b: &Arc<Block>| {
+            let mut any = false;
+            edit::each_para(b, 0, &mut |p| any |= p.objects.iter().any(wants));
+            any
+        };
+        let as_read: Vec<u32> = match self.kept.get() {
+            Some(k) if !k.story_fps.is_empty() && self.parts.values().any(|p| p.blocks.iter().any(holds)) => {
+                k.story_fps.iter().filter(|(id, fp)| **fp == crate::kept::story_fingerprint(self, **id)).map(|(id, _)| *id).collect()
+            }
+            _ => Vec::new(),
+        };
+        let mut moves: Vec<crate::kept::Moved> = Vec::new();
+        let mut stories: Vec<&mut Blocks> = vec![&mut self.body];
+        stories.extend(self.parts.values_mut().map(|p| &mut p.blocks));
+        for blocks in stories {
+            for b in blocks.iter_mut() {
+                let mut any = false;
+                edit::each_para(b, 0, &mut |p| any |= p.objects.iter().any(wants));
+                if !any {
+                    continue;
+                }
+                let before = crate::kept::identities(b);
+                edit::each_para_mut(Arc::make_mut(b), 0, &mut |p| {
+                    if p.objects.iter().any(wants) {
+                        p.objects.iter_mut().filter(|o| wants(o)).for_each(&mut *f);
+                        p.touch();
+                    }
+                });
+                moves.push((before, crate::kept::identities(b)));
+            }
+        }
+        let refreshed: Vec<(u32, u64)> = as_read.into_iter().map(|id| (id, crate::kept::story_fingerprint(self, id))).collect();
+        if let Some(k) = self.kept.get_mut() {
+            for (before, after) in moves {
+                k.follow(&before, &after);
+            }
+            k.story_fps.extend(refreshed);
+        }
+    }
+
     /// Repair every paragraph's runs (see [`Paragraph::normalize`]); for models from untrusted
     /// JSON, whose run lengths may overflow, fall short or split a character.
     pub fn normalize_runs(&mut self) {

crates/doc/src/edit.rs

@@ -609,8 +609,10 @@ impl Document {
             let mut at = at.into_iter().peekable();
             let mut start = 0usize;
             for r in &mut p.runs {
+                crate::work::steps(1);
                 let end = start.saturating_add(r.len);
                 while at.peek().is_some_and(|&o| o < start) {
+                    crate::work::steps(1);
                     at.next();
                 }
                 if at.peek().is_some_and(|&o| o < end) {
@@ -770,6 +772,10 @@ impl Document {
         if a.path.parent() == b.path.parent() {
             // Same container: trim the ends, drop the blocks between, join.
             let (ia, ib) = (a.path.last(), b.path.last());
+            // Removing an already-empty paragraph must not impose its style or mark formatting
+            // on the following content. Test before trimming: deleting actual heading text is
+            // still an ordinary merge that keeps the first paragraph's properties.
+            let empty_start = self.para(a.story, &a.path).is_some_and(Paragraph::is_empty);
             // Kept markers of the paragraphs in between move to the join.
             let mut moved: Vec<(InlineObject, CharProps)> = Vec::new();
             // `kept` is sorted: one walk of each paragraph's objects and runs (#3614).
@@ -817,15 +823,18 @@ impl Document {
             if ib < bl.len() && ia < ib {
                 bl.drain(ia + 1..=ib);
             }
-            tail.props = ParaProps::default();
             // The deleted marks include `a`'s, so a section break there goes too: the joined
             // paragraph ends where `b`'s mark did, and its text takes the following section's
             // setup, as in Word (#2127).
             let sect = tail.section.take();
             let pa = self.para_mut(a.story, &a.path)?;
-            let keep_props = pa.props.clone();
+            let keep_props = if empty_start { tail.props.clone() } else { pa.props.clone() };
+            let keep_mark = empty_start.then(|| tail.mark.clone());
             pa.append(tail);
             pa.props = keep_props;
+            if let Some(mark) = keep_mark {
+                pa.mark = mark;
+            }
             let before = std::mem::replace(&mut pa.section, sect);
             if a.story == crate::StoryRef::Body && a.path.0.len() == 1 {
                 // Nearest first, as "link to previous" resolved them before the delete.
@@ -1054,7 +1063,7 @@ impl Document {
         let mut default_style = None;
         if keep_styles {
             if src.styles.get("Normal").is_none() {
-                src.styles.upsert(crate::styles::Style { id: "Normal".into(), name: "Normal".into(), ..Default::default() });
+                Arc::make_mut(&mut src.styles).upsert(crate::styles::Style { id: "Normal".into(), name: "Normal".into(), ..Default::default() });
             }
             // Ids given out here are taken too: two styles never get the same new one.
             let mut given = std::collections::HashSet::new();
@@ -1089,7 +1098,7 @@ impl Document {
             abs_map.insert(a.id, id);
             let mut a = crate::numbering::AbstractNum { id, ..a.clone() };
             a.levels.iter_mut().for_each(|l| restyle(&mut l.style));
-            self.numbering.abstracts.push(a);
+            Arc::make_mut(&mut self.numbering).abstracts.push(a);
         }
         let next_num = self.numbering.nums.iter().map(|n| n.id.saturating_add(1)).max().unwrap_or(1).max(1);
         let mut nums = BTreeMap::new();
@@ -1099,7 +1108,7 @@ impl Document {
             let abstract_id = abs_map.get(&n.abstract_id).copied().unwrap_or(n.abstract_id);
             let mut n = crate::numbering::Num { id, abstract_id, ..n.clone() };
             n.level_overrides.iter_mut().for_each(|(_, l)| restyle(&mut l.style));
-            self.numbering.nums.push(n);
+            Arc::make_mut(&mut self.numbering).nums.push(n);
         }
         let renumber = |n: &mut Option<crate::props::NumRef>| {
             if let Some(n) = n.as_mut()
@@ -1138,7 +1147,7 @@ impl Document {
             }
             if self.styles.get(&st.id).is_none() {
                 renumber(&mut st.para.numbering);
-                self.styles.upsert(st);
+                Arc::make_mut(&mut self.styles).upsert(st);
             }
         }
         let rev = u32::try_from(self.revisions.len()).unwrap_or(u32::MAX);

crates/doc/src/edit/tests.rs

@@ -663,13 +663,15 @@ fn removing_a_whole_table_keeps_a_control_the_range_only_partly_covers() {
 
 #[test]
 fn a_table_nested_deeper_than_the_paragraph_walk_is_never_removed_whole() {
-    // 20 tables, each in the first cell of the one around it, ahead of that cell's paragraph:
-    // the innermost paragraphs come first in the document but the paragraph walk stops short.
+    // The deepest nesting allowed, each table in the first cell of the one around it, ahead of
+    // that cell's paragraph: the innermost paragraphs come first in the document but the
+    // paragraph walk stops short.
+    const { assert!(MAX_TABLE_DEPTH > crate::PARA_WALK_DEPTH) };
     let mut t = Table::new(1, 1, 200.0);
     if let Some(cell) = t.rows.get_mut(0).and_then(|r| r.cells.get_mut(0)) {
         cell.blocks = vec![para_block(Paragraph::with_text("deep", CharProps::default()))];
     }
-    for _ in 0..20 {
+    for _ in 1..MAX_TABLE_DEPTH {
         let mut outer = Table::new(1, 1, 200.0);
         if let Some(cell) = outer.rows.get_mut(0).and_then(|r| r.cells.get_mut(0)) {
             cell.blocks.insert(0, Arc::new(Block::Table(t)));
@@ -690,19 +692,19 @@ fn adopting_a_document_renumbers_what_it_refers_to() {
     let mut ours = doc(&["ours"]);
     let our_note = ours.add_part(PartKind::Footnote, vec![para_block(Paragraph::with_text("our note", CharProps::default()))]);
     ours.comments.insert(0, Comment { part: our_note, ..Comment::default() });
-    let our_list = ours.numbering.add_list(crate::numbering::ListKind::Bullet);
+    let our_list = Arc::make_mut(&mut ours.numbering).add_list(crate::numbering::ListKind::Bullet);
     ours.revisions.push(Revision { kind: RevisionKind::Insert, ..Revision::default() });
 
     let mut theirs = doc(&["theirs"]);
     let note = theirs.add_part(PartKind::Footnote, vec![para_block(Paragraph::with_text("their note", CharProps::default()))]);
     let said = theirs.add_part(PartKind::Comment, vec![para_block(Paragraph::with_text("their comment", CharProps::default()))]);
     theirs.comments.insert(0, Comment { author: "Them".into(), part: said, durable_id: Some("0A0B0C0D".into()), ..Comment::default() });
-    let list = theirs.numbering.add_list(crate::numbering::ListKind::Bullet);
+    let list = Arc::make_mut(&mut theirs.numbering).add_list(crate::numbering::ListKind::Bullet);
     assert_eq!(list, our_list);
     theirs.revisions.push(Revision { kind: RevisionKind::Delete, ..Revision::default() });
     let mut style = crate::styles::Style { id: "TheirList".into(), name: "Their List".into(), ..Default::default() };
     style.para.numbering = Some(crate::props::NumRef { num: list, level: 0 });
-    theirs.styles.upsert(style);
+    Arc::make_mut(&mut theirs.styles).upsert(style);
     let p = theirs.para_mut(StoryRef::Body, &Path::top(0)).unwrap();
     p.props.numbering = Some(crate::props::NumRef { num: list, level: 0 });
     p.runs[0].props.del = Some(0);
@@ -746,11 +748,11 @@ fn a_kept_character_style_leaves_what_it_doesnt_set_to_the_paragraph_style() {
     use crate::styles::{Style, StyleKind, StyleSheet};
     let mut ours = doc(&["ours"]);
     let mut theirs = doc(&["theirs"]);
-    theirs.styles = StyleSheet::empty();
-    theirs.styles.default_chr = CharProps { size: Some(9.0), ..Default::default() };
+    theirs.styles = Arc::new(StyleSheet::empty());
+    Arc::make_mut(&mut theirs.styles).default_chr = CharProps { size: Some(9.0), ..Default::default() };
     let big = CharProps { size: Some(20.0), ..Default::default() };
-    theirs.styles.upsert(Style { id: "Big".into(), name: "Big".into(), chr: big, ..Default::default() });
-    theirs.styles.upsert(Style { id: "Link".into(), name: "Link".into(), kind: StyleKind::Character, ..Default::default() });
+    Arc::make_mut(&mut theirs.styles).upsert(Style { id: "Big".into(), name: "Big".into(), chr: big, ..Default::default() });
+    Arc::make_mut(&mut theirs.styles).upsert(Style { id: "Link".into(), name: "Link".into(), kind: StyleKind::Character, ..Default::default() });
     let p = theirs.para_mut(StoryRef::Body, &Path::top(0)).unwrap();
     p.props.style = Some("Big".into());
     p.runs[0].props.style = Some("Link".into());
@@ -764,15 +766,15 @@ fn a_kept_character_style_leaves_what_it_doesnt_set_to_the_paragraph_style() {
 fn kept_styles_renamed_on_adoption_are_renamed_in_list_levels_too() {
     use crate::styles::Style;
     let mut ours = doc(&["ours"]);
-    ours.styles.upsert(Style { id: "Heading1".into(), name: "heading 1".into(), ..Default::default() });
+    Arc::make_mut(&mut ours.styles).upsert(Style { id: "Heading1".into(), name: "heading 1".into(), ..Default::default() });
     let mut theirs = doc(&["theirs"]);
-    theirs.styles.upsert(Style { id: "Heading1".into(), name: "heading 1".into(), ..Default::default() });
-    let list = theirs.numbering.add_list(crate::numbering::ListKind::Numbered);
+    Arc::make_mut(&mut theirs.styles).upsert(Style { id: "Heading1".into(), name: "heading 1".into(), ..Default::default() });
+    let list = Arc::make_mut(&mut theirs.numbering).add_list(crate::numbering::ListKind::Numbered);
     let abs = theirs.numbering.nums.iter().find(|n| n.id == list).map(|n| n.abstract_id).unwrap();
-    let a = theirs.numbering.abstracts.iter_mut().find(|a| a.id == abs).unwrap();
+    let a = Arc::make_mut(&mut theirs.numbering).abstracts.iter_mut().find(|a| a.id == abs).unwrap();
     a.levels[0].style = Some("Heading1".into());
     let level = a.levels[0].clone();
-    theirs.numbering.nums.iter_mut().find(|n| n.id == list).unwrap().level_overrides.push((1, level));
+    Arc::make_mut(&mut theirs.numbering).nums.iter_mut().find(|n| n.id == list).unwrap().level_overrides.push((1, level));
 
     let before = ours.numbering.abstracts.len();
     ours.adopt_with(theirs, true);
@@ -812,10 +814,10 @@ fn an_adopted_tracked_format_change_keeps_its_old_styles_and_list() {
     use crate::{Revision, RevisionKind};
     for keep in [false, true] {
         let mut ours = doc(&["ours"]);
-        let our_list = ours.numbering.add_list(crate::numbering::ListKind::Bullet);
+        let our_list = Arc::make_mut(&mut ours.numbering).add_list(crate::numbering::ListKind::Bullet);
         let mut theirs = doc(&["theirs"]);
         theirs.revisions.push(Revision { kind: RevisionKind::Format, ..Revision::default() });
-        let list = theirs.numbering.add_list(crate::numbering::ListKind::Bullet);
+        let list = Arc::make_mut(&mut theirs.numbering).add_list(crate::numbering::ListKind::Bullet);
         assert_eq!(list, our_list);
         let p = theirs.para_mut(StoryRef::Body, &Path::top(0)).unwrap();
         // Live and old props name the same styles (ours has them too) and their list.
@@ -843,8 +845,8 @@ fn adopted_styles_never_become_our_defaults() {
         let mut ours = doc(&["ours"]);
         let mut theirs = doc(&["theirs"]);
         // A localized default character style, and one whose id we have too.
-        theirs.styles.upsert(Style { id: "Absatz".into(), kind: StyleKind::Character, default: true, ..Default::default() });
-        theirs.styles.upsert(Style { id: "Strong".into(), kind: StyleKind::Character, default: true, ..Default::default() });
+        Arc::make_mut(&mut theirs.styles).upsert(Style { id: "Absatz".into(), kind: StyleKind::Character, default: true, ..Default::default() });
+        Arc::make_mut(&mut theirs.styles).upsert(Style { id: "Strong".into(), kind: StyleKind::Character, default: true, ..Default::default() });
         ours.adopt_with(theirs, keep);
         assert!(ours.styles.styles.iter().all(|s| !s.default), "keep_styles {keep}");
     }

crates/doc/src/graphic.rs

@@ -51,6 +51,17 @@ pub struct Embedded {
     pub refs: Vec<EmbeddedRef>,
 }
 
+impl Embedded {
+    /// The root element of an ODF embedded object's markup: its files are kept in
+    /// [`crate::Document::passthrough`] for the ODT writer; DOCX shows it as its picture.
+    pub const ODF_OBJECT: &'static str = "<wordcraft:odf-object";
+
+    /// Is it an ODF embedded object (a spreadsheet in an ODT), not Word markup?
+    pub fn is_odf_object(&self) -> bool {
+        self.xml.trim_start().starts_with(Self::ODF_OBJECT)
+    }
+}
+
 /// One relationship an [`Embedded`] object uses.
 #[derive(Clone, Debug, Default, PartialEq, Serialize, Deserialize)]
 pub struct EmbeddedRef {

crates/doc/src/kept.rs

@@ -0,0 +1,783 @@
+//! Markup kept from the file a document was opened from, so that what isn't edited is written
+//! back as it was: a paragraph keeps its source XML, which a writer uses as long as the
+//! paragraph still is what the reader made of it (its [`fingerprint`] unchanged). Anything the
+//! model has no place for (an unknown element, an attribute nobody reads) survives a save that
+//! way. Never saved with the document (WordCraft's own format), and not part of its content
+//! (`==` ignores it).
+
+use std::collections::{BTreeMap, HashMap};
+use std::hash::{Hash, Hasher};
+use std::sync::Arc;
+
+use crate::Paragraph;
+
+/// Most kept markup per document, bytes (a hostile file can't make the kept copy grow without
+/// bound: past it, paragraphs are written from the model as before).
+pub const MAX_KEPT_BYTES: usize = 64 * 1024 * 1024;
+
+/// The kept markup of a document (see the module docs).
+#[derive(Clone, Debug, Default)]
+pub struct Kept(Option<Arc<KeptXml>>);
+
+/// What a reader kept.
+#[derive(Clone, Debug, Default)]
+pub struct KeptXml {
+    /// The format it came from (`"odt"`): only that format's writer uses it.
+    pub format: String,
+    /// Paragraph [`fingerprint`] → the paragraph's markup, in the order read (identical
+    /// paragraphs keep the first's).
+    pub paras: HashMap<u64, Arc<str>>,
+    /// The same for the body's tables ([`table_fingerprint`]).
+    pub tables: HashMap<u64, Arc<str>>,
+    /// While reading: the markup of the body's `n`th table, until [`Kept::seal`].
+    pub pending_tables: Vec<(usize, Arc<str>)>,
+    /// Wrappers kept as read around paragraphs (sections): for a paragraph's fingerprint, the
+    /// start tags opening before it and how many close after it. Only paragraphs that were the
+    /// only one of their kind have them (a copy can't open a section twice).
+    pub wraps: HashMap<u64, (Vec<Arc<str>>, u32)>,
+    /// While reading: the same by paragraph `rev`, until [`Kept::seal`].
+    pub pending_wraps: HashMap<u64, (Vec<Arc<str>>, u32)>,
+    /// The same for paragraphs with twins (an empty paragraph has many): by fingerprint and
+    /// [`Paragraph::rev`], which stays while the paragraph does (a writer opens each once).
+    pub twin_wraps: HashMap<(u64, u64), (Vec<Arc<str>>, u32)>,
+    /// The same wrappers around the body's tables, by [`table_fingerprint`] (tables that are
+    /// the only one of their kind).
+    pub table_wraps: HashMap<u64, (Vec<Arc<str>>, u32)>,
+    /// While reading: those of the body's `n`th table, until `seal_kept`.
+    pub pending_table_wraps: Vec<(usize, (Vec<Arc<str>>, u32))>,
+    /// Indexes kept as read: by their first paragraph's fingerprint, how many top-level
+    /// paragraphs they take, those paragraphs' fingerprint together ([`run_fingerprint`]),
+    /// and the index's markup. Written as read while those paragraphs are all as read.
+    pub indexes: HashMap<u64, (usize, u64, Arc<str>)>,
+    /// While reading: by the first paragraph's `rev`, how many paragraphs, and the markup.
+    pub pending_indexes: HashMap<u64, (usize, Arc<str>)>,
+    /// The group of paragraphs written as read together (see `wordcraft_formats::model::Group`)
+    /// each paragraph kept belongs to, by its fingerprint: the group's id and size.
+    pub groups: HashMap<u64, (u32, u32)>,
+    /// While reading: the same by paragraph `rev`, until `seal_kept`.
+    pub pending_groups: HashMap<u64, (u32, u32)>,
+    /// Each text box's, shape's or note's story's [`story_fingerprint`] when read: a paragraph
+    /// showing one is written as read only while it is as read.
+    pub story_fps: HashMap<u32, u64>,
+    /// Each top-level comment's [`comments_fingerprint`] when read (it and its replies): a
+    /// paragraph holding comments is written as read only while they are as read.
+    pub comment_fps: HashMap<u32, u64>,
+    /// Named definitions the kept markup refers to (styles, font faces…), by a key the format
+    /// chooses (`"auto:P12"`, `"style:Heading_20_1"`), as read.
+    pub defs: BTreeMap<String, Arc<str>>,
+    /// While reading: paragraph [`Paragraph::rev`] → its markup, until [`Kept::seal`] turns
+    /// them into fingerprints.
+    pub pending: HashMap<u64, Arc<str>>,
+    /// Bytes kept so far.
+    pub bytes: usize,
+    /// The keys of [`KeptXml::defs`] in the order the file had them.
+    pub order: Vec<String>,
+    /// Files of the package kept definitions refer to (a style's fill picture), by path.
+    pub files: BTreeMap<String, Arc<Vec<u8>>>,
+    /// The document's style definitions' [`styles_fingerprint`] when read: while it is the
+    /// same, the file's own style definitions can be written as they were.
+    pub styles_fp: Option<u64>,
+    /// The document's page setup's [`pages_fingerprint`] when read: while it is the same, the
+    /// file's own page layouts and master pages can be written as they were.
+    pub pages_fp: Option<u64>,
+    /// Each section's master page, as the file named it where its section starts (`None`: the
+    /// file's default), for a writer naming the file's own master pages.
+    pub section_masters: Vec<Option<String>>,
+}
+
+impl PartialEq for Kept {
+    /// Not part of the document's content.
+    fn eq(&self, _: &Self) -> bool {
+        true
+    }
+}
+
+impl Kept {
+    pub fn new(x: KeptXml) -> Kept {
+        Kept(Some(Arc::new(x)))
+    }
+    pub fn get(&self) -> Option<&KeptXml> {
+        self.0.as_deref()
+    }
+    /// What `format`'s writer may use, if anything was kept from such a file.
+    pub fn of(&self, format: &str) -> Option<&KeptXml> {
+        self.get().filter(|k| k.format == format)
+    }
+    pub fn get_mut(&mut self) -> Option<&mut KeptXml> {
+        self.0.as_mut().map(Arc::make_mut)
+    }
+    pub fn clear(&mut self) {
+        self.0 = None;
+    }
+    /// After reading: each paragraph still marked pending (by its `rev`, which any change after
+    /// the reader made it would have moved on) gets its markup under its fingerprint.
+    pub fn seal(&mut self, paras: &mut dyn Iterator<Item = &Paragraph>) {
+        let Some(k) = self.get_mut() else { return };
+        if k.pending.is_empty() {
+            return;
+        }
+        for p in paras {
+            if let Some(xml) = k.pending.remove(&p.rev) {
+                k.paras.entry(fingerprint(p)).or_insert(xml);
+            }
+        }
+        k.pending.clear();
+    }
+}
+
+impl KeptXml {
+    /// The markup kept for paragraph `p`, while it still is what was read.
+    pub fn para(&self, p: &Paragraph) -> Option<&Arc<str>> {
+        if self.paras.is_empty() {
+            return None;
+        }
+        self.paras.get(&fingerprint(p))
+    }
+    /// The markup kept for table `t`, while it still is what was read.
+    pub fn table(&self, t: &crate::Table) -> Option<&Arc<str>> {
+        if self.tables.is_empty() {
+            return None;
+        }
+        self.tables.get(&table_fingerprint(t))
+    }
+    /// Whether the comments paragraph `p` holds (those it begins, with their replies) are as
+    /// read and are all the comments its kept markup `xml` holds: a reply written elsewhere
+    /// in the file, or one added since, isn't in it.
+    pub fn comments_as_read(&self, doc: &crate::Document, p: &Paragraph, xml: &str) -> bool {
+        let in_markup = xml.matches("<office:annotation ").count() + xml.matches("<office:annotation>").count();
+        let mut count = 0usize;
+        for o in &p.objects {
+            if let crate::para::InlineObject::CommentStart { id } = o {
+                let (fp, n) = comments_fingerprint(doc, *id);
+                if self.comment_fps.get(id) != Some(&fp) {
+                    return false;
+                }
+                count = count.saturating_add(n);
+            }
+        }
+        count == in_markup
+    }
+    /// Whether the stories paragraph `p` shows (its text boxes' and shapes' text, its notes') are
+    /// as read.
+    pub fn stories_as_read(&self, doc: &crate::Document, p: &Paragraph) -> bool {
+        // A note's citation is its number in the document's numbering, which is the page setup's.
+        let notes = p.objects.iter().any(|o| matches!(o, crate::para::InlineObject::NoteRef { .. }));
+        if notes && !self.pages_as_read(doc) {
+            return false;
+        }
+        p.objects
+            .iter()
+            .flat_map(|o| match o {
+                crate::para::InlineObject::NoteRef { id, .. } => vec![*id],
+                o => o.text_boxes(),
+            })
+            .all(|id| self.story_fps.get(&id).is_some_and(|fp| *fp == story_fingerprint(doc, id)))
+    }
+    /// The wrappers kept around table `t`, while it still is what was read.
+    pub fn table_wrap(&self, t: &crate::Table) -> Option<&(Vec<Arc<str>>, u32)> {
+        if self.table_wraps.is_empty() {
+            return None;
+        }
+        self.table_wraps.get(&table_fingerprint(t))
+    }
+    /// The wrappers kept around paragraph `p`, while it still is what was read.
+    pub fn wrap(&self, p: &Paragraph) -> Option<&(Vec<Arc<str>>, u32)> {
+        if self.wraps.is_empty() && self.twin_wraps.is_empty() {
+            return None;
+        }
+        let fp = fingerprint(p);
+        self.wraps.get(&fp).or_else(|| self.twin_wraps.get(&(fp, p.rev)))
+    }
+    /// The index kept as read starting at paragraph `p`: how many paragraphs it takes, their
+    /// [`run_fingerprint`] when read, its markup.
+    pub fn index(&self, p: &Paragraph) -> Option<&(usize, u64, Arc<str>)> {
+        if self.indexes.is_empty() {
+            return None;
+        }
+        self.indexes.get(&fingerprint(p))
+    }
+    /// Keep an index's markup, starting at the paragraph whose `rev` is `rev` and taking `n`.
+    pub fn keep_pending_index(&mut self, rev: u64, n: usize, xml: Arc<str>) {
+        if self.bytes.saturating_add(xml.len()) <= MAX_KEPT_BYTES {
+            self.bytes += xml.len();
+            self.pending_indexes.insert(rev, (n, xml));
+        }
+    }
+    /// The paragraph whose `rev` is `rev` belongs to group `group` (id, size).
+    pub fn keep_pending_group(&mut self, rev: u64, group: (u32, u32)) {
+        if self.pending_groups.len() < 100_000 {
+            self.pending_groups.insert(rev, group);
+        }
+    }
+    /// The group paragraph `p`'s markup kept as read belongs to (id, size).
+    pub fn group(&self, p: &Paragraph) -> Option<(u32, u32)> {
+        if self.groups.is_empty() {
+            return None;
+        }
+        self.groups.get(&fingerprint(p)).copied()
+    }
+    /// Keep wrapper markup for the paragraph whose `rev` is `rev`, within the budget.
+    pub fn keep_pending_wrap(&mut self, rev: u64, wrap: (Vec<Arc<str>>, u32)) {
+        let n: usize = wrap.0.iter().map(|t| t.len()).sum();
+        if self.bytes.saturating_add(n) <= MAX_KEPT_BYTES {
+            self.bytes += n;
+            self.pending_wraps.insert(rev, wrap);
+        }
+    }
+    /// Keep wrapper markup for the body's `n`th table, within the budget.
+    pub fn keep_pending_table_wrap(&mut self, n: usize, wrap: (Vec<Arc<str>>, u32)) {
+        let size: usize = wrap.0.iter().map(|t| t.len()).sum();
+        if self.bytes.saturating_add(size) <= MAX_KEPT_BYTES {
+            self.bytes += size;
+            self.pending_table_wraps.push((n, wrap));
+        }
+    }
+    /// Keep `xml` for the body's `n`th table, within the budget.
+    pub fn keep_pending_table(&mut self, n: usize, xml: Arc<str>) {
+        if self.bytes.saturating_add(xml.len()) <= MAX_KEPT_BYTES {
+            self.bytes += xml.len();
+            self.pending_tables.push((n, xml));
+        }
+    }
+    /// Keep `xml` for the paragraph whose `rev` is `rev`, within the budget.
+    pub fn keep_pending(&mut self, rev: u64, xml: Arc<str>) {
+        if self.bytes.saturating_add(xml.len()) <= MAX_KEPT_BYTES {
+            self.bytes += xml.len();
+            self.pending.insert(rev, xml);
+        }
+    }
+    /// Keep a named definition, within the budget.
+    pub fn keep_def(&mut self, key: String, xml: Arc<str>) {
+        if self.bytes.saturating_add(xml.len()) <= MAX_KEPT_BYTES && !self.defs.contains_key(&key) {
+            self.bytes += xml.len();
+            self.order.push(key.clone());
+            self.defs.insert(key, xml);
+        }
+    }
+    /// Keep a file kept definitions refer to, within the budget.
+    pub fn keep_file(&mut self, path: String, bytes: Vec<u8>) {
+        if self.bytes.saturating_add(bytes.len()) <= MAX_KEPT_BYTES && !self.files.contains_key(&path) {
+            self.bytes += bytes.len();
+            self.files.insert(path, Arc::new(bytes));
+        }
+    }
+    /// Whether `doc`'s page setup is still as it was read.
+    pub fn pages_as_read(&self, doc: &crate::Document) -> bool {
+        let Some(fp) = self.pages_fp else { return false };
+        let now = in_scope(doc, |sc| *sc.pages.get_or_insert_with(|| pages_fingerprint(doc))).unwrap_or_else(|| pages_fingerprint(doc));
+        fp == now
+    }
+    /// Whether `doc`'s style definitions are still as they were read.
+    pub fn styles_as_read(&self, doc: &crate::Document) -> bool {
+        self.styles_fp.is_some_and(|fp| fp == styles_fingerprint(doc))
+    }
+}
+
+/// What `doc`'s style definitions are: its styles, lists and default tab interval.
+pub fn styles_fingerprint(doc: &crate::Document) -> u64 {
+    let mut h = std::collections::hash_map::DefaultHasher::new();
+    match serde_json::to_vec(&(&doc.styles, &doc.numbering, doc.settings.default_tab)) {
+        Ok(v) => v.hash(&mut h),
+        Err(_) => u64::MAX.hash(&mut h),
+    }
+    h.finish()
+}
+
+/// What a block is, as the kept markup knows it: its paragraphs' fingerprints and
+/// [`Paragraph::rev`]s in order, and a table's [`table_fingerprint`].
+pub struct Identities {
+    paras: Vec<(u64, u64)>,
+    table: Option<u64>,
+}
+
+/// A block's [`Identities`] before and after a change that keeps what it is.
+pub type Moved = (Identities, Identities);
+
+/// The [`Identities`] of block `b`.
+pub fn identities(b: &crate::Block) -> Identities {
+    let mut paras_of = Vec::new();
+    match b {
+        crate::Block::Para(p) => paras_of.push(p),
+        crate::Block::Table(t) => {
+            for r in &t.rows {
+                for c in &r.cells {
+                    paras(&c.blocks, 1, &mut paras_of);
+                }
+            }
+        }
+    }
+    Identities { paras: paras_of.iter().map(|p| (fingerprint(p), p.rev)).collect(), table: b.as_table().map(table_fingerprint) }
+}
+
+impl KeptXml {
+    /// The markup kept for a block's paragraphs and table follows them from `before` to `after`
+    /// (what they were and are: see [`crate::Document::fill_objects`]).
+    pub fn follow(&mut self, before: &Identities, after: &Identities) {
+        for ((fp, rev), (nfp, nrev)) in before.paras.iter().zip(&after.paras) {
+            if fp == nfp && rev == nrev {
+                continue;
+            }
+            if let Some(x) = self.paras.remove(fp) {
+                self.paras.entry(*nfp).or_insert(x);
+            }
+            if let Some(w) = self.wraps.remove(fp) {
+                self.wraps.entry(*nfp).or_insert(w);
+            }
+            if let Some(w) = self.twin_wraps.remove(&(*fp, *rev)) {
+                self.twin_wraps.entry((*nfp, *nrev)).or_insert(w);
+            }
+            // Its group too: all a group's paragraphs are written as read, or none.
+            if let Some(g) = self.groups.remove(fp) {
+                self.groups.entry(*nfp).or_insert(g);
+            }
+        }
+        if let (Some(t), Some(nt)) = (before.table, after.table)
+            && t != nt
+        {
+            if let Some(x) = self.tables.remove(&t) {
+                self.tables.entry(nt).or_insert(x);
+            }
+            if let Some(w) = self.table_wraps.remove(&t) {
+                self.table_wraps.entry(nt).or_insert(w);
+            }
+        }
+    }
+}
+
+/// How deep [`story_fingerprint`] follows stories in stories (text boxes in text boxes, notes).
+const MAX_STORY_DEPTH: usize = 8;
+
+/// What story `id` (a text box's or a shape's text) is: its blocks, and the stories and notes
+/// they hold in turn.
+pub fn story_fingerprint(doc: &crate::Document, id: u32) -> u64 {
+    let mut h = std::collections::hash_map::DefaultHasher::new();
+    let mut todo = vec![(id, 0usize)];
+    let mut seen = std::collections::HashSet::new();
+    while let Some((id, depth)) = todo.pop() {
+        if !seen.insert(id) {
+            continue;
+        }
+        let blocks = doc.parts.get(&id).map(|p| &p.blocks);
+        match serde_json::to_vec(&(id, blocks)) {
+            Ok(v) => v.hash(&mut h),
+            Err(_) => u64::MAX.hash(&mut h),
+        }
+        if depth >= MAX_STORY_DEPTH {
+            continue;
+        }
+        let Some(blocks) = blocks else { continue };
+        let mut held = Vec::new();
+        paras(blocks, 0, &mut held);
+        for p in held {
+            for o in &p.objects {
+                todo.extend(o.text_boxes().into_iter().map(|s| (s, depth + 1)));
+                if let crate::para::InlineObject::NoteRef { id, .. } = o {
+                    todo.push((*id, depth + 1));
+                }
+            }
+        }
+    }
+    h.finish()
+}
+
+/// Most replies a comment's fingerprint follows (hostile input).
+const MAX_REPLIES: usize = 10_000;
+
+/// What comment `root` is with its replies (and theirs): each one's author, date, state and
+/// text, and how many they are.
+pub fn comments_fingerprint(doc: &crate::Document, root: u32) -> (u64, usize) {
+    family_fingerprint(doc, root, &scoped_replies(doc))
+}
+
+/// What a writer asks of a document's kept markup for many paragraphs, worked out once while
+/// [`while_writing`] it: each comment's replies, and its page setup's fingerprint. Without it a
+/// document of C comments took C² steps (and its page setup serialized once per note) to save.
+struct Scope {
+    /// The document's address: it can't move or change while borrowed for the writing.
+    doc: usize,
+    replies: Option<std::rc::Rc<HashMap<u32, Vec<u32>>>>,
+    pages: Option<u64>,
+}
+
+thread_local! {
+    static SCOPE: std::cell::RefCell<Option<Scope>> = const { std::cell::RefCell::new(None) };
+}
+
+#[cfg(test)]
+thread_local! {
+    /// How many times [`replies`] was built (tests).
+    static REPLY_BUILDS: std::cell::Cell<usize> = const { std::cell::Cell::new(0) };
+}
+
+/// Puts the scope before it back when dropped (when `f` returns or unwinds).
+struct Restore(Option<Scope>);
+
+impl Drop for Restore {
+    fn drop(&mut self) {
+        let prev = self.0.take();
+        SCOPE.with(|s| *s.borrow_mut() = prev);
+    }
+}
+
+/// Run `f`, which writes `doc`, with what the kept markup asks of `doc` worked out at most once.
+pub fn while_writing<R>(doc: &crate::Document, f: impl FnOnce() -> R) -> R {
+    let key = std::ptr::from_ref(doc) as usize;
+    let prev = SCOPE.with(|s| s.replace(Some(Scope { doc: key, replies: None, pages: None })));
+    let _restore = Restore(prev);
+    f()
+}
+
+/// The scope for `doc`, if it is being written: `f` of it.
+fn in_scope<R>(doc: &crate::Document, f: impl FnOnce(&mut Scope) -> R) -> Option<R> {
+    let key = std::ptr::from_ref(doc) as usize;
+    SCOPE.with(|s| s.borrow_mut().as_mut().filter(|sc| sc.doc == key).map(f))
+}
+
+/// [`replies`], built once while `doc` is being written.
+fn scoped_replies(doc: &crate::Document) -> std::rc::Rc<HashMap<u32, Vec<u32>>> {
+    in_scope(doc, |sc| sc.replies.get_or_insert_with(|| std::rc::Rc::new(replies(doc))).clone()).unwrap_or_else(|| std::rc::Rc::new(replies(doc)))
+}
+
+/// Each comment's replies, by the comment's id.
+fn replies(doc: &crate::Document) -> HashMap<u32, Vec<u32>> {
+    #[cfg(test)]
+    REPLY_BUILDS.with(|n| n.set(n.get() + 1));
+    let mut out: HashMap<u32, Vec<u32>> = HashMap::new();
+    for (id, c) in &doc.comments {
+        if let Some(p) = c.parent {
+            out.entry(p).or_default().push(*id);
+        }
+    }
+    out
+}
+
+/// [`comments_fingerprint`], with each comment's `replies`.
+fn family_fingerprint(doc: &crate::Document, root: u32, replies: &HashMap<u32, Vec<u32>>) -> (u64, usize) {
+    let mut ids = vec![root];
+    let mut seen = std::collections::HashSet::from([root]);
+    let mut i = 0;
+    while let Some(&id) = ids.get(i) {
+        for r in replies.get(&id).map(Vec::as_slice).unwrap_or(&[]) {
+            // (A reply naming itself, or a circle of them, is followed once.)
+            if ids.len() < MAX_REPLIES && seen.insert(*r) {
+                ids.push(*r);
+            }
+        }
+        i += 1;
+    }
+    let mut h = std::collections::hash_map::DefaultHasher::new();
+    for id in &ids {
+        let c = doc.comments.get(id);
+        let text = c.and_then(|c| doc.parts.get(&c.part)).map(|p| &p.blocks);
+        match serde_json::to_vec(&(c, text)) {
+            Ok(v) => v.hash(&mut h),
+            Err(_) => u64::MAX.hash(&mut h),
+        }
+    }
+    (h.finish(), ids.len())
+}
+
+/// What `doc`'s page setup is: each section's setup (where it starts, its size, margins,
+/// columns, headers and footers), the headers' and footers' content, whether odd and even pages
+/// differ, the page colour and picture, the watermark (one with no text isn't written), and the
+/// notes' numbering.
+pub fn pages_fingerprint(doc: &crate::Document) -> u64 {
+    let mut sections: Vec<&crate::SectionProps> = doc.body.iter().filter_map(|b| b.as_para()).filter_map(|p| p.section.as_deref()).collect();
+    sections.push(&doc.last_section);
+    let parts: Vec<(&u32, &crate::Part)> =
+        doc.parts.iter().filter(|(_, p)| matches!(p.kind, crate::PartKind::Header | crate::PartKind::Footer)).collect();
+    let s = &doc.settings;
+    let mut h = std::collections::hash_map::DefaultHasher::new();
+    match serde_json::to_vec(&(
+        sections,
+        parts,
+        s.even_odd_headers,
+        &s.page_color,
+        &s.page_picture,
+        s.watermark.as_ref().filter(|w| !w.text.trim().is_empty()),
+        (&s.footnote_format, &s.endnote_format, &s.footnote_pr, &s.endnote_pr),
+        (&doc.footnote_separators, &doc.endnote_separators),
+    )) {
+        Ok(v) => v.hash(&mut h),
+        Err(_) => u64::MAX.hash(&mut h),
+    }
+    h.finish()
+}
+
+/// What a run of paragraphs is in the model, together.
+pub fn run_fingerprint<'a>(paras: impl Iterator<Item = &'a Paragraph>) -> u64 {
+    let mut h = std::collections::hash_map::DefaultHasher::new();
+    for p in paras {
+        fingerprint(p).hash(&mut h);
+    }
+    h.finish()
+}
+
+/// What table `t` is in the model.
+pub fn table_fingerprint(t: &crate::Table) -> u64 {
+    let mut h = std::collections::hash_map::DefaultHasher::new();
+    match serde_json::to_vec(t) {
+        Ok(v) => v.hash(&mut h),
+        Err(_) => u64::MAX.hash(&mut h),
+    }
+    h.finish()
+}
+
+/// What paragraph `p` is in the model (its text, runs, objects and properties, not its `rev`):
+/// equal for a paragraph nobody changed.
+pub fn fingerprint(p: &Paragraph) -> u64 {
+    let mut h = std::collections::hash_map::DefaultHasher::new();
+    match serde_json::to_vec(p) {
+        Ok(v) => v.hash(&mut h),
+        // Unserializable (no paragraph is): a value no paragraph's markup is kept under.
+        Err(_) => u64::MAX.hash(&mut h),
+    }
+    h.finish()
+}
+
+/// Deepest table nesting [`crate::Document::seal_kept`] looks into.
+const MAX_SEAL_DEPTH: usize = 16;
+
+fn paras<'a>(bl: &'a crate::Blocks, depth: usize, out: &mut Vec<&'a Paragraph>) {
+    for b in bl {
+        match &**b {
+            crate::Block::Para(p) => out.push(p),
+            crate::Block::Table(t) if depth < MAX_SEAL_DEPTH => {
+                for r in &t.rows {
+                    for c in &r.cells {
+                        paras(&c.blocks, depth + 1, out);
+                    }
+                }
+            }
+            crate::Block::Table(_) => {}
+        }
+    }
+}
+
+impl crate::Document {
+    /// After reading: the markup kept for paragraphs still as the reader made them is filed
+    /// under their fingerprints (see [`Kept::seal`]); the rest is dropped.
+    pub fn seal_kept(&mut self) {
+        // The body's tables kept, by their place among its tables; its indexes, by their run of
+        // top-level paragraphs.
+        if let Some(k) = self.kept.get_mut() {
+            let pending = std::mem::take(&mut k.pending_indexes);
+            for (i, b) in self.body.iter().enumerate() {
+                let Some(p) = b.as_para() else { continue };
+                let Some((n, xml)) = pending.get(&p.rev) else { continue };
+                let run: Option<Vec<&Paragraph>> = self.body.get(i..i + n).map(|r| r.iter().filter_map(|b| b.as_para()).collect());
+                if let Some(run) = run.filter(|r| r.len() == *n) {
+                    k.indexes.entry(fingerprint(p)).or_insert((*n, run_fingerprint(run.into_iter()), xml.clone()));
+                }
+            }
+            let pending = std::mem::take(&mut k.pending_tables);
+            let tables: Vec<&crate::Table> = self.body.iter().filter_map(|b| b.as_table()).collect();
+            for (n, xml) in pending {
+                if let Some(t) = tables.get(n) {
+                    k.tables.entry(table_fingerprint(t)).or_insert(xml);
+                }
+            }
+            // Wrappers, for tables that are the only one of their kind (a copy can't open a
+            // section twice).
+            let pending = std::mem::take(&mut k.pending_table_wraps);
+            if !pending.is_empty() {
+                let fps: Vec<u64> = tables.iter().map(|t| table_fingerprint(t)).collect();
+                for (n, w) in pending {
+                    if let Some(fp) = fps.get(n)
+                        && fps.iter().filter(|f| *f == fp).count() == 1
+                    {
+                        k.table_wraps.insert(*fp, w);
+                    }
+                }
+            }
+        }
+        // The text boxes' and shapes' stories as read, for paragraphs showing them.
+        if self.kept.get().is_some_and(|k| !k.pending.is_empty()) {
+            let fps: HashMap<u32, u64> = self
+                .parts
+                .iter()
+                .filter(|(_, p)| matches!(p.kind, crate::PartKind::TextBox | crate::PartKind::Footnote | crate::PartKind::Endnote))
+                .map(|(id, _)| (*id, story_fingerprint(self, *id)))
+                .collect();
+            if let Some(k) = self.kept.get_mut() {
+                k.story_fps = fps;
+            }
+        }
+        // The comments as read, for paragraphs holding them.
+        if self.kept.get().is_some_and(|k| !k.pending.is_empty()) && !self.comments.is_empty() {
+            let replies = replies(self);
+            let fps: HashMap<u32, u64> =
+                self.comments.iter().filter(|(_, c)| c.parent.is_none()).map(|(id, _)| (*id, family_fingerprint(self, *id, &replies).0)).collect();
+            if let Some(k) = self.kept.get_mut() {
+                k.comment_fps = fps;
+            }
+        }
+        if self.kept.get().is_none_or(|k| k.pending.is_empty() && k.pending_wraps.is_empty()) {
+            return;
+        }
+        let mut all = Vec::new();
+        paras(&self.body, 0, &mut all);
+        for p in self.parts.values() {
+            paras(&p.blocks, 0, &mut all);
+        }
+        let mut kept = std::mem::take(&mut self.kept);
+        // Groups, by fingerprint; a group with a paragraph that has twins (which a writer can't
+        // tell apart) isn't kept.
+        if let Some(k) = kept.get_mut()
+            && !k.pending_groups.is_empty()
+        {
+            let mut count: HashMap<u64, usize> = HashMap::new();
+            let fps: Vec<(u64, u64)> = all.iter().map(|p| (p.rev, fingerprint(p))).collect();
+            for (_, fp) in &fps {
+                *count.entry(*fp).or_default() += 1;
+            }
+            let pending = std::mem::take(&mut k.pending_groups);
+            let mut broken = std::collections::HashSet::new();
+            let mut members: Vec<(u64, u64, (u32, u32))> = Vec::new();
+            for (rev, fp) in &fps {
+                if let Some(g) = pending.get(rev) {
+                    if count.get(fp) != Some(&1) {
+                        broken.insert(g.0);
+                    }
+                    members.push((*rev, *fp, *g));
+                }
+            }
+            for (rev, fp, g) in members {
+                if broken.contains(&g.0) {
+                    k.pending.remove(&rev);
+                } else {
+                    k.groups.insert(fp, g);
+                }
+            }
+        }
+        // Wrappers, for paragraphs that are the only one of their kind.
+        if let Some(k) = kept.get_mut()
+            && !k.pending_wraps.is_empty()
+        {
+            let mut count: HashMap<u64, usize> = HashMap::new();
+            let fps: Vec<(u64, u64)> = all.iter().map(|p| (p.rev, fingerprint(p))).collect();
+            for (_, fp) in &fps {
+                *count.entry(*fp).or_default() += 1;
+            }
+            for (rev, fp) in fps {
+                if let Some(w) = k.pending_wraps.remove(&rev) {
+                    if count.get(&fp) == Some(&1) {
+                        k.wraps.insert(fp, w);
+                    } else {
+                        k.twin_wraps.insert((fp, rev), w);
+                    }
+                }
+            }
+            k.pending_wraps.clear();
+        }
+        kept.seal(&mut all.into_iter());
+        self.kept = kept;
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn a_paragraph_keeps_its_markup_until_it_changes() {
+        let mut p = Paragraph::with_text("Hello", Default::default());
+        let mut kept = Kept::new(KeptXml { format: "odt".into(), ..Default::default() });
+        if let Some(k) = kept.get_mut() {
+            k.keep_pending(p.rev, Arc::from("<text:p>Hello</text:p>"));
+        }
+        kept.seal(&mut std::iter::once(&p));
+        let k = kept.of("odt").unwrap();
+        assert_eq!(k.para(&p).map(|x| &**x), Some("<text:p>Hello</text:p>"));
+        assert!(k.pending.is_empty());
+        // The same paragraph again (a copy, its rev moved on): still what was read.
+        p.touch();
+        assert!(k.para(&p).is_some());
+        // Changed: written from the model.
+        p.text.push('!');
+        assert!(k.para(&p).is_none());
+        assert!(kept.of("docx").is_none());
+    }
+
+    #[test]
+    fn a_paragraph_changed_while_reading_keeps_nothing() {
+        let mut p = Paragraph::with_text("Hi", Default::default());
+        let mut kept = Kept::new(KeptXml { format: "odt".into(), ..Default::default() });
+        if let Some(k) = kept.get_mut() {
+            k.keep_pending(p.rev, Arc::from("<text:p>Hi</text:p>"));
+        }
+        p.touch();
+        kept.seal(&mut std::iter::once(&p));
+        assert!(kept.of("odt").unwrap().paras.is_empty());
+    }
+
+    #[test]
+    fn kept_markup_is_bounded() {
+        let mut k = KeptXml::default();
+        let big: Arc<str> = Arc::from("x".repeat(MAX_KEPT_BYTES / 2 + 1));
+        k.keep_pending(1, big.clone());
+        k.keep_pending(2, big);
+        assert_eq!(k.pending.len(), 1);
+    }
+
+    /// Filling in what objects show (`Document::fill_objects`, a chart drawn from its data)
+    /// doesn't change the document: a paragraph's group and a text box's story stay as read.
+    #[test]
+    fn filling_objects_keeps_groups_and_stories_as_read() {
+        use crate::para::InlineObject;
+        let field =
+            || InlineObject::Field { instr: "X".into(), result: String::new(), locked: false, dirty: false, code: false, form: String::new() };
+        let mut doc = crate::Document::from_text("Body");
+        let mut boxed = Paragraph::with_text("In the box", Default::default());
+        boxed.insert_object(0, field(), &Default::default()).unwrap();
+        let id = doc.add_part(crate::PartKind::TextBox, vec![crate::para_block(boxed)]);
+        doc.para_mut(crate::StoryRef::Body, &crate::Path::top(0)).unwrap().insert_object(0, field(), &Default::default()).unwrap();
+        let fp = fingerprint(doc.body[0].as_para().unwrap());
+        let mut k = KeptXml { format: "odt".into(), ..Default::default() };
+        k.paras.insert(fp, Arc::from("<text:p>Body</text:p>"));
+        k.groups.insert(fp, (1, 1));
+        k.story_fps.insert(id, story_fingerprint(&doc, id));
+        doc.kept = Kept::new(k);
+        doc.fill_objects(&|o| matches!(o, InlineObject::Field { result, .. } if result.is_empty()), &mut |o| {
+            if let InlineObject::Field { result, .. } = o {
+                *result = "filled".into();
+            }
+        });
+        let p = doc.body[0].as_para().unwrap();
+        assert_ne!(fingerprint(p), fp, "the paragraph's fingerprint moved");
+        let k = doc.kept.get().unwrap();
+        assert!(k.para(p).is_some());
+        assert_eq!(k.group(p), Some((1, 1)));
+        assert_eq!(k.story_fps.get(&id), Some(&story_fingerprint(&doc, id)));
+    }
+
+    /// Writing a document checks each kept paragraph's comments against what was read: the
+    /// comments' replies are looked up once for the whole document, not once per comment.
+    #[test]
+    fn writing_many_comments_looks_their_replies_up_once() {
+        use crate::para::InlineObject;
+        let n = 50u32;
+        let text = (0..n).map(|i| format!("Para {i}")).collect::<Vec<_>>().join("\n");
+        let mut doc = crate::Document::from_text(&text);
+        for i in 0..n {
+            let part = doc.add_part(crate::PartKind::Comment, vec![crate::para_block(Paragraph::with_text("Note", Default::default()))]);
+            doc.comments.insert(i, crate::Comment { parent: None, part, ..Default::default() });
+            doc.para_mut(crate::StoryRef::Body, &crate::Path::top(i as usize))
+                .unwrap()
+                .insert_object(0, InlineObject::CommentStart { id: i }, &Default::default())
+                .unwrap();
+        }
+        let mut k = KeptXml { format: "odt".into(), ..Default::default() };
+        for i in 0..n {
+            k.comment_fps.insert(i, comments_fingerprint(&doc, i).0);
+        }
+        let xml = "<text:p><office:annotation>x</office:annotation></text:p>";
+        let before = REPLY_BUILDS.with(std::cell::Cell::get);
+        let all = while_writing(&doc, || doc.body.iter().filter_map(|b| b.as_para()).all(|p| k.comments_as_read(&doc, p, xml)));
+        assert!(all, "every paragraph's comment is as read");
+        assert_eq!(REPLY_BUILDS.with(std::cell::Cell::get) - before, 1);
+        // Outside a writing, nothing is kept between calls.
+        assert!(doc.body.first().and_then(|b| b.as_para()).is_some_and(|p| k.comments_as_read(&doc, p, xml)));
+        assert_eq!(REPLY_BUILDS.with(std::cell::Cell::get) - before, 2);
+    }
+}

crates/doc/src/legacy_form.rs

@@ -120,14 +120,57 @@ fn attr_in(tag: &str) -> Option<String> {
     Some(unescape(rest.get(..end)?))
 }
 
-fn unescape(s: &str) -> String {
-    s.replace("&lt;", "<").replace("&gt;", ">").replace("&quot;", "\"").replace("&apos;", "'").replace("&amp;", "&")
+fn unescape(mut s: &str) -> String {
+    let mut out = String::with_capacity(s.len());
+    while let Some((text, rest)) = s.split_once('&') {
+        out.push_str(text);
+        if let Some(end) = rest.find(['&', ';'])
+            && rest.as_bytes()[end] == b';'
+        {
+            let entity = &rest[..end];
+            let decoded = match entity {
+                "lt" => Some('<'),
+                "gt" => Some('>'),
+                "quot" => Some('"'),
+                "apos" => Some('\''),
+                "amp" => Some('&'),
+                _ => entity.strip_prefix('#').and_then(|n| {
+                    let value = if let Some(hex) = n.strip_prefix('x') {
+                        (!hex.is_empty() && hex.bytes().all(|b| b.is_ascii_hexdigit())).then(|| u32::from_str_radix(hex, 16).ok()).flatten()
+                    } else {
+                        (!n.is_empty() && n.bytes().all(|b| b.is_ascii_digit())).then(|| n.parse::<u32>().ok()).flatten()
+                    }?;
+                    char::from_u32(value)
+                        .filter(|c| matches!(*c, '\t' | '\n' | '\r' | '\u{20}'..='\u{d7ff}' | '\u{e000}'..='\u{fffd}' | '\u{10000}'..='\u{10ffff}'))
+                }),
+            };
+            if let Some(c) = decoded {
+                out.push(c);
+                s = &rest[end + 1..];
+                continue;
+            }
+        }
+        out.push('&');
+        s = rest;
+    }
+    out.push_str(s);
+    out
 }
 
 #[cfg(test)]
 mod tests {
     use super::*;
 
+    #[test]
+    fn drop_down_entities_decode_once_including_xml_whitespace() {
+        let xml = r#"<w:ffData><w:name w:val="N&#x61;me"/><w:ddList><w:listEntry w:val="A&#10;B&#9;C&#13;D&#x1F600;&amp;#10;&lt;&gt;&quot;&apos;"/></w:ddList></w:ffData>"#;
+        let f = LegacyForm::parse(xml).unwrap();
+        assert_eq!(f.name(), "Name");
+        assert_eq!(f.shown_text(), Some("A\nB\tC\rD😀&#10;<>\"'"));
+        let malformed = r#"<w:ffData><w:ddList><w:listEntry w:val="&#0;&#xD800;&#x110000;&#-1;&#xZZ;&unknown;&amp"/></w:ddList></w:ffData>"#;
+        assert_eq!(LegacyForm::parse(malformed).unwrap().shown_text(), Some("&#0;&#xD800;&#x110000;&#-1;&#xZZ;&unknown;&amp"));
+    }
+
     #[test]
     fn check_boxes_text_fields_and_drop_downs() {
         let cb = |inner: &str| format!(r#"<w:ffData><w:name w:val="Box1"/><w:enabled/><w:checkBox>{inner}</w:checkBox></w:ffData>"#);

crates/doc/src/lib.rs

@@ -2,9 +2,9 @@
 //!
 //! A [`Document`] holds the main story (`body`) and secondary stories (`parts`: headers,
 //! footers, notes, comments, text boxes). A story is a list of [`Block`]s: paragraphs and tables;
-//! table cells hold block lists of their own. Blocks are `Arc`-shared, so cloning a document is
-//! cheap (undo snapshots, handing a frozen copy to the renderer) and edits copy only what they
-//! touch (`Arc::make_mut`).
+//! table cells hold block lists of their own. Blocks, the style sheet and the list definitions
+//! are `Arc`-shared, so cloning a document is cheap (undo snapshots, handing a frozen copy to the
+//! renderer) and edits copy only what they touch (`Arc::make_mut`).
 //!
 //! Positions ([`Pos`]) name a story, a [`Path`] to a paragraph and a byte offset in it. Paths
 //! compare in document order.
@@ -28,6 +28,7 @@ pub mod fill;
 pub mod freeform;
 pub mod graphic;
 pub mod ime_text;
+pub mod kept;
 pub mod legacy_form;
 pub mod line_dash;
 pub mod link;
@@ -39,6 +40,7 @@ pub mod math_linear;
 pub mod math_symbols;
 pub mod numbering;
 pub mod para;
+pub mod picfx;
 pub mod props;
 pub mod resolve;
 pub mod section;
@@ -47,6 +49,7 @@ pub mod styles;
 pub mod symbol_fonts;
 pub mod table;
 pub mod wordart;
+pub mod work;
 pub mod wrap;
 pub mod zip_guard;
 
@@ -84,11 +87,11 @@ pub use error::{DocError, Result};
 pub use parts::{MAX_LINKED_BOXES, MAX_TEXT_BOX_INSET, NoteSeparators, Part, PartKind, PermRange, TEXT_BOX_INSET, TextBody, free_id};
 pub use position::{Path, Pos, StoryRef};
 pub use settings::{
-    COMPAT_MODE_CURRENT, CoreProps, CustomProp, DEFAULT_GRID, DEFAULT_HYPHENATION_ZONE, LEGACY_COMPAT_MODE, PagePicture, Settings, Source,
+    COMPAT_MODE_CURRENT, CoreProps, CustomProp, DEFAULT_GRID, DEFAULT_HYPHENATION_ZONE, LEGACY_COMPAT_MODE, OdfLayout, PagePicture, Settings, Source,
     THEME_COLORS, Watermark, Zoom, ZoomFit,
 };
-pub use text_walk::count_words;
 pub(crate) use text_walk::{PARA_WALK_DEPTH, walk_between};
+pub use text_walk::{count_chars, count_chars_no_spaces, count_words};
 
 /// A document.
 #[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
@@ -98,8 +101,12 @@ pub struct Document {
     /// Properties of the last section (OOXML's body-level `w:sectPr`).
     pub last_section: SectionProps,
     pub parts: BTreeMap<u32, Part>,
-    pub styles: StyleSheet,
-    pub numbering: Numbering,
+    /// Shared, like the blocks: an undo step copies it only when a command changes it (a full
+    /// copy is kilobytes a style, so typing into a document with hundreds of styles cost megabytes
+    /// a step).
+    pub styles: Arc<StyleSheet>,
+    /// Shared like [`Document::styles`].
+    pub numbering: Arc<Numbering>,
     pub comments: BTreeMap<u32, Comment>,
     pub revisions: Vec<Revision>,
     pub settings: Settings,
@@ -127,6 +134,10 @@ pub struct Document {
     /// by `==`).
     #[serde(skip)]
     media_index: MediaIndex,
+    /// Markup kept from the file it was opened from, for writing back what isn't edited (see
+    /// [`kept`]); never saved, ignored by `==`.
+    #[serde(skip)]
+    pub kept: kept::Kept,
 }
 
 #[cfg(test)]

crates/doc/src/link.rs

@@ -1,5 +1,6 @@
-//! Which hyperlink targets may leave the document: followed by the OS, or written into an
-//! exported HTML/Markdown file where a browser would run them.
+//! Hyperlink targets: where a `HYPERLINK` field code points, and which targets may leave the
+//! document: followed by the OS, or written into an exported HTML/Markdown file where a browser
+//! would run them.
 
 /// Whether Ctrl/⌘+click may hand a document's hyperlink to the system: web and email links only,
 /// so a document can't launch `file:` paths, programs or custom-scheme handlers. Exact: a leading
@@ -18,6 +19,12 @@ pub fn is_web_or_mail_link(url: &str) -> bool {
 /// names another machine — following it on Windows leaks the user's NTLM credentials — so it is
 /// refused too.
 pub fn is_safe_href(url: &str) -> bool {
+    // Exporters drop other control characters and noncharacters when they write the link
+    // (`file\u{85}:` comes out `file:`), so a link holding one is refused, not judged as it
+    // stands and written as something else.
+    if url.chars().any(|c| (c.is_control() && !c.is_ascii_control()) || matches!(c, '\u{FFFE}' | '\u{FFFF}')) {
+        return false;
+    }
     let cleaned: String = url.chars().filter(|c| !c.is_ascii_control() && *c != ' ').collect();
     // `//host` (protocol-relative) and `\\host` (UNC), with `\` read as `/` the way Windows does.
     if cleaned.replace('\\', "/").starts_with("//") {
@@ -39,10 +46,122 @@ fn scheme(url: &str) -> Option<String> {
     (first.is_ascii_alphabetic() && chars.all(|c| c.is_ascii_alphanumeric() || matches!(c, '+' | '-' | '.'))).then(|| s.to_ascii_lowercase())
 }
 
+/// Where a `HYPERLINK` field code points, as a `w:hyperlink` reads: `url`, `url#bookmark` (`\l`)
+/// or `#bookmark`. None for any other field, or one with no target.
+pub fn hyperlink_target(instr: &str) -> Option<String> {
+    // Only a `HYPERLINK` code is split up (a field's code may run to megabytes).
+    if !instr.trim_start().get(..9).is_some_and(|k| k.eq_ignore_ascii_case("HYPERLINK")) {
+        return None;
+    }
+    let args = field_args(instr);
+    let mut args = args.iter().peekable();
+    if !args.next().is_some_and(|(k, quoted)| !quoted && k.eq_ignore_ascii_case("HYPERLINK")) {
+        return None;
+    }
+    let is_switch = |(a, quoted): &&(String, bool)| !quoted && a.starts_with('\\');
+    let (mut url, mut anchor) = (None, None);
+    while let Some(arg @ (a, _)) = args.next() {
+        if !is_switch(&arg) {
+            url = url.or(Some(a.as_str()));
+            continue;
+        }
+        // `\l` names a bookmark, `\o` a tooltip, `\t` a frame, `\*` `\#` `\@` a format; `\m` and `\n` take none.
+        let takes = matches!(a.to_ascii_lowercase().as_str(), "\\l" | "\\o" | "\\t" | "\\*" | "\\#" | "\\@");
+        let value = takes.then(|| args.next_if(|v| !is_switch(v))).flatten();
+        if a.eq_ignore_ascii_case("\\l")
+            && let Some((v, _)) = value
+        {
+            anchor = Some(v.as_str());
+        }
+    }
+    let url = url.map(str::trim).filter(|u| !u.is_empty());
+    let anchor = anchor.map(|a| a.trim().trim_start_matches('#')).filter(|a| !a.is_empty());
+    match (url, anchor) {
+        // `\l` replaces a fragment the address already has, as in Word.
+        (Some(u), Some(a)) => Some(format!("{}#{a}", u.split_once('#').map_or(u, |(base, _)| base))),
+        (Some(u), None) => Some(u.to_string()),
+        (None, Some(a)) => Some(format!("#{a}")),
+        (None, None) => None,
+    }
+}
+
+/// A field code's words: the field's name, its switches and its arguments, each with whether
+/// it was quoted. In quotes a backslash escapes the next character, as in Word.
+fn field_args(instr: &str) -> Vec<(String, bool)> {
+    let mut out = Vec::new();
+    let mut chars = instr.chars().peekable();
+    while let Some(c) = chars.next() {
+        if c.is_whitespace() {
+            continue;
+        }
+        let mut word = String::new();
+        if c == '"' {
+            while let Some(c) = chars.next() {
+                match c {
+                    '"' => break,
+                    '\\' => word.extend(chars.next()),
+                    _ => word.push(c),
+                }
+            }
+            out.push((word, true));
+        } else {
+            word.push(c);
+            while let Some(c) = chars.next_if(|c| !c.is_whitespace() && *c != '"') {
+                word.push(c);
+            }
+            out.push((word, false));
+        }
+    }
+    out
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
 
+    /// Every reader's `HYPERLINK` cases: switches with and without values, quotes, escapes, spelling.
+    #[test]
+    fn hyperlink_field_codes_name_their_targets() {
+        for (code, want) in [
+            (r#"HYPERLINK "http://example.com" \l "heading""#, Some("http://example.com#heading")),
+            (r#"HYPERLINK \l "heading""#, Some("#heading")),
+            (r#"HYPERLINK  \l  "_Toc1" "#, Some("#_Toc1")),
+            (r#"Hyperlink "http://example.com/a path""#, Some("http://example.com/a path")),
+            ("HYPERLINK https://example.org/b", Some("https://example.org/b")),
+            ("HYPERLINK\t\"http://example.com\"", Some("http://example.com")),
+            (r#"HYPERLINK"http://example.com""#, Some("http://example.com")),
+            (r#"HYPERLINK "http://example.com/path\latest""#, Some("http://example.com/pathlatest")),
+            (r#"HYPERLINK "x \* Upper""#, Some("x * Upper")),
+            (r#"HYPERLINK "C:\\Docs\\My file.docx""#, Some(r"C:\Docs\My file.docx")),
+            (r#"HYPERLINK "say \"hi\"""#, Some(r#"say "hi""#)),
+            (r#"HYPERLINK " http://example.com ""#, Some("http://example.com")),
+            // Switches that take a value, before or after the address, and those that take none.
+            (r#"HYPERLINK "http://example.com" \o "Help text""#, Some("http://example.com")),
+            (r#"hyperlink \o "A tip" \t "_blank" "https://example.org/""#, Some("https://example.org/")),
+            (r#"HYPERLINK \l "bm" \o "tip""#, Some("#bm")),
+            (r#"HYPERLINK \l "Sec" \m"#, Some("#Sec")),
+            (r#"HYPERLINK \n "http://example.com""#, Some("http://example.com")),
+            (r#"HYPERLINK \l "x" \* MERGEFORMAT"#, Some("#x")),
+            (r#"HYPERLINK "http://example.com" \l \o "tip""#, Some("http://example.com")),
+            (r#"HYPERLINK \l "bm" \l"#, Some("#bm")),
+            // A bookmark's own `#`, an address that has a fragment, and an empty address.
+            (r##"HYPERLINK "http://example.com" \l "#top""##, Some("http://example.com#top")),
+            (r#"HYPERLINK "http://example.com/#a" \l "b""#, Some("http://example.com/#b")),
+            (r#"HYPERLINK "" \l "x""#, Some("#x")),
+            // No target, or not a `HYPERLINK` field.
+            ("HYPERLINK", None),
+            (r#"HYPERLINK "" \o "tip""#, None),
+            (r#"HYPERLINKS "https://example.org/""#, None),
+            (r#"HYPERLINKX "http://example.com""#, None),
+            (r#"HYPERLINK\l "x""#, None),
+            ("PAGEREF _Toc1 \\h", None),
+            ("ab日本語", None),
+            ("", None),
+        ] {
+            assert_eq!(hyperlink_target(code).as_deref(), want, "{code:?}");
+        }
+    }
+
     #[test]
     fn only_web_and_mail_links_are_followable() {
         for ok in ["https://example.com/a?b#c", "http://example.com", "HTTPS://EXAMPLE.COM", "mailto:someone@example.com"] {
@@ -87,6 +206,10 @@ mod tests {
             "\\\\attacker.example\\share\\x",
             "/\\attacker.example\\x",
             " //attacker.example/x",
+            // Characters the exporters drop on writing don't hide a scheme.
+            "file\u{85}:///etc/passwd",
+            "java\u{9F}script:alert(1)",
+            "/\u{FFFE}/attacker.example/x",
             "",
         ] {
             assert!(!is_safe_href(bad), "{bad:?}");

crates/doc/src/locale_number.rs

@@ -22,10 +22,29 @@ pub fn decimal_separator(lang: Option<&str>) -> char {
 
 const SPACES: &[char] = &[' ', '\u{a0}', '\u{202f}', '\u{2009}'];
 const CURRENCY: &[char] = &['$', '€', '£', '¥', '¢', '₽', '₴', '₺', '₹', '₩', '₪', '₫', '฿', '₱', '₸', '₼', '₾'];
+/// Currency signs spelled with more than one character: the Panamanian balboa (`B/.`).
+const CURRENCY_SYMBOLS: &[&str] = &["B/."];
 
-/// Strip a currency symbol or a short currency word (`zł`, `kr`, `PLN`) from either end.
+/// Trim spaces and single-character currency signs from either end.
+fn trim_currency(s: &str) -> &str {
+    s.trim_matches(|c: char| c.is_whitespace() || CURRENCY.contains(&c))
+}
+
+/// Strip a currency symbol or a short currency word (`zł`, `kr`, `PLN`, `B/.`) from either end.
+/// As Word does, a multi-character sign is a word of its own (`B/. 1,250.50`, `1,250.50 B/.`);
+/// `B/.1,250.50` and `B/ 1,250.50` are not numbers.
 fn strip_currency(s: &str) -> &str {
-    let mut s = s.trim_matches(|c: char| c.is_whitespace() || CURRENCY.contains(&c));
+    let mut s = trim_currency(s);
+    if let Some(rest) =
+        CURRENCY_SYMBOLS.iter().find_map(|sym| s.strip_prefix(sym).filter(|rest| rest.chars().next().is_some_and(char::is_whitespace)))
+    {
+        s = trim_currency(rest);
+    }
+    if let Some(rest) =
+        CURRENCY_SYMBOLS.iter().find_map(|sym| s.strip_suffix(sym).filter(|rest| rest.chars().next_back().is_some_and(char::is_whitespace)))
+    {
+        s = trim_currency(rest);
+    }
     if let Some((head, word)) = s.rsplit_once(SPACES)
         && (1..=3).contains(&word.chars().count())
         && word.chars().all(|c| c.is_alphabetic() || c == '.')
@@ -137,6 +156,7 @@ mod tests {
         assert_eq!(parse_number("1.5", pl), Some(1.5));
         assert_eq!(parse_number("-2,25", pl), Some(-2.25));
         assert_eq!(parse_number("PLN 12", pl), Some(12.0));
+        assert_eq!(parse_number("-B/. 1 250,50", pl), Some(-1250.5));
         assert_eq!(parse_number("abc", pl), None);
         assert_eq!(parse_number("1,2,3", pl), None);
         // From review: ordinals, a leading zero, percentages and an en dash minus.
@@ -174,11 +194,40 @@ mod tests {
         assert_eq!(format_number(5.75, Some("es-ES")), "5,75");
     }
 
+    #[test]
+    fn balboa_numbers() {
+        let en = Some("en-US");
+        // Word 16.0 reads `B/.` as a word of its own, before or after the number.
+        assert_eq!(parse_number("B/. 1,234.50", en), Some(1234.5));
+        assert_eq!(parse_number("1,234.50 B/.", en), Some(1234.5));
+        assert_eq!(parse_number("B/.\u{a0}1,234.50", en), Some(1234.5), "a no-break space after the sign");
+        assert_eq!(parse_number("B/. 1,234.50 B/.", en), Some(1234.5), "the sign at both ends");
+        assert_eq!(parse_number("-B/. 1,234.50", en), Some(-1234.5), "a minus before the sign");
+        assert_eq!(parse_number("B/. 1,234.50", Some("es-PA")), Some(1234.5), "the balboa's own region");
+        // The sign has no space of its own in Word: touching the number is text.
+        assert_eq!(parse_number("B/.1,234.50", en), None);
+        assert_eq!(parse_number("1,234.50B/.", en), None);
+        assert_eq!(parse_number("B/ 1,234.50", en), None, "the balboa is B/. in Word");
+        assert_eq!(parse_number("B/.", en), None);
+        // The slash and the full stop make `B/.` ambiguous with a fraction: the sign is matched at
+        // the ends of the text only, so a fraction is never read as a balboa amount and a sign in
+        // the middle is not a sign.
+        assert_eq!(parse_number("1/.5", en), None, "a fraction is not a price");
+        assert_eq!(parse_number("1/.5", Some("es-PA")), None);
+        assert_eq!(parse_number("1/.5 B/.", en), None, "the sign at the end does not rescue a fraction");
+        assert_eq!(parse_number("B/. 1/.5", en), None);
+        assert_eq!(parse_number("2 B/. 1,234.50", en), None, "the sign sits before the number");
+        assert_eq!(parse_number("1,234.50 B/. 1,234.50", en), None, "the sign sits between two numbers");
+        assert_eq!(parse_number("1,234.50/2", en), None, "a division is not a balboa amount");
+        assert_eq!(parse_number("B/. 1,234.50,50", en), None, "a second decimal separator");
+    }
+
     #[test]
     fn english_numbers() {
         let en = Some("en-US");
         assert_eq!(parse_number("1,250.50", en), Some(1250.5));
         assert_eq!(parse_number("$1,250", en), Some(1250.0));
+        assert_eq!(parse_number("PAB 1,250.50", en), Some(1250.5), "the ISO code");
         assert_eq!(parse_number("1.5", en), Some(1.5));
         assert_eq!(parse_number("12", None), Some(12.0));
         assert_eq!(parse_number(".5", None), Some(0.5));

crates/doc/src/math.rs

@@ -16,6 +16,12 @@ pub type Arg = Vec<MNode>;
 /// Deepest structure nesting kept (parsers stop here; layout draws nothing deeper).
 pub const MAX_DEPTH: usize = 64;
 
+/// Most matrix columns kept (writers pad every row to the widest, so this bounds a row's cost).
+pub const MAX_MATRIX_COLS: usize = 64;
+
+/// Most matrix rows kept.
+pub const MAX_MATRIX_ROWS: usize = 1000;
+
 /// An equation's structure, plus what's needed to write it back.
 #[derive(Clone, Debug, PartialEq, Default, Serialize, Deserialize)]
 pub struct Math {

crates/doc/src/math_edit.rs

@@ -769,6 +769,30 @@ pub fn autocorrect(root: &mut Arg, pos: &MathPos) -> Option<MathPos> {
     Some(MathPos { path: pos.path, off: pos.off - name.chars().count() })
 }
 
+/// The character just before the caret is an n-ary operator (∫, ∑…) in plain math text, as
+/// AutoCorrect leaves `\int` or `\sum`.
+pub fn nary_before(root: &Arg, pos: &MathPos) -> bool {
+    let pos = clamp(root, pos);
+    let Some(a) = arg_at(root, &pos.path) else { return false };
+    let (i, k) = match locate(a, pos.off) {
+        Loc::Inside(i, k) => (i, k),
+        Loc::Before(i) => match i.checked_sub(1).and_then(|i| a.get(i).map(|n| (i, node_units(n)))) {
+            Some(at) => at,
+            None => return false,
+        },
+    };
+    let Some(MNode::Run(r)) = a.get(i) else { return false };
+    !r.nor && !r.lit && k.checked_sub(1).and_then(|k| r.text.chars().nth(k)).is_some_and(crate::math::is_nary_char)
+}
+
+/// The node just before the caret is a built n-ary operator.
+pub fn nary_node_before(root: &Arg, pos: &MathPos) -> bool {
+    let pos = clamp(root, pos);
+    let Some(a) = arg_at(root, &pos.path) else { return false };
+    let Loc::Before(i) = locate(a, pos.off) else { return false };
+    i.checked_sub(1).and_then(|i| a.get(i)).is_some_and(|n| matches!(n, MNode::Nary { .. }))
+}
+
 /// Characters that make typed linear text build up into structures.
 fn builds(s: &str) -> bool {
     s.chars().any(|c| {
@@ -872,6 +896,28 @@ pub fn build_up(root: &mut Arg, pos: &MathPos, latex: bool) -> Option<MathPos> {
     Some(clamp(root, &out))
 }
 
+/// Typing that just built an n-ary operator without its operand (`∫_0^1` then Space) goes on in
+/// the empty operand, as after inserting one from the gallery, so what is typed next is
+/// integrated or summed (#4917). An operator whose operand ends with another (`∫∫`) leads to the
+/// innermost one. Other carets, and moves or Enter that build up, stay put.
+pub fn into_empty_operand(root: &Arg, pos: &MathPos) -> MathPos {
+    let mut path = pos.path.clone();
+    let mut off = pos.off;
+    while path.len() < MAX_DEPTH
+        && let Some(a) = arg_at(root, &path)
+        && let Loc::Before(i) = locate(a, off)
+        && let Some(k) = i.checked_sub(1)
+        && let Some(MNode::Nary { e, .. }) = a.get(k)
+    {
+        path.push((k, 2));
+        if e.is_empty() {
+            return MathPos { path, off: 0 };
+        }
+        off = units(e);
+    }
+    pos.clone()
+}
+
 /// A top-level `#` numbers the equation: wrap it as a one-row equation array.
 pub fn number_equation(root: &mut Arg) -> bool {
     if has_number_mark(root) && !matches!(root.as_slice(), [MNode::EqArr { .. }]) {

crates/doc/src/numbering.rs

@@ -228,7 +228,7 @@ impl Numbering {
     /// Whether a paragraph in list `num` at `level` gets a label ([`Counters::next_label`] gives
     /// one), without counting it.
     pub fn shows_label(&self, num: u32, level: u8) -> bool {
-        self.abstract_of(num).is_some() && self.level(num, level.min(8)).is_some_and(|d| d.format != NumFormat::None)
+        self.abstract_of(num).is_some() && self.level(num, level.min(8)).is_some()
     }
     pub fn level(&self, num: u32, level: u8) -> Option<&Level> {
         if let Some((_, l)) = self.num(num)?.level_overrides.iter().find(|(i, _)| *i == level) {
@@ -294,6 +294,14 @@ pub struct Counters {
     /// Per abstract list: whether each level has counted yet (a value of 0 can be a real count when a
     /// level starts at 0).
     flags: HashMap<u32, [LevelFlags; 9]>,
+    /// The last explicitly rendered label and template at each abstract-list level.
+    labels: HashMap<(u32, u8), CountedLabel>,
+}
+
+#[derive(Debug, Clone)]
+struct CountedLabel {
+    label: Option<String>,
+    levels: Vec<usize>,
 }
 
 /// Whether a level has counted yet, and the value its next item takes if a start override says so.
@@ -317,11 +325,9 @@ impl Counters {
         let lv = (level as usize).min(8);
         // This list's own definition of a level wins over its abstract list's.
         let level_def = |k: usize| numbering.level(num, k as u8);
+        // A "none" level (Word's numFmt none) still counts and restarts the levels below it; only
+        // its number reads as nothing, so its literal text ("RESPONSE:") shows (#4985).
         let def = level_def(lv)?.clone();
-        // A "none" level shows no label at all (Word's numFmt none) — not even its text.
-        if def.format == NumFormat::None {
-            return None;
-        }
         let st = self.state.entry(abs.id).or_insert([0; 9]);
         let fl = self.flags.entry(abs.id).or_default();
         // Only this level's override, not the num's others: the counters are shared by every
@@ -366,42 +372,108 @@ impl Counters {
             if restarts {
                 *deeper = 0;
                 *f = LevelFlags::default();
+                self.labels.remove(&(abs.id, d as u8));
             }
         }
         let vals = *st;
         let started = *fl;
-        if def.format == NumFormat::Bullet {
-            return Some((def.text.clone(), def));
-        }
-        let mut label = String::new();
-        // Word's level text is short; .docx and the lists command already cap it at 64.
-        let mut chars = def.text.chars().take(MAX_LEVEL_TEXT).peekable();
-        while let Some(c) = chars.next() {
-            if c == '%'
-                && let Some(d) = chars.peek().and_then(|d| d.to_digit(10))
-            {
-                chars.next();
-                let k = (d as usize).saturating_sub(1).min(8);
-                let val = vals.get(k).copied().unwrap_or(0);
-                let counted = started.get(k).is_some_and(|f| f.started);
-                let val = if counted { val } else { level_def(k).map(|l| l.start).unwrap_or(1) };
-                // Legal numbering (w:isLgl) shows every level in Arabic digits, whatever its own
-                // format (二 of a Chinese outline becomes 2 in "2.1").
-                let fmt = match level_def(k).map(|l| l.format).unwrap_or(NumFormat::Decimal) {
-                    NumFormat::DecimalZero => NumFormat::DecimalZero,
-                    _ if def.legal => NumFormat::Decimal,
-                    f => f,
-                };
-                label.push_str(&fmt.format(val));
-                if label.len() > MAX_LABEL_BYTES {
-                    break;
-                }
-            } else {
-                label.push(c);
-            }
-        }
+        let numeric = def.format != NumFormat::Bullet;
+        let label = if numeric { render_label(numbering, num, &def, &vals, &started) } else { def.text.clone() };
+        // A "none" level shows its literal text, but has no number of its own for a reference to
+        // show (#4985): it is kept as one that shows none.
+        let counted = if def.format == NumFormat::None {
+            CountedLabel { label: None, levels: Vec::new() }
+        } else {
+            CountedLabel { label: numeric.then(|| label.clone()), levels: template_levels(&def.text) }
+        };
+        self.labels.insert((abs.id, lv as u8), counted);
         Some((label, def))
     }
+    /// Which levels of `num`'s list have counted a paragraph yet (a level never used has no
+    /// number of its own to show in a reference).
+    pub fn started(&self, numbering: &Numbering, num: u32) -> Option<[bool; 9]> {
+        let abs = numbering.abstract_of(num)?;
+        Some(std::array::from_fn(|k| self.labels.get(&(abs.id, k as u8)).is_some_and(|last| last.label.is_some())))
+    }
+
+    /// The label `level` of `num`'s list shows at this point of the walk, without counting: what
+    /// the last paragraph at that level showed (`"14."` while walking clause 14's sub-clauses).
+    /// `None` for a bullet or "none" level, or a level the list doesn't define.
+    pub fn current_label(&self, numbering: &Numbering, num: u32, level: u8) -> Option<String> {
+        let abs = numbering.abstract_of(num)?;
+        if let Some(last) = self.labels.get(&(abs.id, level.min(8))) {
+            return last.label.clone();
+        }
+        // A deeper item can count a level implicitly without rendering a paragraph there.
+        let def = numbering.level(num, level.min(8))?;
+        if matches!(def.format, NumFormat::None | NumFormat::Bullet) {
+            return None;
+        }
+        let vals = self.state.get(&abs.id).copied().unwrap_or([0; 9]);
+        let started = self.flags.get(&abs.id).copied().unwrap_or_default();
+        Some(render_label(numbering, num, def, &vals, &started))
+    }
+
+    /// Levels shown by the last rendered label, or the current definition for an implicit level.
+    pub fn current_template_levels(&self, numbering: &Numbering, num: u32, level: u8) -> Vec<usize> {
+        if let Some(abs) = numbering.abstract_of(num)
+            && let Some(last) = self.labels.get(&(abs.id, level.min(8)))
+        {
+            return last.levels.clone();
+        }
+        numbering.level(num, level.min(8)).map(|def| template_levels(&def.text)).unwrap_or_default()
+    }
+}
+
+/// The zero-based levels a label template shows (`"%1.%2"` → `[0, 1]`, `"(%3)"` → `[2]`).
+pub fn template_levels(text: &str) -> Vec<usize> {
+    let mut out = Vec::new();
+    let mut chars = text.chars().take(MAX_LEVEL_TEXT).peekable();
+    while let Some(c) = chars.next() {
+        if c == '%'
+            && let Some(d) = chars.peek().and_then(|d| d.to_digit(10))
+        {
+            chars.next();
+            let k = (d as usize).saturating_sub(1).min(8);
+            if !out.contains(&k) {
+                out.push(k);
+            }
+        }
+    }
+    out
+}
+
+/// Render a list label without advancing counters, retaining the existing hostile-input caps.
+fn render_label(numbering: &Numbering, num: u32, def: &Level, vals: &[u32; 9], started: &[LevelFlags; 9]) -> String {
+    let level_def = |k: usize| numbering.level(num, k as u8);
+    let mut label = String::new();
+    // Word's level text is short; .docx and the lists command already cap it at 64.
+    let mut chars = def.text.chars().take(MAX_LEVEL_TEXT).peekable();
+    while let Some(c) = chars.next() {
+        if c == '%'
+            && let Some(d) = chars.peek().and_then(|d| d.to_digit(10))
+        {
+            chars.next();
+            let k = (d as usize).saturating_sub(1).min(8);
+            let val = vals.get(k).copied().unwrap_or(0);
+            let counted = started.get(k).is_some_and(|f| f.started);
+            let val = if counted { val } else { level_def(k).map(|l| l.start).unwrap_or(1) };
+            // Legal numbering (w:isLgl) shows every level in Arabic digits, whatever its own
+            // format (二 of a Chinese outline becomes 2 in "2.1"); a "none" level stays empty.
+            let fmt = match level_def(k).map(|l| l.format).unwrap_or(NumFormat::Decimal) {
+                f @ (NumFormat::DecimalZero | NumFormat::None) => f,
+                _ if def.legal => NumFormat::Decimal,
+                f => f,
+            };
+            label.push_str(&fmt.format(val));
+            if label.len() > MAX_LABEL_BYTES {
+                break;
+            }
+        } else {
+            label.push(c);
+        }
+    }
+    label
 }
 
 /// Longest level text (in characters) expanded into a list label.
@@ -413,6 +485,45 @@ const MAX_LABEL_BYTES: usize = 256;
 mod tests {
     use super::*;
 
+    /// Read-only counter APIs retain the renderer's bounds and never advance a list.
+    #[test]
+    fn reference_labels_are_bounded_and_do_not_advance_counters() {
+        let mut n = Numbering::default();
+        let id = n.add_list(ListKind::Numbered);
+        if let Some(num) = n.nums.iter_mut().find(|x| x.id == id) {
+            num.level_overrides = vec![(0, Level { text: "%1".repeat(30_000), ..Level::default() })];
+        }
+        let mut c = Counters::default();
+        let label = c.next_label(&n, id, 0).unwrap().0;
+        let before = c.values(&n, id);
+        assert_eq!(c.current_label(&n, id, 0).as_deref(), Some(label.as_str()));
+        assert_eq!(c.values(&n, id), before);
+        assert!(label.len() <= 512);
+        assert_eq!(template_levels(&"%1%9".repeat(30_000)), [0, 8]);
+        assert!(c.started(&n, id).unwrap()[0]);
+    }
+
+    #[test]
+    fn last_ancestor_label_keeps_the_instance_that_rendered_it() {
+        let mut numbering = Numbering::default();
+        let first = numbering.add_list(ListKind::Numbered);
+        numbering.abstracts[0].levels[0].text = "Clause %1.".into();
+        let mut second = numbering.num(first).unwrap().clone();
+        second.id = first + 1;
+        second.level_overrides = vec![
+            (0, Level { text: "Part %1.%2.".into(), ..Level::default() }),
+            (1, Level { format: NumFormat::LowerLetter, text: "(%2)".into(), ..Level::default() }),
+        ];
+        let second_id = second.id;
+        numbering.nums.push(second);
+        let mut counters = Counters::default();
+        assert_eq!(counters.next_label(&numbering, first, 0).unwrap().0, "Clause 1.");
+        assert_eq!(counters.next_label(&numbering, second_id, 1).unwrap().0, "(a)");
+        assert_eq!(counters.current_label(&numbering, second_id, 0).as_deref(), Some("Clause 1."));
+        assert_eq!(counters.current_template_levels(&numbering, second_id, 0), vec![0]);
+        assert_eq!(counters.started(&numbering, second_id).unwrap()[..2], [true, true]);
+    }
+
     #[test]
     fn a_huge_level_text_gives_a_bounded_label() {
         let mut n = Numbering::default();
@@ -449,6 +560,27 @@ mod tests {
         assert_eq!(n.level(id, 2).map(|l| l.format), Some(NumFormat::DecimalZero));
     }
 
+    #[test]
+    fn a_none_level_shows_its_text_and_still_counts() {
+        // As Word shows them (#4985): a "none" level's literal text is its label, its own number
+        // reads as nothing (also in legal numbering), and it restarts the levels below it.
+        let mut n = Numbering::default();
+        let id = n.add_list(ListKind::Numbered);
+        let lvl = |format, text: &str| Level { format, text: text.into(), ..Level::default() };
+        let legal = Level { legal: true, ..lvl(NumFormat::Decimal, "%2.%3") };
+        if let Some(num) = n.nums.iter_mut().find(|x| x.id == id) {
+            num.level_overrides = vec![(0, lvl(NumFormat::None, "RESPONSE:")), (1, lvl(NumFormat::None, "[%2]")), (2, legal)];
+        }
+        assert!(n.shows_label(id, 0));
+        let mut c = Counters::default();
+        assert_eq!(c.next_label(&n, id, 0).unwrap().0, "RESPONSE:");
+        assert_eq!(c.next_label(&n, id, 1).unwrap().0, "[]");
+        assert_eq!(c.next_label(&n, id, 2).unwrap().0, ".1");
+        assert_eq!(c.next_label(&n, id, 2).unwrap().0, ".2");
+        assert_eq!(c.next_label(&n, id, 1).unwrap().0, "[]");
+        assert_eq!(c.next_label(&n, id, 2).unwrap().0, ".1");
+    }
+
     #[test]
     fn a_start_override_restarts_only_its_own_level_of_a_shared_list() {
         // One abstract list numbers the headings through two nums: one for levels 1-4, and one

crates/doc/src/para.rs

@@ -231,7 +231,7 @@ impl Paragraph {
 
     /// Number of objects in `text[..off]`.
     fn objects_before(&self, off: usize) -> usize {
-        self.text.get(..off).map(|t| t.chars().filter(|c| *c == OBJ).count()).unwrap_or(0)
+        self.text.get(..off).map(|t| t.chars().inspect(|_| crate::work::steps(1)).filter(|c| *c == OBJ).count()).unwrap_or(0)
     }
 
     /// The object at byte `off` (which must hold U+FFFC).
@@ -250,13 +250,13 @@ impl Paragraph {
     }
     /// Byte offsets of all objects, in order.
     pub fn object_offsets(&self) -> Vec<usize> {
-        self.text.char_indices().filter(|(_, c)| *c == OBJ).map(|(i, _)| i).collect()
+        self.text.char_indices().inspect(|_| crate::work::steps(1)).filter(|(_, c)| *c == OBJ).map(|(i, _)| i).collect()
     }
     /// (byte offset, object) for every object, in text order, in one pass. Prefer it to
     /// [`Paragraph::object_offsets`] + [`Paragraph::object_at`], which counts the objects before
     /// each offset again, so a loop over a paragraph's objects is quadratic (#3614).
     pub fn objects_with_offsets(&self) -> impl Iterator<Item = (usize, &InlineObject)> + '_ {
-        self.text.char_indices().filter(|(_, c)| *c == OBJ).map(|(i, _)| i).zip(self.objects.iter())
+        self.text.char_indices().inspect(|_| crate::work::steps(1)).filter(|(_, c)| *c == OBJ).map(|(i, _)| i).zip(self.objects.iter())
     }
 
     /// Insert plain text (U+FFFC is dropped) at `off` with `props`. Returns the inserted length.

crates/doc/src/para/float.rs

@@ -186,6 +186,11 @@ pub struct Float {
     /// top. 0 = unknown, drawn in document order (objects inserted here, or from other formats).
     #[serde(skip_serializing_if = "is_zero_u32")]
     pub z: u32,
+    /// Turned, but the file keeps no room for the turn (`wp:effectExtent` missing or zero, as
+    /// files from other tools have): Word then keeps an inline one's line to its frame and lets
+    /// the turned object overhang it (see [`Float::line_pad`]). Cleared by [`Float::set_spin`].
+    #[serde(skip_serializing_if = "is_false")]
+    pub bare_turn: bool,
 }
 
 /// Longest alternative text kept (characters): hostile files and commands are capped.
@@ -241,6 +246,7 @@ impl Float {
         self.rot = wordcraft_geom::normalize_degrees(s.deg);
         self.flip_h = s.flip_h;
         self.flip_v = s.flip_v;
+        self.bare_turn = false;
     }
     /// How far a `w` × `h` frame's rotated bounds reach past it on each side (x, y): the room a
     /// rotated object takes beyond its frame (zero unrotated; negative where the turned object
@@ -250,6 +256,23 @@ impl Float {
         let (bw, bh) = self.spin().extent(w, h);
         ((bw - w) / 2.0, (bh - h) / 2.0)
     }
+    /// A [`Float::bare_turn`] still inline and without effects room: made floating, or given room
+    /// for a shadow or glow since, it saves, and so lays out, with the turn's room again.
+    pub fn is_bare_turn(&self) -> bool {
+        self.bare_turn && self.wrap == Wrap::Inline && self.effect_extent() == [0.0; 4]
+    }
+    /// The room an inline `w` × `h` object's line keeps past its frame on each side (x, y) for its
+    /// turn: its rotated bounds ([`Float::spin_pad`]) — or, for a [`Float::is_bare_turn`], none, except
+    /// that Word 2013+ layout (`modern`) exchanges the frame's width and height from 45° up to 135°
+    /// and from 225° up to 315°.
+    pub fn line_pad(&self, w: f32, h: f32, modern: bool) -> (f32, f32) {
+        if !self.is_bare_turn() {
+            return self.spin_pad(w, h);
+        }
+        let (w, h) = (wordcraft_geom::finite(w).clamp(0.0, 4000.0), wordcraft_geom::finite(h).clamp(0.0, 4000.0));
+        let deg = wordcraft_geom::normalize_degrees(self.rot);
+        if modern && ((45.0..135.0).contains(&deg) || (225.0..315.0).contains(&deg)) { ((h - w) / 2.0, (w - h) / 2.0) } else { (0.0, 0.0) }
+    }
     /// The effect extents, finite and clamped (left, top, right, bottom).
     pub fn effect_extent(&self) -> [f32; 4] {
         self.effect.map(|v| if v.is_finite() { v.clamp(0.0, 1584.0) } else { 0.0 })

crates/doc/src/para/inline_object.rs

@@ -19,6 +19,9 @@ pub enum InlineObject {
         /// Crop (left, top, right, bottom) fractions 0..1.
         #[serde(default)]
         crop: [f32; 4],
+        /// Brightness, contrast and transparency read from the file (see [`crate::picfx`]).
+        #[serde(default, skip_serializing_if = "crate::picfx::PicFx::is_identity")]
+        fx: crate::picfx::PicFx,
         /// The OLE object (embedded or linked file) or Word signature line this picture shows, as
         /// read from a file: saving writes the object back, not only its picture. See
         /// [`crate::graphic::Embedded`].
@@ -313,20 +316,34 @@ impl InlineObject {
     /// A group's members placed in a `w` × `h` box at (`x`, `y`): each member's rectangle
     /// (x, y, w, h) and the member. Empty for anything else.
     pub fn group_rects(&self, x: f32, y: f32, w: f32, h: f32) -> Vec<([f32; 4], &InlineObject)> {
-        let InlineObject::Group { ch_w, ch_h, children, .. } = self else { return Vec::new() };
+        let InlineObject::Group { children, .. } = self else { return Vec::new() };
+        self.member_rects(x, y, w, h).into_iter().filter_map(|(i, r)| Some((r, &children.get(i)?.obj))).collect()
+    }
+    /// [`InlineObject::group_rects`] with each member's index in `children` instead of the member.
+    pub fn member_rects(&self, x: f32, y: f32, w: f32, h: f32) -> Vec<(usize, [f32; 4])> {
+        let InlineObject::Group { children, .. } = self else { return Vec::new() };
+        let (sx, sy) = self.group_scale(w, h);
         let fin = |v: f32| if v.is_finite() { v } else { 0.0 };
-        let sx = if fin(*ch_w) > 0.0 { fin(w) / ch_w } else { 1.0 };
-        let sy = if fin(*ch_h) > 0.0 { fin(h) / ch_h } else { 1.0 };
         children
             .iter()
             .take(MAX_GROUP_CHILDREN)
-            .filter_map(|c| {
+            .enumerate()
+            .filter_map(|(i, c)| {
                 let (cw, chh, _) = c.obj.frame()?;
                 let r = [x + fin(c.x) * sx, y + fin(c.y) * sy, (fin(cw) * sx).clamp(0.0, 4000.0), (fin(chh) * sy).clamp(0.0, 4000.0)];
-                (!matches!(c.obj, InlineObject::Group { .. })).then_some((r, &c.obj))
+                (!matches!(c.obj, InlineObject::Group { .. })).then_some((i, r))
             })
             .collect()
     }
+    /// How much a group drawn `w` × `h` scales its members' space (x, y): 1 for a canvas, whose
+    /// members keep their size.
+    pub fn group_scale(&self, w: f32, h: f32) -> (f32, f32) {
+        let InlineObject::Group { ch_w, ch_h, .. } = self else { return (1.0, 1.0) };
+        let fin = |v: f32| if v.is_finite() { v } else { 0.0 };
+        let sx = if fin(*ch_w) > 0.0 { fin(w) / ch_w } else { 1.0 };
+        let sy = if fin(*ch_h) > 0.0 { fin(h) / ch_h } else { 1.0 };
+        (sx, sy)
+    }
     /// A legacy form field's definition (`FORMCHECKBOX`, `FORMTEXT`, `FORMDROPDOWN`), read from
     /// the `w:ffData` it keeps; `None` for anything else.
     pub fn legacy_form(&self) -> Option<crate::legacy_form::LegacyForm> {

crates/doc/src/picfx.rs

@@ -0,0 +1,190 @@
+//! Picture colour corrections read from a file: brightness, contrast and transparency
+//! (DrawingML `a:lum` and `a:alphaModFix` inside `a:blip`, ECMA-376 §20.1.8.42 and §20.1.8.4;
+//! VML `v:imagedata` `gain` and `blacklevel`, which Word's picture watermark "Washout" sets).
+//!
+//! Values are kept in OOXML units: thousandths of a percent, so 100 000 is 100 %. All zero is
+//! the picture as it is.
+//!
+//! The pixel formula was measured on Word (#1711, #1766): Word bakes the corrected pixels into
+//! the picture it embeds in a PDF export, so a synthetic grey and colour ramp exported with
+//! 20 `a:lum` combinations and 6 VML `gain`/`blacklevel` pairs gives the curve. Every channel is
+//! mapped on its own (no hue or luminance coupling) by one straight line, clamped to 0..=255
+//! only at the end:
+//!
+//! `y = s·(x − 127.5) + 127.5 + 127.5·b·(1 + s)`, with `b = bright / 100 000`, `c = contrast /
+//! 100 000`, `s = 1 + c` when `c ≤ 0` and `s = 1 / (1 − c)` when `c > 0`
+//!
+//! (as if half the brightness shift were applied before the contrast and half after). Measured
+//! slope and intercept of Word's output for grey input `x` (PNG pictures, exact to ±1 level):
+//! bright 20 000 → 1.00, +51; bright −40 000 → 1.00, −103; contrast 80 000 → 5.00, −510;
+//! contrast −80 000 → 0.20, +101.4; (40 000, −40 000) → 0.60, +132.2; (80 000, −80 000) →
+//! 0.20, +223.8 (also the #1766 header picture: fitted 0.197..0.201, +223.6..224.0);
+//! (−40 000, 40 000) → 1.667, −221.7; (30 000, 60 000) → 2.50, −57.8.
+//!
+//! VML: `gain` is the slope itself and `blacklevel` half the brightness — Word draws `gain` 0.3,
+//! `blacklevel` 0.35 (Washout) as 0.30, +205.3, the same as `a:lum` (70 000, −70 000), and
+//! `gain` 1, `blacklevel` 0.2 as the very same picture as `bright="40000"`.
+
+use serde::{Deserialize, Serialize};
+
+/// The largest magnitude of a correction (100 %).
+pub const FX_MAX: i32 = 100_000;
+
+/// Brightness, contrast and transparency of a picture (see the module docs).
+#[derive(Clone, Copy, Debug, Default, PartialEq, Eq, Hash, Serialize, Deserialize)]
+#[serde(default)]
+pub struct PicFx {
+    /// −100 000 (black) ..= 100 000 (white).
+    pub bright: i32,
+    /// −100 000 (flat grey) ..= 100 000 (hard threshold).
+    pub contrast: i32,
+    /// 0 (opaque) ..= 100 000 (invisible).
+    pub transparency: i32,
+}
+
+impl PicFx {
+    /// Corrections clamped to their ranges.
+    pub fn new(bright: i32, contrast: i32, transparency: i32) -> PicFx {
+        PicFx { bright: bright.clamp(-FX_MAX, FX_MAX), contrast: contrast.clamp(-FX_MAX, FX_MAX), transparency: transparency.clamp(0, FX_MAX) }
+    }
+
+    /// The same corrections, each clamped to its range (for values that came from anywhere).
+    pub fn clamped(self) -> PicFx {
+        PicFx::new(self.bright, self.contrast, self.transparency)
+    }
+
+    /// No correction at all.
+    pub fn is_identity(&self) -> bool {
+        *self == PicFx::default()
+    }
+
+    /// Brightness or contrast changes the pixels (transparency alone doesn't).
+    pub fn changes_pixels(&self) -> bool {
+        let f = self.clamped();
+        f.bright != 0 || f.contrast != 0
+    }
+
+    /// The picture's opacity, 0..=1 (1 − transparency).
+    pub fn opacity(&self) -> f32 {
+        1.0 - self.transparency.clamp(0, FX_MAX) as f32 / FX_MAX as f32
+    }
+
+    /// The corrections a VML `v:imagedata` gives with `gain` and `blacklevel` (already parsed to
+    /// plain numbers, 1 and 0 when missing): contrast from the slope `gain`, brightness twice the
+    /// black level (see the module docs). Non-finite values count as missing.
+    pub fn from_vml(gain: f64, blacklevel: f64) -> PicFx {
+        let gain = if gain.is_finite() { gain.clamp(0.0, 1.0e6) } else { 1.0 };
+        let black = if blacklevel.is_finite() { blacklevel.clamp(-0.5, 0.5) } else { 0.0 };
+        // s = 1 + c (c ≤ 0) or 1 / (1 − c) (c > 0).
+        let c = if gain <= 1.0 { gain - 1.0 } else { 1.0 - 1.0 / gain };
+        let fx = |v: f64| (v * FX_MAX as f64).round().clamp(-(FX_MAX as f64), FX_MAX as f64) as i32;
+        PicFx::new(fx(2.0 * black), fx(c), 0)
+    }
+
+    /// `gain` and `blacklevel` for VML (the inverse of [`PicFx::from_vml`]).
+    pub fn to_vml(&self) -> (f64, f64) {
+        (self.slope(), self.clamped().bright as f64 / FX_MAX as f64 / 2.0)
+    }
+
+    /// The contrast slope `s` (capped at 255: contrast 100 % is a threshold).
+    fn slope(&self) -> f64 {
+        let c = self.contrast.clamp(-FX_MAX, FX_MAX) as f64 / FX_MAX as f64;
+        if c <= 0.0 { 1.0 + c } else { (1.0 / (1.0 - c).max(1.0 / 255.0)).min(255.0) }
+    }
+
+    /// The lookup table every colour channel goes through.
+    pub fn table(&self) -> [u8; 256] {
+        let s = self.slope();
+        let b = self.bright.clamp(-FX_MAX, FX_MAX) as f64 / FX_MAX as f64;
+        let mut t = [0u8; 256];
+        for (x, slot) in t.iter_mut().enumerate() {
+            let y = s * (x as f64 - 127.5) + 127.5 + 127.5 * b * (1.0 + s);
+            *slot = if y.is_finite() { y.round().clamp(0.0, 255.0) as u8 } else { 0 };
+        }
+        t
+    }
+
+    /// Apply brightness and contrast to straight (not premultiplied) RGBA8 pixels in place. The
+    /// alpha byte is left alone (transparency is drawn as an opacity, see [`PicFx::opacity`]); a
+    /// trailing partial pixel is ignored.
+    pub fn apply_rgba(&self, rgba: &mut [u8]) {
+        if !self.changes_pixels() {
+            return;
+        }
+        let t = self.table();
+        for px in rgba.as_chunks_mut::<4>().0 {
+            for c in px.iter_mut().take(3) {
+                *c = t.get(usize::from(*c)).copied().unwrap_or(*c);
+            }
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    /// Word's own output for grey input 0, 128 and 255 (PNG pictures in a PDF export).
+    #[test]
+    fn matches_word() {
+        let cases: [(PicFx, [u8; 3]); 6] = [
+            (PicFx::new(40_000, -40_000, 0), [132, 209, 255]),
+            (PicFx::new(80_000, -80_000, 0), [224, 249, 255]),
+            (PicFx::new(-40_000, 40_000, 0), [0, 0, 203]),
+            (PicFx::new(0, 80_000, 0), [0, 130, 255]),
+            (PicFx::new(20_000, 0, 0), [51, 179, 255]),
+            (PicFx::from_vml(19661.0 / 65536.0, 22938.0 / 65536.0), [205, 244, 255]),
+        ];
+        for (fx, want) in cases {
+            let t = fx.table();
+            for (x, w) in [0usize, 128, 255].into_iter().zip(want) {
+                let got = t[x];
+                assert!(got.abs_diff(w) <= 2, "{fx:?}: {x} → {got}, Word {w}");
+            }
+        }
+    }
+
+    #[test]
+    fn channels_apart_alpha_kept() {
+        let fx = PicFx::new(80_000, -80_000, 0);
+        let mut px = [255, 0, 0, 77, 0, 128, 255, 255];
+        fx.apply_rgba(&mut px);
+        // Word: red (255, 0, 0) → (255, 224, 224); 128 → 249 (the formula gives 250.0).
+        assert_eq!(px, [255, 224, 224, 77, 224, 250, 255, 255]);
+        let mut odd = [0u8; 6];
+        fx.apply_rgba(&mut odd);
+        assert_eq!(odd[4..], [0, 0]);
+    }
+
+    #[test]
+    fn vml_mapping() {
+        // 16.16 fixed point: 19661f and 22938f are 0.3 and 0.35 to within 1e-5.
+        let washout = PicFx::from_vml(19661.0 / 65536.0, 22938.0 / 65536.0);
+        assert!((washout.bright - 70_000).abs() <= 2 && (washout.contrast + 70_000).abs() <= 2, "{washout:?}");
+        assert_eq!(PicFx::from_vml(1.0, 0.2), PicFx::new(40_000, 0, 0));
+        assert_eq!(PicFx::from_vml(2.0, 0.0), PicFx::new(0, 50_000, 0));
+        assert!(PicFx::from_vml(1.0, 0.0).is_identity());
+        for (g, b) in [(f64::NAN, f64::INFINITY), (-1.0e30, 1.0e30), (1.0e300, -7.0), (0.0, 0.0)] {
+            let fx = PicFx::from_vml(g, b);
+            assert_eq!(fx, fx.clamped(), "{g} {b}");
+            let _ = fx.table();
+        }
+    }
+
+    #[test]
+    fn hostile_values_clamp() {
+        for v in [i32::MIN, -100_001, 0, 100_001, i32::MAX] {
+            let fx = PicFx { bright: v, contrast: v, transparency: v };
+            let _ = fx.table();
+            assert!((0.0..=1.0).contains(&fx.opacity()));
+            let mut px = [1u8, 2, 3, 4];
+            fx.apply_rgba(&mut px);
+            assert_eq!(px[3], 4);
+        }
+        assert_eq!(PicFx::new(0, 0, 25_000).opacity(), 0.75);
+        assert!(!PicFx::new(0, 0, 25_000).changes_pixels());
+        // Contrast 100 %: a threshold at mid-grey, no NaN.
+        let t = PicFx::new(0, FX_MAX, 0).table();
+        assert_eq!((t[0], t[127], t[128], t[255]), (0, 0, 255, 255));
+    }
+}

crates/doc/src/props.rs

@@ -23,12 +23,16 @@ mod para_props;
 mod table_props;
 
 pub use char_props::{
-    CharProps, CombineBrackets, DEFAULT_FONT_SIZE, EastAsianLayout, Highlight, Ligatures, NumChange, PropChange, Underline, VertAlign,
+    CharProps, CombineBrackets, DEFAULT_FONT_SIZE, EastAsianLayout, EmphasisMark, Highlight, Ligatures, NumChange, PropChange, Underline, VertAlign,
 };
 pub use color::{Rgb, TextColor};
-pub use para_props::{Align, Border, BorderStyle, Borders, LineSpacing, NumRef, ParaProps, TAB_POS_EPS, TabAlign, TabLeader, TabStop, TextFrame};
+pub use para_props::{
+    Align, Border, BorderStyle, Borders, LineSpacing, NumRef, ParaProps, ParagraphTextDirection, TAB_POS_EPS, TabAlign, TabLeader, TabStop,
+    TextAlignment, TextFrame,
+};
 pub use table_props::{
-    CellProps, HeightRule, RowProps, TableFloat, TableLook, TableProps, TextDirection, TextVert, VAlign, VMerge, anchor_ooxml, valign_from_anchor,
+    CellPicture, CellProps, HeightRule, PictureRepeat, RowProps, TableFloat, TableLook, TableProps, TextDirection, TextVert, VAlign, VMerge,
+    anchor_ooxml, valign_from_anchor,
 };
 
 #[cfg(test)]

crates/doc/src/props/char_props.rs

@@ -291,6 +291,10 @@ pub struct CharProps {
     /// Kerning for fonts at or above this size (points); 0 = off.
     #[serde(skip_serializing_if = "Option::is_none")]
     pub kern: Option<f32>,
+    /// A mark over (or under) each character (`w:em`, Font › Emphasis mark), as East Asian text
+    /// stresses words; `Some(EmphasisMark::None)` turns off one a style sets.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub emphasis: Option<EmphasisMark>,
     /// OpenType ligatures (`w14:ligatures`). `None` leaves WordCraft's default: standard and
     /// contextual ligatures.
     #[serde(skip_serializing_if = "Option::is_none")]
@@ -381,6 +385,41 @@ pub struct EastAsianLayout {
     pub vert_compress: Option<bool>,
 }
 
+/// A mark over or under each character (`ST_Em`, ECMA-376 §17.18.24): Word's Font › Emphasis
+/// mark.
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash, Serialize, Deserialize, Default)]
+#[serde(rename_all = "camelCase")]
+pub enum EmphasisMark {
+    #[default]
+    None,
+    /// A dot above.
+    Dot,
+    /// A comma (sesame) above.
+    Comma,
+    /// A hollow circle above.
+    Circle,
+    /// A dot below.
+    UnderDot,
+}
+
+impl EmphasisMark {
+    pub const ALL: [EmphasisMark; 5] = [EmphasisMark::None, EmphasisMark::Dot, EmphasisMark::Comma, EmphasisMark::Circle, EmphasisMark::UnderDot];
+    /// OOXML `w:em` value.
+    pub fn ooxml(self) -> &'static str {
+        match self {
+            EmphasisMark::None => "none",
+            EmphasisMark::Dot => "dot",
+            EmphasisMark::Comma => "comma",
+            EmphasisMark::Circle => "circle",
+            EmphasisMark::UnderDot => "underDot",
+        }
+    }
+    /// The mark an OOXML `w:em` value names; unknown values are none.
+    pub fn from_ooxml(v: &str) -> EmphasisMark {
+        EmphasisMark::ALL.into_iter().find(|m| m.ooxml() == v.trim()).unwrap_or_default()
+    }
+}
+
 /// Brackets around two lines in one (`ST_CombineBrackets`).
 #[derive(Clone, Copy, Debug, PartialEq, Eq, Hash, Serialize, Deserialize)]
 #[serde(rename_all = "camelCase")]
@@ -466,7 +505,7 @@ impl CharProps {
     }
 
     fn minus_run_visibility(&mut self, base: &CharProps) {
-        drop_same_fields!(self, base; web_hidden, snap_to_grid);
+        drop_same_fields!(self, base; web_hidden, snap_to_grid, emphasis);
     }
 
     fn minus_script_style(&mut self, base: &CharProps) {
@@ -497,7 +536,7 @@ impl CharProps {
     }
 
     fn overlay_run_visibility(&mut self, patch: &CharProps) {
-        overlay_fields!(self, patch; web_hidden, snap_to_grid);
+        overlay_fields!(self, patch; web_hidden, snap_to_grid, emphasis);
     }
 
     fn overlay_script_style(&mut self, patch: &CharProps) {

crates/doc/src/props/para_props.rs

@@ -26,6 +26,101 @@ impl Align {
     }
 }
 
+/// Paragraph-level text flow (`w:pPr/w:textDirection`, ECMA-376 §17.3.1.41).
+///
+/// Kept separate from [`super::TextDirection`], whose three values model the table-cell direction
+/// control. These values preserve the exact OOXML direction through JSON and DOCX saves.
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash, Serialize, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub enum ParagraphTextDirection {
+    LrTb,
+    LrTbV,
+    Tb,
+    TbRl,
+    TbRlV,
+    TbLr,
+    TbLrV,
+    BtLr,
+    TbV,
+    Rl,
+    RlV,
+    Lr,
+    LrV,
+}
+
+impl ParagraphTextDirection {
+    pub fn from_ooxml(value: &str) -> Option<Self> {
+        match value {
+            "lrTb" => Some(Self::LrTb),
+            "lrTbV" => Some(Self::LrTbV),
+            "tb" => Some(Self::Tb),
+            "tbRl" => Some(Self::TbRl),
+            "tbRlV" => Some(Self::TbRlV),
+            "tbLr" => Some(Self::TbLr),
+            "tbLrV" => Some(Self::TbLrV),
+            "btLr" => Some(Self::BtLr),
+            "tbV" => Some(Self::TbV),
+            "rl" => Some(Self::Rl),
+            "rlV" => Some(Self::RlV),
+            "lr" => Some(Self::Lr),
+            "lrV" => Some(Self::LrV),
+            _ => None,
+        }
+    }
+
+    pub fn ooxml(self) -> &'static str {
+        match self {
+            Self::LrTb => "lrTb",
+            Self::LrTbV => "lrTbV",
+            Self::Tb => "tb",
+            Self::TbRl => "tbRl",
+            Self::TbRlV => "tbRlV",
+            Self::TbLr => "tbLr",
+            Self::TbLrV => "tbLrV",
+            Self::BtLr => "btLr",
+            Self::TbV => "tbV",
+            Self::Rl => "rl",
+            Self::RlV => "rlV",
+            Self::Lr => "lr",
+            Self::LrV => "lrV",
+        }
+    }
+}
+
+/// Vertical alignment of characters of different sizes on a line (`w:textAlignment`,
+/// ECMA-376 §17.3.1.39).
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash, Serialize, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub enum TextAlignment {
+    Auto,
+    Top,
+    Center,
+    Baseline,
+    Bottom,
+}
+
+impl TextAlignment {
+    pub fn ooxml(self) -> &'static str {
+        match self {
+            TextAlignment::Auto => "auto",
+            TextAlignment::Top => "top",
+            TextAlignment::Center => "center",
+            TextAlignment::Baseline => "baseline",
+            TextAlignment::Bottom => "bottom",
+        }
+    }
+    pub fn from_ooxml(s: &str) -> Option<TextAlignment> {
+        match s {
+            "auto" => Some(TextAlignment::Auto),
+            "top" => Some(TextAlignment::Top),
+            "center" => Some(TextAlignment::Center),
+            "baseline" => Some(TextAlignment::Baseline),
+            "bottom" => Some(TextAlignment::Bottom),
+            _ => None,
+        }
+    }
+}
+
 /// Line spacing rule.
 #[derive(Clone, Copy, Debug, PartialEq, Serialize, Deserialize)]
 #[serde(rename_all = "camelCase", tag = "rule", content = "value")]
@@ -107,6 +202,18 @@ pub enum BorderStyle {
     Triple,
     DotDash,
     Wave,
+    /// The compound lines of thin and thick strokes (`thinThickSmallGap` to
+    /// `thinThickThinLargeGap`), kept so a file writes back the line it read; drawn as a single
+    /// line of the border's width.
+    ThinThickSmallGap,
+    ThickThinSmallGap,
+    ThinThickThinSmallGap,
+    ThinThickMediumGap,
+    ThickThinMediumGap,
+    ThinThickThinMediumGap,
+    ThinThickLargeGap,
+    ThickThinLargeGap,
+    ThinThickThinLargeGap,
 }
 
 impl BorderStyle {
@@ -121,6 +228,15 @@ impl BorderStyle {
             BorderStyle::Triple => "triple",
             BorderStyle::DotDash => "dotDash",
             BorderStyle::Wave => "wave",
+            BorderStyle::ThinThickSmallGap => "thinThickSmallGap",
+            BorderStyle::ThickThinSmallGap => "thickThinSmallGap",
+            BorderStyle::ThinThickThinSmallGap => "thinThickThinSmallGap",
+            BorderStyle::ThinThickMediumGap => "thinThickMediumGap",
+            BorderStyle::ThickThinMediumGap => "thickThinMediumGap",
+            BorderStyle::ThinThickThinMediumGap => "thinThickThinMediumGap",
+            BorderStyle::ThinThickLargeGap => "thinThickLargeGap",
+            BorderStyle::ThickThinLargeGap => "thickThinLargeGap",
+            BorderStyle::ThinThickThinLargeGap => "thinThickThinLargeGap",
         }
     }
     pub fn from_ooxml(s: &str) -> BorderStyle {
@@ -133,6 +249,15 @@ impl BorderStyle {
             "triple" => BorderStyle::Triple,
             "dotDash" | "dotDotDash" => BorderStyle::DotDash,
             "wave" | "doubleWave" => BorderStyle::Wave,
+            "thinThickSmallGap" => BorderStyle::ThinThickSmallGap,
+            "thickThinSmallGap" => BorderStyle::ThickThinSmallGap,
+            "thinThickThinSmallGap" => BorderStyle::ThinThickThinSmallGap,
+            "thinThickMediumGap" => BorderStyle::ThinThickMediumGap,
+            "thickThinMediumGap" => BorderStyle::ThickThinMediumGap,
+            "thinThickThinMediumGap" => BorderStyle::ThinThickThinMediumGap,
+            "thinThickLargeGap" => BorderStyle::ThinThickLargeGap,
+            "thickThinLargeGap" => BorderStyle::ThickThinLargeGap,
+            "thinThickThinLargeGap" => BorderStyle::ThinThickThinLargeGap,
             _ => BorderStyle::Single,
         }
     }
@@ -183,6 +308,11 @@ impl Borders {
     pub fn any_visible(&self) -> bool {
         [self.top, self.left, self.bottom, self.right, self.between, self.inside_v].iter().flatten().any(Border::is_visible)
     }
+    /// A side is drawn, or a paragraph side without a line holds padding (ODF's, which keeps
+    /// its room in LibreOffice's layout).
+    pub fn takes_room(&self) -> bool {
+        self.any_visible() || [self.top, self.left, self.bottom, self.right].iter().flatten().any(|b| b.space.is_finite() && b.space > 0.0)
+    }
     /// Apply every `Some` side of `patch` on top of `self`.
     pub fn overlay(&mut self, patch: &Borders) {
         if patch.top.is_some() {
@@ -236,6 +366,9 @@ pub struct TextFrame {
     pub v_anchor: String,
     /// `around`, `auto`, `notBeside`, `none`, `tight` or `through`.
     pub wrap: String,
+    /// Whether the frame stays fixed to its page anchor when its paragraph moves.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub anchor_lock: Option<bool>,
     /// Distance from the text left and right, and above and below.
     pub h_space: f32,
     pub v_space: f32,
@@ -290,6 +423,9 @@ pub struct ParaProps {
     /// outside ones, so they swap sides on even (left-hand) pages.
     #[serde(skip_serializing_if = "Option::is_none")]
     pub mirror_indents: Option<bool>,
+    /// `w:adjustRightInd`: automatically adjust the right indent for the document grid when enabled; keep the specified right indent when disabled.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub adjust_right_indent: Option<bool>,
     #[serde(skip_serializing_if = "Option::is_none")]
     pub keep_next: Option<bool>,
     #[serde(skip_serializing_if = "Option::is_none")]
@@ -316,6 +452,9 @@ pub struct ParaProps {
     pub suppress_line_numbers: Option<bool>,
     #[serde(skip_serializing_if = "Option::is_none")]
     pub bidi: Option<bool>,
+    /// Paragraph text direction (`w:textDirection`); `None` inherits from the paragraph style.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub text_direction: Option<ParagraphTextDirection>,
     /// Drop cap: lines to drop (0 = none) — applies to the first character(s).
     #[serde(skip_serializing_if = "Option::is_none")]
     pub drop_cap: Option<u8>,
@@ -363,6 +502,10 @@ pub struct ParaProps {
     /// `w:snapToGrid`: lines snap to the section's document grid. Unset = on.
     #[serde(skip_serializing_if = "Option::is_none")]
     pub snap_to_grid: Option<bool>,
+    /// `w:textAlignment`: how characters of different sizes line up on a line. Kept for the
+    /// file (not laid out yet); Word sizes a line by it, so dropping it changes Word's pages.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub text_alignment: Option<TextAlignment>,
 }
 
 impl ParaProps {
@@ -374,7 +517,7 @@ impl ParaProps {
 
     fn overlay_layout(&mut self, patch: &ParaProps) {
         overlay_fields!(self, patch; style, align, indent_left, indent_right, indent_first, indent_left_chars, indent_right_chars, indent_first_chars,
-            space_before, space_after, space_before_auto, space_after_auto, line_spacing, contextual_spacing, mirror_indents, tabs, shading, borders);
+            space_before, space_after, space_before_auto, space_after_auto, line_spacing, contextual_spacing, mirror_indents, adjust_right_indent, tabs, shading, borders);
     }
 
     fn overlay_pagination(&mut self, patch: &ParaProps) {
@@ -383,7 +526,8 @@ impl ParaProps {
     }
 
     fn overlay_script_layout(&mut self, patch: &ParaProps) {
-        overlay_fields!(self, patch; bidi, kinsoku, word_wrap, overflow_punct, top_line_punct, auto_space_de, auto_space_dn);
+        overlay_fields!(self, patch; bidi, text_direction, kinsoku, word_wrap, overflow_punct, top_line_punct, auto_space_de, auto_space_dn);
+        overlay_fields!(self, patch; bidi, kinsoku, word_wrap, overflow_punct, top_line_punct, auto_space_de, auto_space_dn, text_alignment);
     }
     pub fn overlaid(mut self, patch: &ParaProps) -> ParaProps {
         self.overlay(patch);

crates/doc/src/props/table_props.rs

@@ -158,6 +158,12 @@ pub struct TableProps {
     /// the right margin.
     #[serde(skip_serializing_if = "std::ops::Not::not")]
     pub bidi_visual: bool,
+    /// Space above and below a table in the text flow, points (ODF's table `fo:margin-top` and
+    /// `fo:margin-bottom`; Word has none). Added to the spacing of the paragraphs around it.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub space_before: Option<f32>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub space_after: Option<f32>,
 }
 
 /// Where a floating table sits; the text after it wraps around it.
@@ -294,6 +300,38 @@ pub struct CellProps {
     /// Tracked formatting change (`w:tcPrChange`).
     #[serde(skip_serializing_if = "Option::is_none")]
     pub fmt_change: Option<Box<PropChange<CellProps>>>,
+    /// A picture behind the cell's content (an ODF cell's `style:background-image`; Word has
+    /// none).
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub picture: Option<Box<CellPicture>>,
+}
+
+/// A cell's background picture: drawn at its own size where `align` puts it, tiled, or stretched
+/// over the cell, and cut off at the cell's edges.
+#[derive(Clone, Debug, PartialEq, Default, Serialize, Deserialize)]
+#[serde(default, rename_all = "camelCase")]
+pub struct CellPicture {
+    /// Key into [`crate::Document::media`].
+    pub media: String,
+    /// Its own size, points.
+    pub size: [f32; 2],
+    pub repeat: PictureRepeat,
+    /// Where an unrepeated picture sits: 0 the left (top) edge, 0.5 the middle, 1 the right
+    /// (bottom) edge.
+    pub align: [f32; 2],
+}
+
+/// How a background picture fills its area (ODF §20.348 `style:repeat`).
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Default, Serialize, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub enum PictureRepeat {
+    /// Once, at its own size.
+    #[default]
+    None,
+    /// Repeated across the area.
+    Tile,
+    /// Stretched over the area.
+    Stretch,
 }
 
 impl CellProps {

crates/doc/src/resolve.rs

@@ -41,6 +41,9 @@ pub struct ResolvedChar {
     /// Kerning: the smallest font size, in points, that is kerned (`w:kern`); 0 = never kerned.
     #[serde(default)]
     pub kern: f32,
+    /// A mark over or under each character (Font › Emphasis mark).
+    #[serde(default)]
+    pub emphasis: crate::props::EmphasisMark,
     /// OpenType ligatures (`w14:ligatures`); `None` = the shaper's default (standard and
     /// contextual).
     #[serde(default)]
@@ -97,6 +100,7 @@ impl ResolvedChar {
             scale: c.scale.unwrap_or(100.0).clamp(1.0, 600.0),
             position: c.position.unwrap_or(0.0).clamp(-800.0, 800.0),
             kern: c.kern.filter(|k| k.is_finite()).unwrap_or(0.0).clamp(0.0, 1638.0),
+            emphasis: c.emphasis.unwrap_or_default(),
             ligatures: c.ligatures,
             outline: c.outline.unwrap_or(false),
             shadow: c.shadow.unwrap_or(false),
@@ -133,9 +137,12 @@ impl ResolvedChar {
     pub fn uses_complex(&self, c: char) -> bool {
         self.rtl || self.cs || crate::bidi::is_complex_script(c)
     }
-    /// The size glyphs are drawn at (super/subscript shrink to ~2/3).
+    /// The size glyphs are drawn and measured at. Word sets superscript and subscript text at
+    /// 0.65 of the run's size rounded to the half point, a font size the format can store
+    /// (measured in Word for Mac from digit advances: 8 pt → 5, 9 → 6, 11 → 7, 12 → 8,
+    /// 16 → 10.5, 24 → 15.5, 36 → 23.5).
     pub fn draw_size(&self) -> f32 {
-        if self.vert_align == VertAlign::Baseline { self.size } else { self.size * 0.65 }
+        if self.vert_align == VertAlign::Baseline { self.size } else { (self.size * 0.65 * 2.0).round() / 2.0 }
     }
     /// Baseline shift, points (positive = up).
     pub fn baseline_shift(&self) -> f32 {
@@ -149,6 +156,10 @@ impl ResolvedChar {
 }
 
 /// Concrete paragraph formatting.
+fn default_adjust_right_indent() -> bool {
+    true
+}
+
 #[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
 #[serde(rename_all = "camelCase")]
 pub struct ResolvedPara {
@@ -169,6 +180,9 @@ pub struct ResolvedPara {
     pub contextual_spacing: bool,
     /// Left/right indents are inside/outside: they swap on even pages.
     pub mirror_indents: bool,
+    /// Automatically adjust the right indent for the document grid when enabled; keep the specified right indent when disabled.
+    #[serde(default = "default_adjust_right_indent")]
+    pub adjust_right_indent: bool,
     pub keep_next: bool,
     pub keep_lines: bool,
     pub page_break_before: bool,
@@ -178,6 +192,10 @@ pub struct ResolvedPara {
     pub tabs: Vec<TabStop>,
     pub shading: Option<Rgb>,
     pub borders: Option<Borders>,
+    /// Padding on the sides without a line (top, left, bottom, right), points. ODF's box model
+    /// (XSL's) keeps it for documents opened from ODF (`Settings::odf`); Word drops it.
+    #[serde(default)]
+    pub padding: [f32; 4],
     pub bidi: bool,
     pub drop_cap: u8,
     /// The drop cap hangs in the margin (see `ParaProps::drop_cap_margin`).
@@ -256,6 +274,7 @@ impl ResolvedPara {
             },
             contextual_spacing: p.contextual_spacing.unwrap_or(false),
             mirror_indents: p.mirror_indents.unwrap_or(false),
+            adjust_right_indent: p.adjust_right_indent.unwrap_or(true),
             keep_next: p.keep_next.unwrap_or(false),
             keep_lines: p.keep_lines.unwrap_or(false),
             page_break_before: p.page_break_before.unwrap_or(false),
@@ -265,6 +284,10 @@ impl ResolvedPara {
             tabs,
             shading: p.shading,
             borders: p.borders.filter(Borders::any_visible),
+            padding: p.borders.map_or([0.0; 4], |b| {
+                [b.top, b.left, b.bottom, b.right]
+                    .map(|s| s.filter(|l| !l.is_visible() && l.space.is_finite() && l.space > 0.0).map_or(0.0, |l| l.space.min(1584.0)))
+            }),
             bidi: p.bidi.unwrap_or(false),
             drop_cap: p.drop_cap.unwrap_or(0).min(10),
             drop_cap_margin: p.drop_cap_margin.unwrap_or(false),
@@ -323,6 +346,38 @@ impl StyleSheet {
         (pp, cp)
     }
 
+    /// What a table style's paragraph formatting `table` changes for a paragraph with `p`, once
+    /// the paragraph style chain is applied over it (ECMA-376 §17.7.2): each property it gives
+    /// that the chain doesn't set. Formats without table styles state these directly.
+    pub fn table_para_patch(&self, p: &ParaProps, table: &ParaProps) -> ParaProps {
+        let style = p.style.clone().unwrap_or_else(|| "Normal".into());
+        let mut with = self.default_para.clone();
+        with.overlay(&ParaProps { style: None, ..table.clone() });
+        let (with, _) = self.para_style_props_over(&style, with, CharProps::default());
+        let (without, _) = self.para_style_props_over(&style, self.default_para.clone(), CharProps::default());
+        diff_patch(&with, &without)
+    }
+
+    /// What a table style's run formatting `table` changes for a run of character style
+    /// `run_style` in a paragraph of style `para_style`, once the paragraph and character style
+    /// chains are applied over it (ECMA-376 §17.7.2).
+    pub fn table_char_patch(&self, para_style: Option<&str>, run_style: Option<&str>, table: &CharProps) -> CharProps {
+        let chain = |defaults: CharProps| {
+            let (_, mut cp) = self.para_style_props_over(para_style.unwrap_or("Normal"), ParaProps::default(), defaults);
+            if let Some(cs) = run_style {
+                for s in self.chain(cs) {
+                    if s.kind == StyleKind::Character || s.kind == StyleKind::Paragraph {
+                        cp.overlay(&s.chr);
+                    }
+                }
+            }
+            cp
+        };
+        let mut with = self.default_chr.clone();
+        with.overlay(&CharProps { style: None, ..table.clone() });
+        diff_patch(&chain(with), &chain(self.default_chr.clone()))
+    }
+
     /// Resolve a paragraph's properties.
     pub fn resolve_para(&self, p: &ParaProps) -> ResolvedPara {
         self.resolve_para_in(p, None)
@@ -387,10 +442,46 @@ impl StyleSheet {
     }
 }
 
+/// The properties of `with` that differ from `without` (both fully resolved property sets of one
+/// kind), as a set of only those; empty when serialisation fails.
+fn diff_patch<T: Serialize + serde::de::DeserializeOwned + Default>(with: &T, without: &T) -> T {
+    let (Ok(serde_json::Value::Object(w)), Ok(serde_json::Value::Object(o))) = (serde_json::to_value(with), serde_json::to_value(without)) else {
+        return T::default();
+    };
+    let patch: serde_json::Map<String, serde_json::Value> = w.into_iter().filter(|(k, v)| o.get(k) != Some(v)).collect();
+    serde_json::from_value(serde_json::Value::Object(patch)).unwrap_or_default()
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
 
+    /// Superscript and subscript sizes as Word for Mac sets them (measured from digit advances in
+    /// its PDF export): 0.65 of the size, rounded to the half point.
+    #[test]
+    fn raised_and_lowered_text_is_sized_to_the_half_point() {
+        for (size, want) in [
+            (8.0, 5.0),
+            (9.0, 6.0),
+            (10.0, 6.5),
+            (11.0, 7.0),
+            (12.0, 8.0),
+            (13.0, 8.5),
+            (14.0, 9.0),
+            (16.0, 10.5),
+            (20.0, 13.0),
+            (24.0, 15.5),
+            (36.0, 23.5),
+        ] {
+            for vert_align in [VertAlign::Superscript, VertAlign::Subscript] {
+                let rc = ResolvedChar { size, vert_align, ..StyleSheet::builtin().resolve_char(None, &CharProps::default()) };
+                assert_eq!(rc.draw_size(), want, "{size} pt {vert_align:?}");
+            }
+        }
+        let rc = StyleSheet::builtin().resolve_char(None, &CharProps { size: Some(11.0), ..Default::default() });
+        assert_eq!(rc.draw_size(), 11.0, "baseline text keeps its size");
+    }
+
     /// #2036: only a character itself (direct formatting) has a symbol font of its own; a
     /// style never gives one to its text.
     #[test]
@@ -524,4 +615,15 @@ mod tests {
         assert_eq!(p.tabs.len(), 3);
         assert_eq!(p.tabs[0].pos, 100.0);
     }
+
+    /// Resolved formatting saved before paragraph padding existed still reads: no padding.
+    #[test]
+    fn resolved_para_without_padding_reads() {
+        let p = StyleSheet::builtin().resolve_para(&ParaProps::default());
+        let mut v = serde_json::to_value(&p).unwrap();
+        v.as_object_mut().unwrap().remove("padding");
+        let back: ResolvedPara = serde_json::from_value(v).unwrap();
+        assert_eq!(back.padding, [0.0; 4]);
+        assert_eq!(back, ResolvedPara { padding: [0.0; 4], ..p });
+    }
 }

crates/doc/src/section.rs

@@ -115,6 +115,40 @@ pub enum NumFormat {
     DollarText,
 }
 
+/// Chapter separator for page numbers (`w:chapSep`, ECMA-376 §17.18.12).
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash, Serialize, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub enum ChapterSep {
+    Hyphen,
+    Period,
+    Colon,
+    EmDash,
+    EnDash,
+}
+
+impl ChapterSep {
+    pub fn from_ooxml(value: &str) -> Option<Self> {
+        match value {
+            "hyphen" => Some(Self::Hyphen),
+            "period" => Some(Self::Period),
+            "colon" => Some(Self::Colon),
+            "emDash" => Some(Self::EmDash),
+            "enDash" => Some(Self::EnDash),
+            _ => None,
+        }
+    }
+
+    pub const fn ooxml(self) -> &'static str {
+        match self {
+            Self::Hyphen => "hyphen",
+            Self::Period => "period",
+            Self::Colon => "colon",
+            Self::EmDash => "emDash",
+            Self::EnDash => "enDash",
+        }
+    }
+}
+
 impl NumFormat {
     /// Every format, in the order ST_NumberFormat lists them (which is also the order of the
     /// binary .doc number format codes up to 0x27).
@@ -255,13 +289,14 @@ impl NumFormat {
     pub fn from_ooxml(s: &str) -> NumFormat {
         NumFormat::ALL.into_iter().find(|f| f.ooxml() == s).unwrap_or(NumFormat::Decimal)
     }
-    /// Format `n` (1-based). Formats WordCraft doesn't draw yet (Hebrew, Hindi, Thai,
-    /// Vietnamese, text amounts) show decimal, as do values past a sequence's end
-    /// (the 21st item of ①…⑳).
+    /// Format `n` (1-based). Formats WordCraft doesn't draw yet (Hebrew, Hindi, Thai letters
+    /// and counting, Vietnamese, text amounts) show decimal, as do values past a sequence's end
+    /// (the 21st item of ①…⑳). Thai numbers are the decimal number in Thai digits (๑, ๒ … ๑๐).
     pub fn format(self, n: u32) -> String {
         use crate::cjk_numbers as cjk;
         match self {
             NumFormat::DecimalZero => format!("{n:02}"),
+            NumFormat::ThaiNumbers => thai_digits(n),
             NumFormat::UpperRoman => roman(n).to_uppercase(),
             NumFormat::LowerRoman => roman(n),
             NumFormat::UpperLetter => letters(n).to_uppercase(),
@@ -315,7 +350,6 @@ impl NumFormat {
             | NumFormat::HindiNumbers
             | NumFormat::HindiCounting
             | NumFormat::ThaiLetters
-            | NumFormat::ThaiNumbers
             | NumFormat::ThaiCounting
             | NumFormat::BahtText
             | NumFormat::DollarText => n.to_string(),
@@ -323,6 +357,11 @@ impl NumFormat {
     }
 }
 
+/// `n` in Thai digits, ๐ (U+0E50) to ๙ (#5757).
+fn thai_digits(n: u32) -> String {
+    n.to_string().chars().filter_map(|c| c.to_digit(10).and_then(|d| char::from_u32(0x0E50 + d))).collect()
+}
+
 /// Chicago Manual of Style note marks: *, †, ‡, §, then doubled (**, ††, …) and so on.
 fn chicago(n: u32) -> String {
     const MARKS: [char; 4] = ['*', '\u{2020}', '\u{2021}', '\u{a7}'];
@@ -733,6 +772,13 @@ pub struct SectionProps {
     /// Distance from the page edge to the header / footer.
     pub header: f32,
     pub footer: f32,
+    /// The space kept between a header and the body, and between the body and a footer, points
+    /// (ODF's header and footer spacing; Word keeps none). The margin already holds the band and
+    /// its spacing, so it only moves the body once content and spacing reach past the margin.
+    #[serde(skip_serializing_if = "is_zero")]
+    pub header_gap: f32,
+    #[serde(skip_serializing_if = "is_zero")]
+    pub footer_gap: f32,
     pub gutter: f32,
     /// Mirror margins (inside/outside) — set from the document settings.
     pub columns: Columns,
@@ -742,6 +788,9 @@ pub struct SectionProps {
     pub title_page: bool,
     pub page_num_start: Option<u32>,
     pub page_num_format: NumFormat,
+    /// Chapter heading level and separator in `w:pgNumType`.
+    pub chapter_style: Option<u8>,
+    pub chapter_sep: Option<ChapterSep>,
     /// The file carries a `w:pgNumType` element that sets nothing else (no start, decimal
     /// format), so saving writes it back rather than dropping it.
     #[serde(skip_serializing_if = "std::ops::Not::not")]
@@ -772,6 +821,18 @@ pub struct SectionProps {
     pub footnote_pr: NoteProps,
     #[serde(skip_serializing_if = "NoteProps::is_empty")]
     pub endnote_pr: NoteProps,
+    /// First-page and other-page printer trays (`w:paperSrc`).
+    pub paper_src_first: Option<u16>,
+    pub paper_src_other: Option<u16>,
+    /// Section-specific suppression of endnotes (`w:noEndnote`).
+    pub no_endnote: Option<bool>,
+    /// Independent footnote column count (`w15:footnoteColumns`).
+    pub footnote_columns: Option<u32>,
+    /// Whether its columns end balanced (an ODF section's `text:dont-balance-text-columns`):
+    /// `None` is Word's rule, balanced when a continuous section follows; `Some(true)` also at
+    /// the end of the document, `Some(false)` never.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub balance_columns: Option<bool>,
 }
 
 impl Default for SectionProps {
@@ -788,6 +849,8 @@ impl Default for SectionProps {
             margin_right: 72.0,
             header: 36.0,
             footer: 36.0,
+            header_gap: 0.0,
+            footer_gap: 0.0,
             gutter: 0.0,
             columns: Columns::default(),
             headers: HeaderSet::default(),
@@ -795,6 +858,8 @@ impl Default for SectionProps {
             title_page: false,
             page_num_start: None,
             page_num_format: NumFormat::Decimal,
+            chapter_style: None,
+            chapter_sep: None,
             page_num_set: false,
             line_numbers: None,
             valign: PageVAlign::Top,
@@ -807,10 +872,19 @@ impl Default for SectionProps {
             doc_grid: None,
             footnote_pr: NoteProps::default(),
             endnote_pr: NoteProps::default(),
+            paper_src_first: None,
+            paper_src_other: None,
+            no_endnote: None,
+            footnote_columns: None,
+            balance_columns: None,
         }
     }
 }
 
+fn is_zero(v: &f32) -> bool {
+    *v == 0.0
+}
+
 fn is_horizontal(d: &crate::props::TextDirection) -> bool {
     !d.is_turned()
 }
@@ -930,6 +1004,18 @@ mod tests {
         assert_eq!(NumFormat::from_ooxml("klingon"), NumFormat::Decimal, "unknown formats are decimal");
     }
 
+    /// #5757: Thai numbers are the decimal number in Thai digits, every digit, at any length.
+    #[test]
+    fn thai_numbers_use_thai_digits() {
+        let labels: Vec<String> = [1, 2, 3, 9, 10, 11, 20, 105, 2569].iter().map(|n| NumFormat::ThaiNumbers.format(*n)).collect();
+        assert_eq!(labels.join(" "), "๑ ๒ ๓ ๙ ๑๐ ๑๑ ๒๐ ๑๐๕ ๒๕๖๙");
+        assert_eq!(NumFormat::ThaiNumbers.format(0), "๐");
+        assert_eq!(NumFormat::ThaiNumbers.format(u32::MAX), "๔๒๙๔๙๖๗๒๙๕");
+        assert_eq!(NumFormat::from_ooxml("thaiNumbers"), NumFormat::ThaiNumbers);
+        // Formats not drawn yet still show decimal.
+        assert_eq!(NumFormat::ThaiLetters.format(3), "3");
+    }
+
     #[test]
     fn russian_letters_follow_word() {
         // Word's own labels for a russianLower list: no ё, й, ъ or ь, then doubled letters.

crates/doc/src/settings.rs

@@ -46,6 +46,10 @@ pub struct CustomProp {
     /// Bookmark associated with this property (`linkTarget`), retained without evaluating it.
     #[serde(skip_serializing_if = "Option::is_none")]
     pub link_target: Option<String>,
+    /// Declared in an ODF document as a user field (§7.4.7 `text:user-field-decl`) rather than
+    /// a document property: written back as one.
+    #[serde(default, skip_serializing_if = "std::ops::Not::not")]
+    pub odf_user_field: bool,
 }
 
 /// Document properties (File › Info).
@@ -162,6 +166,23 @@ pub struct Settings {
     /// comments and tracked changes are saved without `w:date`.
     #[serde(skip_serializing_if = "std::ops::Not::not")]
     pub remove_date_time: bool,
+    /// The document was opened from ODF: the standard's rules that differ from Word's apply as
+    /// it is laid out.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub odf: Option<OdfLayout>,
+}
+
+/// A document opened from ODF, for the layout: rules of the standard that differ from Word's (a
+/// paragraph's padding takes room on a side without a border, XSL's box model) apply, and what
+/// the file says about laying it out.
+#[derive(Clone, Copy, Debug, Default, PartialEq, Eq, Serialize, Deserialize)]
+#[serde(default, rename_all = "camelCase")]
+pub struct OdfLayout {
+    /// Tab stops are measured from the paragraph's left indent, default ones too (LibreOffice's
+    /// `TabsRelativeToIndent` in the file; from the margin when it doesn't say). The document's
+    /// explicit stops are held from the margin either way; the default ones are counted from the
+    /// indent.
+    pub tabs_from_indent: bool,
 }
 
 /// A saved zoom: a percentage, or a fit mode (the percentage it last came to).
@@ -307,6 +328,7 @@ impl Default for Settings {
             zoom: None,
             remove_personal_info: false,
             remove_date_time: false,
+            odf: None,
         }
     }
 }

crates/doc/src/smart_art.rs

@@ -12,6 +12,8 @@
 
 use serde::{Deserialize, Serialize};
 
+use crate::props::Rgb;
+
 /// Most items in a SmartArt graphic.
 pub const MAX_ITEMS: usize = 100;
 /// Longest item text, characters.
@@ -198,11 +200,14 @@ pub struct SmartArtItem {
     pub text: String,
     #[serde(default)]
     pub level: u8,
+    /// Fill of the item's shape in place of the style's colour.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub color: Option<Rgb>,
 }
 
 impl SmartArtItem {
     pub fn new(text: impl Into<String>, level: u8) -> SmartArtItem {
-        SmartArtItem { text: text.into(), level }
+        SmartArtItem { text: text.into(), level, color: None }
     }
 }
 
@@ -257,6 +262,35 @@ impl SmartArtSpec {
         }
     }
 
+    /// Whether any item has its own colour.
+    pub fn has_custom_colors(&self) -> bool {
+        self.items.iter().any(|it| it.color.is_some())
+    }
+
+    /// Give every item the style's colour again.
+    pub fn clear_custom_colors(&mut self) {
+        for it in &mut self.items {
+            it.color = None;
+        }
+    }
+
+    /// The theme accent (1–6) the style gives item `i`; `None` when the item has no shape of
+    /// its own (it is text inside another item's shape).
+    pub fn style_accent(&self, i: usize) -> Option<usize> {
+        let it = self.items.get(i)?;
+        let tops = || self.items.iter().take(i).filter(|o| o.level == 0).count();
+        match self.layout {
+            SmartArtLayout::Hierarchy => Some(self.colors.accent(i)),
+            SmartArtLayout::Radial if i == 0 => Some(self.colors.accent(0)),
+            SmartArtLayout::Radial => {
+                let parents = self.parents();
+                let spoke = |j: usize| j > 0 && matches!(parents.get(j), Some(Some(0)) | Some(None));
+                spoke(i).then(|| self.colors.accent((1..i).filter(|j| spoke(*j)).count() + 1))
+            }
+            _ => (it.level == 0).then(|| self.colors.accent(tops())),
+        }
+    }
+
     /// The graphic after `sanitize`.
     pub fn sanitized(mut self) -> SmartArtSpec {
         self.sanitize();

crates/doc/src/styles.rs

@@ -232,6 +232,40 @@ impl PartSet {
     }
 }
 
+/// A cell's borders from the table style's parts it is in, later parts winning. Row parts give
+/// their left/right edges at the table's sides and `inside_v` between cells; column parts their
+/// top/bottom edges at the table's top and bottom and `between` between rows; corner cells their
+/// four edges.
+pub fn part_borders(parts: &[(TablePart, Borders)], (ri, nrows): (usize, usize), (first_g, last_g): (bool, bool)) -> Borders {
+    let mut out = Borders::default();
+    for (part, b) in parts {
+        let b = *b;
+        let mapped = if part.is_row() {
+            Borders {
+                top: b.top,
+                bottom: b.bottom,
+                left: if first_g { b.left } else { b.inside_v },
+                right: if last_g { b.right } else { b.inside_v },
+                between: None,
+                inside_v: None,
+            }
+        } else if part.is_column() {
+            Borders {
+                top: if ri == 0 { b.top } else { b.between },
+                bottom: if ri + 1 >= nrows { b.bottom } else { b.between },
+                left: b.left,
+                right: b.right,
+                between: None,
+                inside_v: None,
+            }
+        } else {
+            Borders { between: None, inside_v: None, ..b }
+        };
+        out.overlay(&mapped);
+    }
+    out
+}
+
 /// What one table style part gives its cells.
 #[derive(Clone, Debug, PartialEq, Serialize, Deserialize, Default)]
 #[serde(default, rename_all = "camelCase")]
@@ -932,6 +966,31 @@ fn table_styles() -> Vec<Style> {
 #[cfg(test)]
 mod tests {
     use super::*;
+    use crate::resolve::ResolvedPara;
+
+    #[test]
+    fn adjust_right_indent_inherits_and_can_be_overridden() {
+        let resolved = StyleSheet::empty().resolve_para(&ParaProps::default());
+        assert!(resolved.adjust_right_indent);
+
+        let mut older_json = serde_json::to_value(&resolved).unwrap();
+        older_json.as_object_mut().unwrap().remove("adjustRightIndent");
+        assert!(serde_json::from_value::<ResolvedPara>(older_json).unwrap().adjust_right_indent);
+
+        let mut s = StyleSheet::empty();
+        s.styles.push(Style {
+            id: "Normal".into(),
+            name: "Normal".into(),
+            para: ParaProps { adjust_right_indent: Some(true), ..Default::default() },
+            ..Default::default()
+        });
+        assert!(s.resolve_para(&ParaProps::default()).adjust_right_indent);
+        assert!(!s.resolve_para(&ParaProps { adjust_right_indent: Some(false), ..Default::default() }).adjust_right_indent);
+
+        let json = serde_json::to_value(ParaProps { adjust_right_indent: Some(false), ..Default::default() }).unwrap();
+        assert_eq!(json["adjustRightIndent"], false);
+        assert_eq!(serde_json::from_value::<ParaProps>(json).unwrap().adjust_right_indent, Some(false));
+    }
 
     #[test]
     fn builtins_present() {

crates/doc/src/symbol_fonts.rs

@@ -19,8 +19,8 @@
 //!
 //! Tables are indexed by code − 0x20; 0 = no Unicode equivalent.
 
-/// The symbol fonts this module knows, by lowercase family name.
-fn table(family: &str) -> Option<&'static [u32; 224]> {
+/// The symbol table for a single family name, by lowercase family name.
+fn single_table(family: &str) -> Option<&'static [u32; 224]> {
     let f = family.trim();
     if f.eq_ignore_ascii_case("symbol") || f.eq_ignore_ascii_case("symbol (pcl6)") {
         Some(&SYMBOL)
@@ -37,6 +37,11 @@ fn table(family: &str) -> Option<&'static [u32; 224]> {
     }
 }
 
+/// The symbol fonts this module knows, by lowercase family name.
+fn table(family: &str) -> Option<&'static [u32; 224]> {
+    if family.contains(';') { family.split(';').map(str::trim).filter(|s| !s.is_empty()).find_map(single_table) } else { single_table(family) }
+}
+
 /// Whether `family` is one of Word's symbol fonts (Symbol, Wingdings, Wingdings 2, Wingdings 3,
 /// Webdings), whose characters [`symbol_char`] maps.
 pub fn is_symbol_font(family: &str) -> bool {

crates/doc/src/table.rs

@@ -71,10 +71,12 @@ pub struct Table {
 /// Hard limits that keep hostile input from exhausting memory.
 pub const MAX_ROWS: usize = 32_767;
 pub const MAX_COLS: usize = 63;
-/// Most tables nested one inside another: the depth the DOCX reader and writer keep. Pasting
-/// and inserting stop there too, so no edit builds a document deep enough to overflow the stack
-/// when it is saved, copied or dropped (#1596).
-pub const MAX_TABLE_DEPTH: usize = 24;
+/// Most tables nested one inside another. Pasting and inserting stop there, so no edit builds a
+/// document deep enough to overflow the stack when it is saved, copied or dropped (#1596), or
+/// deeper than a `.json` save reads back: serde_json reads 127 nested containers, a table level
+/// takes six, and the deepest cell of a header, note or text box needs up to 15 more for its
+/// paragraph and the richest content measured (#3618). The DOCX reader and writer keep 24.
+pub const MAX_TABLE_DEPTH: usize = 18;
 
 impl Table {
     /// Grid columns `lo..hi` that [`Table::slice`] copies for `rows`/`cols` (widened over merged cells),

crates/doc/src/tests.rs

@@ -1,4 +1,5 @@
 use super::*;
+use std::sync::Arc;
 
 mod note_conversion;
 
@@ -18,6 +19,25 @@ fn word_count() {
     assert_eq!(count_words(" -- "), 0);
 }
 
+/// #2797: the status bar's character counts, worded like the Word Count dialog's: characters
+/// with spaces leave the paragraph breaks out, characters without spaces leave every
+/// whitespace character out.
+#[test]
+fn character_counts() {
+    assert_eq!(count_chars(""), 0);
+    assert_eq!(count_chars("Hello"), 5);
+    // Paragraph breaks join the paragraphs but aren't characters.
+    assert_eq!(count_chars("Hello\nworld\r\n"), 10);
+    assert_eq!(count_chars("a\tb"), 3, "a tab is a character");
+    // Every kind of whitespace goes, spaces, tabs and (non-ASCII) spaces between words.
+    assert_eq!(count_chars_no_spaces("Hello world"), 10);
+    assert_eq!(count_chars_no_spaces("a\tb c\u{a0}d"), 4, "including a non-breaking space");
+    assert_eq!(count_chars_no_spaces(" \n\r\t "), 0);
+    // Accented letters and emoji count as one character each, as they are.
+    assert_eq!(count_chars("café 🚀"), 6);
+    assert_eq!(count_chars_no_spaces("café 🚀"), 5);
+}
+
 /// #4547: each Han ideograph and kana counts as a word; punctuation and Hangul spacing don't change.
 #[test]
 fn word_count_counts_each_chinese_and_japanese_character() {
@@ -35,8 +55,8 @@ fn word_count_counts_each_chinese_and_japanese_character() {
 #[test]
 fn word_count_includes_numbered_list_labels_but_not_bullets() {
     let mut d = Document::new();
-    let num_id = d.numbering.add_list(ListKind::Numbered);
-    let bul_id = d.numbering.add_list(ListKind::Bullet);
+    let num_id = Arc::make_mut(&mut d.numbering).add_list(ListKind::Numbered);
+    let bul_id = Arc::make_mut(&mut d.numbering).add_list(ListKind::Bullet);
 
     let mut p1 = Paragraph::with_text("Apple", CharProps::default());
     p1.props.numbering = Some(props::NumRef { num: num_id, level: 0 });
@@ -60,9 +80,9 @@ fn word_count_includes_numbered_list_labels_but_not_bullets() {
 #[test]
 fn word_count_includes_numbered_labels_with_nothing_suffix() {
     let mut d = Document::new();
-    let num_id = d.numbering.add_list(ListKind::Numbered);
+    let num_id = Arc::make_mut(&mut d.numbering).add_list(ListKind::Numbered);
     let aid = d.numbering.num(num_id).expect("list num exists").abstract_id;
-    let a = d.numbering.abstracts.iter_mut().find(|a| a.id == aid).expect("abstract num exists");
+    let a = Arc::make_mut(&mut d.numbering).abstracts.iter_mut().find(|a| a.id == aid).expect("abstract num exists");
     a.levels[0].suffix = numbering::LevelSuffix::Nothing;
 
     let mut p1 = Paragraph::with_text("Apple", CharProps::default());
@@ -81,9 +101,9 @@ fn word_count_includes_numbered_labels_with_nothing_suffix() {
 #[test]
 fn word_count_keeps_a_cjk_list_label_one_word() {
     let mut d = Document::new();
-    let num_id = d.numbering.add_list(ListKind::Numbered);
+    let num_id = Arc::make_mut(&mut d.numbering).add_list(ListKind::Numbered);
     let aid = d.numbering.num(num_id).expect("list num exists").abstract_id;
-    let a = d.numbering.abstracts.iter_mut().find(|a| a.id == aid).expect("abstract num exists");
+    let a = Arc::make_mut(&mut d.numbering).abstracts.iter_mut().find(|a| a.id == aid).expect("abstract num exists");
     a.levels[0].format = section::NumFormat::ChineseCounting;
     a.levels[0].text = "第%1章".into();
     a.levels[0].start = 12;
@@ -187,7 +207,7 @@ fn parts_and_lists_after_the_largest_ids_take_free_ones() {
     assert_eq!(d.add_part(PartKind::Footer, Vec::new()), 3);
     assert!(parts.iter().all(|(k, p)| d.parts.get(k) == Some(p)));
 
-    let n = &mut d.numbering;
+    let n = Arc::make_mut(&mut d.numbering);
     n.add_list(ListKind::Bullet);
     n.add_list(ListKind::Bullet);
     n.abstracts[1].id = u32::MAX;
@@ -229,11 +249,11 @@ fn same_content_sees_every_change() {
     assert!(!d.same_content(&e), "the formatting changed");
     // The styles changed.
     let mut e = d.clone();
-    e.styles.get_mut("Normal").unwrap().chr.bold = Some(true);
+    Arc::make_mut(&mut e.styles).get_mut("Normal").unwrap().chr.bold = Some(true);
     assert!(!d.same_content(&e), "a style changed");
     // The lists changed.
     let mut e = d.clone();
-    e.numbering.add_list(ListKind::Numbered);
+    Arc::make_mut(&mut e.numbering).add_list(ListKind::Numbered);
     assert!(!d.same_content(&e), "a list was added");
     // The settings changed.
     let mut e = d.clone();
@@ -452,9 +472,9 @@ fn box_para(story: u32, id: u32) -> Blocks {
 fn adopting_relinks_lists_in_deeply_nested_tables() {
     // As #1339: a list paragraph 20 tables down takes the list it is brought in with.
     let mut ours = Document::from_text("ours");
-    let our_list = ours.numbering.add_list(numbering::ListKind::Bullet);
+    let our_list = Arc::make_mut(&mut ours.numbering).add_list(numbering::ListKind::Bullet);
     let mut theirs = Document::from_text("theirs");
-    let list = theirs.numbering.add_list(numbering::ListKind::Bullet);
+    let list = Arc::make_mut(&mut theirs.numbering).add_list(numbering::ListKind::Bullet);
     assert_eq!(list, our_list);
     let mut p = Paragraph::with_text("item", CharProps::default());
     p.props.numbering = Some(props::NumRef { num: list, level: 0 });

crates/doc/src/tests/note_conversion.rs

@@ -102,18 +102,6 @@ fn dense(n: usize) -> Document {
     d
 }
 
-// Thread CPU excludes scheduler delays. Missing or nonadvancing counters provide no timing
-// evidence; the semantic assertions still run on those platforms.
-fn cpu_cost(f: impl FnOnce()) -> Option<f64> {
-    let ran = || std::fs::read_to_string("/proc/thread-self/schedstat").ok()?.split_whitespace().next()?.parse::<u64>().ok();
-    let before = ran();
-    f();
-    match (before, ran()) {
-        (Some(a), Some(b)) if b > a => Some((b - a) as f64 * 1e-9),
-        _ => None,
-    }
-}
-
 #[test]
 fn conversion_work_scales_with_reference_and_formatting_runs() {
     let (small, big) = (dense(2_000), dense(16_000));
@@ -128,18 +116,13 @@ fn conversion_work_scales_with_reference_and_formatting_runs() {
     };
     convert(&small);
     convert(&big);
-    let (mut t1, mut t8) = (f64::INFINITY, f64::INFINITY);
-    for _ in 0..3 {
-        let a = cpu_cost(|| (0..8).for_each(|_| convert(&small)));
-        let b = cpu_cost(|| convert(&big));
-        if let (Some(a), Some(b)) = (a, b) {
-            t1 = t1.min(a / 8.0);
-            t8 = t8.min(b);
-        }
-    }
-    if t1.is_finite() && t8.is_finite() {
-        let ratio = t8 / t1;
-        eprintln!("note conversion CPU: small={t1:.6}s big={t8:.6}s ratio={ratio:.2}");
-        assert!(ratio < 24.0, "8x the references and runs cost {ratio:.2}x CPU");
-    }
+    // The characters and runs the conversion walks (`crate::work`), not a timing: about 8x for 8x
+    // the references and runs when they are walked together, about 64x when each run rescans the
+    // references.
+    let s1 = crate::work::count(|| convert(&small));
+    let s8 = crate::work::count(|| convert(&big));
+    let ratio = s8 as f64 / s1.max(1) as f64;
+    eprintln!("note conversion steps: {s1} -> {s8}, {ratio:.1}x for 8x references and runs");
+    assert!(s1 > 0, "no steps counted, so the walk this test watches no longer counts them (crate::work)");
+    assert!(ratio <= 12.0, "8x the references and runs cost {ratio:.1}x the steps ({s1} -> {s8})");
 }

crates/doc/src/text_walk.rs

@@ -35,6 +35,18 @@ fn is_cjk_word(c: char) -> bool {
     ) && c.is_alphanumeric()
 }
 
+/// Characters in `s`, as the Word Count dialog counts them ("Characters (with spaces)"):
+/// paragraph breaks aren't characters (#2797).
+pub fn count_chars(s: &str) -> usize {
+    s.chars().filter(|c| *c != '\n' && *c != '\r').count()
+}
+
+/// Characters in `s` ignoring every whitespace character, as the Word Count dialog counts them
+/// ("Characters (no spaces)"; #2797).
+pub fn count_chars_no_spaces(s: &str) -> usize {
+    s.chars().filter(|c| !c.is_whitespace()).count()
+}
+
 /// How deep in nested tables [`Document::para_paths`] goes: cells below this are skipped.
 pub(crate) const PARA_WALK_DEPTH: usize = 16;
 
@@ -51,6 +63,7 @@ pub(crate) fn walk(bl: &Blocks, prefix: &mut Vec<u32>, out: &mut Vec<Path>, dept
         return;
     }
     for (i, b) in bl.iter().enumerate() {
+        crate::work::steps(1);
         prefix.push(i as u32);
         match &**b {
             Block::Para(_) => out.push(Path(prefix.clone())),

crates/doc/src/work.rs

@@ -0,0 +1,30 @@
+//! A count of the characters a paragraph's object lookups walk, the blocks
+//! [`crate::Document::para_paths`] visits, and each block lookup by path with the table levels it
+//! descends, for the linear-time tests of the crates that use them: an input 8x the size must cost
+//! about 8x the steps, which a busy machine can't skew the way it skews a timing. Counted only in
+//! tests (this crate's, and other crates' through the `test-support` feature, which they turn on
+//! as a dev-dependency); in the app the count is nothing.
+
+#[cfg(any(test, feature = "test-support"))]
+thread_local! {
+    /// The steps counted on this thread.
+    static STEPS: std::cell::Cell<u64> = const { std::cell::Cell::new(0) };
+}
+
+/// Count `n` steps of a walk (nothing outside tests). Public for the walks other crates do over a
+/// paragraph's text themselves (Markdown and HTML comment export, for one).
+#[inline(always)]
+pub fn steps(n: usize) {
+    #[cfg(any(test, feature = "test-support"))]
+    STEPS.with(|s| s.set(s.get().saturating_add(n as u64)));
+    #[cfg(not(any(test, feature = "test-support")))]
+    let _ = n;
+}
+
+/// The steps `f` takes.
+#[cfg(any(test, feature = "test-support"))]
+pub fn count(f: impl FnOnce()) -> u64 {
+    let before = STEPS.with(std::cell::Cell::get);
+    f();
+    STEPS.with(std::cell::Cell::get).saturating_sub(before)
+}

crates/doc/tests/empty_paragraph_merge.rs

@@ -0,0 +1,158 @@
+use std::sync::Arc;
+use wordcraft_doc::props::{Align, CharProps, ParaProps};
+use wordcraft_doc::{Block, Document, Paragraph, Path, Pos, StoryRef, Table, para_block};
+
+/// Build an empty heading followed by explicitly formatted, mixed-run body text.
+fn fixture(first: &str, style: Option<&str>) -> Document {
+    let mut d = Document::from_text(&format!("{first}\nBody paragraph."));
+    let p = d.para_mut(StoryRef::Body, &Path::top(0)).unwrap();
+    p.props.style = Some("Heading1".into());
+    p.props.outline_level = Some(0);
+    p.mark = CharProps { size: Some(40.0), bold: Some(true), ..Default::default() };
+    let p = d.para_mut(StoryRef::Body, &Path::top(1)).unwrap();
+    p.props = ParaProps {
+        style: style.map(str::to_string),
+        align: Some(Align::Right),
+        indent_left: Some(18.0),
+        space_after: Some(7.0),
+        keep_next: Some(false),
+        ..Default::default()
+    };
+    p.mark = CharProps { size: Some(11.0), bold: Some(false), lang: Some("en-GB".into()), ..Default::default() };
+    p.format(0, 4, &|c| c.italic = Some(true)).unwrap();
+    d
+}
+
+/// Compare paragraph, mark and inline-run formatting independently of text.
+fn assert_format(p: &Paragraph, expected: &Paragraph) {
+    assert_eq!(p.props, expected.props, "the following paragraph keeps all its properties");
+    assert_eq!(p.mark, expected.mark, "the empty heading's mark must not replace the following mark");
+    assert_eq!(serde_json::to_value(&p.runs).unwrap(), serde_json::to_value(&expected.runs).unwrap());
+}
+
+/// Deleting an empty paragraph preserves the following style, including implicit Normal.
+#[test]
+fn deleting_an_empty_paragraph_keeps_the_following_format() {
+    for first_style in ["Heading1", "Heading3", "Normal"] {
+        for style in [None, Some("Normal"), Some("Heading2")] {
+            let mut d = fixture("", style);
+            d.para_mut(StoryRef::Body, &Path::top(0)).unwrap().props.style = Some(first_style.into());
+            let expected = d.para_at(&Pos::body(1, 0)).unwrap().clone();
+            assert_eq!(d.delete_range(&Pos::body(0, 0), &Pos::body(1, 0)).unwrap(), Pos::body(0, 0));
+            let p = d.para_at(&Pos::body(0, 0)).unwrap();
+            assert_eq!(p.text, expected.text);
+            assert_format(p, &expected);
+        }
+    }
+}
+
+/// A selection spanning several paragraphs keeps the surviving tail format in both directions.
+#[test]
+fn an_empty_start_preserves_the_last_paragraph_over_a_multi_paragraph_selection() {
+    for backwards in [false, true] {
+        let mut d = fixture("", None);
+        d.body.insert(1, para_block(Paragraph::with_text("Removed middle paragraph", Default::default()).styled("Heading2")));
+        let mut expected = d.para_at(&Pos::body(2, 0)).unwrap().clone();
+        expected.delete(0, 5).unwrap();
+        let (a, b) = (Pos::body(0, 0), Pos::body(2, 5));
+        d.delete_range(if backwards { &b } else { &a }, if backwards { &a } else { &b }).unwrap();
+        assert_eq!(d.body.len(), 1);
+        let p = d.para_at(&Pos::body(0, 0)).unwrap();
+        assert_eq!(p.text, "paragraph.");
+        assert_format(p, &expected);
+    }
+}
+
+/// A control whose closing marker survives must move across the join without replacing tail formatting.
+#[test]
+fn an_empty_start_keeps_a_control_from_a_deleted_middle_paragraph() {
+    use wordcraft_doc::control::ContentControl;
+    use wordcraft_doc::{InlineObject, Revision, RevisionKind};
+
+    let mut d = fixture("", Some("Normal"));
+    d.body.insert(1, para_block(Paragraph::with_text("Removed", Default::default())));
+    d.body.push(para_block(Paragraph::with_text("After", Default::default())));
+    d.revisions.push(Revision { kind: RevisionKind::Insert, ..Default::default() });
+    d.revisions.push(Revision { kind: RevisionKind::Delete, ..Default::default() });
+    let p = d.para_mut(StoryRef::Body, &Path::top(2)).unwrap();
+    p.mark.ins = Some(0);
+    p.mark.del = Some(1);
+    let expected = p.clone();
+    let marker = CharProps { bold: Some(true), ins: Some(0), ..Default::default() };
+    d.insert_object(&Pos::body(1, 0), InlineObject::ControlStart { control: Box::new(ContentControl::default()) }, &marker).unwrap();
+    d.insert_object(&Pos::body(3, 5), InlineObject::ControlEnd, &Default::default()).unwrap();
+    assert_eq!(d.control_ranges(StoryRef::Body).len(), 1);
+
+    d.delete_range(&Pos::body(0, 0), &Pos::body(2, 0)).unwrap();
+    let p = d.para_at(&Pos::body(0, 0)).unwrap();
+    assert_eq!(p.props, expected.props);
+    assert_eq!(p.mark, expected.mark, "tail revision IDs and formatting survive the moved marker");
+    assert_eq!(p.props_of_char(0), &marker);
+    assert!(matches!(p.objects.as_slice(), [InlineObject::ControlStart { .. }]));
+    assert_eq!(p.text, format!("\u{fffc}{}", expected.text));
+    let ranges = d.control_ranges(StoryRef::Body);
+    assert_eq!(ranges.len(), 1);
+    assert_eq!(ranges[0].start, Pos::body(0, 0));
+    assert_eq!(ranges[0].end, Pos::body(1, 5));
+    assert!(!d.has_unbalanced_controls());
+}
+
+/// The empty-paragraph rule applies in header stories and table-cell containers.
+#[test]
+fn empty_paragraph_merges_work_in_a_header_and_a_table_cell() {
+    for cell in [false, true] {
+        let mut d = fixture("", Some("Normal"));
+        let expected = d.para_at(&Pos::body(1, 0)).unwrap().clone();
+        let (story, a, b) = if cell {
+            let mut table = Table::new(1, 1, 300.0);
+            table.rows[0].cells[0].blocks = d.body.clone();
+            d.body = vec![Arc::new(Block::Table(table)), para_block(Paragraph::new())];
+            (StoryRef::Body, Path(vec![0, 0, 0, 0]), Path(vec![0, 0, 0, 1]))
+        } else {
+            let id = d.add_part(wordcraft_doc::PartKind::Header, d.body.clone());
+            (StoryRef::Part(id), Path::top(0), Path::top(1))
+        };
+        d.delete_range(&Pos { story, path: a.clone(), off: 0 }, &Pos { story, path: b, off: 0 }).unwrap();
+        assert_format(d.para(story, &a).unwrap(), &expected);
+    }
+}
+
+/// Absorbing an empty section keeps the following format and inherited header.
+#[test]
+fn deleting_an_empty_section_paragraph_keeps_format_and_linked_headers() {
+    let mut d = fixture("", None);
+    let expected = d.para_at(&Pos::body(1, 0)).unwrap().clone();
+    let header = d.add_part(wordcraft_doc::PartKind::Header, vec![para_block(Paragraph::with_text("Header", Default::default()))]);
+    let mut section = d.last_section.clone();
+    section.headers.default = Some(header);
+    d.para_mut(StoryRef::Body, &Path::top(0)).unwrap().section = Some(Box::new(section));
+    d.last_section.margin_left = 96.0;
+    d.delete_range(&Pos::body(0, 0), &Pos::body(1, 0)).unwrap();
+    assert_format(d.para_at(&Pos::body(0, 0)).unwrap(), &expected);
+    assert!(d.para_at(&Pos::body(0, 0)).unwrap().section.is_none());
+    assert_eq!(d.last_section.headers.default, Some(header));
+    assert_eq!(d.last_section.margin_left, 96.0);
+}
+
+/// Deleting actual text continues to use the first paragraph merge rule.
+#[test]
+fn joining_nonempty_paragraphs_keeps_the_existing_first_paragraph_rule() {
+    for off in [0, 7] {
+        let mut d = fixture("Heading", None);
+        let expected = d.para_at(&Pos::body(0, 0)).unwrap().props.clone();
+        d.delete_range(&Pos::body(0, off), &Pos::body(1, 0)).unwrap();
+        assert_eq!(d.para_at(&Pos::body(0, 0)).unwrap().props, expected);
+    }
+}
+
+/// Removing a soft line break must not change paragraph formatting.
+#[test]
+fn a_soft_break_is_not_an_empty_paragraph_merge() {
+    let mut d = fixture("", None);
+    let p = d.para_mut(StoryRef::Body, &Path::top(0)).unwrap();
+    p.insert_text(0, "\nHeading", &Default::default()).unwrap();
+    let expected = p.props.clone();
+    d.delete_range(&Pos::body(0, 0), &Pos::body(0, 1)).unwrap();
+    assert_eq!(d.body.len(), 2);
+    assert_eq!(d.para_at(&Pos::body(0, 0)).unwrap().props, expected);
+}

crates/docbin/src/drawing.rs

@@ -185,7 +185,7 @@ fn leaf(sp: &Sp, w: f32, h: f32, float: Float, b: &Build, pics: &mut Pictures) -
     let p = &sp.props;
     if let Some(pib) = p.pib {
         let media = pics.bse(pib, b.bstore, b.delay)?;
-        return Some(InlineObject::Image { media, w, h, alt: p.alt.clone(), float, crop: p.crop, ole: None });
+        return Some(InlineObject::Image { media, w, h, alt: p.alt.clone(), float, crop: p.crop, fx: Default::default(), ole: None });
     }
     let story = b.stories.get(&sp.spid).copied().filter(|_| sp.textbox || sp.spt == 202);
     let mut kind = match sp.spt {
@@ -211,7 +211,11 @@ fn leaf(sp: &Sp, w: f32, h: f32, float: Float, b: &Build, pics: &mut Pictures) -
         kind = ShapeKind::TextBox;
     }
     let line = kind == ShapeKind::Line;
-    let fill = if line || p.filled == Some(false) { None } else { Some(Fill::solid(p.fill_color.unwrap_or(Rgb::WHITE))) };
+    // Brackets and braces (85 to 88, bracket and brace pairs 185, 186) are unfilled unless
+    // `fFilled` says otherwise, as Word's VML shapetypes for them say (`filled="f"`). They are
+    // drawn as their box here, which a default fill would make an opaque block over the text.
+    let unfilled = p.filled == Some(false) || (matches!(sp.spt, 85..=88 | 185 | 186) && p.filled != Some(true));
+    let fill = if line || unfilled { None } else { Some(Fill::solid(p.fill_color.unwrap_or(Rgb::WHITE))) };
     let stroke = if p.line == Some(false) { None } else { Some(p.line_color.unwrap_or(Rgb::BLACK)) };
     // Line widths are EMUs; Word's default is 0.75 pt.
     let stroke_width = p.line_width.map_or(0.75, |e| (e as f32 / 12_700.0).clamp(0.0, 100.0));

crates/docbin/src/fkp.rs

@@ -7,9 +7,26 @@ use crate::DocbinError;
 /// Largest bin table (entries) we accept.
 const MAX_BINS: usize = 100_000;
 
+/// A Chpx grpprl and the FCs `[lo, hi)` for which it is the one [`Bins::chpx`] returns.
+#[derive(Clone, Copy)]
+pub(crate) struct ChpxRun<'w> {
+    pub(crate) grpprl: &'w [u8],
+    lo: u32,
+    hi: u32,
+}
+
+impl ChpxRun<'_> {
+    pub(crate) fn contains(&self, fc: u32) -> bool {
+        self.lo <= fc && fc < self.hi
+    }
+}
+
 /// A parsed `PlcBte*`: FC breakpoints and the FKP page each range lives in.
 pub(crate) struct Bins {
     fcs: Vec<u32>,
+    /// `fcs` ascends, so the bin of an FC holds every FC up to the next breakpoint (a corrupt
+    /// table that does not gets no cached ranges, [`ChpxRun`]).
+    sorted: bool,
     /// Page number (× 512 = byte offset in the WordDocument stream).
     pns: Vec<u32>,
 }
@@ -20,7 +37,7 @@ impl Bins {
         let fc = fc as usize;
         let lcb = lcb as usize;
         if lcb < 4 || !(lcb - 4).is_multiple_of(8) {
-            return Ok(Bins { fcs: Vec::new(), pns: Vec::new() });
+            return Ok(Bins { fcs: Vec::new(), sorted: true, pns: Vec::new() });
         }
         let end = fc.checked_add(lcb).ok_or_else(|| DocbinError::Malformed("bin table overflow".into()))?;
         let plc = table.get(fc..end).ok_or_else(|| DocbinError::Malformed("bin table outside the Table stream".into()))?;
@@ -38,36 +55,58 @@ impl Bins {
             // PnBte* keeps the page number in its low bits; the rest must be zero.
             pns.push(at(n + 1 + i).ok_or_else(|| DocbinError::Malformed("truncated bin aPn".into()))? & 0x003F_FFFF);
         }
-        Ok(Bins { fcs, pns })
+        let sorted = fcs.windows(2).all(|w| w.first() <= w.get(1));
+        Ok(Bins { fcs, sorted, pns })
     }
 
-    /// The page containing `fc`, as a slice of the WordDocument stream.
-    fn page<'w>(&self, word: &'w [u8], fc: u32) -> Option<&'w [u8]> {
+    /// The page containing `fc`, as a slice of the WordDocument stream, and the FC range
+    /// `[lo, hi)` of its bin (`hi` is `u32::MAX` for the last; an empty range when the table
+    /// is not ascending, so nothing is cached from it).
+    fn page<'w>(&self, word: &'w [u8], fc: u32) -> Option<(&'w [u8], u32, u32)> {
         let i = self.fcs.partition_point(|&f| f <= fc).checked_sub(1)?;
         let pn = *self.pns.get(i)?;
         let at = pn.checked_mul(512)? as usize;
-        word.get(at..at.checked_add(512)?)
+        let page = word.get(at..at.checked_add(512)?)?;
+        if !self.sorted {
+            return Some((page, 1, 0));
+        }
+        let lo = self.fcs.get(i).copied().unwrap_or(0);
+        let hi = self.fcs.get(i + 1).copied().unwrap_or(u32::MAX);
+        Some((page, lo, hi))
     }
 
     /// The grpprl of the Chpx covering `fc` ([MS-DOC] ChpxFkp: rgfc u32s, rgb offsets ×2,
     /// crun in the last byte).
     pub(crate) fn chpx<'w>(&self, word: &'w [u8], fc: u32) -> &'w [u8] {
-        let Some(page) = self.page(word, fc) else { return &[] };
-        let Some(&crun) = page.get(511) else { return &[] };
+        self.chpx_run(word, fc).grpprl
+    }
+
+    /// [`Bins::chpx`] with the FC range over which the same grpprl is the answer, so a walk
+    /// over consecutive characters looks it up once per run, not once per character.
+    pub(crate) fn chpx_run<'w>(&self, word: &'w [u8], fc: u32) -> ChpxRun<'w> {
+        let none = ChpxRun { grpprl: &[], lo: 1, hi: 0 };
+        let Some((page, page_lo, page_hi)) = self.page(word, fc) else { return none };
+        let Some(&crun) = page.get(511) else { return none };
         let crun = crun.min(0x65) as usize;
         let rgfc = |i: usize| -> Option<u32> { page.get(i * 4..i * 4 + 4).map(|b| u32::from_le_bytes([b[0], b[1], b[2], b[3]])) };
         // rgb array sits right after rgfc (crun+1 u32s); each byte is a half-offset into the
         // region between rgb's end and crun.
         let rgb_at = (crun + 1) * 4;
+        // The last run starting at or before `fc`. It stays the answer for every FC from its own
+        // start up to the smallest start after it (also in a page whose starts do not ascend).
         let j = (0..crun).rev().find(|&j| rgfc(j).is_some_and(|f| f <= fc));
-        let Some(j) = j else { return &[] };
-        let Some(&off) = page.get(rgb_at + j) else { return &[] };
+        let first_after = j.map_or(0, |j| j + 1);
+        let hi = (first_after..crun).filter_map(rgfc).min().map_or(page_hi, |m| m.min(page_hi));
+        let lo = j.and_then(rgfc).map_or(page_lo, |f| f.max(page_lo));
+        let run = |grpprl: &'w [u8]| ChpxRun { grpprl, lo, hi };
+        let Some(j) = j else { return run(&[]) };
+        let Some(&off) = page.get(rgb_at + j) else { return run(&[]) };
         if off == 0 {
-            return &[];
+            return run(&[]);
         }
         let at = off as usize * 2;
-        let Some(&cb) = page.get(at) else { return &[] };
-        page.get(at + 1..at + 1 + cb as usize).unwrap_or(&[])
+        let Some(&cb) = page.get(at) else { return run(&[]) };
+        run(page.get(at + 1..at + 1 + cb as usize).unwrap_or(&[]))
     }
 
     /// The (istd, grpprl) of the Papx of the paragraph whose mark is just before `fc`
@@ -75,7 +114,7 @@ impl Bins {
     /// entries (1 offset byte + 12 PHE bytes), crun in the last byte; the PAPX itself is a
     /// PapxInFkp with the cb/cb' size quirk). Query with the FC of the character after the mark.
     pub(crate) fn papx<'w>(&self, word: &'w [u8], fc: u32) -> (u16, &'w [u8]) {
-        let Some(page) = self.page(word, fc) else { return (0, &[]) };
+        let Some((page, _, _)) = self.page(word, fc) else { return (0, &[]) };
         let Some(&crun) = page.get(511) else { return (0, &[]) };
         // BX is 13 bytes; cap crun at what fits before byte 511.
         let max_run = (511usize.saturating_sub((crun as usize + 1) * 4)) / 13;

crates/docbin/src/fmt.rs

@@ -109,12 +109,13 @@ pub(crate) fn ico(i: u8) -> Option<Rgb> {
     ICO.get(i as usize).copied()
 }
 
-/// The model border style a `brcType` stands for ([MS-DOC] §2.9.22 BrcType). Types the model
-/// doesn't carry (the gap variants, 3-D, inset/outset) draw as single lines, as the DOCX
-/// reader maps them.
+/// The model border style a `brcType` stands for ([MS-DOC] §2.9.22 BrcType; 0x02, thick, is in
+/// the BRC of Microsoft's Word 97-2007 specification). Types the model doesn't carry (the gap
+/// variants, 3-D, inset/outset) draw as single lines, as the DOCX reader maps them.
 fn brc_style(t: u8) -> BorderStyle {
     match t {
         0x00 => BorderStyle::None,
+        0x02 => BorderStyle::Thick,
         0x03 => BorderStyle::Double,
         0x06 => BorderStyle::Dotted,
         0x07 | 0x16 => BorderStyle::Dashed,
@@ -125,15 +126,15 @@ fn brc_style(t: u8) -> BorderStyle {
     }
 }
 
-/// The border width in points: `dptLineWidth` in 1/8-point steps (values below 2 count as 2;
-/// 1-point steps for page borders, which never reach a paragraph property).
+/// The border width in points: `dptLineWidth` in 1/8-point steps from 2 to 96, as DOCX `w:sz`
+/// reads (1-point steps for page borders, which never reach a paragraph property).
 fn brc_width(t: u8, dpt: u8) -> f32 {
-    if t >= 0x40 { f32::from(dpt) } else { f32::from(dpt.max(2)) / 8.0 }
+    if t >= 0x40 { f32::from(dpt) } else { (f32::from(dpt.max(2)) / 8.0).min(12.0) }
 }
 
 /// One border from a Brc80 operand ([MS-DOC] §2.9.18). All bits set means "no border"; a
 /// truncated operand is ignored.
-fn brc80(op: &[u8]) -> Option<Border> {
+pub(crate) fn brc80(op: &[u8]) -> Option<Border> {
     let [dpt, ty, ico_index, tail, ..] = op else { return None };
     if [*dpt, *ty, *ico_index, *tail] == [0xFF; 4] {
         return Some(Border::default());

crates/docbin/src/lib.rs

@@ -25,7 +25,9 @@ mod stsh;
 mod table;
 
 use std::io::{Read, Seek};
+use std::sync::Arc;
 
+use wordcraft_doc::link::hyperlink_target;
 use wordcraft_doc::para::InlineObject;
 use wordcraft_doc::para::NoteKind;
 use wordcraft_doc::para::Paragraph;
@@ -49,6 +51,14 @@ const MAX_PARTS: usize = 50_000;
 /// headers, notes, text boxes), whatever the file's size. The working budget is tied to the
 /// WordDocument stream (see [`read`]); this caps even a maximal stream.
 const MAX_TOTAL_CHARS: u64 = 64 << 20;
+/// Most paragraph and cell marks across every story walk of one document. The character
+/// budget does not see them: a mark is one byte of the file but about 1 KB of document, so a
+/// file of nothing but marks would grow memory about a thousand times faster than its size.
+/// The cap bounds a flood to about 2 GB while reading (about 4 GB at the very cap, when the
+/// blocks are assembled). Past it the read fails with [`DocbinError::Limit`]. Real documents stay far below it:
+/// 50 paragraphs on each of 10,000 pages is 500,000, and a 10,000-page one-column table (46
+/// rows, each a cell mark plus a row mark) is about 920,000.
+const MAX_PARAGRAPHS: usize = 2_000_000;
 /// Bytes of repeated work per document beyond the text itself: piece Prm grpprls (Prcs, up
 /// to 64 KB each) interpreted (#3692), hyperlink targets (#3708), picture alt text (#3709)
 /// and floating objects (#3710) copied into every run or anchor. Past it Prms are skipped,
@@ -193,6 +203,12 @@ pub fn read(bytes: &[u8]) -> Result<Document, DocbinError> {
 /// verifier ([`DocbinError::WrongPassword`] when it does not match) and then decrypted. The
 /// password is ignored for an unprotected file.
 pub fn read_with_password(bytes: &[u8], password: Option<&str>) -> Result<Document, DocbinError> {
+    read_limited(bytes, password, MAX_PARAGRAPHS)
+}
+
+/// [`read_with_password`] with the paragraph cap as a parameter, so tests need not build a
+/// file of [`MAX_PARAGRAPHS`] marks.
+fn read_limited(bytes: &[u8], password: Option<&str>, max_paragraphs: usize) -> Result<Document, DocbinError> {
     let mut comp = open_container(bytes)?;
     let mut word = stream(&mut comp, "WordDocument").ok_or_else(|| DocbinError::NotWord("no WordDocument stream".into()))?;
     if word.len() as u64 > MAX_STREAM {
@@ -244,13 +260,16 @@ pub fn read_with_password(bytes: &[u8], password: Option<&str>) -> Result<Docume
     };
 
     let mut doc = Document::new();
+    // Word opens Word 97-2003 files in compatibility mode 11 (Word 2003); the model's default
+    // 15 would apply Word 2013+ justification and line-breaking rules to them.
+    doc.settings.compat_mode = 11;
     if let Some(core) = metadata::read(&mut comp) {
         doc.core = core;
     }
     doc.body.clear();
-    doc.styles = fmt::stylesheet(&raw_styles, &fonts, standard_fonts);
-    doc.numbering = list::parse(&table, &fib, &fonts);
-    let sheet: StyleSheet = doc.styles.clone();
+    doc.styles = Arc::new(fmt::stylesheet(&raw_styles, &fonts, standard_fonts));
+    doc.numbering = Arc::new(list::parse(&table, &fib, &fonts));
+    let sheet: StyleSheet = StyleSheet::clone(&doc.styles);
     let sects = sections::parse(&word, &table, &fib);
     let stories = sections::header_stories(&table, &fib);
     // Subdocument layout in the CP space: main document, footnotes, headers, comments,
@@ -268,6 +287,7 @@ pub fn read_with_password(bytes: &[u8], password: Option<&str>) -> Result<Docume
     // tight against aliasing — and never past the absolute ceiling.
     let budget = std::cell::Cell::new((word.len() as u64).saturating_mul(2).min(MAX_TOTAL_CHARS));
     let work = Work::new(MAX_WORK);
+    let paragraphs = ParaBudget { left: std::cell::Cell::new(max_paragraphs), exceeded: std::cell::Cell::new(false) };
 
     let ctx = WalkCtx {
         word: &word,
@@ -283,6 +303,7 @@ pub fn read_with_password(bytes: &[u8], password: Option<&str>) -> Result<Docume
         drawn: &[],
         sect_ends: &[],
         budget: &budget,
+        paragraphs: &paragraphs,
         work: &work,
     };
 
@@ -302,7 +323,7 @@ pub fn read_with_password(bytes: &[u8], password: Option<&str>) -> Result<Docume
             if doc.parts.len() >= MAX_PARTS || tb_stories.contains_key(&spid) {
                 continue;
             }
-            let blocks = table::assemble(walk(&ctx, a, b, &mut pics));
+            let blocks = story_blocks(&ctx, a, b, &mut pics, max_paragraphs)?;
             tb_stories.insert(spid, doc.add_part(PartKind::TextBox, blocks));
         }
     }
@@ -327,7 +348,7 @@ pub fn read_with_password(bytes: &[u8], password: Option<&str>) -> Result<Docume
                 0 | 1 | 4 => PartKind::Header,
                 _ => PartKind::Footer,
             };
-            let blocks = table::assemble(walk(&ctx, a, b, &mut pics));
+            let blocks = story_blocks(&ctx, a, b, &mut pics, max_paragraphs)?;
             if !blocks.is_empty() {
                 let id = doc.add_part(kind, blocks);
                 hf_ids.push((si, k, id));
@@ -340,7 +361,7 @@ pub fn read_with_password(bytes: &[u8], password: Option<&str>) -> Result<Docume
     let mut note_links: Vec<(u32, NoteKind, u32)> = Vec::new();
     for (cp, ab) in notes::parse_notes(&table, &fib, fib::pair::PLCF_FND_REF, fib::pair::PLCF_FND_TXT) {
         let Some((a, b)) = story(ftn_base, fib.ccp.ftn, ab) else { continue };
-        let blocks = table::assemble(walk(&ctx, a, b, &mut pics));
+        let blocks = story_blocks(&ctx, a, b, &mut pics, max_paragraphs)?;
         if !blocks.is_empty() && doc.parts.len() < MAX_PARTS {
             let id = doc.add_part(PartKind::Footnote, blocks);
             note_links.push((cp, NoteKind::Footnote, id));
@@ -348,7 +369,7 @@ pub fn read_with_password(bytes: &[u8], password: Option<&str>) -> Result<Docume
     }
     for (cp, ab) in notes::parse_notes(&table, &fib, fib::pair::PLCF_END_REF, fib::pair::PLCF_END_TXT) {
         let Some((a, b)) = story(edn_base, fib.ccp.edn, ab) else { continue };
-        let blocks = table::assemble(walk(&ctx, a, b, &mut pics));
+        let blocks = story_blocks(&ctx, a, b, &mut pics, max_paragraphs)?;
         if !blocks.is_empty() && doc.parts.len() < MAX_PARTS {
             let id = doc.add_part(PartKind::Endnote, blocks);
             note_links.push((cp, NoteKind::Endnote, id));
@@ -391,6 +412,9 @@ pub fn read_with_password(bytes: &[u8], password: Option<&str>) -> Result<Docume
         doc.last_section = l;
     }
 
+    if paragraphs.exceeded.get() {
+        return Err(paragraph_limit(max_paragraphs));
+    }
     doc.media.extend(pics.media);
     doc.body = table::assemble(paras);
     doc.ensure_nonempty();
@@ -398,6 +422,20 @@ pub fn read_with_password(bytes: &[u8], password: Option<&str>) -> Result<Docume
     Ok(doc)
 }
 
+fn paragraph_limit(max_paragraphs: usize) -> DocbinError {
+    DocbinError::Limit(format!("the document has more than {max_paragraphs} paragraphs and table cells"))
+}
+
+/// Walks one story into blocks. A walk that spent the shared paragraph budget is refused before
+/// its partial list is assembled, so a flood does not build blocks it will throw away.
+fn story_blocks(ctx: &WalkCtx<'_>, a: u32, b: u32, pics: &mut media::Pictures, max_paragraphs: usize) -> Result<wordcraft_doc::Blocks, DocbinError> {
+    let paras = walk(ctx, a, b, pics);
+    if ctx.paragraphs.exceeded.get() {
+        return Err(paragraph_limit(max_paragraphs));
+    }
+    Ok(table::assemble(paras))
+}
+
 /// A story's CP range `[a, b)`, relative to a subdocument starting at `base` with `len`
 /// CPs, as absolute CPs clamped to that subdocument; `None` when empty.
 fn story(base: u32, len: u32, (a, b): (u32, u32)) -> Option<(u32, u32)> {
@@ -431,10 +469,35 @@ struct WalkCtx<'a> {
     /// footnotes, text boxes) are hostile and can each span a whole subdocument, so thousands of
     /// them could otherwise materialise the same text thousands of times; this bounds the total.
     budget: &'a std::cell::Cell<u64>,
+    /// Paragraph and cell marks still allowed across every story walk ([`MAX_PARAGRAPHS`]).
+    paragraphs: &'a ParaBudget,
     /// Repeated-work bytes left ([`MAX_WORK`]), and what they left out.
     work: &'a Work,
 }
 
+/// The paragraph and cell marks still allowed in this document, and whether a walk ran into
+/// the end of them (the read then fails).
+struct ParaBudget {
+    left: std::cell::Cell<usize>,
+    exceeded: std::cell::Cell<bool>,
+}
+
+impl ParaBudget {
+    /// Spends one mark; false (and the overrun noted) when none are left.
+    fn spend(&self) -> bool {
+        match self.left.get().checked_sub(1) {
+            Some(left) => {
+                self.left.set(left);
+                true
+            }
+            None => {
+                self.exceeded.set(true);
+                false
+            }
+        }
+    }
+}
+
 /// State while assembling one paragraph from per-CP formatting.
 #[derive(Default)]
 struct ParaBuild {
@@ -442,29 +505,37 @@ struct ParaBuild {
     runs: Vec<Run>,
     props: wordcraft_doc::props::ParaProps,
     mark: CharProps,
-    /// The current run's properties, so runs only break when formatting changes.
-    cur: Option<CharProps>,
+    /// Which resolved formatting ([`FmtState::generation`]) the last character came from, while
+    /// the last run is still the one it extended: the same generation means equal properties.
+    run_gen: Option<u64>,
     /// Inline objects, each anchored at a U+FFFC pushed into `text`.
     objects: Vec<InlineObject>,
 }
 
 impl ParaBuild {
     fn push(&mut self, c: char, props: CharProps) {
+        self.run_gen = None;
         let len = c.len_utf8();
-        match &self.cur {
-            Some(p) if *p == props => {
-                if let Some(r) = self.runs.last_mut() {
-                    r.len += len;
-                }
-            }
-            _ => {
-                self.runs.push(Run { len, props: props.clone() });
-                self.cur = Some(props);
-            }
+        match self.runs.last_mut() {
+            Some(r) if r.props == props => r.len += len,
+            _ => self.runs.push(Run { len, props }),
         }
         self.text.push(c);
     }
 
+    /// [`ParaBuild::push`] for properties that are resolved once and shared by many characters:
+    /// they are cloned only when a run starts, and not even compared while `generation` stays
+    /// the same.
+    fn push_shared(&mut self, c: char, props: &CharProps, generation: u64) {
+        let len = c.len_utf8();
+        match self.runs.last_mut() {
+            Some(r) if self.run_gen == Some(generation) || r.props == *props => r.len += len,
+            _ => self.runs.push(Run { len, props: props.clone() }),
+        }
+        self.run_gen = Some(generation);
+        self.text.push(c);
+    }
+
     /// Append `c` to the current run, whose properties the caller knows it matches.
     fn extend(&mut self, c: char) {
         if let Some(r) = self.runs.last_mut() {
@@ -492,7 +563,7 @@ fn walk<'a>(ctx: &WalkCtx<'a>, cp_start: u32, cp_end: u32, pics: &mut media::Pic
     // Field assembly state: instruction characters are swallowed, the result is kept until
     // the end character, then emitted as one object (hyperlinks keep their result text).
     let mut field: Option<FieldBuild> = None;
-    let mut memo: Option<FmtMemo<'a>> = None;
+    let mut fmt = FmtState::default();
     while cp < cp_end {
         // Bookmark boundaries at this CP (main-document walks only; subdocument walks get
         // an empty table and `cp` never matches the main-document CPs).
@@ -525,6 +596,9 @@ fn walk<'a>(ctx: &WalkCtx<'a>, cp_start: u32, cp_end: u32, pics: &mut media::Pic
         let section_mark = c == '\u{C}' && ctx.sect_ends.binary_search(&cp.saturating_add(1)).is_ok();
         match c {
             '\r' | '\u{7}' | '\u{C}' if c != '\u{C}' || section_mark => {
+                if !ctx.paragraphs.spend() {
+                    break;
+                }
                 // Paragraph mark: its PAPX covers the FC range of the paragraph it ends, so
                 // querying at the mark's own FC finds that paragraph's properties.
                 let fc_mark = ctx.pieces.fc_of_cp(cp).unwrap_or(0);
@@ -534,6 +608,9 @@ fn walk<'a>(ctx: &WalkCtx<'a>, cp_start: u32, cp_end: u32, pics: &mut media::Pic
                 }
                 let mut row = table::decode(papx);
                 row.cell_mark = c == '\u{7}';
+                // Only table marks keep their table context; a plain paragraph would carry
+                // a `RowInfo` (about 300 bytes) it never uses.
+                let row = row.is_table().then(|| Box::new(row));
                 let mut ilvl = 0u8;
                 let mut ilfo = 0i32;
                 for prl in sprm::iter(papx) {
@@ -587,7 +664,7 @@ fn walk<'a>(ctx: &WalkCtx<'a>, cp_start: u32, cp_end: u32, pics: &mut media::Pic
                     && f.depth == 1
                     && f.sep.is_none()
                 {
-                    f.sep = Some(Vec::new());
+                    f.sep = Some(FieldResult::default());
                 }
                 cp = cp.saturating_add(1);
                 continue;
@@ -621,6 +698,7 @@ fn walk<'a>(ctx: &WalkCtx<'a>, cp_start: u32, cp_end: u32, pics: &mut media::Pic
                             alt,
                             float: Default::default(),
                             crop: p.crop,
+                            fx: Default::default(),
                             ole: None,
                         });
                     }
@@ -648,16 +726,16 @@ fn walk<'a>(ctx: &WalkCtx<'a>, cp_start: u32, cp_end: u32, pics: &mut media::Pic
                     pb.objects.push(InlineObject::NoteRef { kind: *kind, id: *id, custom: String::new() });
                 }
             }
-            '\t' => push_formatted(&mut pb, cp, '\t', ctx, field.as_mut(), &mut memo),
-            '\u{B}' => push_formatted(&mut pb, cp, '\n', ctx, field.as_mut(), &mut memo),
-            '\u{C}' => push_formatted(&mut pb, cp, '\u{C}', ctx, field.as_mut(), &mut memo),
-            '\u{E}' => push_formatted(&mut pb, cp, '\u{E}', ctx, field.as_mut(), &mut memo),
-            '\u{1E}' => push_formatted(&mut pb, cp, '\u{2011}', ctx, field.as_mut(), &mut memo),
-            '\u{1F}' => push_formatted(&mut pb, cp, '\u{AD}', ctx, field.as_mut(), &mut memo),
+            '\t' => push_formatted(&mut pb, cp, '\t', ctx, field.as_mut(), &mut fmt),
+            '\u{B}' => push_formatted(&mut pb, cp, '\n', ctx, field.as_mut(), &mut fmt),
+            '\u{C}' => push_formatted(&mut pb, cp, '\u{C}', ctx, field.as_mut(), &mut fmt),
+            '\u{E}' => push_formatted(&mut pb, cp, '\u{E}', ctx, field.as_mut(), &mut fmt),
+            '\u{1E}' => push_formatted(&mut pb, cp, '\u{2011}', ctx, field.as_mut(), &mut fmt),
+            '\u{1F}' => push_formatted(&mut pb, cp, '\u{AD}', ctx, field.as_mut(), &mut fmt),
             // Object anchors, comment references and other special characters are dropped
             // (or handled by later passes).
             c if (c as u32) < 0x20 => {}
-            c => push_formatted(&mut pb, cp, c, ctx, field.as_mut(), &mut memo),
+            c => push_formatted(&mut pb, cp, c, ctx, field.as_mut(), &mut fmt),
         }
         cp = cp.saturating_add(units);
     }
@@ -666,7 +744,7 @@ fn walk<'a>(ctx: &WalkCtx<'a>, cp_start: u32, cp_end: u32, pics: &mut media::Pic
         out.push(ParaOut {
             para: Paragraph { text: done.text, runs: done.runs, props: done.props, mark: done.mark, objects: done.objects, ..Default::default() },
             end_cp: cp_end,
-            row: table::RowInfo::default(),
+            row: None,
         });
     }
     out
@@ -677,7 +755,28 @@ fn walk<'a>(ctx: &WalkCtx<'a>, cp_start: u32, cp_end: u32, pics: &mut media::Pic
 struct FieldBuild {
     depth: u32,
     instr: String,
-    sep: Option<Vec<(char, CharProps)>>,
+    sep: Option<FieldResult>,
+}
+
+/// A field's cached result: its text and, as runs, the formatting of that text. A run is added
+/// only when the formatting changes, so a result of millions of equally formatted characters
+/// costs one `CharProps`, not one per character (365 bytes each).
+#[derive(Default)]
+struct FieldResult {
+    text: String,
+    /// (byte length in `text`, formatting) in order; adjacent runs differ.
+    runs: Vec<(usize, CharProps)>,
+}
+
+impl FieldResult {
+    fn push(&mut self, c: char, props: CharProps) {
+        let len = c.len_utf8();
+        match self.runs.last_mut() {
+            Some((n, last)) if *last == props => *n += len,
+            _ => self.runs.push((len, props)),
+        }
+        self.text.push(c);
+    }
 }
 
 /// Bytes a clone of `obj` copies: the object itself plus a picture's alt text, a group with all
@@ -702,103 +801,26 @@ fn finish_field(pb: &mut ParaBuild, f: FieldBuild, work: &Work) {
     let result = f.sep.unwrap_or_default();
     if let Some(target) = hyperlink_target(&instr) {
         let cost = target.len() as u64;
-        let mut run: Option<CharProps> = None;
-        for (c, props) in result {
-            if run.as_ref() == Some(&props) {
-                pb.extend(c);
-                continue;
-            }
-            let mut linked = props.clone();
+        let mut chars = result.text.chars();
+        for (len, props) in result.runs {
+            let mut linked = props;
             if work.charge(cost, Cost::LinkTarget) {
                 linked.link = Some(target.clone());
             }
-            pb.push(c, linked);
-            run = Some(props);
+            let mut left = len;
+            let Some(first) = chars.next() else { break };
+            pb.push(first, linked);
+            left = left.saturating_sub(first.len_utf8());
+            while left > 0 {
+                let Some(c) = chars.next() else { break };
+                pb.extend(c);
+                left = left.saturating_sub(c.len_utf8());
+            }
         }
         return;
     }
-    let result_text: String = result.iter().map(|(c, _)| *c).collect();
     pb.push('\u{FFFC}', CharProps::default());
-    pb.objects.push(InlineObject::Field { instr, result: result_text, locked: false, dirty: false, code: false, form: String::new() });
-}
-
-/// `HYPERLINK "url"` / `HYPERLINK url \l "bookmark"` → the model link target
-/// (`#bookmark` for internal links); `None` when not a hyperlink.
-fn hyperlink_target(instr: &str) -> Option<String> {
-    let t = instr.trim();
-    let rest = if t.get(..9).is_some_and(|h| h.eq_ignore_ascii_case("hyperlink")) {
-        t.get(9..).filter(|r| r.is_empty() || r.starts_with(char::is_whitespace))?.trim()
-    } else {
-        return None;
-    };
-    if rest.is_empty() {
-        return None;
-    }
-
-    let mut url: Option<String> = None;
-    let mut local: Option<String> = None;
-    let mut switch: Option<&str> = None;
-
-    let mut i = 0;
-    let bytes = rest.as_bytes();
-    while i < bytes.len() {
-        while i < bytes.len() && bytes[i].is_ascii_whitespace() {
-            i += 1;
-        }
-        if i >= bytes.len() {
-            break;
-        }
-        if bytes[i] == b'\\' {
-            let start = i;
-            i += 1;
-            while i < bytes.len() && (bytes[i].is_ascii_alphabetic() || bytes[i] == b'*') {
-                i += 1;
-            }
-            switch = Some(rest.get(start..i).unwrap_or(""));
-            continue;
-        }
-        let arg = if bytes[i] == b'"' {
-            i += 1;
-            let start = i;
-            while i < bytes.len() && bytes[i] != b'"' {
-                i += 1;
-            }
-            let val = rest.get(start..i).unwrap_or("");
-            if i < bytes.len() && bytes[i] == b'"' {
-                i += 1;
-            }
-            val.to_string()
-        } else {
-            let start = i;
-            while i < bytes.len() && !bytes[i].is_ascii_whitespace() {
-                i += 1;
-            }
-            rest.get(start..i).unwrap_or("").to_string()
-        };
-        let clean_arg = arg.replace("%22", "\"");
-        match switch.take() {
-            Some("\\l") => local = Some(clean_arg),
-            Some(_) => {}
-            None => {
-                if url.is_none() {
-                    url = Some(clean_arg);
-                }
-            }
-        }
-    }
-
-    match (url, local) {
-        (Some(u), Some(l)) if !u.is_empty() && !l.is_empty() => {
-            let frag = l.trim_start_matches('#');
-            if u.contains('#') { Some(u) } else { Some(format!("{u}#{frag}")) }
-        }
-        (Some(u), _) if !u.is_empty() => Some(u),
-        (None, Some(l)) if !l.is_empty() => {
-            let frag = l.trim_start_matches('#');
-            Some(format!("#{frag}"))
-        }
-        _ => None,
-    }
+    pb.objects.push(InlineObject::Field { instr, result: result.text, locked: false, dirty: false, code: false, form: String::new() });
 }
 
 /// `sprmCPicLocation` from a CHPX grpprl: the fc of the picture in the Data stream.
@@ -843,34 +865,53 @@ fn num_of(ilfo: i32) -> Option<u32> {
 /// plus the piece's Prm, over the paragraph style's base character props. Inside a field,
 /// the character is recorded into the instruction or the cached result instead of (or as
 /// well as) the visible text.
-fn push_formatted<'a>(pb: &mut ParaBuild, cp: u32, c: char, ctx: &WalkCtx<'a>, field: Option<&mut FieldBuild>, memo: &mut Option<FmtMemo<'a>>) {
-    let fc = ctx.pieces.fc_of_cp(cp).unwrap_or(0);
-    let grpprl = ctx.chpx_bins.chpx(ctx.word, fc);
-    let prm = ctx.pieces.prm_of(cp);
-    let style = pb.props.style.as_deref();
-    let mut props = match memo.as_ref().filter(|m| std::ptr::eq(m.chpx, grpprl) && m.prm == prm && m.style.as_deref() == style) {
-        Some(m) => m.props.clone(),
+fn push_formatted<'a>(pb: &mut ParaBuild, cp: u32, c: char, ctx: &WalkCtx<'a>, field: Option<&mut FieldBuild>, st: &mut FmtState<'a>) {
+    // The piece and the CHPX run of the previous character usually cover this one too.
+    let span = match st.span.filter(|s| s.contains(cp)) {
+        Some(s) => Some(s),
         None => {
-            let default;
-            let base = match style.and_then(|id| ctx.sheet.get(id)) {
-                Some(st) => &st.chr,
-                None => {
-                    default = CharProps::default();
-                    &default
-                }
-            };
-            let mut props = char_props(grpprl, ctx.fonts, base);
-            apply_prm(&mut props, ctx.pieces.piece_prm(cp), ctx.fonts, base, ctx.work);
-            *memo = Some(FmtMemo { chpx: grpprl, prm, style: style.map(str::to_owned), props: props.clone() });
-            props
+            st.span = ctx.pieces.span(cp);
+            st.span
         }
     };
+    let fc = span.and_then(|s| s.fc_at(cp)).unwrap_or(0);
+    let run = match st.chpx.filter(|r| r.contains(fc)) {
+        Some(r) => r,
+        None => {
+            let r = ctx.chpx_bins.chpx_run(ctx.word, fc);
+            st.chpx = Some(r);
+            r
+        }
+    };
+    let grpprl = run.grpprl;
+    let prm = span.map_or(0, |s| s.prm());
+    let style = pb.props.style.as_deref();
+    let hit = st.memo.as_ref().is_some_and(|m| std::ptr::eq(m.chpx, grpprl) && m.prm == prm && m.style.as_deref() == style);
+    if !hit {
+        let default;
+        let base = match style.and_then(|id| ctx.sheet.get(id)) {
+            Some(sty) => &sty.chr,
+            None => {
+                default = CharProps::default();
+                &default
+            }
+        };
+        let mut props = char_props(grpprl, ctx.fonts, base);
+        apply_prm(&mut props, ctx.pieces.piece_prm(cp), ctx.fonts, base, ctx.work);
+        st.memo = Some(FmtMemo { chpx: grpprl, prm, style: style.map(str::to_owned), props });
+        st.generation += 1;
+    }
+    let Some(m) = st.memo.as_ref() else { return };
     // A symbol character (Insert › Symbol) is stored as the placeholder U+0028 with
     // sprmCSymbol carrying the font and the character code ([MS-DOC] §2.6.1).
+    let symbol = if c == '(' { symbol_operand(grpprl) } else { None };
+    if field.is_none() && symbol.is_none() {
+        pb.push_shared(c, &m.props, st.generation);
+        return;
+    }
+    let mut props = m.props.clone();
     let mut c = c;
-    if c == '('
-        && let Some((ftc, xchar)) = symbol_operand(grpprl)
-    {
+    if let Some((ftc, xchar)) = symbol {
         if let Some(font) = ctx.fonts.get(usize::from(ftc)) {
             props.font = Some(font.clone());
         }
@@ -882,13 +923,25 @@ fn push_formatted<'a>(pb: &mut ParaBuild, cp: u32, c: char, ctx: &WalkCtx<'a>, f
         Some(f) => match f.sep.as_mut() {
             // Inside a field, characters are recorded for the field object instead of shown;
             // `finish_field` emits the visible text (hyperlink runs or one object anchor).
-            Some(result) => result.push((c, props)),
+            Some(result) => result.push(c, props),
             None => f.instr.push(c),
         },
         None => pb.push(c, props),
     }
 }
 
+/// What the walk carries from one character to the next while it resolves formatting: the piece
+/// and the CHPX run the last character was in (so most characters need no search), and the last
+/// formatting resolved.
+#[derive(Default)]
+struct FmtState<'a> {
+    span: Option<piece::PieceSpan<'a>>,
+    chpx: Option<fkp::ChpxRun<'a>>,
+    memo: Option<FmtMemo<'a>>,
+    /// Counts the memo's replacements, so equal values mean the same properties.
+    generation: u64,
+}
+
 /// The last character formatting resolved, reused while the CHPX, the piece's Prm and the
 /// paragraph style stay the same, so a Prm grpprl (up to 64 KB) is interpreted once per
 /// formatting change, not once per character (#3692).

crates/docbin/src/piece.rs

@@ -117,12 +117,6 @@ impl PieceTable {
         }
     }
 
-    /// The raw Prm of the piece holding `cp` (0, no properties, when there is none): two CPs
-    /// with the same value get the same [`PrmRef`].
-    pub(crate) fn prm_of(&self, cp: u32) -> u16 {
-        self.piece_of(cp).map_or(0, |p| p.prm)
-    }
-
     fn piece_of(&self, cp: u32) -> Option<&Pcd> {
         let i = self.cps.partition_point(|&c| c <= cp).checked_sub(1)?;
         if i >= self.pieces.len() {
@@ -143,6 +137,17 @@ impl PieceTable {
         if p.compressed { p.fc.checked_add(step)?.checked_div(2) } else { p.fc.checked_add(step) }
     }
 
+    /// The piece holding `cp`, with the CP range it covers: for every CP in that range the
+    /// answers of [`PieceTable::fc_of_cp`] and the piece's Prm come from this span alone,
+    /// so a sequential walk searches the piece table once per piece.
+    pub(crate) fn span(&self, cp: u32) -> Option<PieceSpan<'_>> {
+        let i = self.cps.partition_point(|&c| c <= cp).checked_sub(1)?;
+        let pcd = self.pieces.get(i)?;
+        let cp_lo = *self.cps.get(i)?;
+        let cp_hi = *self.cps.get(i + 1)?;
+        Some(PieceSpan { cp_lo, cp_hi, pcd })
+    }
+
     /// The CP just past the last piece: no text exists at or beyond it.
     pub(crate) fn cp_end(&self) -> u32 {
         self.cps.last().copied().unwrap_or(0)
@@ -187,6 +192,31 @@ impl PieceTable {
     }
 }
 
+/// One piece and the CPs `[cp_lo, cp_hi)` it covers ([`PieceTable::span`]).
+#[derive(Clone, Copy)]
+pub(crate) struct PieceSpan<'a> {
+    cp_lo: u32,
+    cp_hi: u32,
+    pcd: &'a Pcd,
+}
+
+impl PieceSpan<'_> {
+    pub(crate) fn contains(&self, cp: u32) -> bool {
+        self.cp_lo <= cp && cp < self.cp_hi
+    }
+
+    /// [`PieceTable::fc_of_cp`] for a CP inside the span.
+    pub(crate) fn fc_at(&self, cp: u32) -> Option<u32> {
+        let step = cp.checked_sub(self.cp_lo)?.checked_mul(2)?;
+        if self.pcd.compressed { self.pcd.fc.checked_add(step)?.checked_div(2) } else { self.pcd.fc.checked_add(step) }
+    }
+
+    /// The raw Prm of the piece: two CPs with the same value get the same [`PrmRef`].
+    pub(crate) fn prm(&self) -> u16 {
+        self.pcd.prm
+    }
+}
+
 /// What a piece's `Prm` refers to.
 pub(crate) enum PrmRef<'a> {
     None,

crates/docbin/src/table.rs

@@ -5,7 +5,7 @@ use std::sync::Arc;
 
 use wordcraft_doc::Rgb;
 use wordcraft_doc::para::Paragraph;
-use wordcraft_doc::props::{Align, Border, BorderStyle, Borders, CellProps, HeightRule, RowProps, TableProps, VAlign, VMerge};
+use wordcraft_doc::props::{Align, Borders, CellProps, HeightRule, RowProps, TableProps, VAlign, VMerge};
 use wordcraft_doc::table::{Cell, Row, Table};
 use wordcraft_doc::{Block, Blocks, para_block};
 
@@ -83,6 +83,11 @@ pub(crate) struct RowInfo {
 
 /// Read the table context out of a paragraph mark's PAPX grpprl.
 impl RowInfo {
+    /// Does the mark belong to a table: inside one, ending a row, or ending a cell?
+    pub(crate) fn is_table(&self) -> bool {
+        self.in_table || self.ttp || self.cell_mark
+    }
+
     /// The row height from `sprmTDyaRowHeight`, for tests.
     #[cfg(test)]
     pub(crate) fn height(&self) -> Option<f32> {
@@ -263,8 +268,8 @@ pub(crate) struct ParaOut {
     pub(crate) para: Paragraph,
     /// CP just past the terminating mark.
     pub(crate) end_cp: u32,
-    /// Table context from the mark's PAPX (empty default = not in a table).
-    pub(crate) row: RowInfo,
+    /// Table context from the mark's PAPX; `None` for a paragraph outside a table.
+    pub(crate) row: Option<Box<RowInfo>>,
 }
 
 /// Group consecutive in-table paragraphs into [`Table`] blocks; the rest pass through.
@@ -369,9 +374,7 @@ pub(crate) fn assemble(paras: Vec<ParaOut>) -> Blocks {
     }
 
     for p in paras {
-        let info = p.row;
-        let is_table = info.in_table || info.ttp || info.cell_mark;
-        if !is_table {
+        let Some(info) = p.row else {
             if open {
                 flush_table(&mut out, &mut rows, &mut cells, &mut cell, &grid, &props, depth_seen > 1);
                 grid.clear();
@@ -386,7 +389,7 @@ pub(crate) fn assemble(paras: Vec<ParaOut>) -> Blocks {
             }
             out.push(para_block(p.para));
             continue;
-        }
+        };
         if !open {
             open = true;
         }
@@ -458,32 +461,12 @@ fn def_table_edges(operand: &[u8]) -> Vec<f32> {
 }
 
 /// Table borders from `sprmTTableBorders80`: cb (0x18) then six Brc80s in the order top,
-/// left, bottom, right, inside-horizontal, inside-vertical.
+/// left, bottom, right, inside-horizontal, inside-vertical, read as paragraph borders are.
 fn table_borders(operand: &[u8]) -> Option<Borders> {
-    let b = |i: usize| operand.get(1 + i * 4..1 + i * 4 + 4).and_then(brc80);
+    let b = |i: usize| operand.get(1 + i * 4..1 + i * 4 + 4).and_then(fmt::brc80);
     Some(Borders { top: b(0), left: b(1), bottom: b(2), right: b(3), between: b(4), inside_v: b(5) })
 }
 
-/// One Brc80: width in 1/8 pt, type, palette colour, space/flags byte.
-fn brc80(b: &[u8]) -> Option<Border> {
-    let [w, ty, ico, flags] = b else { return None };
-    let style = match ty {
-        0x00 => BorderStyle::None,
-        0x02 | 0x15 => BorderStyle::Thick,
-        0x03 => BorderStyle::Double,
-        0x04 | 0x05 => BorderStyle::Dotted,
-        0x06 | 0x07 => BorderStyle::Dashed,
-        0x08 | 0x09 => BorderStyle::DotDash,
-        0x0A | 0x18 => BorderStyle::Wave,
-        0x11 => BorderStyle::Triple,
-        _ => BorderStyle::Single,
-    };
-    if *ty == 0 && *w == 0 {
-        return None;
-    }
-    Some(Border { style, width: (*w as f32 / 8.0).clamp(0.25, 12.0), color: fmt::ico(*ico), space: (flags & 0x1F) as f32 })
-}
-
 /// Background colour from a 10-byte `Shd`: cvFore, cvBack (COLORREFs), ipat. Solid patterns
 /// show the foreground; clear (0) and the other patterns show the background (Word writes a
 /// plain cell fill as clear with cvBack); nil (0xFFFF) and automatic colours are no fill.

crates/docbin/src/tests.rs

@@ -116,6 +116,12 @@ fn piece_bytes(p: &Piece) -> Vec<u8> {
 
 /// A document whose main text is the given pieces (extra subdocument text may follow).
 fn text_doc(main: &[Piece], extra: &[Piece], ccp_hdd: u32) -> Vec<u8> {
+    text_doc_with(main, extra, FibSpec { ccp_hdd, ..Default::default() }, &[])
+}
+
+/// [`text_doc`] with the FIB knobs given (`ccp_text` is filled in from `main`; `pairs` must be empty) and extra
+/// Table-stream data placed after the piece table: `(pair index, bytes)` for each.
+fn text_doc_with(main: &[Piece], extra: &[Piece], spec: FibSpec, table_data: &[(usize, &[u8])]) -> Vec<u8> {
     let all: Vec<Piece> = main.iter().chain(extra.iter()).cloned().collect();
     let ccp_text: u32 = main.iter().map(|p| p.text.chars().count() as u32).sum();
     let mut word = vec![0u8; TEXT_FC as usize];
@@ -128,7 +134,13 @@ fn text_doc(main: &[Piece], extra: &[Piece], ccp_hdd: u32) -> Vec<u8> {
     clx.extend_from_slice(&plc);
     let mut table = vec![0u8; CLX_AT as usize];
     table.extend_from_slice(&clx);
-    let spec = FibSpec { ccp_text, ccp_hdd, pairs: vec![(33, CLX_AT, clx.len() as u32)], ..Default::default() };
+    let mut pairs = vec![(33, CLX_AT, clx.len() as u32)];
+    for (pair, bytes) in table_data {
+        pairs.push((*pair, table.len() as u32, bytes.len() as u32));
+        table.extend_from_slice(bytes);
+    }
+    debug_assert!(spec.pairs.is_empty(), "text_doc_with sets the pairs itself");
+    let spec = FibSpec { ccp_text, pairs, ..spec };
     word[..fib_bytes(&spec).len()].copy_from_slice(&fib_bytes(&spec));
     cfb_file(&[("WordDocument", &word), ("0Table", &table)])
 }
@@ -247,6 +259,77 @@ fn empty_paragraphs_kept() {
     assert_eq!(body_texts(&doc), ["a", "", "b"]);
 }
 
+/// A document of `n` copies of `mark` as one compressed piece.
+fn marks_doc(mark: char, n: usize) -> Vec<u8> {
+    text_doc(&[Piece { compressed: true, text: mark.to_string().repeat(n) }], &[], 0)
+}
+
+/// Paragraph marks are one byte each in the file but about 1 KB each in memory, so they have
+/// their own cap ([`super::MAX_PARAGRAPHS`]); a file past it is refused instead of growing
+/// until the machine runs out. A document exactly at the cap still opens.
+#[test]
+fn a_flood_of_paragraph_marks_is_refused() {
+    let at_cap = super::read_limited(&marks_doc('\r', 1000), None, 1000).expect("opens");
+    assert_eq!(at_cap.body.len(), 1000);
+    let over = super::read_limited(&marks_doc('\r', 1001), None, 1000);
+    assert!(matches!(over, Err(crate::DocbinError::Limit(_))), "{over:?}");
+}
+
+/// The paragraph cap is one budget for the whole document, not one per story: marks in the
+/// body and in a footnote add up. (A walk that got its own budget would open both over-cap files.)
+#[test]
+fn the_paragraph_cap_is_shared_by_the_body_and_a_footnote() {
+    // One footnote at CP 0 whose text is `note` marks, stored right after the body in the stream.
+    let doc_of = |body: usize, note: usize| {
+        let cps = |v: &[u32]| v.iter().flat_map(|c| c.to_le_bytes()).collect::<Vec<u8>>();
+        let mut refs = cps(&[0, 1]);
+        refs.extend_from_slice(&[0, 0]); // the one reference's data
+        let txt = cps(&[0, note as u32, note as u32]); // story 0..note, then the limit
+        let spec = FibSpec { ccp_ftn: note as u32, ..Default::default() };
+        text_doc_with(
+            &[Piece { compressed: true, text: "\r".repeat(body) }],
+            &[Piece { compressed: true, text: "\r".repeat(note) }],
+            spec,
+            &[(2, &refs), (3, &txt)],
+        )
+    };
+    let at_cap = super::read_limited(&doc_of(600, 400), None, 1000).expect("opens");
+    assert_eq!(at_cap.body.len(), 600);
+    assert_eq!(at_cap.parts.len(), 1);
+    let over = super::read_limited(&doc_of(600, 401), None, 1000);
+    assert!(matches!(over, Err(crate::DocbinError::Limit(_))), "{over:?}");
+    let over = super::read_limited(&doc_of(601, 400), None, 1000);
+    assert!(matches!(over, Err(crate::DocbinError::Limit(_))), "{over:?}");
+}
+
+/// Cell marks are paragraph marks too: a never-closed row of them is refused the same way.
+#[test]
+fn a_flood_of_cell_marks_is_refused() {
+    let over = super::read_limited(&marks_doc('\u{7}', 5000), None, 1000);
+    assert!(matches!(over, Err(crate::DocbinError::Limit(_))), "{over:?}");
+}
+
+/// The budget counts down across spends and notes the overrun (`walk` receives one shared by
+/// every story of the document).
+#[test]
+fn the_paragraph_budget_counts_down_and_notes_the_overrun() {
+    let budget = super::ParaBudget { left: std::cell::Cell::new(2), exceeded: std::cell::Cell::new(false) };
+    assert!(budget.spend() && budget.spend());
+    assert!(!budget.exceeded.get());
+    assert!(!budget.spend());
+    assert!(budget.exceeded.get());
+}
+
+/// The cap is far above any real document (10,000 pages of 50 paragraphs is 500,000), and a
+/// paragraph outside a table no longer carries a table context (a `ParaOut` was 1,224 bytes).
+#[test]
+fn the_paragraph_cap_leaves_room_and_a_paragraph_stays_small() {
+    const { assert!(super::MAX_PARAGRAPHS >= 1_000_000) };
+    assert!(std::mem::size_of::<crate::table::ParaOut>() < 1000, "{}", std::mem::size_of::<crate::table::ParaOut>());
+    let doc = read(&marks_doc('\r', 10)).expect("opens");
+    assert_eq!(doc.body.len(), 10);
+}
+
 #[test]
 fn empty_document_gets_a_paragraph() {
     let f = text_doc(&[Piece { compressed: false, text: "".into() }], &[], 0);
@@ -254,6 +337,15 @@ fn empty_document_gets_a_paragraph() {
     assert_eq!(doc.body.len(), 1);
 }
 
+#[test]
+fn a_doc_opens_in_word_97_2003_compatibility_mode() {
+    let f = text_doc(&[Piece { compressed: false, text: "body\r".into() }], &[], 0);
+    let doc = read(&f).expect("opens");
+    // Word 97-2003 files are mode 11 in Word 2013+; mode 15 would change justification and
+    // line breaking.
+    assert_eq!(doc.settings.compat_mode, 11);
+}
+
 #[test]
 fn header_text_stays_out_of_body() {
     let f = text_doc(
@@ -1209,6 +1301,53 @@ fn table_merges_borders_shading() {
     assert_eq!(top.width, 0.5);
 }
 
+/// Table borders read the BrcType byte as paragraph borders do; a nil Brc80 is no border.
+#[test]
+fn table_borders_read_border_types_as_paragraphs_do() {
+    use wordcraft_doc::props::BorderStyle as S;
+    let text = "a\u{7}\u{7}tail\r";
+    let in_tbl = grpprl(&[(0x2416, &[0x01]), (0x6649, &[1, 0, 0, 0])]);
+    let row_end = {
+        let mut v = [grpprl(&[(0x2416, &[0x01]), (0x2417, &[0x01]), (0x6649, &[1, 0, 0, 0])]), t_def_edges(&[0, 2880])].concat();
+        v.extend_from_slice(&0xD605u16.to_le_bytes());
+        v.push(0x18);
+        // Dotted, triple, double wave, nil, 3-D emboss 255/8 pt wide, all zero.
+        for brc in [[8, 0x06, 1, 0], [8, 0x0A, 1, 0], [8, 0x15, 1, 0], [0xFF; 4], [0xFF, 0x18, 1, 0], [0, 0, 0, 0]] {
+            v.extend_from_slice(&brc);
+        }
+        v
+    };
+    let spec = StructSpec {
+        text,
+        header_stories: vec![],
+        sections: vec![(text.chars().count() as u32, Vec::new())],
+        chpx_runs: vec![(0, Vec::new())],
+        papx_entries: vec![(0, 0, in_tbl), (2, 0, row_end), (3, 0, grpprl(&[(0x6424, &[0xFF, 0x01, 1, 0])]))],
+        plf_lst: Vec::new(),
+        plf_lfo: Vec::new(),
+        note_stories: vec![],
+        endnote_stories: vec![],
+        extra_pairs: Vec::new(),
+        data_stream: Vec::new(),
+    };
+    let doc = read(&struct_doc(&spec)).expect("opens");
+    let tbl = doc.body.iter().find_map(|b| b.as_table()).expect("table");
+    let b = tbl.props.borders.expect("table borders");
+    let seen = [b.top, b.left, b.bottom, b.right, b.between, b.inside_v].map(|s| s.map(|s| (s.style, s.is_visible(), s.width)));
+    let want = [
+        (S::Dotted, true, 1.0),
+        (S::Triple, true, 1.0),
+        (S::Wave, true, 1.0),
+        (S::None, false, 0.0),
+        (S::Single, true, 12.0),
+        (S::None, false, 0.25),
+    ];
+    assert_eq!(seen, want.map(Some));
+    // A paragraph border is at most 12 points too, as a table border and a DOCX `w:sz` are.
+    let tail = doc.body.iter().filter_map(|b| b.as_para()).find(|p| p.text == "tail").expect("tail paragraph");
+    assert_eq!(tail.props.borders.and_then(|b| b.top).map(|t| t.width), Some(12.0));
+}
+
 // ---------------------------------------------------------------------------------------------
 // D3: lists (PlfLst + PlfLfo) and the combined document.
 
@@ -1469,6 +1608,16 @@ fn hyperlink_target_needs_the_whole_field_name() {
     assert_eq!(super::hyperlink_target(r#"HYPERLINK	"http://example.com""#), Some("http://example.com".into()));
 }
 
+/// A bare `\m` or `\n` keeps the address after it; escapes, an empty address and `%22` read as in DOCX.
+#[test]
+fn hyperlink_target_keeps_the_address_after_a_bare_switch() {
+    assert_eq!(super::hyperlink_target(r#"HYPERLINK \m "http://example.com""#), Some("http://example.com".into()));
+    assert_eq!(super::hyperlink_target(r#"HYPERLINK \n "http://example.com""#), Some("http://example.com".into()));
+    assert_eq!(super::hyperlink_target(r#"HYPERLINK "C:\\Docs\\f.doc""#), Some(r"C:\Docs\f.doc".into()));
+    assert_eq!(super::hyperlink_target(r#"HYPERLINK "" \l "bm""#), Some("#bm".into()));
+    assert_eq!(super::hyperlink_target(r#"HYPERLINK "http://example.com/a%22b""#), Some("http://example.com/a%22b".into()));
+}
+
 #[test]
 fn bookmarks_become_objects() {
     let text = "Hello world\r";
@@ -2482,17 +2631,21 @@ fn symbol_runs_read_the_character_and_font_from_sprm_c_symbol() {
 }
 
 /// Paragraph borders around a block are read from both storage forms: the Word 97 Brc80
-/// (palette colour) and the Word 2000+ BrcOperand (cb + Brc with a full colour).
+/// (palette colour) and the Word 2000+ BrcOperand (cb + Brc with a full colour). A thick
+/// border (type 0x02) stays thick in both, as tables read it too.
 #[test]
 fn paragraph_borders_read_from_both_forms() {
+    use wordcraft_doc::props::BorderStyle as S;
     let b80 = [8u8, 0x01, 0x06, 0x01]; // 1 pt single line, palette red (6), 1 pt from the text
+    let thick80 = [8u8, 0x02, 0x06, 0x01];
     let brc = [8u8, 0xFF, 0x00, 0x00, 0x00, 8, 0x01, 0x02, 0x00]; // cb 8, red COLORREF, 1 pt single, space 2
+    let thick = [8u8, 0xFF, 0x00, 0x00, 0x00, 8, 0x02, 0x02, 0x00];
     let text = "boxed\r\r";
     let spec = StructSpec {
         text,
         papx_entries: vec![
-            (0, 0, grpprl(&[(0x6424, &b80), (0x6425, &b80), (0x6426, &b80), (0x6427, &b80), (0x6428, &b80)])),
-            (6, 0, grpprl(&[(0xC64E, &brc)])),
+            (0, 0, grpprl(&[(0x6424, &b80), (0x6425, &thick80), (0x6426, &b80), (0x6427, &b80), (0x6428, &b80)])),
+            (6, 0, grpprl(&[(0xC64E, &brc), (0xC64F, &thick)])),
         ],
         ..structured_spec(text, text.chars().count() as u32)
     };
@@ -2500,13 +2653,16 @@ fn paragraph_borders_read_from_both_forms() {
     let p0 = doc.body.first().and_then(|b| b.as_para()).expect("first paragraph");
     let b = p0.props.borders.expect("borders");
     let top = b.top.expect("top border");
-    assert_eq!((top.style, top.width, top.space), (wordcraft_doc::props::BorderStyle::Single, 1.0, 1.0));
+    assert_eq!((top.style, top.width, top.space), (S::Single, 1.0, 1.0));
     assert_eq!(top.color, Some(wordcraft_doc::Rgb(0xFF, 0, 0)));
-    assert!(b.left.is_some() && b.bottom.is_some() && b.right.is_some() && b.between.is_some());
+    assert_eq!(b.left.map(|l| (l.style, l.width)), Some((S::Thick, 1.0)), "Brc80 thick");
+    assert!(b.bottom.is_some() && b.right.is_some() && b.between.is_some());
     let p1 = doc.body.get(1).and_then(|b| b.as_para()).expect("second paragraph");
-    let top = p1.props.borders.expect("borders").top.expect("top border");
-    assert_eq!((top.style, top.width, top.space), (wordcraft_doc::props::BorderStyle::Single, 1.0, 2.0));
+    let b = p1.props.borders.expect("borders");
+    let top = b.top.expect("top border");
+    assert_eq!((top.style, top.width, top.space), (S::Single, 1.0, 2.0));
     assert_eq!(top.color, Some(wordcraft_doc::Rgb(0xFF, 0, 0)));
+    assert_eq!(b.left.map(|l| (l.style, l.width)), Some((S::Thick, 1.0)), "BrcOperand thick");
 }
 
 /// #3642: a small compound file whose DIFAT names one FAT sector thousands of times made `cfb`
@@ -2596,7 +2752,8 @@ fn a_long_hyperlink_over_a_long_result_is_one_linked_run() {
     assert_eq!(linked.first().map(|r| r.len), Some(n));
     // One copy of the target for the one run, not one per character (~2.5 s here): counted by
     // what the copies charge to the work budget, which a busy machine can't skew.
-    let result = vec![('y', wordcraft_doc::props::CharProps::default()); n];
+    let mut result = super::FieldResult::default();
+    (0..n).for_each(|_| result.push('y', wordcraft_doc::props::CharProps::default()));
     let f = super::FieldBuild { depth: 1, instr: format!("HYPERLINK \"http://{target}\""), sep: Some(result) };
     let work = super::Work::new(u64::MAX);
     super::finish_field(&mut super::ParaBuild::default(), f, &work);
@@ -2609,7 +2766,8 @@ fn a_long_hyperlink_over_a_long_result_is_one_linked_run() {
 fn hyperlink_target_copies_are_charged_to_the_work_budget() {
     use wordcraft_doc::props::CharProps;
     let bold = CharProps { bold: Some(true), ..Default::default() };
-    let result: Vec<(char, CharProps)> = (0..1000).map(|i| ('a', if i % 2 == 0 { bold.clone() } else { CharProps::default() })).collect();
+    let mut result = super::FieldResult::default();
+    (0..1000).for_each(|i| result.push('a', if i % 2 == 0 { bold.clone() } else { CharProps::default() }));
     let f = super::FieldBuild { depth: 1, instr: format!("HYPERLINK \"{}\"", "u".repeat(100)), sep: Some(result) };
     let mut pb = super::ParaBuild::default();
     let budget = super::Work::new(10_000);
@@ -2683,3 +2841,201 @@ fn prm_grpprls_past_the_work_budget_are_skipped() {
     assert_eq!(work.left.get(), 1);
     assert_eq!(work.dropped(), [super::Cost::Formatting]);
 }
+
+/// A field's cached result keeps one `CharProps` per formatting run, not one per character
+/// (365 bytes each: 4 million result characters took 1.4 GB).
+#[test]
+fn a_field_result_stores_one_props_per_run() {
+    use wordcraft_doc::props::CharProps;
+    let bold = CharProps { bold: Some(true), ..Default::default() };
+    let mut r = super::FieldResult::default();
+    for _ in 0..200_000 {
+        r.push('x', CharProps::default());
+    }
+    assert_eq!(r.runs.len(), 1);
+    assert_eq!(r.runs.first().map(|(n, _)| *n), Some(200_000));
+    r.push('\u{e9}', bold.clone());
+    r.push('y', bold);
+    r.push('z', CharProps::default());
+    assert_eq!(r.runs.iter().map(|(n, _)| *n).collect::<Vec<_>>(), [200_000, 3, 1], "a two-byte char counts its bytes");
+    assert_eq!(r.text.len(), 200_004);
+}
+
+/// A hyperlink whose result changes formatting, with multi-byte text, keeps every run's text and
+/// links all of it; a plain field keeps its result text whole.
+#[test]
+fn field_results_keep_their_text_and_runs() {
+    use wordcraft_doc::props::CharProps;
+    let bold = CharProps { bold: Some(true), ..Default::default() };
+    let mut result = super::FieldResult::default();
+    for (c, p) in
+        [('a', CharProps::default()), ('\u{e9}', CharProps::default()), ('\u{20ac}', bold.clone()), ('b', bold), ('c', CharProps::default())]
+    {
+        result.push(c, p);
+    }
+    let f = super::FieldBuild { depth: 1, instr: "HYPERLINK \"http://x\"".into(), sep: Some(result) };
+    let mut pb = super::ParaBuild::default();
+    super::finish_field(&mut pb, f, &super::Work::new(u64::MAX));
+    assert_eq!(pb.text, "a\u{e9}\u{20ac}bc");
+    assert_eq!(pb.runs.iter().map(|r| r.len).collect::<Vec<_>>(), [3, 4, 1]);
+    assert!(pb.runs.iter().all(|r| r.props.link.as_deref() == Some("http://x")));
+    let doc = read(&text_doc(&[Piece { compressed: false, text: "\u{13} PAGE \u{14}1\u{e9}2\u{15}\r".into() }], &[], 0)).expect("opens");
+    let p = doc.body.first().and_then(|b| b.as_para()).expect("paragraph");
+    assert!(
+        p.objects
+            .iter()
+            .any(|o| matches!(o, wordcraft_doc::para::InlineObject::Field { instr, result, .. } if instr == "PAGE" && result == "1\u{e9}2")),
+        "{:?}",
+        p.objects
+    );
+}
+
+/// A one-page CHPX bin table over `page`: FCs 0..`end` live in page 1 of the returned stream.
+fn one_page_bins(page: &[u8], end: u32) -> (Vec<u8>, crate::fkp::Bins) {
+    let mut word = vec![0u8; 512];
+    word.extend_from_slice(page);
+    let plc = [0u32.to_le_bytes().as_slice(), end.to_le_bytes().as_slice(), 1u32.to_le_bytes().as_slice()].concat();
+    let bins = crate::fkp::Bins::parse(&plc, 0, 12).expect("bins");
+    (word, bins)
+}
+
+/// The CHPX run a lookup reports covers exactly the FCs that give the same grpprl: the walk
+/// reuses it for the following characters without asking again.
+#[test]
+fn a_chpx_run_covers_the_fcs_with_the_same_answer() {
+    let runs = [(0u32, grpprl(&[(0x0835, &[1])])), (5, Vec::new()), (9, grpprl(&[(0x0836, &[1])])), (30, grpprl(&[(0x0835, &[1])]))];
+    let (word, bins) = one_page_bins(&chpx_fkp(&runs, 60), 0x2000);
+    for fc in (fc_of(0) - 4)..(fc_of(70)) {
+        let run = bins.chpx_run(&word, fc);
+        assert_eq!(run.grpprl, bins.chpx(&word, fc), "fc {fc}");
+        for other in (fc_of(0) - 4)..(fc_of(70)) {
+            if run.contains(other) {
+                assert_eq!(bins.chpx(&word, other), run.grpprl, "fc {fc} run reused at {other}");
+            }
+        }
+    }
+    assert!(bins.chpx_run(&word, fc_of(6)).contains(fc_of(8)) && !bins.chpx_run(&word, fc_of(6)).contains(fc_of(9)));
+}
+
+/// A page whose run starts do not ascend (a corrupt file) gives the same answers as before and
+/// a range that never claims an FC with a different answer.
+#[test]
+fn a_chpx_run_is_sound_on_a_page_whose_starts_do_not_ascend() {
+    let runs = [(10u32, grpprl(&[(0x0835, &[1])])), (3, grpprl(&[(0x0836, &[1])])), (20, Vec::new()), (7, grpprl(&[(0x0835, &[1])]))];
+    let (word, bins) = one_page_bins(&chpx_fkp(&runs, 60), 0x2000);
+    for fc in (fc_of(0) - 4)..(fc_of(70)) {
+        let run = bins.chpx_run(&word, fc);
+        assert_eq!(run.grpprl, bins.chpx(&word, fc));
+        for other in (fc_of(0) - 4)..(fc_of(70)) {
+            if run.contains(other) {
+                assert_eq!(bins.chpx(&word, other), run.grpprl, "fc {fc} run reused at {other}");
+            }
+        }
+    }
+}
+
+/// A piece span answers exactly what the per-CP lookup answers, in every piece (compressed
+/// and UTF-16) and past the last.
+#[test]
+fn a_piece_span_matches_the_per_cp_lookups() {
+    let doc = text_doc(
+        &[
+            Piece { compressed: true, text: "abc".into() },
+            Piece { compressed: false, text: "d\u{e9}fgh".into() },
+            Piece { compressed: true, text: "ij".into() },
+        ],
+        &[],
+        0,
+    );
+    let mut comp = crate::open_container(&doc).expect("container");
+    let table = crate::stream(&mut comp, "0Table").expect("table");
+    let pieces = crate::piece::PieceTable::parse(&table, CLX_AT, table.len() as u32 - CLX_AT).expect("pieces");
+    for cp in 0..14 {
+        let span = pieces.span(cp);
+        assert_eq!(span.and_then(|s| s.fc_at(cp)), pieces.fc_of_cp(cp), "cp {cp}");
+        for other in 0..14 {
+            if span.is_some_and(|s| s.contains(other)) {
+                assert_eq!(pieces.span(other).and_then(|s| s.fc_at(other)), pieces.fc_of_cp(other));
+            }
+        }
+    }
+    assert!(pieces.span(10).is_none(), "past the last piece");
+}
+
+/// Characters of one formatting share one `CharProps`: it is cloned when a run starts, and not
+/// even compared while the formatting stays the same.
+#[test]
+fn shared_props_start_one_run_per_formatting() {
+    use wordcraft_doc::props::CharProps;
+    let bold = CharProps { bold: Some(true), ..Default::default() };
+    let mut pb = super::ParaBuild::default();
+    for _ in 0..100_000 {
+        pb.push_shared('a', &bold, 1);
+    }
+    pb.push_shared('b', &CharProps::default(), 2);
+    pb.push_shared('c', &bold, 3);
+    pb.push_shared('d', &bold, 4);
+    assert_eq!(pb.runs.iter().map(|r| r.len).collect::<Vec<_>>(), [100_000, 1, 2], "equal props under a new generation join the run");
+    pb.push('e', bold.clone());
+    pb.push_shared('f', &bold, 4);
+    assert_eq!(
+        pb.runs.iter().map(|r| r.len).collect::<Vec<_>>(),
+        [100_000, 1, 4],
+        "a plain push forgets the generation but still joins an equal run"
+    );
+    assert_eq!(pb.text.len(), 100_005);
+}
+
+/// Several bins, equal and descending breakpoints, a missing page and the last FC: wherever a
+/// run is reported as valid it gives the answer a fresh lookup gives.
+#[test]
+fn chpx_runs_are_sound_across_bins() {
+    let page_a = chpx_fkp(&[(0, grpprl(&[(0x0835, &[1])])), (8, grpprl(&[(0x0836, &[1])]))], 20);
+    let page_b = chpx_fkp(&[(20, Vec::new()), (26, grpprl(&[(0x0835, &[1])]))], 40);
+    let mut word = vec![0u8; 512];
+    word.extend_from_slice(&page_a);
+    word.extend_from_slice(&page_b);
+    let (a, b, end) = (fc_of(20), fc_of(40), fc_of(60));
+    let plc = |fcs: &[u32], pns: &[u32]| {
+        [fcs.iter().flat_map(|f| f.to_le_bytes()).collect::<Vec<u8>>(), pns.iter().flat_map(|p| p.to_le_bytes()).collect()].concat()
+    };
+    let tables = [
+        ("two bins", plc(&[0, a, end], &[1, 2])),
+        ("an empty bin", plc(&[0, a, a, end], &[1, 2, 2])),
+        ("a missing page", plc(&[0, a, end], &[1, 9])),
+        ("descending breakpoints", plc(&[0, b, a, end], &[1, 2, 1])),
+        ("fcs up to the maximum", plc(&[0, a, u32::MAX], &[1, 2])),
+    ];
+    for (what, t) in &tables {
+        let bins = crate::fkp::Bins::parse(t, 0, t.len() as u32).expect("bins");
+        let probes: Vec<u32> = (0..fc_of(50)).step_by(2).chain([u32::MAX - 1, u32::MAX]).collect();
+        for &fc in &probes {
+            let run = bins.chpx_run(&word, fc);
+            assert_eq!(run.grpprl, bins.chpx(&word, fc), "{what}: fc {fc}");
+            for &other in &probes {
+                if run.contains(other) {
+                    assert_eq!(bins.chpx(&word, other), run.grpprl, "{what}: fc {fc} run reused at {other}");
+                }
+            }
+        }
+    }
+}
+
+/// The same formatting changes through a whole read: runs break exactly where the CHPX changes
+/// (with a field and a symbol character in between), and nowhere else.
+#[test]
+fn formatting_runs_break_where_the_chpx_changes() {
+    let spec = RichSpec {
+        text: "aaaabbbb\u{13}PAGE\u{14}12\u{15}ccdd(ee\r",
+        styles: vec![(1, 0x0FFF, "Normal", Vec::new(), Vec::new())],
+        fonts: vec!["Times New Roman", "Arial"],
+        chpx_runs: vec![(0, grpprl(&[(0x0835, &[1])])), (4, Vec::new()), (8, grpprl(&[(0x0835, &[1])])), (20, grpprl(&[(0x0836, &[1])]))],
+        papx_entries: vec![(30, 0, Vec::new())],
+    };
+    let doc = read(&rich_doc(&spec)).expect("opens");
+    let p = para0(&doc);
+    let shape: Vec<(usize, Option<bool>, Option<bool>)> = p.runs.iter().map(|r| (r.len, r.props.bold, r.props.italic)).collect();
+    // "aaaa" bold, "bbbb" plain plus the field's anchor (3 bytes, default formatting), "ccd" bold
+    // (the run from the field to CP 20), "d(ee" italic.
+    assert_eq!(shape, [(4, Some(true), None), (7, None, None), (3, Some(true), None), (4, None, Some(true))], "{shape:?}");
+}

crates/docbin/src/tests/drawing.rs

@@ -183,6 +183,41 @@ fn shapes_keep_kind_colours_wrap_and_text_box_text() {
     assert_eq!(part.blocks.first().and_then(|b| b.as_para()).map(|p| p.text.as_str()), Some("Inside"));
 }
 
+#[test]
+fn braces_and_brackets_are_unfilled_unless_filled_is_set() {
+    // A right brace (88) in front of the text with no fill properties, and a left bracket (85)
+    // whose fFilled is set: only the bracket is filled. A plain rectangle (1) keeps its default
+    // white fill.
+    let brace = sp(&[fsp(88, 1026, 0xA00)]);
+    let bracket = sp(&[fsp(85, 1027, 0xA00), fopt(&[(0x01BF, (1 << 20) | (1 << 4))], &[])]);
+    let rect = sp(&[fsp(1, 1028, 0xA00)]);
+    let art = art_content(&[], &[brace, bracket, rect]);
+    let flags = spa_flags(0, 0, 3, false);
+    let spas =
+        plc(&[1, 2, 3, 5], &[spa(1026, [0, 0, 240, 1140], flags), spa(1027, [0, 0, 240, 1140], flags), spa(1028, [0, 0, 240, 1140], flags)].concat());
+    let text = "A\u{8}\u{8}\u{8}B\r";
+    let ccp = text.chars().count() as u32;
+    let spec = StructSpec { text, extra_pairs: vec![(40, spas), (50, art)], ..structured_spec(text, ccp) };
+    let doc = read(&struct_doc(&spec)).expect("opens");
+    let p0 = doc.body.first().and_then(|b| b.as_para()).expect("para");
+    let fills: Vec<_> = p0
+        .objects
+        .iter()
+        .map(|o| match o {
+            InlineObject::Shape { fill, stroke, w, h, .. } => (fill.clone(), *stroke, *w, *h),
+            other => panic!("{other:?}"),
+        })
+        .collect();
+    assert_eq!(
+        fills,
+        [
+            (None, Some(Rgb::BLACK), 12.0, 57.0),
+            (Some(Fill::solid(Rgb::WHITE)), Some(Rgb::BLACK), 12.0, 57.0),
+            (Some(Fill::solid(Rgb::WHITE)), Some(Rgb::BLACK), 12.0, 57.0),
+        ]
+    );
+}
+
 #[test]
 fn text_boxes_pointing_at_the_same_story_text_read_it_once() {
     // Two text boxes whose PlcftxbxTxt stories are both [0, 7) (the table steps back to CP 0
@@ -391,6 +426,7 @@ fn a_floating_copy_is_charged_its_alt_text() {
         alt,
         float: Default::default(),
         crop: Default::default(),
+        fx: Default::default(),
         ole: None,
     };
     let bare = crate::object_cost(&image(String::new()));

crates/docx/Cargo.toml

@@ -32,6 +32,7 @@ getrandom = { workspace = true, features = ["wasm_js"] }
 
 [dev-dependencies]
 wordcraft-doc = { workspace = true, features = ["test-support"] }
+serde_json = { workspace = true }
 proptest = { workspace = true }
 image = { workspace = true }
 zip = { workspace = true }

crates/docx/src/custom.rs

@@ -25,11 +25,23 @@ pub fn read(root: &El) -> Vec<CustomProp> {
         let Some(v) = p.els().next() else { continue };
         let kind = v.local();
         let prop = if SIMPLE.contains(&kind) {
-            CustomProp { name: name.to_string(), kind: kind.to_string(), value: v.text(), link_target: p.attr("linkTarget").map(str::to_string) }
+            CustomProp {
+                name: name.to_string(),
+                kind: kind.to_string(),
+                value: v.text(),
+                link_target: p.attr("linkTarget").map(str::to_string),
+                odf_user_field: false,
+            }
         } else {
             let mut xml = String::new();
             el_xml(v, &mut xml, 0);
-            CustomProp { name: name.to_string(), kind: "raw".into(), value: xml, link_target: p.attr("linkTarget").map(str::to_string) }
+            CustomProp {
+                name: name.to_string(),
+                kind: "raw".into(),
+                value: xml,
+                link_target: p.attr("linkTarget").map(str::to_string),
+                odf_user_field: false,
+            }
         };
         out.push(prop);
     }

crates/docx/src/package.rs

@@ -350,6 +350,7 @@ impl Package {
                 let path = resolve(dir, target);
                 self.canonical(&path).map_or(path, str::to_string)
             };
+            rels.ids.entry(id.to_string()).or_insert(rels.list.len());
             rels.list.push(Rel { id: id.to_string(), kind: r.attr("Type").unwrap_or("").to_string(), target: resolved, external });
         }
         rels
@@ -405,11 +406,13 @@ pub struct Rel {
 #[derive(Clone, Debug, Default)]
 pub struct Rels {
     pub list: Vec<Rel>,
+    /// Each id's first relationship in `list`, so a lookup doesn't scan every relationship.
+    ids: HashMap<String, usize>,
 }
 
 impl Rels {
     pub fn by_id(&self, id: &str) -> Option<&Rel> {
-        self.list.iter().find(|r| r.id == id)
+        self.ids.get(id).and_then(|&i| self.list.get(i))
     }
     pub fn by_type(&self, full: &str) -> Option<&Rel> {
         self.list.iter().find(|r| rel_is(&r.kind, full))
@@ -547,6 +550,22 @@ mod tests {
         assert_eq!(crate::package_title(std::io::Cursor::new(z), 1 << 20).as_deref(), Some("Plan"));
     }
 
+    #[test]
+    fn relationships_are_looked_up_by_id_the_first_one_winning() {
+        let mut rels = String::from(r#"<Relationships xmlns="http://schemas.openxmlformats.org/package/2006/relationships">"#);
+        for i in 0..20_000 {
+            rels.push_str(&format!(r#"<Relationship Id="rId{i}" Type="hyperlink" Target="https://example.com/{i}" TargetMode="External"/>"#));
+        }
+        rels.push_str(r#"<Relationship Id="rId7" Type="hyperlink" Target="https://example.com/dup" TargetMode="External"/></Relationships>"#);
+        let z = zip_of(&[("word/_rels/document.xml.rels", rels.into_bytes())]);
+        let r = Package::open(&z).unwrap().rels("word/document.xml");
+        assert_eq!(r.by_id("rId7").map(|r| r.target.as_str()), Some("https://example.com/7"), "the first rId7 wins");
+        assert_eq!(r.by_id("rId19999").map(|r| r.target.as_str()), Some("https://example.com/19999"));
+        assert!(r.by_id("rId20000").is_none());
+        // Hyperlink-style lookups, one per relationship, are each a map lookup.
+        assert_eq!((0..20_000).filter(|i| r.by_id(&format!("rId{i}")).is_some()).count(), 20_000);
+    }
+
     #[test]
     fn decompression_bomb_is_rejected() {
         let z = zip_of(&[("word/document.xml", vec![b' '; 2_000_000])]);

crates/docx/src/read/chart/model.rs

@@ -66,6 +66,9 @@ pub(super) struct Axis {
 
 pub(super) struct Series {
     pub(super) name: String,
+    /// The series' `c:idx` (an `xsd:unsignedInt`), which picks its automatic colour; `None`
+    /// when missing or not a valid number.
+    pub(super) idx: Option<usize>,
     pub(super) color: Option<Rgb>,
     pub(super) line_w: f32,
     /// Whether the series is drawn as a line (scatter series without one get markers).

crates/docx/src/read/chart/parse.rs

@@ -91,8 +91,11 @@ impl Chart {
         self.accents.get(i % 6).copied().unwrap_or(Rgb::BLACK)
     }
 
+    /// Colour of series `j`: its own, else the accent its `c:idx` picks (as Word does; `c:order`
+    /// doesn't matter), else the one its place in the file picks.
     pub(super) fn series_colour(&self, j: usize) -> Rgb {
-        self.series.get(j).and_then(|s| s.color).unwrap_or_else(|| self.accent(j))
+        let s = self.series.get(j);
+        s.and_then(|s| s.color).unwrap_or_else(|| self.accent(s.and_then(|s| s.idx).unwrap_or(j)))
     }
 
     /// Colour of point `i` of series `j`: its `c:dPt` override, else the varied accent (pie) or
@@ -199,6 +202,7 @@ pub(super) fn text_size(space: &El) -> f32 {
 impl Series {
     fn parse(ser: &El, scatter: bool, theme: &[Rgb], cap: usize) -> Series {
         let name = ser.child("c:tx").and_then(|t| t.find("c:v")).map(|v| label(v.text().trim())).unwrap_or_default();
+        let idx = ser.child("c:idx").and_then(|e| e.attr("val")).and_then(|v| v.trim().parse::<u32>().ok()).and_then(|v| usize::try_from(v).ok());
         let sp = ser.child("c:spPr");
         let ln = sp.and_then(|p| p.child("a:ln"));
         let marker = ser.child("c:marker").and_then(|m| m.child("c:spPr")).and_then(|p| fill_colour(p, theme));
@@ -219,7 +223,7 @@ impl Series {
             }
         }
         let labels = show_values(ser.child("c:dLbls"));
-        Series { name, color, line_w, line, values, xs, points, fmt, x_fmt, labels, raw: Vec::new() }
+        Series { name, idx, color, line_w, line, values, xs, points, fmt, x_fmt, labels, raw: Vec::new() }
     }
 }
 

crates/docx/src/read/chart/tests.rs

@@ -528,3 +528,65 @@ fn negative_sections_and_scientific_numbers() {
     assert_eq!(format_value(0.0, "0.00E+00", 1.0), "0.00E+00");
     assert_eq!(format_value(-12345.0, "0.0E-00", 1.0), "-1.2E04");
 }
+
+// An automatic series colour follows the series' `c:idx`, not its place in the file (#5963), as
+// measured in Word: `idx` 3 then 0 are accent4 then accent1, a lone `idx` 2 is accent3, and
+// `c:order` changes nothing.
+
+/// The fill colours of the bars of a column chart of `sers`, and of its legend swatches.
+fn column_colours(sers: &[String]) -> (Vec<Rgb>, Vec<Rgb>) {
+    let plot = format!(
+        r#"<c:barChart><c:barDir val="col"/><c:grouping val="clustered"/><c:varyColors val="0"/>{}<c:axId val="1"/><c:axId val="2"/></c:barChart><c:catAx><c:axId val="1"/></c:catAx><c:valAx><c:axId val="2"/></c:valAx>"#,
+        sers.concat()
+    );
+    let items = with_legend(&plot, "r", 300.0, 200.0);
+    let bars = filled(&items).iter().map(|b| b.0).collect();
+    let keys = items
+        .iter()
+        .filter_map(|it| if let GraphicItem::Shape { rect, fill, .. } = it { (rect[2] == SWATCH).then_some(*fill).flatten() } else { None })
+        .collect();
+    (bars, keys)
+}
+
+/// `ser` with its `c:idx` / `c:order` replaced by `idx_order` (an empty string drops them).
+fn renumbered(ser: &str, idx_order: &str) -> String {
+    let start = ser.find("<c:idx").expect("ser writes c:idx");
+    let end = ser.find("<c:tx>").expect("ser writes c:tx");
+    format!("{}{idx_order}{}", &ser[..start], &ser[end..])
+}
+
+#[test]
+fn automatic_series_colours_follow_c_idx() {
+    let accent = |k: usize| theme()[4 + k];
+    let a = ser(0, "A", None, &["Q1"], &["1"]);
+    let b = ser(1, "B", None, &["Q1"], &["2"]);
+    // idx 3 first, idx 0 second: accent4, accent1, in the bars and the legend.
+    let sers = [renumbered(&a, r#"<c:idx val="3"/><c:order val="0"/>"#), renumbered(&b, r#"<c:idx val="0"/><c:order val="1"/>"#)];
+    assert_eq!(column_colours(&sers), (vec![accent(3), accent(0)], vec![accent(3), accent(0)]));
+    // A lone series with idx 2 is accent3.
+    let lone = [renumbered(&a, r#"<c:idx val="2"/><c:order val="0"/>"#)];
+    assert_eq!(column_colours(&lone).0, [accent(2)]);
+    // c:order doesn't matter: idx 0 and 1 stay accent1 and accent2 with their orders swapped.
+    let swapped = [renumbered(&a, r#"<c:idx val="0"/><c:order val="1"/>"#), renumbered(&b, r#"<c:idx val="1"/><c:order val="0"/>"#)];
+    assert_eq!(column_colours(&swapped).0, [accent(0), accent(1)]);
+    // A series' own colour still wins.
+    let own = [renumbered(&ser(0, "A", Some("FF0000"), &["Q1"], &["1"]), r#"<c:idx val="3"/><c:order val="0"/>"#)];
+    assert_eq!(column_colours(&own).0, [Rgb(0xFF, 0, 0)]);
+}
+
+#[test]
+fn series_without_a_usable_c_idx_take_their_place() {
+    let accent = |k: usize| theme()[4 + k];
+    let a = ser(0, "A", None, &["Q1"], &["1"]);
+    let b = ser(1, "B", None, &["Q1"], &["2"]);
+    // No c:idx at all: the place in the file.
+    assert_eq!(column_colours(&[renumbered(&a, ""), renumbered(&b, "")]).0, [accent(0), accent(1)]);
+    // Junk, negative and out-of-range values: no panic, the place in the file.
+    for junk in ["x", "-1", "1e3", "99999999999999999999999"] {
+        let sers = [renumbered(&a, &format!(r#"<c:idx val="{junk}"/>"#)), renumbered(&b, &format!(r#"<c:idx val="{junk}"/>"#))];
+        assert_eq!(column_colours(&sers).0, [accent(0), accent(1)], "{junk:?}");
+    }
+    // The largest unsignedInt `c:idx` allows draws, without a panic.
+    let big = [renumbered(&a, &format!(r#"<c:idx val="{}"/>"#, u32::MAX))];
+    assert_eq!(column_colours(&big).0.len(), 1);
+}

crates/docx/src/read/embed.rs

@@ -239,7 +239,13 @@ impl Reader<'_> {
     /// Keep one part's bytes and content type (not its relationships).
     fn keep_part(&mut self, part: &str) -> bool {
         let lower = part.to_ascii_lowercase();
-        if !manifest_field_ok(part) || lower.ends_with(".rels") || lower == "[content_types].xml" || part.eq_ignore_ascii_case(&self.main) {
+        // `wordcraft:` names are WordCraft's own bookkeeping in `passthrough`, never a file's part.
+        if !manifest_field_ok(part)
+            || lower.ends_with(".rels")
+            || lower == "[content_types].xml"
+            || lower.starts_with("wordcraft:")
+            || part.eq_ignore_ascii_case(&self.main)
+        {
             log::warn!("docx: not carrying part {part:?} for an embedded object");
             return false;
         }
@@ -388,6 +394,23 @@ mod tests {
         assert!(roots.contains("keep.bin"), "the part is dropped: {roots}");
     }
 
+    /// A part named like WordCraft's own bookkeeping (`wordcraft:unmodelled.roots`) is not carried:
+    /// kept, it would be read back as that bookkeeping and add relationships the file chose.
+    #[test]
+    fn a_part_named_like_bookkeeping_is_not_carried() {
+        let forged = b"D\thttp://schemas.openxmlformats.org/officeDocument/2006/relationships/officeDocument\tword/evil.xml\n";
+        let bytes = package(
+            r#"<w:p><w:r><w:t>text</w:t></w:r></w:p>"#,
+            &[("rIdDoc1", "/wordcraft:unmodelled.roots"), ("rIdDoc2", "/wordcraft:embedded.parts")],
+            &[("wordcraft:unmodelled.roots", forged), ("wordcraft:embedded.parts", b"P\tword/evil.xml\tapplication/xml\n")],
+        );
+        let doc = crate::read(&bytes).unwrap();
+        let roots = unmodelled_roots(&doc);
+        assert!(!roots.contains("wordcraft:") && !roots.contains("evil"), "{roots}");
+        let manifest = doc.passthrough.get(crate::package::EMBEDDED_PARTS).map(|v| String::from_utf8_lossy(v).to_string()).unwrap_or_default();
+        assert!(!manifest.contains("evil"), "{manifest}");
+    }
+
     /// An unqualified attribute is a relationship only where the markup names a relationship
     /// that way: a hyperlink's `id` names the part, a VML shape's `id` is its own.
     #[test]

crates/docx/src/read/parts.rs

@@ -55,6 +55,8 @@ impl Reader<'_> {
             .children("w:font")
             .filter(|f| f.child("w:family").and_then(|v| v.attr("w:val")) == Some("roman"))
             .filter_map(|f| f.attr("w:name"))
+            .flat_map(|n| n.split(';'))
+            .map(str::trim)
             .filter(|n| !n.is_empty() && n.len() <= MAX_FONT_NAME)
             .take(MAX_TABLE_FONTS)
             .map(str::to_string)

crates/docx/src/read/props.rs

@@ -5,16 +5,17 @@ use std::collections::HashMap;
 use wordcraft_doc::para::cap_alt;
 use wordcraft_doc::props::{
     Align, Border, BorderStyle, Borders, CellProps, CharProps, CombineBrackets, EastAsianLayout, HeightRule, Highlight, LineSpacing, NumRef,
-    ParaProps, Rgb, RowProps, TabAlign, TabLeader, TabStop, TableFloat, TableLook, TableProps, TextColor, TextDirection, Underline, VAlign, VMerge,
-    VertAlign,
+    ParaProps, ParagraphTextDirection, Rgb, RowProps, TabAlign, TabLeader, TabStop, TableFloat, TableLook, TableProps, TextAlignment, TextColor,
+    TextDirection, Underline, VAlign, VMerge, VertAlign,
 };
 use wordcraft_doc::section::{
     Columns, DocGrid, DocGridType, LineNumberRestart, LineNumbering, MAX_CHAR_SPACE, MAX_GRID_PITCH, NotePos, NoteProps, NoteRestart, NumFormat,
     PageVAlign, SectionProps, SectionStart,
 };
 
+use super::MAX_FONT_NAME;
 use super::drawing_color::adjust_lum;
-use crate::units::{int, line_multiple, measure, on_off, tw, u32_of};
+use crate::units::{int, line_multiple, measure, on_off, tw, tw_i16, tw_u16, u32_of};
 
 /// A text frame (`w:framePr` without a drop cap, ECMA-376 §17.3.1.11), sizes and offsets in
 /// points (clamped), names kept as OOXML writes them.
@@ -33,6 +34,7 @@ fn text_frame(k: &El) -> wordcraft_doc::props::TextFrame {
         h_anchor: word("w:hAnchor", &["text", "margin", "page"], "page"),
         v_anchor: word("w:vAnchor", &["text", "margin", "page"], "page"),
         wrap: word("w:wrap", &["around", "auto", "notBeside", "none", "tight", "through"], "auto"),
+        anchor_lock: k.attr("w:anchorLock").map(|v| !matches!(v.trim(), "0" | "false" | "off" | "False" | "FALSE")),
         h_space: len("w:hSpace").unwrap_or(0.0).max(0.0),
         v_space: len("w:vSpace").unwrap_or(0.0).max(0.0),
     }
@@ -144,6 +146,29 @@ pub(crate) fn resolve_level_only(sheet: &wordcraft_doc::styles::StyleSheet, p: &
     }
 }
 
+/// A `w:rFonts` value as the document keeps it: `None` when it names no family, else its leading
+/// families up to [`MAX_FONT_NAME`] bytes, cut at a `;` so no name is split. A LibreOffice file
+/// names its Latin font as a semicolon-separated fallback list, and a real list names a few
+/// families; a longer value is corrupt or hostile, and every piece of text drawn in it would pay
+/// for its whole length as the list is classified and resolved. Keeping the leading families keeps
+/// the family such a list picks, and the rest of the value is dropped like `w:cs`/`w:eastAsia`
+/// drop an over-long name.
+fn bounded_font_list(v: &str) -> Option<String> {
+    if v.len() <= MAX_FONT_NAME {
+        return (!v.is_empty()).then(|| v.to_string());
+    }
+    let mut end = 0usize;
+    for part in v.split_inclusive(';') {
+        let next = end + part.len();
+        if next > MAX_FONT_NAME {
+            break;
+        }
+        end = next;
+    }
+    let kept = v.get(..end)?.trim_end_matches(';');
+    (!kept.is_empty()).then(|| kept.to_string())
+}
+
 impl PropCtx {
     pub fn style_id(&self, id: &str) -> String {
         self.alias.get(id).cloned().unwrap_or_else(|| id.to_string())
@@ -172,8 +197,9 @@ impl PropCtx {
     }
 
     /// The theme colour `e` names in `attr` (`w:themeColor`, `w:themeFill`), through the colour
-    /// scheme mapping, with its `tint` and `shade` (hex bytes) applied to the HSL luminance as
-    /// Word does. When given it supersedes the fixed colour; `None` leaves the fixed one.
+    /// scheme mapping, with its `tint` or else its `shade` (hex bytes) applied to the HSL
+    /// luminance as Word does: given both, Word applies the tint alone (MS-OI29500 17.3.2.6 b,
+    /// 17.4.32 a). When given it supersedes the fixed colour; `None` leaves the fixed one.
     fn theme_color(&self, e: &El, attr: &str, tint: &str, shade: &str) -> Option<Rgb> {
         let logical = |i: usize| self.color_map.and_then(|m| m.get(i).copied()).unwrap_or(i);
         let slot = match e.attr(attr)? {
@@ -199,8 +225,7 @@ impl PropCtx {
         let byte = |a: &str| e.attr(a).and_then(|v| u8::from_str_radix(v.trim(), 16).ok()).map(|v| f64::from(v) / 255.0);
         if let Some(t) = byte(tint) {
             c = adjust_lum(c, |l| l * t + (1.0 - t));
-        }
-        if let Some(s) = byte(shade) {
+        } else if let Some(s) = byte(shade) {
             c = adjust_lum(c, |l| l * s);
         }
         Some(c)
@@ -208,6 +233,10 @@ impl PropCtx {
 
     /// `w:shd`: its fill colour (a theme fill first), or a solid pattern's colour.
     pub fn shd_fill(&self, e: &El) -> Option<Rgb> {
+        // `nil` counts as no shading, whatever fill it carries (MS-OI29500 17.18.78).
+        if e.attr("w:val") == Some("nil") {
+            return None;
+        }
         if let Some(c) = self.theme_color(e, "w:themeFill", "w:themeFillTint", "w:themeFillShade") {
             return Some(c);
         }
@@ -221,6 +250,8 @@ impl PropCtx {
         shd_fill(e)
     }
 
+    /// The character properties `e` (a `w:rPr`) sets: direct formatting, plus the fonts, size,
+    /// weight and posture of the complex-script and East Asian text beside them.
     pub fn rpr(&self, e: &El) -> CharProps {
         let mut c = CharProps::default();
         for k in e.els() {
@@ -229,7 +260,11 @@ impl PropCtx {
                 "w:rFonts" => {
                     // A theme font supersedes the fixed font beside it (`w:asciiTheme` over
                     // `w:ascii`, `w:hAnsiTheme` over `w:hAnsi`).
-                    let named = |a: &str| k.attr(a).filter(|f| !f.is_empty()).map(str::to_string);
+                    // A LibreOffice file names the Latin font as a semicolon-separated fallback
+                    // list. Names are bounded like `w:cs`/`w:eastAsia` below: a longer one is
+                    // corrupt or hostile, and every piece of text drawn in it would pay for its
+                    // whole length as the list is classified (see `bounded_font_list`).
+                    let named = |a: &str| k.attr(a).and_then(bounded_font_list);
                     let themed = |a: &str| k.attr(a).and_then(|t| self.theme_font(t));
                     // `w:eastAsia` and `w:cs` name fonts for other scripts and stay in `font_ea` /
                     // `font_cs`: `w:hint` only picks between them for ambiguous characters, never for
@@ -250,6 +285,7 @@ impl PropCtx {
                 "w:cs" => c.cs = Some(on_off(k)),
                 "w:caps" => c.caps = Some(on_off(k)),
                 "w:smallCaps" => c.small_caps = Some(on_off(k)),
+                "w:em" => c.emphasis = Some(k.attr("w:val").map(wordcraft_doc::props::EmphasisMark::from_ooxml).unwrap_or_default()),
                 "w:strike" => c.strike = Some(on_off(k)),
                 "w:dstrike" => c.double_strike = Some(on_off(k)),
                 "w:outline" => c.outline = Some(on_off(k)),
@@ -282,7 +318,7 @@ impl PropCtx {
                     c.underline_color =
                         self.theme_color(k, "w:themeColor", "w:themeTint", "w:themeShade").or_else(|| k.attr("w:color").and_then(Rgb::parse));
                 }
-                "w:bdr" => c.border = Some(border(k)),
+                "w:bdr" => c.border = Some(self.border(k)),
                 "w:shd" => c.shading = self.shd_fill(k),
                 "w:vertAlign" => {
                     c.vert_align = k.attr("w:val").map(|v| match v {
@@ -328,9 +364,11 @@ impl PropCtx {
                 "w:widowControl" => p.widow_control = Some(on_off(k)),
                 "w:contextualSpacing" => p.contextual_spacing = Some(on_off(k)),
                 "w:mirrorIndents" => p.mirror_indents = Some(on_off(k)),
+                "w:adjustRightInd" => p.adjust_right_indent = Some(on_off(k)),
                 "w:suppressAutoHyphens" => p.suppress_hyphens = Some(on_off(k)),
                 "w:suppressLineNumbers" => p.suppress_line_numbers = Some(on_off(k)),
                 "w:bidi" => p.bidi = Some(on_off(k)),
+                "w:textDirection" => p.text_direction = k.attr("w:val").and_then(ParagraphTextDirection::from_ooxml),
                 "w:kinsoku" => p.kinsoku = Some(on_off(k)),
                 "w:wordWrap" => p.word_wrap = Some(on_off(k)),
                 "w:overflowPunct" => p.overflow_punct = Some(on_off(k)),
@@ -338,6 +376,7 @@ impl PropCtx {
                 "w:autoSpaceDE" => p.auto_space_de = Some(on_off(k)),
                 "w:autoSpaceDN" => p.auto_space_dn = Some(on_off(k)),
                 "w:snapToGrid" => p.snap_to_grid = Some(on_off(k)),
+                "w:textAlignment" => p.text_alignment = k.attr("w:val").and_then(TextAlignment::from_ooxml),
                 "w:framePr" => {
                     if let Some(kind @ ("drop" | "margin")) = k.attr("w:dropCap") {
                         p.drop_cap = Some(k.attr("w:lines").and_then(u32_of).unwrap_or(3).clamp(1, 10) as u8);
@@ -361,17 +400,17 @@ impl PropCtx {
                         (None, None) => {}
                     }
                 }
-                "w:pBdr" => p.borders = Some(borders(k)),
+                "w:pBdr" => p.borders = Some(self.borders(k)),
                 "w:shd" => p.shading = self.shd_fill(k),
                 "w:tabs" => {
                     let tabs: Vec<TabStop> = k.children("w:tab").take(64).filter_map(tab_stop).collect();
                     p.tabs = Some(tabs);
                 }
                 "w:spacing" => {
-                    if let Some(v) = tw(k, "w:before") {
+                    if let Some(v) = tw_u16(k, "w:before") {
                         p.space_before = Some(v.max(0.0));
                     }
-                    if let Some(v) = tw(k, "w:after") {
+                    if let Some(v) = tw_u16(k, "w:after") {
                         p.space_after = Some(v.max(0.0));
                     }
                     // HTML-style automatic spacing; "0" switches an inherited one off.
@@ -394,15 +433,15 @@ impl PropCtx {
                     }
                 }
                 "w:ind" => {
-                    if let Some(v) = tw(k, "w:left").or_else(|| tw(k, "w:start")) {
+                    if let Some(v) = tw_i16(k, "w:left").or_else(|| tw_i16(k, "w:start")) {
                         p.indent_left = Some(v);
                     }
-                    if let Some(v) = tw(k, "w:right").or_else(|| tw(k, "w:end")) {
+                    if let Some(v) = tw_i16(k, "w:right").or_else(|| tw_i16(k, "w:end")) {
                         p.indent_right = Some(v);
                     }
-                    if let Some(v) = tw(k, "w:hanging") {
+                    if let Some(v) = tw_i16(k, "w:hanging") {
                         p.indent_first = Some(-v);
-                    } else if let Some(v) = tw(k, "w:firstLine") {
+                    } else if let Some(v) = tw_i16(k, "w:firstLine") {
                         p.indent_first = Some(v);
                     }
                     // Character units (hundredths of a character), which win when non-zero.
@@ -476,29 +515,32 @@ fn shd_fill(e: &El) -> Option<Rgb> {
     None
 }
 
-pub fn border(e: &El) -> Border {
-    let style = BorderStyle::from_ooxml(e.attr("w:val").unwrap_or("single"));
-    let width = e.attr("w:sz").and_then(|v| measure(v, 8.0)).unwrap_or(0.5).clamp(0.0, 12.0);
-    let color = e.attr("w:color").and_then(Rgb::parse);
-    let space = e.attr("w:space").and_then(|v| measure(v, 1.0)).unwrap_or(0.0).clamp(0.0, 31.0);
-    Border { style, width, color, space }
-}
+impl PropCtx {
+    /// One border (`CT_Border`); its theme colour supersedes `w:color`, as for text.
+    pub fn border(&self, e: &El) -> Border {
+        let style = BorderStyle::from_ooxml(e.attr("w:val").unwrap_or("single"));
+        let width = e.attr("w:sz").and_then(|v| measure(v, 8.0)).unwrap_or(0.5).clamp(0.0, 12.0);
+        let color = self.theme_color(e, "w:themeColor", "w:themeTint", "w:themeShade").or_else(|| e.attr("w:color").and_then(Rgb::parse));
+        let space = e.attr("w:space").and_then(|v| measure(v, 1.0)).unwrap_or(0.0).clamp(0.0, 31.0);
+        Border { style, width, color, space }
+    }
 
-/// `w:pBdr`, `w:tblBorders`, `w:tcBorders`, `w:pgBorders`.
-pub fn borders(e: &El) -> Borders {
-    let mut b = Borders::default();
-    for k in e.els() {
-        match k.local() {
-            "top" => b.top = Some(border(k)),
-            "left" | "start" => b.left = Some(border(k)),
-            "bottom" => b.bottom = Some(border(k)),
-            "right" | "end" => b.right = Some(border(k)),
-            "between" | "insideH" => b.between = Some(border(k)),
-            "insideV" => b.inside_v = Some(border(k)),
-            _ => {}
+    /// `w:pBdr`, `w:tblBorders`, `w:tcBorders`, `w:pgBorders`.
+    pub fn borders(&self, e: &El) -> Borders {
+        let mut b = Borders::default();
+        for k in e.els() {
+            match k.local() {
+                "top" => b.top = Some(self.border(k)),
+                "left" | "start" => b.left = Some(self.border(k)),
+                "bottom" => b.bottom = Some(self.border(k)),
+                "right" | "end" => b.right = Some(self.border(k)),
+                "between" | "insideH" => b.between = Some(self.border(k)),
+                "insideV" => b.inside_v = Some(self.border(k)),
+                _ => {}
+            }
         }
+        b
     }
-    b
 }
 
 /// Margins element (`w:tblCellMar`, `w:tcMar`) → [top, left, bottom, right].
@@ -532,7 +574,7 @@ impl PropCtx {
                 }
                 "w:jc" => t.align = k.attr("w:val").map(align),
                 "w:tblInd" => t.indent = tw(k, "w:w"),
-                "w:tblBorders" => t.borders = Some(borders(k)),
+                "w:tblBorders" => t.borders = Some(self.borders(k)),
                 "w:shd" => t.shading = self.shd_fill(k),
                 "w:tblLayout" => t.fixed = k.attr("w:type") == Some("fixed"),
                 "w:bidiVisual" => t.bidi_visual = on_off(k),
@@ -638,7 +680,7 @@ pub fn tcpr(pc: &PropCtx, e: &El) -> (CellProps, bool) {
             "w:gridSpan" => c.span = k.attr("w:val").and_then(u32_of).unwrap_or(1).clamp(1, 63),
             "w:hMerge" => hcont = k.attr("w:val") != Some("restart"),
             "w:vMerge" => c.vmerge = if k.attr("w:val") == Some("restart") { VMerge::Restart } else { VMerge::Continue },
-            "w:tcBorders" => c.borders = Some(borders(k)),
+            "w:tcBorders" => c.borders = Some(pc.borders(k)),
             "w:shd" => c.shading = pc.shd_fill(k),
             "w:noWrap" => c.no_wrap = on_off(k),
             "w:hideMark" => c.hide_mark = on_off(k),
@@ -675,7 +717,7 @@ pub struct HfRef {
 }
 
 /// `w:sectPr` (header/footer references are returned for the caller to resolve).
-pub fn sectpr(e: &El) -> (SectionProps, Vec<HfRef>) {
+pub fn sectpr(pc: &PropCtx, e: &El) -> (SectionProps, Vec<HfRef>) {
     let mut s = SectionProps::default();
     let mut refs = Vec::new();
     for k in e.els() {
@@ -754,8 +796,26 @@ pub fn sectpr(e: &El) -> (SectionProps, Vec<HfRef>) {
                 if let Some(f) = k.attr("w:fmt") {
                     s.page_num_format = NumFormat::from_ooxml(f);
                 }
+                s.chapter_style = k.attr("w:chapStyle").and_then(u32_of).filter(|v| *v <= u8::MAX as u32).map(|v| v as u8);
+                s.chapter_sep = k.attr("w:chapSep").and_then(wordcraft_doc::section::ChapterSep::from_ooxml);
                 s.page_num_set = s.page_num_start.is_none() && s.page_num_format == NumFormat::Decimal;
             }
+            "w:paperSrc" => {
+                let tray = |name| {
+                    let value = k.attr(name)?.trim();
+                    let digits = value.strip_prefix('+').unwrap_or(value);
+                    if digits.is_empty() || !digits.bytes().all(|b| b.is_ascii_digit()) {
+                        return None;
+                    }
+                    u32_of(digits).filter(|v| *v <= u16::MAX as u32).map(|v| v as u16)
+                };
+                s.paper_src_first = tray("w:first");
+                s.paper_src_other = tray("w:other");
+            }
+            "w:noEndnote" => s.no_endnote = Some(on_off(k)),
+            "w15:footnoteColumns" => {
+                s.footnote_columns = k.attr("w:val").and_then(u32_of).filter(|v| (1..=45).contains(v));
+            }
             "w:lnNumType" => {
                 let mut l = LineNumbering::default();
                 if let Some(v) = k.attr("w:countBy").and_then(u32_of) {
@@ -780,7 +840,7 @@ pub fn sectpr(e: &El) -> (SectionProps, Vec<HfRef>) {
                     _ => PageVAlign::Top,
                 }
             }
-            "w:pgBorders" => s.page_borders = Some(borders(k)),
+            "w:pgBorders" => s.page_borders = Some(pc.borders(k)),
             "w:docGrid" => {
                 let mut g = DocGrid::default();
                 if let Some(t) = k.attr("w:type").and_then(DocGridType::from_ooxml) {

crates/docx/src/read/story/blocks.rs

@@ -31,7 +31,7 @@ impl Reader<'_> {
                         let cap_font = first.runs.first().and_then(|r| r.props.font.clone());
                         let text_font = p.runs.first().and_then(|r| r.props.font.clone());
                         if cap_font.is_some() && cap_font != text_font {
-                            for r in &mut first.runs {
+                            for r in first.runs.iter_mut().filter(|r| r.props.font == cap_font) {
                                 r.props.font = text_font.clone();
                             }
                             props.drop_cap_font = cap_font;
@@ -289,10 +289,10 @@ impl Reader<'_> {
     }
 
     pub fn read_section(&mut self, s: &El, rels: &Rels) -> wordcraft_doc::SectionProps {
-        let (mut sp, refs) = sectpr(s);
+        let (mut sp, refs) = sectpr(&self.pc, s);
         // §17.13.5.32: the section's properties before a tracked change (no header references).
         if let Some(ch) = s.child("w:sectPrChange") {
-            let old = ch.child("w:sectPr").map(|o| sectpr(o).0).unwrap_or_default();
+            let old = ch.child("w:sectPr").map(|o| sectpr(&self.pc, o).0).unwrap_or_default();
             sp.fmt_change = PropChange::boxed(self.revision(RevisionKind::Format, ch), old);
         }
         for r in refs {

crates/docx/src/read/story/drawing.rs

@@ -41,6 +41,7 @@ impl Reader<'_> {
             }
             let source = self.keep_embedded(c.child("a:graphic")?, rels, &[])?;
             float.set_spin(xfrm_spin(content.child("w14:xfrm")));
+            float.bare_turn = !anchored && float.rot != 0.0 && float.effect == [0.0; 4];
             let (px, py) = float.spin_pad(w, h);
             for (e, p) in float.effect.iter_mut().zip([px, py, px, py]) {
                 if *e != 0.0 {
@@ -106,7 +107,11 @@ impl Reader<'_> {
         self.set_shape_id(&mut obj, c.child("wp:docPr").and_then(|p| p.attr("id")));
         // Word's effect extent also covers a rotated object's overhang; the model keeps that
         // apart (it follows from the angle), so take it back out.
+        // Shadow or glow room the file didn't record: saving records it (and the turn's room).
+        let fx_room = matches!(&obj, InlineObject::Shape { effects, .. } if effects.extent() != [0.0; 4]);
         if let Some(f) = obj.float_mut() {
+            // None recorded: Word keeps an inline one's line to the frame (see `Float::bare_turn`).
+            f.bare_turn = !anchored && !fx_room && f.rot != 0.0 && f.effect == [0.0; 4];
             let (px, py) = f.spin_pad(w, h);
             for (e, p) in f.effect.iter_mut().zip([px, py, px, py]) {
                 // Zero adds no effects room when rotation contracts this axis.
@@ -118,9 +123,33 @@ impl Reader<'_> {
         Some(obj)
     }
 
+    /// The hyperlink Word keeps on a drawing itself: its `wp:docPr`'s `a:hlinkClick` (ECMA-376
+    /// Part 1 §20.4.2.5, §21.1.2.3.5) naming an external hyperlink relationship. Word stores a
+    /// picture's link there rather than in a `w:hyperlink` around its run (#5267). Its `#name`
+    /// part, when the address has one, stays part of the address.
+    pub(super) fn drawing_link(d: &El, rels: &Rels) -> Option<String> {
+        let frame = d.child("wp:inline").or_else(|| d.child("wp:anchor"))?;
+        let id = frame.child("wp:docPr")?.child("a:hlinkClick")?.attr("r:id")?;
+        let rel = rels.by_id(id).filter(|r| r.external && crate::package::rel_is(&r.kind, rt::HYPERLINK))?;
+        Some(rel.target.trim().to_string()).filter(|u| !u.is_empty())
+    }
+
+    /// Native ink or an Office 2016 chart kept with its fallback (`mc:AlternateContent`), and the
+    /// [`Self::drawing_link`] of the choice it was read from, else of its fallback: another
+    /// choice's link is not this object's.
+    pub(super) fn read_alternate(&mut self, sc: &mut StoryCtx, ac: &El, rels: &Rels) -> Option<(InlineObject, Option<String>)> {
+        let (object, choice) = match self.read_ink_alternate(sc, ac, rels) {
+            Some(o) => (o, children_of_choice(ac).filter(|e| e.name == "mc:Choice")),
+            None => (self.read_chartex_alternate(sc, ac, rels)?, chartex_choice(ac)),
+        };
+        let link =
+            [choice, ac.child("mc:Fallback")].into_iter().flatten().filter_map(|e| e.child("w:drawing")).find_map(|d| Self::drawing_link(d, rels));
+        Some((object, link))
+    }
+
     /// Keep the selected native-ink branch and its picture fallback as one object. Reading only
     /// the fallback flattens the ink; reading only the choice loses its preview on the next save.
-    pub(super) fn read_ink_alternate(&mut self, sc: &mut StoryCtx, ac: &El, rels: &Rels) -> Option<InlineObject> {
+    fn read_ink_alternate(&mut self, sc: &mut StoryCtx, ac: &El, rels: &Rels) -> Option<InlineObject> {
         let choice = children_of_choice(ac).filter(|e| e.name == "mc:Choice")?;
         let drawing = choice.child("w:drawing")?;
         if drawing.find("a:graphicData")?.attr("uri") != Some(INK_URI) {
@@ -149,10 +178,8 @@ impl Reader<'_> {
     /// An Office 2016 chart (`cx:chart`: waterfall, sunburst, box & whisker…) drawn from its
     /// fallback picture, keeping both branches so saving writes the chart back rather than only
     /// the picture. `None` (read the fallback alone) when it can't be kept whole.
-    pub(super) fn read_chartex_alternate(&mut self, sc: &mut StoryCtx, ac: &El, rels: &Rels) -> Option<InlineObject> {
-        let choice = ac
-            .children("mc:Choice")
-            .find(|c| c.child("w:drawing").and_then(|d| d.find("a:graphicData")).and_then(|g| g.attr("uri")) == Some(CHARTEX_URI))?;
+    fn read_chartex_alternate(&mut self, sc: &mut StoryCtx, ac: &El, rels: &Rels) -> Option<InlineObject> {
+        let choice = chartex_choice(ac)?;
         // A requirement in a namespace we can't name couldn't be written back.
         let requires = choice.attr("Requires").unwrap_or("");
         if requires.split_whitespace().next().is_none() || requires.split_whitespace().any(|p| p == "?") {
@@ -263,7 +290,7 @@ impl Reader<'_> {
                 }
             }
         }
-        Some(InlineObject::Image { media, w, h, alt, float, crop, ole: None })
+        Some(InlineObject::Image { media, w, h, alt, float, crop, fx: crate::units::blip_fx(blip), ole: None })
     }
 
     /// A `wpg:wgp` group (`w` × `h`): its pictures, shapes and text boxes, nested groups
@@ -563,6 +590,11 @@ impl Reader<'_> {
     }
 }
 
+/// The `mc:Choice` holding an Office 2016 chart (`cx:chart`) in an `mc:AlternateContent`.
+fn chartex_choice(ac: &El) -> Option<&El> {
+    ac.children("mc:Choice").find(|c| c.child("w:drawing").and_then(|d| d.find("a:graphicData")).and_then(|g| g.attr("uri")) == Some(CHARTEX_URI))
+}
+
 /// The alternative text of a `wp:docPr`, `pic:cNvPr` or `wps:cNvPr` (ECMA-376
 /// `CT_NonVisualDrawingProps`): the description (the title when there is none, as before), and
 /// the title and decorative mark (the `adec:decorative` extension of [MS-ODRAWXML]).

crates/docx/src/read/story/fields.rs

@@ -1,4 +1,5 @@
 use super::{CharProps, FieldState, InlineObject, Item, MAX_FIELD_DEPTH, PB, Phase, Reader, StoryCtx};
+use wordcraft_doc::link::hyperlink_target;
 
 /// Enclosed characters (`eq \o\ac(○,字)`), with their style from the code's font sizes: Word
 /// enlarges the symbol (bigger than the text, which keeps the field's size) or shrinks the text
@@ -24,69 +25,6 @@ fn is_range_instr(instr: &str) -> bool {
     crate::read::is_toc(instr) || instr.trim_start().get(..5).is_some_and(|k| k.eq_ignore_ascii_case("ADDIN"))
 }
 
-/// Where a `HYPERLINK` field goes, as a `w:hyperlink` reads: `url`, `url#bookmark` (`\l`) or
-/// `#bookmark`. None for any other field, or one with no target.
-fn hyperlink_target(instr: &str) -> Option<String> {
-    // Only a `HYPERLINK` code is split up (a field's code may run to megabytes).
-    if !instr.split_whitespace().next().is_some_and(|k| k.eq_ignore_ascii_case("HYPERLINK")) {
-        return None;
-    }
-    let args = field_args(instr);
-    let mut args = args.iter();
-    if !args.next().is_some_and(|(k, quoted)| !quoted && k.eq_ignore_ascii_case("HYPERLINK")) {
-        return None;
-    }
-    let (mut url, mut anchor) = (None, None);
-    while let Some((a, quoted)) = args.next() {
-        if *quoted || !a.starts_with('\\') {
-            url = url.or(Some(a.as_str()));
-            continue;
-        }
-        // `\l` names a bookmark, `\o` a tooltip and `\t` a target frame; `\m` and `\n` take none.
-        let value = matches!(a.to_ascii_lowercase().as_str(), "\\l" | "\\o" | "\\t").then(|| args.next()).flatten();
-        if a.eq_ignore_ascii_case("\\l") {
-            anchor = value.map(|(v, _)| v.as_str());
-        }
-    }
-    let (url, anchor) = (url.map(str::trim).filter(|u| !u.is_empty()), anchor.map(str::trim).filter(|a| !a.is_empty()));
-    match (url, anchor) {
-        (Some(u), Some(a)) => Some(format!("{u}#{a}")),
-        (Some(u), None) => Some(u.to_string()),
-        (None, Some(a)) => Some(format!("#{a}")),
-        (None, None) => None,
-    }
-}
-
-/// A field code's words: the field's name, its switches and its arguments, each with whether
-/// it was quoted. In quotes, `\"` is a quote and `\\` a backslash.
-fn field_args(instr: &str) -> Vec<(String, bool)> {
-    let mut out = Vec::new();
-    let mut chars = instr.chars().peekable();
-    while let Some(c) = chars.next() {
-        if c.is_whitespace() {
-            continue;
-        }
-        let mut word = String::new();
-        if c == '"' {
-            while let Some(c) = chars.next() {
-                match c {
-                    '"' => break,
-                    '\\' if matches!(chars.peek(), Some('"' | '\\')) => word.extend(chars.next()),
-                    _ => word.push(c),
-                }
-            }
-            out.push((word, true));
-        } else {
-            word.push(c);
-            while let Some(c) = chars.next_if(|c| !c.is_whitespace() && *c != '"') {
-                word.push(c);
-            }
-            out.push((word, false));
-        }
-    }
-    out
-}
-
 impl Reader<'_> {
     // ---- fields ----
 

crates/docx/src/read/story/runs.rs

@@ -98,9 +98,9 @@ impl Reader<'_> {
                 }
             }
             "mc:AlternateContent" => {
-                if let Some(o) = self.read_ink_alternate(sc, k, rels).or_else(|| self.read_chartex_alternate(sc, k, rels)) {
+                if let Some((o, link)) = self.read_alternate(sc, k, rels) {
                     let props = self.run_props_none(ctx);
-                    self.emit_obj(sc, pb, o, &props);
+                    self.emit_linked_obj(sc, pb, o, &props, link);
                 } else if let Some(c) = children_of_choice(k) {
                     self.read_inline_children(sc, pb, c, rels, ctx, depth + 1);
                 }
@@ -237,7 +237,7 @@ impl Reader<'_> {
             },
             "w:drawing" => {
                 if let Some(o) = self.read_drawing(sc, k, rels) {
-                    self.emit_obj(sc, pb, o, props);
+                    self.emit_linked_obj(sc, pb, o, props, Self::drawing_link(k, rels));
                 }
             }
             "w:pict" => {
@@ -298,8 +298,8 @@ impl Reader<'_> {
             ),
             "w:ruby" => self.read_ruby(sc, pb, k, props),
             "mc:AlternateContent" => {
-                if let Some(o) = self.read_ink_alternate(sc, k, rels).or_else(|| self.read_chartex_alternate(sc, k, rels)) {
-                    self.emit_obj(sc, pb, o, props);
+                if let Some((o, link)) = self.read_alternate(sc, k, rels) {
+                    self.emit_linked_obj(sc, pb, o, props, link);
                 } else if let Some(c) = children_of_choice(k) {
                     for x in c.els() {
                         self.read_run_child(sc, pb, x, (props, rels), note, depth + 1);
@@ -376,6 +376,14 @@ impl Reader<'_> {
         }
     }
 
+    /// [`Self::emit_obj`] for a drawing that carries its own `link`: a link around the run wins.
+    fn emit_linked_obj(&mut self, sc: &mut StoryCtx, pb: &mut PB, o: InlineObject, props: &CharProps, link: Option<String>) {
+        match link.filter(|_| props.link.is_none()) {
+            Some(link) => self.emit_obj(sc, pb, o, &CharProps { link: Some(link), ..props.clone() }),
+            None => self.emit_obj(sc, pb, o, props),
+        }
+    }
+
     pub(super) fn emit_obj(&mut self, sc: &mut StoryCtx, pb: &mut PB, o: InlineObject, props: &CharProps) {
         match sink(sc) {
             Some(f) => {

crates/docx/src/read/story/vml.rs

@@ -134,7 +134,10 @@ impl Reader<'_> {
             let id = img.attr("r:id").or_else(|| img.attr("r:pict"))?;
             let media = self.media_for(rels, id)?;
             let alt = shape.attr("alt").or_else(|| img.attr("o:title")).unwrap_or("").to_string();
-            return Some(InlineObject::Image { media, w, h, alt, float, crop: [0.0; 4], ole: None });
+            // Brightness and contrast (Word's picture watermark "Washout" sets them, #1711).
+            let frac = |k: &str| img.attr(k).and_then(crate::units::vml_number);
+            let fx = wordcraft_doc::picfx::PicFx::from_vml(frac("gain").unwrap_or(1.0), frac("blacklevel").unwrap_or(0.0));
+            return Some(InlineObject::Image { media, w, h, alt, float, crop: [0.0; 4], fx, ole: None });
         }
         // The shapetype the shape is drawn from (its `type`, ECMA-376 Part 4 VML), when written in
         // the same picture: what the shape leaves out, it takes from there.
@@ -221,6 +224,15 @@ impl Reader<'_> {
         {
             let id = self.read_textbox(sc, t, rels);
             if let Some(part) = self.doc.parts.get_mut(&id) {
+                // Text direction (#5193), as Word reads it: `mso-layout-flow-alt:bottom-to-top` is
+                // `vert270` whatever `layout-flow` says; otherwise `layout-flow:vertical…` (also
+                // `vertical-ideographic`, our `eaVert`) is `vert`, and anything else horizontal.
+                let tb_style = tb.attr("style").unwrap_or("");
+                if vml_style(tb_style, "mso-layout-flow-alt").is_some_and(|v| v.eq_ignore_ascii_case("bottom-to-top")) {
+                    part.body.vert = wordcraft_doc::props::TextVert::Vert270;
+                } else if vml_style(tb_style, "layout-flow").is_some_and(|v| v.to_ascii_lowercase().starts_with("vertical")) {
+                    part.body.vert = wordcraft_doc::props::TextVert::Vert;
+                }
                 // Internal margins `inset="l,t,r,b"` (ECMA-376 Part 4 §19.1.2.22), each one left
                 // out keeping Word's default; a bare number is in EMUs, VML's base unit.
                 // `o:insetmode="auto"` leaves them to the application.

crates/docx/src/read/styles.rs

@@ -2,6 +2,8 @@ use super::{
     AbstractNum, DocxError, El, Level, LevelSuffix, Num, NumFormat, PartFormat, Reader, Style, StyleKind, StyleSheet, TablePart, TableStyleParts,
     VAlign, flag, int, props, tw, u32_of,
 };
+use crate::units::tw_i16;
+use std::sync::Arc;
 
 /// Built-in style names Word stores in lower case, and how the UI shows them.
 pub(crate) const LOWER_NAMES: &[(&str, &str)] = &[
@@ -186,7 +188,7 @@ impl Reader<'_> {
         if !sheet.styles.iter().any(|s| s.quick) {
             sheet.add_default_gallery();
         }
-        self.doc.styles = sheet;
+        self.doc.styles = Arc::new(sheet);
         Ok(())
     }
 
@@ -196,7 +198,7 @@ impl Reader<'_> {
         let mut t = TableStyleParts::default();
         if let Some(tp) = s.child("w:tblPr") {
             if let Some(b) = tp.child("w:tblBorders") {
-                t.borders = Some(props::borders(b));
+                t.borders = Some(self.pc.borders(b));
             }
             t.fill = tp.child("w:shd").and_then(|e| self.pc.shd_fill(e));
             t.cell_margins = tp.child("w:tblCellMar").map(props::margins);
@@ -210,7 +212,7 @@ impl Reader<'_> {
             let f = PartFormat {
                 chr: c.child("w:rPr").map(|r| self.pc.rpr(r)).unwrap_or_default(),
                 fill: tc.and_then(|p| p.child("w:shd")).and_then(|e| self.pc.shd_fill(e)),
-                borders: tc.and_then(|p| p.child("w:tcBorders")).map(props::borders),
+                borders: tc.and_then(|p| p.child("w:tcBorders")).map(|b| self.pc.borders(b)),
                 align: ppr.as_ref().and_then(|p| p.align),
                 v_align: tc.and_then(|p| p.child_val("w:vAlign")).map(|v| match v {
                     "center" => VAlign::Center,
@@ -229,7 +231,7 @@ impl Reader<'_> {
                     t.fill = f.fill;
                 }
                 if let Some(b) = c.child("w:tblPr").and_then(|p| p.child("w:tblBorders")) {
-                    t.borders = Some(props::borders(b));
+                    t.borders = Some(self.pc.borders(b));
                 }
                 if f.v_align.is_some() {
                     t.regions.insert(part, PartFormat { v_align: f.v_align, ..PartFormat::default() });
@@ -298,8 +300,8 @@ impl Reader<'_> {
                 a.levels = levels;
             }
         }
-        self.doc.numbering.abstracts = abstracts;
-        self.doc.numbering.nums = nums;
+        Arc::make_mut(&mut self.doc.numbering).abstracts = abstracts;
+        Arc::make_mut(&mut self.doc.numbering).nums = nums;
         Ok(())
     }
 
@@ -316,12 +318,12 @@ impl Reader<'_> {
             lv.align = props::align(j);
         }
         if let Some(ind) = l.child("w:pPr").and_then(|p| p.child("w:ind")) {
-            if let Some(v) = tw(ind, "w:left").or_else(|| tw(ind, "w:start")) {
+            if let Some(v) = tw_i16(ind, "w:left").or_else(|| tw_i16(ind, "w:start")) {
                 lv.indent = v;
             }
-            if let Some(h) = tw(ind, "w:hanging") {
+            if let Some(h) = tw_i16(ind, "w:hanging") {
                 lv.hanging = h;
-            } else if let Some(f) = tw(ind, "w:firstLine") {
+            } else if let Some(f) = tw_i16(ind, "w:firstLine") {
                 lv.hanging = -f;
             }
         }

crates/docx/src/read/toc.rs

@@ -83,13 +83,14 @@ pub(super) fn toc_control_to_toc(body: &mut Blocks) {
     }
     // An empty paragraph opening the control takes the field. It may end a section, as above
     // entries in columns (#3889).
-    if engine.get(s).and_then(|b| b.as_para()).is_some_and(|p| p.is_empty()) {
+    let empty_heading = engine.get(s).and_then(|b| b.as_para()).is_some_and(|p| p.is_empty());
+    if empty_heading {
         field_to_heading(&mut engine, s + 1);
     } else {
         toc_field_to_heading(&mut engine);
     }
     if crate::write::written_toc_span(&engine) == Some((s, last)) {
-        *body = toc;
+        *body = if empty_heading { toc } else { engine };
     }
 }
 
@@ -97,7 +98,7 @@ pub(super) fn toc_control_to_toc(body: &mut Blocks) {
 /// reproduce exactly. Other imported TOCs keep their explicit boundaries, independent of styles.
 pub(super) fn toc_field_to_heading(body: &mut Blocks) {
     for i in 1..body.len() {
-        if body[i - 1].as_para().is_some_and(|p| p.props.style.as_deref() == Some("TOCHeading")) {
+        if body.get(i - 1).and_then(|b| b.as_para()).is_some_and(|p| p.props.style.as_deref() == Some("TOCHeading")) {
             field_to_heading(body, i);
         }
     }

crates/docx/src/smart_art.rs

@@ -86,6 +86,8 @@ enum Paint {
     None,
     /// Theme accent `n` (1–6), lightened to `tint` (thousandths of a percent kept) when set.
     Accent(usize, Option<u32>),
+    /// A colour of its own, lightened to `tint` when set.
+    Rgb(Rgb, Option<u32>),
     /// The theme's light background colour.
     Light,
 }
@@ -150,6 +152,22 @@ impl Sh {
         }
     }
 
+    /// Accent fills and outlines in item `i`'s own colour, when it has one. Light outlines
+    /// keep the theme's background colour so adjacent nodes remain separated.
+    fn recolor(&mut self, spec: &SmartArtSpec, i: usize) {
+        let Some(c) = spec.items.get(i).and_then(|it| it.color) else { return };
+        for p in [&mut self.fill, &mut self.line] {
+            if let Paint::Accent(_, tint) = *p {
+                *p = Paint::Rgb(c, tint);
+            }
+        }
+    }
+
+    fn item(mut self, spec: &SmartArtSpec, i: usize) -> Sh {
+        self.recolor(spec, i);
+        self
+    }
+
     fn with_text(mut self, text: Vec<(String, bool)>) -> Sh {
         self.text = text;
         self
@@ -226,7 +244,7 @@ fn basic_list(spec: &SmartArtSpec, w: f32, h: f32) -> Vec<Sh> {
         let in_row = if r + 1 == rows { n - r * cols } else { cols };
         let x0 = (w - (in_row as f32 * bw + (in_row - 1) as f32 * gx)) / 2.0;
         let b = [x0 + c as f32 * (bw + gx), y0 + r as f32 * (bh + gy), bw, bh];
-        out.push(Sh::node("rect", b, spec.colors.accent(k)).with_text(paras(spec, i, true)));
+        out.push(Sh::node("rect", b, spec.colors.accent(k)).item(spec, i).with_text(paras(spec, i, true)));
     }
     out
 }
@@ -247,11 +265,14 @@ fn process(spec: &SmartArtSpec, w: f32, h: f32) -> Vec<Sh> {
         let mut a = Sh::node("rightArrow", [gx + (gap - aw) / 2.0, h / 2.0 - ah / 2.0, aw, ah], spec.colors.accent(k - 1));
         a.fill = Paint::Accent(spec.colors.accent(k - 1), Some(60_000));
         a.line = Paint::None;
+        if let Some(&i) = tops.get(k - 1) {
+            a.recolor(spec, i);
+        }
         out.push(a);
     }
     for (k, &i) in tops.iter().enumerate() {
         let b = [k as f32 * (nw + gap), y, nw, nh];
-        out.push(Sh::node("roundRect", b, spec.colors.accent(k)).with_text(paras(spec, i, true)));
+        out.push(Sh::node("roundRect", b, spec.colors.accent(k)).item(spec, i).with_text(paras(spec, i, true)));
     }
     out
 }
@@ -294,6 +315,9 @@ fn cycle(spec: &SmartArtSpec, w: f32, h: f32) -> Vec<Sh> {
                 a.rot = (mid + 90.0).rem_euclid(360.0);
                 a.fill = Paint::Accent(spec.colors.accent(k), Some(60_000));
                 a.line = Paint::None;
+                if let Some(&i) = tops.get(k) {
+                    a.recolor(spec, i);
+                }
                 out.push(a);
             }
         }
@@ -301,7 +325,7 @@ fn cycle(spec: &SmartArtSpec, w: f32, h: f32) -> Vec<Sh> {
     for (k, &i) in tops.iter().enumerate() {
         let (x, y) = if n == 1 { (cx, cy) } else { polar(cx, cy, r, angle(k, n)) };
         let b = [x - d / 2.0, y - d / 2.0, d, d];
-        out.push(Sh::node("ellipse", b, spec.colors.accent(k)).with_text(paras(spec, i, true)).text_box(inset(b, ROUND_INSET)));
+        out.push(Sh::node("ellipse", b, spec.colors.accent(k)).item(spec, i).with_text(paras(spec, i, true)).text_box(inset(b, ROUND_INSET)));
     }
     out
 }
@@ -360,19 +384,19 @@ fn hierarchy(spec: &SmartArtSpec, w: f32, h: f32) -> Vec<Sh> {
         let (px, pb) = (cx(i), top(i) + nh);
         let ct = ks.first().map_or(pb, |&k| top(k));
         let mid = (pb + ct) / 2.0;
-        out.push(Sh::line(px, pb, px, mid, accent));
+        out.push(Sh::line(px, pb, px, mid, accent).item(spec, i));
         let xs: Vec<f32> = ks.iter().map(|&k| cx(k)).collect();
         let (lo, hi) = xs.iter().fold((px, px), |(lo, hi), &x| (lo.min(x), hi.max(x)));
         if hi - lo > 0.5 {
-            out.push(Sh::line(lo, mid, hi, mid, accent));
+            out.push(Sh::line(lo, mid, hi, mid, accent).item(spec, i));
         }
         for x in xs {
-            out.push(Sh::line(x, mid, x, ct, accent));
+            out.push(Sh::line(x, mid, x, ct, accent).item(spec, i));
         }
     }
     for i in 0..n {
         let b = [cx(i) - nw / 2.0, top(i), nw, nh];
-        out.push(Sh::node("roundRect", b, spec.colors.accent(i)).with_text(paras(spec, i, false)));
+        out.push(Sh::node("roundRect", b, spec.colors.accent(i)).item(spec, i).with_text(paras(spec, i, false)));
     }
     out
 }
@@ -400,7 +424,7 @@ fn pyramid(spec: &SmartArtSpec, w: f32, h: f32) -> Vec<Sh> {
             let mid_w = (top_w + bot_w) / 2.0 * 0.8;
             s.text_box([w / 2.0 - mid_w / 2.0, b[1] + bh * 0.08, mid_w, bh * 0.84])
         };
-        out.push(sh.with_text(paras(spec, i, true)));
+        out.push(sh.item(spec, i).with_text(paras(spec, i, true)));
     }
     out
 }
@@ -418,7 +442,7 @@ fn radial(spec: &SmartArtSpec, w: f32, h: f32) -> Vec<Sh> {
     if n == 0 {
         let d = m * 0.5;
         let b = [cx - d / 2.0, cy - d / 2.0, d, d];
-        out.push(Sh::node("ellipse", b, spec.colors.accent(0)).with_text(paras(spec, 0, false)).text_box(inset(b, ROUND_INSET)));
+        out.push(Sh::node("ellipse", b, spec.colors.accent(0)).item(spec, 0).with_text(paras(spec, 0, false)).text_box(inset(b, ROUND_INSET)));
         return out;
     }
     let step = if n == 1 { 1.0 } else { (std::f32::consts::PI / n as f32).sin() };
@@ -432,14 +456,14 @@ fn radial(spec: &SmartArtSpec, w: f32, h: f32) -> Vec<Sh> {
     let at = |k: usize| polar(cx, cy, r, angle(k, n));
     for k in 0..n {
         let (x, y) = at(k);
-        out.push(Sh::line(cx, cy, x, y, spec.colors.accent(0)));
+        out.push(Sh::line(cx, cy, x, y, spec.colors.accent(0)).item(spec, 0));
     }
     let bc = [cx - dc / 2.0, cy - dc / 2.0, dc, dc];
-    out.push(Sh::node("ellipse", bc, spec.colors.accent(0)).with_text(paras(spec, 0, false)).text_box(inset(bc, ROUND_INSET)));
+    out.push(Sh::node("ellipse", bc, spec.colors.accent(0)).item(spec, 0).with_text(paras(spec, 0, false)).text_box(inset(bc, ROUND_INSET)));
     for (k, &i) in spokes.iter().enumerate() {
         let (x, y) = at(k);
         let b = [x - ds / 2.0, y - ds / 2.0, ds, ds];
-        out.push(Sh::node("ellipse", b, spec.colors.accent(k + 1)).with_text(paras(spec, i, true)).text_box(inset(b, ROUND_INSET)));
+        out.push(Sh::node("ellipse", b, spec.colors.accent(k + 1)).item(spec, i).with_text(paras(spec, i, true)).text_box(inset(b, ROUND_INSET)));
     }
     out
 }
@@ -457,7 +481,7 @@ fn matrix(spec: &SmartArtSpec, w: f32, h: f32) -> Vec<Sh> {
     for (k, &i) in tops.iter().enumerate() {
         let (r, c) = (k / cols, k % cols);
         let b = [x0 + c as f32 * cell + gap / 2.0, y0 + r as f32 * cell + gap / 2.0, cell - gap, cell - gap];
-        out.push(Sh::node("roundRect", b, spec.colors.accent(k)).with_text(paras(spec, i, true)).text_box(inset(b, 0.1)));
+        out.push(Sh::node("roundRect", b, spec.colors.accent(k)).item(spec, i).with_text(paras(spec, i, true)).text_box(inset(b, 0.1)));
     }
     out
 }
@@ -478,22 +502,22 @@ fn venn(spec: &SmartArtSpec, w: f32, h: f32) -> Vec<Sh> {
         (x, y, deg)
     };
     let mut out = Vec::new();
-    for (j, _) in tops.iter().enumerate() {
+    for (j, &i) in tops.iter().enumerate() {
         let (x, y, _) = place(j);
         let mut s = Sh::node("ellipse", [x - r, y - r, 2.0 * r, 2.0 * r], spec.colors.accent(j));
         s.fill = Paint::Accent(spec.colors.accent(j), Some(40_000));
         s.alpha = Some(60_000);
         s.line = Paint::None;
-        out.push(s);
+        out.push(s.item(spec, i));
     }
     // Every outline over every fill, so each circle shows whole.
-    for (j, _) in tops.iter().enumerate() {
+    for (j, &i) in tops.iter().enumerate() {
         let (x, y, _) = place(j);
         let mut s = Sh::node("ellipse", [x - r, y - r, 2.0 * r, 2.0 * r], spec.colors.accent(j));
         s.fill = Paint::None;
         s.line = Paint::Accent(spec.colors.accent(j), None);
         s.line_w = 1.5;
-        out.push(s);
+        out.push(s.item(spec, i));
     }
     for (j, &i) in tops.iter().enumerate() {
         let (x, y, deg) = place(j);
@@ -554,6 +578,23 @@ fn paint(w: &mut W, p: Paint, alpha: Option<u32>) {
             w.empty("a:schemeClr", &[("val", "lt1")]);
             w.close("a:solidFill");
         }
+        Paint::Rgb(c, tint) => {
+            w.open("a:solidFill", &[]);
+            let val = c.hex();
+            if tint.is_none() && alpha.is_none() {
+                w.empty("a:srgbClr", &[("val", &val)]);
+            } else {
+                w.open("a:srgbClr", &[("val", &val)]);
+                if let Some(t) = tint {
+                    w.empty("a:tint", &[("val", &t.to_string())]);
+                }
+                if let Some(a) = alpha {
+                    w.empty("a:alpha", &[("val", &a.to_string())]);
+                }
+                w.close("a:srgbClr");
+            }
+            w.close("a:solidFill");
+        }
         Paint::Accent(n, tint) => {
             w.open("a:solidFill", &[]);
             let val = format!("accent{}", n.clamp(1, 6));
@@ -743,7 +784,16 @@ pub fn data_xml(spec: &SmartArtSpec, drawing_rel: &str) -> String {
         let (node, cxn, par, sib) = item_ids(i);
         w.open("dgm:pt", &[("modelId", &node)]);
         w.empty("dgm:prSet", &[]);
-        w.empty("dgm:spPr", &[]);
+        match it.color {
+            Some(c) => {
+                w.open("dgm:spPr", &[]);
+                w.open("a:solidFill", &[]);
+                w.empty("a:srgbClr", &[("val", &c.hex())]);
+                w.close("a:solidFill");
+                w.close("dgm:spPr");
+            }
+            None => w.empty("dgm:spPr", &[]),
+        }
         w.open("dgm:t", &[]);
         w.empty("a:bodyPr", &[]);
         w.empty("a:lstStyle", &[]);
@@ -1062,7 +1112,9 @@ pub(crate) fn spec_of(data: &El) -> Option<SmartArtSpec> {
         .zip(&levels)
         .map(|(p, l)| {
             let text: String = p.child("dgm:t")?.child("a:p")?.children("a:r").filter_map(|r| r.child("a:t")).map(|t| t.text()).collect();
-            Some(SmartArtItem { text, level: (*l)? })
+            let color =
+                p.child("dgm:spPr").and_then(|s| s.child("a:solidFill")).and_then(|f| f.child("a:srgbClr")).and_then(|c| Rgb::parse(c.attr("val")?));
+            Some(SmartArtItem { text, level: (*l)?, color })
         })
         .collect::<Option<Vec<_>>>()?;
     let spec = SmartArtSpec { layout, items, colors }.sanitized();
@@ -1138,6 +1190,18 @@ mod tests {
         }
     }
 
+    #[test]
+    fn an_items_own_colour_is_saved_drawn_and_read_back() {
+        let mut spec = SmartArtSpec::sample(SmartArtLayout::Process);
+        spec.items[1].color = Some(Rgb(0x12, 0x34, 0x56));
+        let spec = spec.sanitized();
+        assert_eq!(spec_of(&parse(&data_xml(&spec, "rId9"))).as_ref(), Some(&spec));
+        let drawing = drawing_xml(&spec, 400.0, 250.0);
+        assert_eq!(drawing.matches("<a:srgbClr val=\"123456\"").count(), 2, "the step and the arrow after it");
+        let plain = drawing_xml(&SmartArtSpec::sample(SmartArtLayout::Process), 400.0, 250.0);
+        assert!(!plain.contains("srgbClr"));
+    }
+
     #[test]
     fn a_diagram_with_more_than_the_model_stays_read_only() {
         let spec = SmartArtSpec::sample(SmartArtLayout::Process);

crates/docx/src/units.rs

@@ -84,6 +84,29 @@ pub fn int(s: &str) -> Option<i64> {
     v.is_finite().then(|| v.round().clamp(i64::MIN as f64, i64::MAX as f64) as i64)
 }
 
+/// A VML number (`gain`, `blacklevel`), unclamped: 16.16 fixed point with an `f` suffix
+/// (`19661f` is 0.3) or a plain decimal (`0.3`). None for anything else or non-finite.
+pub fn vml_number(s: &str) -> Option<f64> {
+    let t = s.trim();
+    if t.is_empty() || t.len() > 40 {
+        return None;
+    }
+    let v = match t.strip_suffix('f') {
+        Some(fixed) => fixed.trim().parse::<f64>().ok()? / 65_536.0,
+        None => t.parse::<f64>().ok()?,
+    };
+    v.is_finite().then_some(v)
+}
+
+/// A picture's brightness, contrast and transparency: `a:lum` (ECMA-376 §20.1.8.42) and
+/// `a:alphaModFix` (§20.1.8.4) in its `a:blip` (#1766).
+pub fn blip_fx(blip: &El) -> wordcraft_doc::picfx::PicFx {
+    let val = |el: Option<&El>, k: &str| el.and_then(|e| e.attr(k)).and_then(int).map(|v| v.clamp(-1_000_000, 1_000_000) as i32);
+    let lum = blip.child("a:lum");
+    let amt = val(blip.child("a:alphaModFix"), "amt").unwrap_or(100_000).clamp(0, 100_000);
+    wordcraft_doc::picfx::PicFx::new(val(lum, "bright").unwrap_or(0), val(lum, "contrast").unwrap_or(0), 100_000 - amt)
+}
+
 pub fn u32_of(s: &str) -> Option<u32> {
     int(s).map(|v| v.clamp(0, u32::MAX as i64) as u32)
 }
@@ -93,6 +116,27 @@ pub fn tw(e: &El, attr: &str) -> Option<f32> {
     e.attr(attr).and_then(|v| measure(v, 20.0))
 }
 
+/// Signed twips Word keeps in 16 bits (paragraph and list indents, ECMA-376 Part 1 §17.3.1.12):
+/// a plain integer outside `i16` keeps its low 16 bits, two's complement, as Word reads it
+/// (66256 → 720). Values with a unit or a fraction read as [`tw`].
+pub fn tw_i16(e: &El, attr: &str) -> Option<f32> {
+    let v = e.attr(attr)?;
+    match v.trim().parse::<i64>() {
+        Ok(i) => Some(f32::from(i as i16) / 20.0),
+        Err(_) => measure(v, 20.0),
+    }
+}
+
+/// Unsigned twips Word keeps in 16 bits (paragraph spacing, §17.3.1.33): a plain integer above
+/// 65535 keeps its low 16 bits (65537 → 1). Other values read as [`tw`].
+pub fn tw_u16(e: &El, attr: &str) -> Option<f32> {
+    let v = e.attr(attr)?;
+    match v.trim().parse::<i64>() {
+        Ok(i) if i > 0xFFFF => Some(f32::from((i & 0xFFFF) as u16) / 20.0),
+        _ => measure(v, 20.0),
+    }
+}
+
 /// ST_OnOff: absent `w:val` = on; `0`, `false`, `off` = off.
 pub fn on_off(e: &El) -> bool {
     match e.attr("w:val") {
@@ -125,6 +169,24 @@ pub fn emu(pt: f32) -> String {
 mod tests {
     use super::*;
 
+    #[test]
+    fn twips_beyond_16_bits_wrap() {
+        let el = |a: &str| El { attrs: vec![("w:v".into(), a.into())], ..El::default() };
+        let s = |a: &str| tw_i16(&el(a), "w:v");
+        let u = |a: &str| tw_u16(&el(a), "w:v");
+        assert_eq!(s("66256"), Some(36.0));
+        assert_eq!(s("-65516"), Some(1.0));
+        assert_eq!(s("98304"), Some(-1638.4));
+        assert_eq!(s("32767"), Some(1638.35));
+        assert_eq!(s("-32768"), Some(-1638.4));
+        assert_eq!(s("1in"), Some(72.0));
+        assert_eq!(s("x"), None);
+        assert_eq!(u("65537"), Some(0.05));
+        assert_eq!(u("65535"), Some(3276.75));
+        assert_eq!(u("12pt"), Some(12.0));
+        assert_eq!(u("-20"), Some(-1.0));
+    }
+
     #[test]
     fn measures() {
         assert_eq!(measure("1440", 20.0), Some(72.0));

crates/docx/src/wordml2003_write.rs

@@ -317,7 +317,8 @@ impl<'a> Conv<'a> {
 
     /// A DrawingML picture as a `w:pict` (inline, at its extent); any other drawing is dropped.
     fn picture(&mut self, drawing: &El, rels: &Rels) -> Vec<Node> {
-        let Some(id) = drawing.find("a:blip").and_then(|b| b.attr("r:embed")) else { return Vec::new() };
+        let Some(blip) = drawing.find("a:blip") else { return Vec::new() };
+        let Some(id) = blip.attr("r:embed") else { return Vec::new() };
         let Some(target) = rels.by_id(id).filter(|r| !r.external).map(|r| r.target.clone()) else { return Vec::new() };
         let Some(src) = self.bin_for(&target) else { return Vec::new() };
         let pt = |k: &str| {
@@ -329,7 +330,19 @@ impl<'a> Conv<'a> {
         let shape_id = format!("_x0000_i{}", self.next_shape);
         self.next_shape = self.next_shape.saturating_add(1);
         let mut kids: Vec<Node> = std::mem::take(&mut self.bins).into_iter().map(Node::El).collect();
-        kids.push(Node::El(self.shape_frame(&shape_id, &style, &src, &alt)));
+        let mut shape = self.shape_frame(&shape_id, &style, &src, &alt);
+        // Brightness and contrast as VML `gain` and `blacklevel` (16.16 fixed point, #1711).
+        // `v:imagedata` has no opacity, so a picture's transparency (`a:alphaModFix`) is dropped.
+        let fx = crate::units::blip_fx(blip);
+        if fx.changes_pixels()
+            && let Some(Node::El(data)) = shape.kids.iter_mut().find(|k| matches!(k, Node::El(e) if e.name == "v:imagedata"))
+        {
+            let (gain, black) = fx.to_vml();
+            let fixed = |v: f64| format!("{}f", (v * 65_536.0).round().clamp(-1.0e12, 1.0e12) as i64);
+            data.attrs.push(("gain".into(), fixed(gain)));
+            data.attrs.push(("blacklevel".into(), fixed(black)));
+        }
+        kids.push(Node::El(shape));
         vec![Node::El(plain("w:pict", kids))]
     }
 

crates/docx/src/write/embed.rs

@@ -353,11 +353,11 @@ impl super::Writer<'_> {
         };
         self.embeds.markup_bytes = left;
         let mut root = xml::parse(src.xml.as_bytes()).ok()?;
-        let mut used = Vec::new();
+        let mut used = HashSet::new();
         rel_attr_values(&root, &mut used, 0);
         // Every relationship the markup uses must be writable, or the object would be broken.
-        let writable = |id: &String| src.refs.iter().any(|r| r.id == *id && (r.external || self.embeds.available(&r.target)));
-        if !used.iter().all(writable) {
+        let writable: HashSet<&str> = src.refs.iter().filter(|r| r.external || self.embeds.available(&r.target)).map(|r| r.id.as_str()).collect();
+        if !used.iter().all(|id| writable.contains(id.as_str())) {
             log::warn!("docx: an embedded object's parts are missing; it is left out");
             return None;
         }
@@ -499,8 +499,9 @@ impl super::Writer<'_> {
                 }
             }
             "wp:effectExtent" => {
-                // Word's effect extent also covers a rotated object's overhang (its rotated bounds).
-                let (px, py) = float.spin_pad(w, h);
+                // Word's effect extent also covers a rotated object's overhang (its rotated bounds),
+                // unless the file it came from kept none.
+                let (px, py) = if float.is_bare_turn() { (0.0, 0.0) } else { float.spin_pad(w, h) };
                 let [el, et, er, eb] = float.effect_extent();
                 let ext = [el + px, et + py, er + px, eb + py].map(emu);
                 e.attrs.retain(|(k, _)| !matches!(k.as_str(), "l" | "t" | "r" | "b"));
@@ -637,13 +638,15 @@ fn set_attr(e: &mut El, k: &str, v: &str) {
     }
 }
 
-fn rel_attr_values(e: &El, out: &mut Vec<String>, depth: usize) {
+/// The distinct relationship ids `e` uses: a set, so a large tree (a native document's markup
+/// never went through the reader's id limit) costs no scan per attribute.
+fn rel_attr_values(e: &El, out: &mut HashSet<String>, depth: usize) {
     if depth > MAX_DEPTH {
         return;
     }
     for (k, v) in &e.attrs {
         if xml::is_rel_attr(&e.name, k) && !out.contains(v) {
-            out.push(v.clone());
+            out.insert(v.clone());
         }
     }
     for c in e.els() {
@@ -851,6 +854,24 @@ fn valid_part_name(name: &str) -> String {
 mod tests {
     use super::*;
 
+    #[test]
+    fn relationship_ids_are_collected_once_each() {
+        // A native document's embedded markup skips the reader's id limit: 40,000 distinct ids
+        // (each named twice) were a scan of every id collected so far, per attribute.
+        let mut xml = String::from(
+            r#"<w:p xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main" xmlns:r="http://schemas.openxmlformats.org/officeDocument/2006/relationships">"#,
+        );
+        for i in 0..40_000 {
+            xml.push_str(&format!(r#"<w:r r:id="rId{i}"/><w:r r:id="rId{i}"/>"#));
+        }
+        xml.push_str("</w:p>");
+        let root = xml::parse(xml.as_bytes()).unwrap();
+        let mut used = HashSet::new();
+        rel_attr_values(&root, &mut used, 0);
+        assert_eq!(used.len(), 40_000);
+        assert!(used.contains("rId39999"));
+    }
+
     #[test]
     fn charts_cut_to_the_limits_on_save_say_so_once() {
         // A model deserialised as is can hold more than a chart part keeps; the save left the rest

crates/docx/src/write/math.rs

@@ -1,7 +1,8 @@
 //! [`wordcraft_doc::math`] tree → Office Math (OMML).
 
 use wordcraft_doc::math::{
-    Arg, BrkBin, BrkBinSub, ColJc, FracKind, LimLoc, MAX_DEPTH, MNode, MRun, MScr, MSty, Math, MathJc, MathProps, ScriptKind, parse_linear,
+    Arg, BrkBin, BrkBinSub, ColJc, FracKind, LimLoc, MAX_DEPTH, MAX_MATRIX_COLS, MAX_MATRIX_ROWS, MNode, MRun, MScr, MSty, Math, MathJc, MathProps,
+    ScriptKind, parse_linear,
 };
 use wordcraft_doc::props::Rgb;
 
@@ -181,7 +182,10 @@ fn char_prop(w: &mut W, name: &str, c: Option<char>) {
 
 fn write_matrix(w: &mut W, n: &MNode, depth: usize) {
     let MNode::Matrix { rows, col_jc } = n else { return };
-    let cols = rows.iter().map(|r| r.len()).max().unwrap_or(0).max(1);
+    // Every row is padded to the widest one, so a ragged matrix is clamped to the reader's
+    // caps rather than written as rows × widest cells (#3643).
+    let rows = rows.get(..rows.len().min(MAX_MATRIX_ROWS)).unwrap_or_default();
+    let cols = rows.iter().map(|r| r.len()).max().unwrap_or(0).clamp(1, MAX_MATRIX_COLS);
     w.open("m:m", &[]);
     w.open("m:mPr", &[]);
     w.open("m:mcs", &[]);

crates/docx/src/write/parts.rs

@@ -294,8 +294,10 @@ fn write_document_body(doc: &Document, wr: &mut Writer<'_>, rels: &mut PartRels,
             wr.used_media.insert(p.media.clone());
             w.open("v:background", &[("id", "_x0000_s1025"), ("o:bwmode", "white")]);
             let mut fa: Vec<(&str, &str)> = vec![("r:id", &rid)];
-            if !p.fill_type.is_empty() {
-                fa.push(("type", &p.fill_type));
+            // VML has no centred fill (an ODT page background's): stretched to the page.
+            let kind = if p.fill_type.eq_ignore_ascii_case("center") { "frame" } else { p.fill_type.as_str() };
+            if !kind.is_empty() {
+                fa.push(("type", kind));
             }
             if !p.title.is_empty() {
                 fa.push(("o:title", &p.title));

crates/docx/src/write/props.rs

@@ -108,6 +108,10 @@ pub fn rpr_inner(w: &mut W, c: &CharProps) {
     }
     toggle(w, "w:rtl", c.rtl);
     toggle(w, "w:cs", c.cs);
+    // Emphasis mark (§17.3.2.12), after `w:cs` and before `w:lang`.
+    if let Some(m) = c.emphasis {
+        w.val("w:em", m.ooxml());
+    }
     let mut lang_attrs: Vec<(&str, &str)> = Vec::with_capacity(3);
     if let Some(l) = &c.lang {
         lang_attrs.push(("w:val", l));
@@ -283,6 +287,12 @@ pub fn ppr_inner(w: &mut W, p: &ParaProps, framed: bool, num_change: Option<&Cha
     if let Some(a) = p.align {
         w.val("w:jc", align_val(a));
     }
+    if let Some(direction) = p.text_direction {
+        w.val("w:textDirection", direction.ooxml());
+    }
+    if let Some(t) = p.text_alignment {
+        w.val("w:textAlignment", t.ooxml());
+    }
     if let Some(l) = p.outline_level {
         w.val("w:outlineLvl", &n(l.min(9) as i64));
     }
@@ -319,6 +329,9 @@ fn write_frame(w: &mut W, p: &ParaProps, framed: bool) {
             a.push(("w:vSpace", &vs));
         }
         a.push(("w:wrap", &f.wrap));
+        if let Some(anchor_lock) = f.anchor_lock {
+            a.push(("w:anchorLock", if anchor_lock { "1" } else { "0" }));
+        }
         a.push(("w:vAnchor", &f.v_anchor));
         a.push(("w:hAnchor", &f.h_anchor));
         match &f.x_align {
@@ -366,6 +379,7 @@ fn write_text_flow(w: &mut W, p: &ParaProps) {
     toggle(w, "w:autoSpaceDE", p.auto_space_de);
     toggle(w, "w:autoSpaceDN", p.auto_space_dn);
     toggle(w, "w:bidi", p.bidi);
+    toggle(w, "w:adjustRightInd", p.adjust_right_indent);
     toggle(w, "w:snapToGrid", p.snap_to_grid);
 }
 

crates/docx/src/write/sdt.rs

@@ -86,11 +86,28 @@ pub fn plan(bl: &Blocks) -> Plan<'_> {
                 && (0..off).step_by(OBJ.len_utf8()).all(|o| matches!(p.object_at(o), Some(wordcraft_doc::InlineObject::ControlStart { .. })))
         })
     };
+    // Bookmark, comment and permission markers after the end may stay there: a body-level
+    // bookmark after a block control (Word's `_GoBack`) is read onto the end of its paragraph,
+    // and the control still covers the whole paragraph.
     let at_end = |b: usize, off: usize| {
+        use wordcraft_doc::InlineObject as O;
         para(b).is_some_and(|p| {
             let from = off + OBJ.len_utf8();
             p.text.get(from..).is_some_and(|t| t.chars().all(|c| c == OBJ))
-                && (from..p.len()).step_by(OBJ.len_utf8()).all(|o| matches!(p.object_at(o), Some(wordcraft_doc::InlineObject::ControlEnd)))
+                && (from..p.len()).step_by(OBJ.len_utf8()).all(|o| {
+                    matches!(
+                        p.object_at(o),
+                        Some(
+                            O::ControlEnd
+                                | O::BookmarkStart { .. }
+                                | O::BookmarkEnd { .. }
+                                | O::CommentStart { .. }
+                                | O::CommentEnd { .. }
+                                | O::PermStart { .. }
+                                | O::PermEnd { .. }
+                        )
+                    )
+                })
         })
     };
     let mut mode = vec![Mode::Inline; ranges.len()];