Merge remote-tracking branch 'origin/main' into HEAD

; Conflicts:
;	ROADMAP.md

.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 }}

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;
         }

apps/wordcraft-cli/src/tests.rs

@@ -211,6 +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
@@ -228,6 +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/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/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;
@@ -421,14 +423,15 @@ 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);
@@ -449,9 +452,22 @@ fn main() -> eframe::Result {
     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;
@@ -566,6 +582,10 @@ 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);
@@ -605,8 +625,9 @@ 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);
                 }
             }

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)]

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/kept.rs

@@ -478,8 +478,9 @@ fn family_fingerprint(doc: &crate::Document, root: u32, replies: &HashMap<u32, V
 }
 
 /// 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, the page colour and
-/// picture, and the notes' numbering.
+/// 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);
@@ -490,8 +491,10 @@ pub fn pages_fingerprint(doc: &crate::Document) -> u64 {
     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),
     )) {

crates/doc/src/lib.rs

@@ -90,8 +90,8 @@ pub use settings::{
     COMPAT_MODE_CURRENT, CoreProps, CustomProp, DEFAULT_GRID, DEFAULT_HYPHENATION_ZONE, LEGACY_COMPAT_MODE, OdfLayout, PagePicture, Settings, Source,
     THEME_COLORS, Watermark, Zoom, ZoomFit,
 };
-pub(crate) use text_walk::walk_between;
-pub use text_walk::{PARA_WALK_DEPTH, count_chars, count_chars_no_spaces, 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)]

crates/doc/src/link.rs

@@ -125,6 +125,7 @@ mod tests {
             ("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")),

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) {
@@ -325,12 +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 {
-            self.labels.insert((abs.id, lv as u8), CountedLabel { label: None, levels: Vec::new() });
-            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
@@ -382,7 +379,14 @@ impl Counters {
         let started = *fl;
         let numeric = def.format != NumFormat::Bullet;
         let label = if numeric { render_label(numbering, num, &def, &vals, &started) } else { def.text.clone() };
-        self.labels.insert((abs.id, lv as u8), CountedLabel { label: numeric.then(|| label.clone()), levels: template_levels(&def.text) });
+        // 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
@@ -455,9 +459,9 @@ fn render_label(numbering: &Numbering, num: u32, def: &Level, vals: &[u32; 9], s
             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").
+            // 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) {
-                NumFormat::DecimalZero => NumFormat::DecimalZero,
+                f @ (NumFormat::DecimalZero | NumFormat::None) => f,
                 _ if def.legal => NumFormat::Decimal,
                 f => f,
             };
@@ -556,6 +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/section.rs

@@ -289,13 +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(),
@@ -349,7 +350,6 @@ impl NumFormat {
             | NumFormat::HindiNumbers
             | NumFormat::HindiCounting
             | NumFormat::ThaiLetters
-            | NumFormat::ThaiNumbers
             | NumFormat::ThaiCounting
             | NumFormat::BahtText
             | NumFormat::DollarText => n.to_string(),
@@ -357,6 +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}'];
@@ -999,6 +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/text_walk.rs

@@ -48,7 +48,7 @@ pub fn count_chars_no_spaces(s: &str) -> usize {
 }
 
 /// How deep in nested tables [`Document::para_paths`] goes: cells below this are skipped.
-pub const PARA_WALK_DEPTH: usize = 16;
+pub(crate) const PARA_WALK_DEPTH: usize = 16;
 
 /// The note and text box stories an object shows, for `Document::reachable`.
 pub(crate) fn notes_and_boxes(o: &InlineObject) -> Vec<(u32, &'static [PartKind])> {

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/lib.rs

@@ -502,29 +502,37 @@ 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() {
@@ -552,7 +560,7 @@ 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).
@@ -715,16 +723,16 @@ 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);
     }
@@ -854,34 +862,53 @@ 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());
         }
@@ -900,6 +927,18 @@ fn push_formatted<'a>(pb: &mut ParaBuild, cp: u32, c: char, ctx: &WalkCtx<'a>, f
     }
 }
 
+/// 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/tests.rs

@@ -2880,3 +2880,153 @@ fn field_results_keep_their_text_and_runs() {
         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/docx/src/read/story/drawing.rs

@@ -118,9 +118,33 @@ 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 +173,8 @@ 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 == "?") {
@@ -563,6 +585,11 @@ 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/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/write/story/drawing.rs

@@ -10,9 +10,19 @@ pub(super) struct DrawingInfo<'a> {
     pub docpr: &'a str,
     pub name: &'a str,
     pub alt: &'a str,
+    /// The relationship id of the drawing's hyperlink (`a:hlinkClick`), see [`Writer::drawing_link`].
+    pub link: Option<&'a str>,
 }
 
 impl Writer<'_> {
+    /// The relationship of the hyperlink a drawing in a run with `props` carries on itself, as
+    /// Word writes a picture's link (`wp:docPr/a:hlinkClick`, #5267): a web or file address. A
+    /// link to a place in the document (`#name`) is only the `w:hyperlink` around the run.
+    pub(super) fn drawing_link(props: &wordcraft_doc::CharProps, rels: &mut PartRels) -> Option<String> {
+        let url = props.link.as_deref().filter(|l| !l.is_empty() && !l.starts_with('#'))?;
+        Some(rels.add(rt::HYPERLINK, url, true))
+    }
+
     /// A picture's `pic:pic` (the image in `media/{file}`), at `off` in its group (or 0, 0).
     pub(super) fn pic(&mut self, w: &mut W, o: &InlineObject, file: &str, off: (f32, f32), rels: &mut PartRels) {
         let InlineObject::Image { media, w: iw, h: ih, alt, crop, float, fx, .. } = o else { return };
@@ -20,7 +30,7 @@ impl Writer<'_> {
         self.used_media.insert(media.clone());
         w.open("pic:pic", &[]);
         w.open("pic:nvPicPr", &[]);
-        nv_props(w, "pic:cNvPr", "0", file, alt, &float.alt, float.hidden);
+        nv_props(w, "pic:cNvPr", ("0", file), alt, &float.alt, float.hidden, None);
         w.open("pic:cNvPicPr", &[]);
         w.empty("a:picLocks", &[("noChangeAspect", "1"), ("noChangeArrowheads", "1")]);
         w.close("pic:cNvPicPr");
@@ -66,7 +76,7 @@ impl Writer<'_> {
         w.open("wps:wsp", &[]);
         if let Some(id) = id {
             let name = shape_name(*kind, freeform.as_deref(), id);
-            nv_props(w, "wps:cNvPr", id, &name, &float.alt.text, &float.alt, float.hidden);
+            nv_props(w, "wps:cNvPr", (id, &name), &float.alt.text, &float.alt, float.hidden, None);
         }
         if kind.is_connector() {
             // A connector's glued ends (ECMA-376 §20.1.2.2.8 `cNvCxnSpPr`).
@@ -235,7 +245,7 @@ impl Writer<'_> {
     }
 
     pub(super) fn drawing_open(&mut self, w: &mut W, info: DrawingInfo<'_>) {
-        let DrawingInfo { float, size: (iw, ih), docpr, name, alt } = info;
+        let DrawingInfo { float, size: (iw, ih), docpr, name, alt, link } = info;
         let (cx, cy) = (emu(iw.max(0.0)), emu(ih.max(0.0)));
         if float.wrap == Wrap::Inline {
             w.open("wp:inline", &[("distT", "0"), ("distB", "0"), ("distL", "0"), ("distR", "0")]);
@@ -316,7 +326,7 @@ impl Writer<'_> {
             Wrap::TopAndBottom => w.empty("wp:wrapTopAndBottom", &[]),
             Wrap::BehindText | Wrap::InFrontOfText => w.empty("wp:wrapNone", &[]),
         }
-        nv_props(w, "wp:docPr", docpr, name, alt, &float.alt, float.hidden);
+        nv_props(w, "wp:docPr", (docpr, name), alt, &float.alt, float.hidden, link);
     }
 }
 
@@ -502,8 +512,9 @@ pub(in super::super) fn spin_attrs(spin: wordcraft_geom::Spin) -> Vec<(&'static
 /// A drawing's `wp:docPr` (or a group member's `pic:cNvPr`/`wps:cNvPr`, ECMA-376
 /// `CT_NonVisualDrawingProps`) with its alt text `descr`, its title, and the decorative mark as
 /// the `adec:decorative` extension ([MS-ODRAWXML]). A description that is only the title (read
-/// from a file with no description) is not written twice.
-pub(in super::super) fn nv_props(w: &mut W, tag: &str, id: &str, name: &str, descr: &str, alt: &AltText, hidden: bool) {
+/// from a file with no description) is not written twice. `link` is the relationship id of the
+/// drawing's hyperlink (`a:hlinkClick`, the first child the schema allows).
+pub(in super::super) fn nv_props(w: &mut W, tag: &str, (id, name): (&str, &str), descr: &str, alt: &AltText, hidden: bool, link: Option<&str>) {
     let mut attrs = vec![("id", id), ("name", name)];
     if descr != alt.title || descr.is_empty() {
         attrs.push(("descr", descr));
@@ -514,16 +525,21 @@ pub(in super::super) fn nv_props(w: &mut W, tag: &str, id: &str, name: &str, des
     if hidden {
         attrs.push(("hidden", "1"));
     }
-    if !alt.decorative {
+    if !alt.decorative && link.is_none() {
         w.empty(tag, &attrs);
         return;
     }
     w.open(tag, &attrs);
-    w.open("a:extLst", &[]);
-    w.open("a:ext", &[("uri", crate::xml::DECORATIVE_EXT)]);
-    w.empty("adec:decorative", &[("xmlns:adec", crate::xml::DECORATIVE_NS), ("val", "1")]);
-    w.close("a:ext");
-    w.close("a:extLst");
+    if let Some(rid) = link {
+        w.empty("a:hlinkClick", &[("r:id", rid)]);
+    }
+    if alt.decorative {
+        w.open("a:extLst", &[]);
+        w.open("a:ext", &[("uri", crate::xml::DECORATIVE_EXT)]);
+        w.empty("adec:decorative", &[("xmlns:adec", crate::xml::DECORATIVE_NS), ("val", "1")]);
+        w.close("a:ext");
+        w.close("a:extLst");
+    }
     w.close(tag);
 }
 

crates/docx/src/write/story/object.rs

@@ -220,13 +220,18 @@ impl Writer<'_> {
                 w.open("mc:Choice", &[("Requires", requires)]);
             }
             let docpr = self.next_docpr();
+            let link = Self::drawing_link(props, rels);
             let name = match graphic.kind {
                 wordcraft_doc::graphic::GraphicKind::Chart => format!("Chart {docpr}"),
                 wordcraft_doc::graphic::GraphicKind::Diagram => format!("Diagram {docpr}"),
                 wordcraft_doc::graphic::GraphicKind::Ink => format!("Ink {docpr}"),
             };
             let z = self.z;
-            self.graphic_frame(w, DrawingInfo { float, size: (*gw, *gh), docpr: &docpr, name: &name, alt }, native.unwrap_or(&root));
+            self.graphic_frame(
+                w,
+                DrawingInfo { float, size: (*gw, *gh), docpr: &docpr, name: &name, alt, link: link.as_deref() },
+                native.unwrap_or(&root),
+            );
             if let Some(fallback) = fallback {
                 w.close("mc:Choice");
                 match picture {
@@ -234,7 +239,11 @@ impl Writer<'_> {
                         // Both branches are one object: same id and stacking order.
                         self.z = z;
                         w.open("mc:Fallback", &[]);
-                        self.graphic_frame(w, DrawingInfo { float, size: (*gw, *gh), docpr: &docpr, name: &name, alt }, picture);
+                        self.graphic_frame(
+                            w,
+                            DrawingInfo { float, size: (*gw, *gh), docpr: &docpr, name: &name, alt, link: link.as_deref() },
+                            picture,
+                        );
                         w.close("mc:Fallback");
                     }
                     None => w.raw(&fallback.to_xml()),
@@ -266,8 +275,9 @@ impl Writer<'_> {
         self.rpr(w, props);
         w.open("w:drawing", &[]);
         let docpr = self.next_docpr();
+        let link = Self::drawing_link(props, rels);
         let name = format!("Picture {docpr}");
-        self.drawing_open(w, DrawingInfo { float, size: (*iw, *ih), docpr: &docpr, name: &name, alt });
+        self.drawing_open(w, DrawingInfo { float, size: (*iw, *ih), docpr: &docpr, name: &name, alt, link: link.as_deref() });
         w.empty("wp:cNvGraphicFramePr", &[]);
         w.open("a:graphic", &[]);
         w.open("a:graphicData", &[("uri", "http://schemas.openxmlformats.org/drawingml/2006/picture")]);
@@ -355,6 +365,7 @@ impl Writer<'_> {
         self.rpr(w, props);
         w.open("w:drawing", &[]);
         let docpr = self.shape_docpr(extra.id);
+        let link = Self::drawing_link(props, rels);
         // Ink is told apart by its name (and its pen), which reading looks for.
         let name = shape_name(*kind, freeform.as_deref(), &docpr);
         // The effect extent leaves room for the shadow and glow.
@@ -369,7 +380,7 @@ impl Writer<'_> {
                 .map(std::sync::Arc::new);
         }
         let float = &float;
-        self.drawing_open(w, DrawingInfo { float, size: (*sw, *sh), docpr: &docpr, name: &name, alt: o.alt_text() });
+        self.drawing_open(w, DrawingInfo { float, size: (*sw, *sh), docpr: &docpr, name: &name, alt: o.alt_text(), link: link.as_deref() });
         w.empty("wp:cNvGraphicFramePr", &[]);
         w.open("a:graphic", &[]);
         w.open("a:graphicData", &[("uri", "http://schemas.microsoft.com/office/word/2010/wordprocessingShape")]);
@@ -389,8 +400,9 @@ impl Writer<'_> {
         self.rpr(w, props);
         w.open("w:drawing", &[]);
         let docpr = self.next_docpr();
+        let link = Self::drawing_link(props, rels);
         let name = format!("{} {docpr}", if canvas.is_some() { "Canvas" } else { "Group" });
-        self.drawing_open(w, DrawingInfo { float, size: (*gw, *gh), docpr: &docpr, name: &name, alt: o.alt_text() });
+        self.drawing_open(w, DrawingInfo { float, size: (*gw, *gh), docpr: &docpr, name: &name, alt: o.alt_text(), link: link.as_deref() });
         w.empty("wp:cNvGraphicFramePr", &[]);
         w.open("a:graphic", &[]);
         let outer = match canvas {

crates/docx/src/xml.rs

@@ -94,6 +94,49 @@ fn prefix_for(uri: &[u8], extra: &[(&'static str, &str)]) -> &'static str {
     "?"
 }
 
+/// Word's 2006 namespace (`wne`). A Strict document from Word keeps a text box's content in
+/// `wne:txbxContent`; a Transitional one uses `w:txbxContent`.
+const WNE: &[u8] = b"http://schemas.microsoft.com/office/word/2006/wordml";
+
+/// The canonical prefix of an element named `local` in namespace `uri`, inside an element
+/// named `parent` (canonical, `None` at the root). This is [`prefix_for`] with two exceptions
+/// for Strict documents (#3548):
+///
+/// - Shapes, groups and canvases (`wsp`, `wgp`, `wpc`) and their parts are elements of the
+///   WordprocessingDrawing namespace in ECMA-376 Part 1, 5th edition (`dml-wordprocessingDrawing.xsd`,
+///   Strict and Transitional alike). A Strict document from Word puts them there; a Transitional
+///   one uses Word 2010's own namespaces, whose names the reader knows. So they read as `wps:`,
+///   `wpg:` and `wpc:` (see [`drawing_prefix`]).
+/// - A shape's text box content reads as `w:txbxContent` from the drawing namespace (the
+///   schema's name) and from Word's `wne:txbxContent`.
+fn element_prefix(uri: &[u8], local: &str, parent: Option<&str>, extra: &[(&'static str, &str)]) -> &'static str {
+    match prefix_for(uri, extra) {
+        "wp" => drawing_prefix(local, parent),
+        "?" if uri == WNE && local == "txbxContent" && parent == Some("wps:txbx") => "w",
+        p => p,
+    }
+}
+
+/// The prefix for `local`, an element of the WordprocessingDrawing namespace inside `parent`.
+/// Word 2010's namespaces split what this one holds: a shape and its parts are `wps:`, a group
+/// (`wgp`, a nested `grpSp`) and a graphic frame in it are `wpg:`, and a canvas is `wpc:`. Its
+/// members are named for what they are. A text box's content is `w:txbxContent`, and anything
+/// else (`inline`, `anchor`, `docPr`…) stays `wp:`.
+fn drawing_prefix(local: &str, parent: Option<&str>) -> &'static str {
+    match (parent, local) {
+        (_, "wsp") => "wps",
+        (_, "wgp" | "grpSp" | "graphicFrame") => "wpg",
+        (_, "wpc") => "wpc",
+        (Some("wps:wsp"), "cNvPr" | "cNvSpPr" | "cNvCnPr" | "spPr" | "style" | "extLst" | "txbx" | "linkedTxbx" | "bodyPr") => "wps",
+        (Some("wps:txbx"), "txbxContent") => "w",
+        (Some("wps:txbx" | "wps:linkedTxbx"), "extLst") => "wps",
+        (Some("wpg:wgp" | "wpg:grpSp"), "cNvPr" | "cNvGrpSpPr" | "grpSpPr" | "extLst") => "wpg",
+        (Some("wpg:graphicFrame"), "cNvPr" | "cNvFrPr" | "xfrm" | "extLst") => "wpg",
+        (Some("wpc:wpc"), "bg" | "whole" | "extLst") => "wpc",
+        _ => "wp",
+    }
+}
+
 /// Compatibility feature namespaces whose elements are not otherwise handled by this tree.
 /// Keep the reader's existing feature set when canonicalizing `mc:Choice/@Requires` aliases.
 fn requirement_prefix(uri: &[u8]) -> &'static str {
@@ -342,8 +385,8 @@ fn parse_with(bytes: &[u8], extra: &[(&'static str, &str)], strict: bool) -> Res
     loop {
         buf.clear();
         let (res, ev) = r.read_resolved_event_into(&mut buf).map_err(|e| DocxError::Xml(e.to_string()))?;
-        let el_prefix = match &res {
-            ResolveResult::Bound(ns) => Some(prefix_for(ns.as_ref(), extra)),
+        let el_ns = match &res {
+            ResolveResult::Bound(ns) => Some(ns.as_ref()),
             _ => None,
         };
         match ev {
@@ -356,8 +399,8 @@ fn parse_with(bytes: &[u8], extra: &[(&'static str, &str)], strict: bool) -> Res
                     return Err(DocxError::Limit("too many XML elements".into()));
                 }
                 let local = String::from_utf8_lossy(e.local_name().as_ref()).into_owned();
-                let name = match el_prefix {
-                    Some(p) => format!("{p}:{local}"),
+                let name = match el_ns {
+                    Some(ns) => format!("{}:{local}", element_prefix(ns, &local, stack.last().map(|p| p.name.as_str()), extra)),
                     None => local,
                 };
                 let mut el = El { name, attrs: Vec::new(), kids: Vec::new() };
@@ -593,6 +636,42 @@ mod tests {
         assert!(parse(b"<a></a><b></b>").is_ok(), "lenient parsing still stops at the root");
     }
 
+    /// Every element name in `e`, depth first.
+    fn names(e: &El, out: &mut Vec<String>) {
+        out.push(e.name.clone());
+        for c in e.els() {
+            names(c, out);
+        }
+    }
+
+    #[test]
+    fn strict_drawing_names_read_as_word_2010_names() {
+        // A Strict shape with a text box, a group with a nested group and a graphic frame, and
+        // a canvas, each in the one Strict WordprocessingDrawing namespace (#3548).
+        let xml = br#"<w:body xmlns:w="http://purl.oclc.org/ooxml/wordprocessingml/main" xmlns:a="http://purl.oclc.org/ooxml/drawingml/main" xmlns:d="http://purl.oclc.org/ooxml/drawingml/wordprocessingDrawing" xmlns:wne="http://schemas.microsoft.com/office/word/2006/wordml">
+<d:anchor><d:extent/><d:docPr/><d:cNvGraphicFramePr/><a:graphic><a:graphicData>
+<d:wsp><d:cNvPr/><d:cNvSpPr/><d:spPr/><d:style/><d:extLst/><d:txbx><wne:txbxContent><w:p/></wne:txbxContent></d:txbx><d:bodyPr/></d:wsp>
+<d:wgp><d:cNvPr/><d:cNvGrpSpPr/><d:grpSpPr/><d:grpSp><d:cNvGrpSpPr/><d:grpSpPr/><d:wsp><d:cNvCnPr/><d:spPr/><d:linkedTxbx/><d:bodyPr/></d:wsp></d:grpSp>
+<d:graphicFrame><d:cNvPr/><d:cNvFrPr/><d:xfrm/><a:graphic/><d:extLst/></d:graphicFrame><d:extLst/></d:wgp>
+<d:wpc><d:bg/><d:whole/><d:wsp><d:spPr/><d:txbx><d:txbxContent/><d:extLst/></d:txbx></d:wsp><d:wgp><d:grpSpPr/></d:wgp><d:extLst/></d:wpc>
+</a:graphicData></a:graphic></d:anchor>
+<wne:txbxContent/><d:inline><d:docPr/><d:bodyPr/><d:bg/></d:inline></w:body>"#;
+        let mut got = Vec::new();
+        names(&parse(xml).unwrap(), &mut got);
+        let want = "w:body wp:anchor wp:extent wp:docPr wp:cNvGraphicFramePr a:graphic a:graphicData \
+            wps:wsp wps:cNvPr wps:cNvSpPr wps:spPr wps:style wps:extLst wps:txbx w:txbxContent w:p wps:bodyPr \
+            wpg:wgp wpg:cNvPr wpg:cNvGrpSpPr wpg:grpSpPr wpg:grpSp wpg:cNvGrpSpPr wpg:grpSpPr wps:wsp wps:cNvCnPr wps:spPr wps:linkedTxbx wps:bodyPr \
+            wpg:graphicFrame wpg:cNvPr wpg:cNvFrPr wpg:xfrm a:graphic wpg:extLst wpg:extLst \
+            wpc:wpc wpc:bg wpc:whole wps:wsp wps:spPr wps:txbx w:txbxContent wps:extLst wpg:wgp wpg:grpSpPr wpc:extLst \
+            ?:txbxContent wp:inline wp:docPr wp:bodyPr wp:bg";
+        assert_eq!(got.join(" "), want);
+        // The Transitional namespaces read as before.
+        let xml = br#"<w:body xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main" xmlns:wp="http://schemas.openxmlformats.org/drawingml/2006/wordprocessingDrawing" xmlns:wps="http://schemas.microsoft.com/office/word/2010/wordprocessingShape" xmlns:wne="http://schemas.microsoft.com/office/word/2006/wordml"><wp:inline><wp:docPr/></wp:inline><wps:wsp><wps:txbx><w:txbxContent/></wps:txbx></wps:wsp><wne:toolbars/></w:body>"#;
+        let mut got = Vec::new();
+        names(&parse(xml).unwrap(), &mut got);
+        assert_eq!(got.join(" "), "w:body wp:inline wp:docPr wps:wsp wps:txbx w:txbxContent ?:toolbars");
+    }
+
     #[test]
     fn a_relationship_attribute_is_the_namespaced_one_or_the_alias_of_one() {
         // The relationship namespace, and VML's `o:relid`.

crates/docx/tests/it/fixtures.rs

@@ -1766,12 +1766,27 @@ const CT_CHARTEX: &str = "application/vnd.ms-office.chartex+xml";
 /// A package holding one box & whisker chart the way Word writes it: the `cx:chart` in an
 /// `mc:Choice` requiring `requires` (bound to `requires_ns`), a picture of it in the `mc:Fallback`.
 fn chartex_docx(requires: &str, requires_ns: &str) -> Vec<u8> {
-    let frame = r#"<wp:inline distT="0" distB="0" distL="0" distR="0"><wp:extent cx="2743200" cy="1371600"/><wp:docPr id="1" name="Chart 1"/><wp:cNvGraphicFramePr/>"#;
+    chartex_docx_linked(requires, requires_ns, false)
+}
+
+/// [`chartex_docx`], its drawing linked to a web address (`wp:docPr/a:hlinkClick`) when `link`,
+/// after a first choice linked elsewhere, which needs a namespace no reader knows, so a reader
+/// takes the chart's choice.
+fn chartex_docx_linked(requires: &str, requires_ns: &str, link: bool) -> Vec<u8> {
+    let click = if link { r#"<a:hlinkClick r:id="rIdWeb"/>"# } else { "" };
+    let other = if link {
+        r#"<mc:Choice xmlns:w99="urn:example:unsupported" Requires="w99"><w:drawing><wp:inline distT="0" distB="0" distL="0" distR="0"><wp:extent cx="254000" cy="254000"/><wp:docPr id="9" name="Other"><a:hlinkClick r:id="rIdOther"/></wp:docPr><a:graphic><a:graphicData uri="http://schemas.microsoft.com/office/word/2010/wordprocessingShape"><wps:wsp><wps:spPr><a:prstGeom prst="rect"><a:avLst/></a:prstGeom></wps:spPr><wps:bodyPr/></wps:wsp></a:graphicData></a:graphic></wp:inline></w:drawing></mc:Choice>"#
+    } else {
+        ""
+    };
+    let frame = format!(
+        r#"<wp:inline distT="0" distB="0" distL="0" distR="0"><wp:extent cx="2743200" cy="1371600"/><wp:docPr id="1" name="Chart 1">{click}</wp:docPr><wp:cNvGraphicFramePr/>"#
+    );
     let body = format!(
-        r#"<w:p><w:r><w:t>Spread</w:t></w:r><w:r><mc:AlternateContent><mc:Choice xmlns:{requires}="{requires_ns}" Requires="{requires}"><w:drawing>{frame}<a:graphic><a:graphicData uri="{CHARTEX_NS}"><cx:chart xmlns:cx="{CHARTEX_NS}" r:id="rIdCx"/></a:graphicData></a:graphic></wp:inline></w:drawing></mc:Choice><mc:Fallback><w:drawing>{frame}<a:graphic><a:graphicData uri="http://schemas.openxmlformats.org/drawingml/2006/picture"><pic:pic><pic:nvPicPr><pic:cNvPr id="1" name="Chart 1"/><pic:cNvPicPr/></pic:nvPicPr><pic:blipFill><a:blip r:embed="rIdPic"/><a:stretch><a:fillRect/></a:stretch></pic:blipFill><pic:spPr><a:xfrm><a:off x="0" y="0"/><a:ext cx="2743200" cy="1371600"/></a:xfrm><a:prstGeom prst="rect"><a:avLst/></a:prstGeom></pic:spPr></pic:pic></a:graphicData></a:graphic></wp:inline></w:drawing></mc:Fallback></mc:AlternateContent></w:r></w:p>"#
+        r#"<w:p><w:r><w:t>Spread</w:t></w:r><w:r><mc:AlternateContent>{other}<mc:Choice xmlns:{requires}="{requires_ns}" Requires="{requires}"><w:drawing>{frame}<a:graphic><a:graphicData uri="{CHARTEX_NS}"><cx:chart xmlns:cx="{CHARTEX_NS}" r:id="rIdCx"/></a:graphicData></a:graphic></wp:inline></w:drawing></mc:Choice><mc:Fallback><w:drawing>{frame}<a:graphic><a:graphicData uri="http://schemas.openxmlformats.org/drawingml/2006/picture"><pic:pic><pic:nvPicPr><pic:cNvPr id="1" name="Chart 1"/><pic:cNvPicPr/></pic:nvPicPr><pic:blipFill><a:blip r:embed="rIdPic"/><a:stretch><a:fillRect/></a:stretch></pic:blipFill><pic:spPr><a:xfrm><a:off x="0" y="0"/><a:ext cx="2743200" cy="1371600"/></a:xfrm><a:prstGeom prst="rect"><a:avLst/></a:prstGeom></pic:spPr></pic:pic></a:graphicData></a:graphic></wp:inline></w:drawing></mc:Fallback></mc:AlternateContent></w:r></w:p>"#
     );
     let doc = format!(r#"<?xml version="1.0" encoding="UTF-8" standalone="yes"?><w:document {W_NS}><w:body>{body}</w:body></w:document>"#);
-    let doc_rels = r#"<Relationships xmlns="http://schemas.openxmlformats.org/package/2006/relationships"><Relationship Id="rIdCx" Type="http://schemas.microsoft.com/office/2014/relationships/chartEx" Target="charts/chartEx1.xml"/><Relationship Id="rIdPic" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/image" Target="media/image1.png"/></Relationships>"#;
+    let doc_rels = r#"<Relationships xmlns="http://schemas.openxmlformats.org/package/2006/relationships"><Relationship Id="rIdCx" Type="http://schemas.microsoft.com/office/2014/relationships/chartEx" Target="charts/chartEx1.xml"/><Relationship Id="rIdPic" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/image" Target="media/image1.png"/><Relationship Id="rIdWeb" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/hyperlink" Target="https://example.com/chart" TargetMode="External"/><Relationship Id="rIdOther" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/hyperlink" Target="https://example.com/other" TargetMode="External"/></Relationships>"#;
     let chart_rels = r#"<Relationships xmlns="http://schemas.openxmlformats.org/package/2006/relationships"><Relationship Id="rId1" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/package" Target="../embeddings/Microsoft_Excel_Worksheet.xlsx"/><Relationship Id="rId2" Type="http://schemas.microsoft.com/office/2011/relationships/chartColorStyle" Target="colors1.xml"/></Relationships>"#;
     let types = format!(
         r#"<Types xmlns="http://schemas.openxmlformats.org/package/2006/content-types"><Default Extension="rels" ContentType="application/vnd.openxmlformats-package.relationships+xml"/><Default Extension="xml" ContentType="application/xml"/><Default Extension="png" ContentType="image/png"/><Default Extension="xlsx" ContentType="{CT_XLSX}"/><Override PartName="/word/document.xml" ContentType="application/vnd.openxmlformats-officedocument.wordprocessingml.document.main+xml"/><Override PartName="/word/charts/chartEx1.xml" ContentType="{CT_CHARTEX}"/><Override PartName="/word/charts/colors1.xml" ContentType="application/vnd.ms-office.chartcolorstyle+xml"/></Types>"#
@@ -1827,6 +1842,25 @@ fn office_2016_chart_keeps_its_chart_part_and_fallback_through_saves() {
     }
 }
 
+/// #5267: a link on an Office 2016 chart's drawing is the chart's link (not an earlier choice's),
+/// and saving keeps it on both branches (the writer rebuilds their frames).
+#[test]
+fn a_link_on_an_office_2016_charts_drawing_is_kept() {
+    let mut d = wordcraft_docx::read(&chartex_docx_linked("cx1", "http://schemas.microsoft.com/office/drawing/2015/9/8/chartex", true)).unwrap();
+    for _ in 0..2 {
+        let p = d.body[0].as_para().unwrap();
+        let off = p.object_offsets().into_iter().find(|o| matches!(p.object_at(*o), Some(InlineObject::Graphic { .. }))).unwrap();
+        assert_eq!(p.props_of_char(off).link.as_deref(), Some("https://example.com/chart"));
+        let files = unzip(&wordcraft_docx::write(&d).unwrap());
+        let body = text(&files, "word/document.xml");
+        let ids = attr_values(&body, "a:hlinkClick", "r:id");
+        assert_eq!(ids.len(), 2, "on the chart and its fallback picture: {body}");
+        assert_eq!(rel(&text(&files, "word/_rels/document.xml.rels"), &ids[0]).1, "https://example.com/chart");
+        let entries: Vec<_> = files.iter().map(|(name, bytes)| (name.as_str(), bytes.as_slice())).collect();
+        d = wordcraft_docx::read(&zip(&entries)).unwrap();
+    }
+}
+
 #[test]
 fn resized_office_2016_chart_resizes_its_fallback_too() {
     let mut d = wordcraft_docx::read(&chartex_docx("cx1", "http://schemas.microsoft.com/office/drawing/2015/9/8/chartex")).unwrap();
@@ -2462,6 +2496,31 @@ fn east_asian_number_formats_read_label_and_round_trip() {
     assert_eq!(formats(&back), want);
 }
 
+/// #4985: a level with numFmt none and literal text (`RESPONSE:`) labels its paragraph with that
+/// text, as in Word, and keeps both through a save.
+#[test]
+fn a_none_level_s_literal_text_is_its_label() {
+    use wordcraft_doc::numbering::Counters;
+    use wordcraft_doc::section::NumFormat;
+    let numbering = format!(
+        r#"<w:numbering {W_NS}><w:abstractNum w:abstractNumId="0"><w:lvl w:ilvl="0"><w:start w:val="1"/><w:numFmt w:val="none"/><w:lvlText w:val="RESPONSE:"/><w:pPr><w:ind w:left="1440" w:hanging="1440"/></w:pPr></w:lvl></w:abstractNum><w:num w:numId="1"><w:abstractNumId w:val="0"/></w:num></w:numbering>"#
+    );
+    let body = r#"<w:p><w:pPr><w:numPr><w:ilvl w:val="0"/><w:numId w:val="1"/></w:numPr></w:pPr><w:r><w:t>X</w:t></w:r></w:p>"#;
+    let bytes = docx(body, &[("rId1", "numbering", "numbering.xml")], &[("word/numbering.xml", &numbering)]);
+    let label = |d: &Document| {
+        let n = paras(d).first().and_then(|p| p.props.numbering)?;
+        Counters::default().next_label(&d.numbering, n.num, n.level).map(|l| l.0)
+    };
+    let d = wordcraft_docx::read(&bytes).unwrap();
+    assert_eq!(label(&d).as_deref(), Some("RESPONSE:"));
+    let back = wordcraft_docx::read(&wordcraft_docx::write(&d).unwrap()).unwrap();
+    assert_eq!(
+        back.numbering.level(1, 0).map(|l| (l.format, l.text.as_str(), l.indent, l.hanging)),
+        Some((NumFormat::None, "RESPONSE:", 72.0, 72.0))
+    );
+    assert_eq!(label(&back).as_deref(), Some("RESPONSE:"));
+}
+
 /// #464: a bookmark whose end follows the document's final table keeps its end (it goes into
 /// the table's last cell) and is saved as a matched pair.
 #[test]

crates/docx/tests/it/main.rs

@@ -23,6 +23,7 @@ mod ligatures;
 mod malformed;
 mod note_comments;
 mod passthrough;
+mod picture_links;
 mod positional_tabs;
 mod privacy;
 mod roundtrip;
@@ -34,6 +35,7 @@ mod section_page_number_type;
 mod section_text_direction;
 mod section_valign;
 mod settings_merge;
+mod strict_drawings;
 mod text_alignment;
 mod text_boxes;
 mod toc_ranges;

crates/docx/tests/it/picture_links.rs

@@ -0,0 +1,127 @@
+//! A picture's hyperlink (#5267): Word keeps it on the drawing (`wp:docPr/a:hlinkClick`,
+//! ECMA-376 Part 1 §20.4.2.5) rather than in a `w:hyperlink` around the run. Original
+//! synthetic packages; the 2×2 picture is generated in code.
+use std::io::{Read, Write};
+use std::sync::Arc;
+use wordcraft_doc::{CharProps, Document, InlineObject, Pos};
+
+fn png() -> Vec<u8> {
+    let mut out = Vec::new();
+    image::RgbaImage::from_pixel(2, 2, image::Rgba([20, 90, 200, 255]))
+        .write_to(&mut std::io::Cursor::new(&mut out), image::ImageFormat::Png)
+        .unwrap();
+    out
+}
+
+/// An inline picture's run; `link` goes into its `wp:docPr`.
+fn picture_run(id: u32, link: &str) -> String {
+    format!(
+        r#"<w:r><w:drawing><wp:inline distT="0" distB="0" distL="0" distR="0"><wp:extent cx="254000" cy="254000"/><wp:docPr id="{id}" name="Picture {id}">{link}</wp:docPr><a:graphic><a:graphicData uri="http://schemas.openxmlformats.org/drawingml/2006/picture"><pic:pic><pic:nvPicPr><pic:cNvPr id="0" name="square.png"/><pic:cNvPicPr/></pic:nvPicPr><pic:blipFill><a:blip r:embed="rIdPic"/></pic:blipFill><pic:spPr><a:xfrm><a:off x="0" y="0"/><a:ext cx="254000" cy="254000"/></a:xfrm><a:prstGeom prst="rect"/></pic:spPr></pic:pic></a:graphicData></a:graphic></wp:inline></w:drawing></w:r>"#
+    )
+}
+
+fn package(body: &str) -> Vec<u8> {
+    let xml = format!(
+        r#"<?xml version="1.0" encoding="UTF-8" standalone="yes"?><w:document xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main" xmlns:r="http://schemas.openxmlformats.org/officeDocument/2006/relationships" xmlns:wp="http://schemas.openxmlformats.org/drawingml/2006/wordprocessingDrawing" xmlns:a="http://schemas.openxmlformats.org/drawingml/2006/main" xmlns:pic="http://schemas.openxmlformats.org/drawingml/2006/picture"><w:body>{body}<w:sectPr/></w:body></w:document>"#
+    );
+    let rels = r#"<Relationships xmlns="http://schemas.openxmlformats.org/package/2006/relationships"><Relationship Id="rId1" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/officeDocument" Target="word/document.xml"/></Relationships>"#;
+    let doc_rels = r#"<Relationships xmlns="http://schemas.openxmlformats.org/package/2006/relationships"><Relationship Id="rIdPic" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/image" Target="media/square.png"/><Relationship Id="rIdWeb" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/hyperlink" Target="https://example.com/picture" TargetMode="External"/><Relationship Id="rIdRun" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/hyperlink" Target="https://example.com/run" TargetMode="External"/></Relationships>"#;
+    let types = r#"<Types xmlns="http://schemas.openxmlformats.org/package/2006/content-types"><Default Extension="rels" ContentType="application/vnd.openxmlformats-package.relationships+xml"/><Default Extension="png" ContentType="image/png"/><Override PartName="/word/document.xml" ContentType="application/vnd.openxmlformats-officedocument.wordprocessingml.document.main+xml"/></Types>"#;
+    let png = png();
+    let mut zip = zip::ZipWriter::new(std::io::Cursor::new(Vec::new()));
+    for (name, bytes) in [
+        ("[Content_Types].xml", types.as_bytes()),
+        ("_rels/.rels", rels.as_bytes()),
+        ("word/document.xml", xml.as_bytes()),
+        ("word/_rels/document.xml.rels", doc_rels.as_bytes()),
+        ("word/media/square.png", png.as_slice()),
+    ] {
+        zip.start_file(name, zip::write::SimpleFileOptions::default()).unwrap();
+        zip.write_all(bytes).unwrap();
+    }
+    zip.finish().unwrap().into_inner()
+}
+
+/// The link on each picture of the first paragraph, in order.
+fn picture_links(doc: &Document) -> Vec<Option<String>> {
+    let p = doc.body[0].as_para().unwrap();
+    p.object_offsets()
+        .into_iter()
+        .filter(|off| matches!(p.object_at(*off), Some(InlineObject::Image { .. })))
+        .map(|off| p.props_of_char(off).link.clone())
+        .collect()
+}
+
+fn part(bytes: &[u8], name: &str) -> String {
+    let mut z = zip::ZipArchive::new(std::io::Cursor::new(bytes)).unwrap();
+    let mut s = String::new();
+    z.by_name(name).unwrap().read_to_string(&mut s).unwrap();
+    s
+}
+
+#[test]
+fn a_link_on_a_pictures_drawing_is_read() {
+    let body = format!(
+        "<w:p>{}{}{}<w:hyperlink r:id=\"rIdRun\">{}</w:hyperlink></w:p>",
+        // Word's way: the link on the drawing.
+        picture_run(1, r#"<a:hlinkClick r:id="rIdWeb"/>"#),
+        // A click naming a relationship that is no hyperlink, and one naming none.
+        picture_run(2, r#"<a:hlinkClick r:id="rIdPic"/>"#),
+        picture_run(3, r#"<a:hlinkClick r:id="rIdMissing"/>"#),
+        // A link around the run wins over the drawing's.
+        picture_run(4, r#"<a:hlinkClick r:id="rIdWeb"/>"#),
+    );
+    let doc = wordcraft_docx::read(&package(&body)).unwrap();
+    assert_eq!(picture_links(&doc), [Some("https://example.com/picture".to_string()), None, None, Some("https://example.com/run".to_string())]);
+}
+
+#[test]
+fn a_linked_picture_is_saved_with_its_link_on_the_drawing_too() {
+    let mut doc = Document::new();
+    doc.media.insert("square.png".into(), Arc::new(png()));
+    let picture = || InlineObject::Image {
+        media: "square.png".into(),
+        w: 20.0,
+        h: 20.0,
+        alt: String::new(),
+        float: Default::default(),
+        crop: [0.0; 4],
+        ole: None,
+        fx: Default::default(),
+    };
+    let linked = |l: &str| CharProps { link: Some(l.to_string()), ..Default::default() };
+    // A place in this document stays a `w:hyperlink w:anchor` only: `a:hlinkClick` names a relationship.
+    doc.insert_object(&Pos::body(0, 0), picture(), &linked("#Spot")).unwrap();
+    doc.insert_object(&Pos::body(0, 0), picture(), &linked("https://example.com/saved")).unwrap();
+    let bytes = wordcraft_docx::write(&doc).unwrap();
+    let xml = part(&bytes, "word/document.xml");
+    let clicks: Vec<&str> = xml.match_indices("<a:hlinkClick ").map(|(i, _)| &xml[i..]).collect();
+    assert_eq!(clicks.len(), 1, "{xml}");
+    let rid = clicks[0].split('"').nth(1).unwrap();
+    let docpr = xml.find("<wp:docPr ").unwrap();
+    assert!(docpr < xml.find("<a:hlinkClick ").unwrap() && xml.find("<a:hlinkClick ").unwrap() < xml.find("</wp:docPr>").unwrap(), "{xml}");
+    let rels = part(&bytes, "word/_rels/document.xml.rels");
+    assert!(rels.contains(&format!(r#"Id="{rid}""#)) && rels.contains(r#"Target="https://example.com/saved""#), "{rels}");
+    // The run's own hyperlink names the same relationship, so the address is stored once.
+    assert_eq!(rels.matches("https://example.com/saved").count(), 1, "{rels}");
+    let back = wordcraft_docx::read(&bytes).unwrap();
+    assert_eq!(picture_links(&back), [Some("https://example.com/saved".to_string()), Some("#Spot".to_string())]);
+    // The drawing's link alone (as Word writes it) reads back the same.
+    let without_run_link = xml.replace(&format!(r#"<w:hyperlink r:id="{rid}" w:history="1">"#), "").replacen("</w:hyperlink>", "", 1);
+    assert_ne!(without_run_link, xml);
+    let mut z = zip::ZipArchive::new(std::io::Cursor::new(bytes.as_slice())).unwrap();
+    let mut out = zip::ZipWriter::new(std::io::Cursor::new(Vec::new()));
+    for i in 0..z.len() {
+        let mut f = z.by_index(i).unwrap();
+        let name = f.name().to_string();
+        let mut data = Vec::new();
+        f.read_to_end(&mut data).unwrap();
+        if name == "word/document.xml" {
+            data = without_run_link.clone().into_bytes();
+        }
+        out.start_file(name, zip::write::SimpleFileOptions::default()).unwrap();
+        out.write_all(&data).unwrap();
+    }
+    let edited = out.finish().unwrap().into_inner();
+    assert_eq!(picture_links(&wordcraft_docx::read(&edited).unwrap())[0], Some("https://example.com/saved".to_string()));
+}

crates/docx/tests/it/strict_drawings.rs

@@ -0,0 +1,182 @@
+//! Shapes, text boxes, groups and canvases in a Strict document (#3548). ISO/IEC 29500 Strict
+//! puts them in the one WordprocessingDrawing namespace, where Transitional documents use Word
+//! 2010's `wps`, `wpg` and `wpc` namespaces. Fixtures are original and hand-written from
+//! ECMA-376 Part 1, 5th edition (`dml-wordprocessingDrawing.xsd`, Strict and Transitional): each
+//! drawing is written once and filled in with either set of namespaces.
+
+use std::io::{Read, Write};
+
+use wordcraft_doc::{Document, InlineObject, PartKind};
+
+const STRICT_NS: &str = r#"xmlns:w="http://purl.oclc.org/ooxml/wordprocessingml/main" xmlns:r="http://purl.oclc.org/ooxml/officeDocument/relationships" xmlns:wp="http://purl.oclc.org/ooxml/drawingml/wordprocessingDrawing" xmlns:a="http://purl.oclc.org/ooxml/drawingml/main" xmlns:wne="http://schemas.microsoft.com/office/word/2006/wordml" w:conformance="strict""#;
+const TRANSITIONAL_NS: &str = r#"xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main" xmlns:r="http://schemas.openxmlformats.org/officeDocument/2006/relationships" xmlns:wp="http://schemas.openxmlformats.org/drawingml/2006/wordprocessingDrawing" xmlns:a="http://schemas.openxmlformats.org/drawingml/2006/main" xmlns:wps="http://schemas.microsoft.com/office/word/2010/wordprocessingShape" xmlns:wpg="http://schemas.microsoft.com/office/word/2010/wordprocessingGroup" xmlns:wpc="http://schemas.microsoft.com/office/word/2010/wordprocessingCanvas""#;
+const STRICT_RELS: &str = r#"<?xml version="1.0" encoding="UTF-8"?><Relationships xmlns="http://schemas.openxmlformats.org/package/2006/relationships"><Relationship Id="rId1" Type="http://purl.oclc.org/ooxml/officeDocument/relationships/officeDocument" Target="word/document.xml"/></Relationships>"#;
+const TRANSITIONAL_RELS: &str = r#"<?xml version="1.0" encoding="UTF-8"?><Relationships xmlns="http://schemas.openxmlformats.org/package/2006/relationships"><Relationship Id="rId1" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/officeDocument" Target="word/document.xml"/></Relationships>"#;
+
+/// How a fixture names its drawing elements: the shape, group and canvas prefixes, and the
+/// text box content element.
+struct Names {
+    ns: &'static str,
+    rels: &'static str,
+    shape: &'static str,
+    group: &'static str,
+    canvas: &'static str,
+    content: &'static str,
+}
+
+/// Word's Strict output: everything in `wp:`, text box content in `wne:txbxContent`.
+const STRICT_WORD: Names = Names { ns: STRICT_NS, rels: STRICT_RELS, shape: "wp", group: "wp", canvas: "wp", content: "wne:txbxContent" };
+/// The Strict schema's own names: the text box content in `wp:` too.
+const STRICT_SCHEMA: Names = Names { ns: STRICT_NS, rels: STRICT_RELS, shape: "wp", group: "wp", canvas: "wp", content: "wp:txbxContent" };
+/// The same drawings in a Transitional document.
+const TRANSITIONAL: Names =
+    Names { ns: TRANSITIONAL_NS, rels: TRANSITIONAL_RELS, shape: "wps", group: "wpg", canvas: "wpc", content: "w:txbxContent" };
+
+fn zip(entries: &[(&str, &[u8])]) -> Vec<u8> {
+    let mut zw = zip::ZipWriter::new(std::io::Cursor::new(Vec::new()));
+    for (name, bytes) in entries {
+        zw.start_file(*name, zip::write::SimpleFileOptions::default()).unwrap();
+        zw.write_all(bytes).unwrap();
+    }
+    zw.finish().unwrap().into_inner()
+}
+
+fn document_xml(bytes: &[u8]) -> String {
+    let mut z = zip::ZipArchive::new(std::io::Cursor::new(bytes)).unwrap();
+    let mut s = String::new();
+    z.by_name("word/document.xml").unwrap().read_to_string(&mut s).unwrap();
+    s
+}
+
+fn solid(rgb: &str) -> String {
+    format!(r#"<a:solidFill><a:srgbClr val="{rgb}"/></a:solidFill>"#)
+}
+
+/// A `wsp` at (`x`, `y`) EMUs in its group, `cx` × `cy`. `text` gives it a text box.
+fn shape(n: &Names, prst: &str, (x, y, cx, cy): (u32, u32, u32, u32), fill: &str, text: Option<&str>) -> String {
+    let s = n.shape;
+    let connector = prst.contains("Connector");
+    let nv = if connector {
+        format!("<{s}:cNvCnPr/>")
+    } else if text.is_some() {
+        format!(r#"<{s}:cNvSpPr txBox="1"/>"#)
+    } else {
+        format!("<{s}:cNvSpPr/>")
+    };
+    let fill = if connector { String::new() } else { solid(fill) };
+    let line =
+        if connector { r#"<a:ln w="25400"><a:solidFill><a:srgbClr val="117A65"/></a:solidFill><a:tailEnd type="triangle"/></a:ln>"# } else { "" };
+    let txbx = match text {
+        Some(t) => {
+            let c = n.content;
+            format!(r#"<{s}:txbx><{c}><w:p><w:r><w:t>{t}</w:t></w:r></w:p></{c}></{s}:txbx>"#)
+        }
+        None => String::new(),
+    };
+    format!(
+        r#"<{s}:wsp><{s}:cNvPr id="{cx}" name="{prst}"/>{nv}<{s}:spPr><a:xfrm><a:off x="{x}" y="{y}"/><a:ext cx="{cx}" cy="{cy}"/></a:xfrm><a:prstGeom prst="{prst}"><a:avLst/></a:prstGeom>{fill}{line}</{s}:spPr>{txbx}<{s}:bodyPr anchor="ctr"/></{s}:wsp>"#
+    )
+}
+
+/// Drawing `id`, anchored `x` EMUs from the page's left edge, 1 in × ½ in.
+fn anchor(id: u32, x: u32, uri: &str, payload: &str) -> String {
+    format!(
+        r#"<w:r><w:drawing><wp:anchor distT="0" distB="0" distL="0" distR="0" simplePos="0" relativeHeight="{id}" behindDoc="0" locked="0" layoutInCell="1" allowOverlap="1"><wp:simplePos x="0" y="0"/><wp:positionH relativeFrom="page"><wp:posOffset>{x}</wp:posOffset></wp:positionH><wp:positionV relativeFrom="page"><wp:posOffset>914400</wp:posOffset></wp:positionV><wp:extent cx="914400" cy="457200"/><wp:effectExtent l="0" t="0" r="0" b="0"/><wp:wrapNone/><wp:docPr id="{id}" name="Drawing {id}"/><wp:cNvGraphicFramePr/><a:graphic><a:graphicData uri="{uri}">{payload}</a:graphicData></a:graphic></wp:anchor></w:drawing></w:r>"#
+    )
+}
+
+/// A rectangle, a text box, an arrow, a group holding a nested group, and a canvas holding a
+/// shape and a group, each 1 in × ½ in.
+fn package(n: &Names) -> Vec<u8> {
+    const WPS: &str = "http://schemas.microsoft.com/office/word/2010/wordprocessingShape";
+    const WPG: &str = "http://schemas.microsoft.com/office/word/2010/wordprocessingGroup";
+    const WPC: &str = "http://schemas.microsoft.com/office/word/2010/wordprocessingCanvas";
+    let full = (0, 0, 914_400, 457_200);
+    let half = (0, 0, 457_200, 457_200);
+    let g = n.group;
+    let group_xfrm = |(x, y, cx, cy): (u32, u32, u32, u32)| {
+        format!(
+            r#"<{g}:cNvGrpSpPr/><{g}:grpSpPr><a:xfrm><a:off x="{x}" y="{y}"/><a:ext cx="{cx}" cy="{cy}"/><a:chOff x="0" y="0"/><a:chExt cx="{cx}" cy="{cy}"/></a:xfrm>{}</{g}:grpSpPr>"#,
+            solid("F39C12")
+        )
+    };
+    let group = format!(
+        "<{g}:wgp>{}{}<{g}:grpSp>{}{}</{g}:grpSp></{g}:wgp>",
+        group_xfrm(full),
+        shape(n, "ellipse", half, "28B463", None),
+        group_xfrm((457_200, 0, 457_200, 457_200)),
+        shape(n, "triangle", half, "8E44AD", None),
+    );
+    let c = n.canvas;
+    let canvas = format!(
+        r#"<{c}:wpc><{c}:bg>{}</{c}:bg><{c}:whole><a:ln w="12700">{}</a:ln></{c}:whole>{}<{g}:wgp>{}{}</{g}:wgp></{c}:wpc>"#,
+        solid("FDF2E9"),
+        solid("A04000"),
+        shape(n, "roundRect", (45_720, 45_720, 365_760, 365_760), "16A085", None),
+        group_xfrm((457_200, 45_720, 365_760, 365_760)),
+        shape(n, "diamond", (0, 0, 365_760, 365_760), "2E86C1", None),
+    );
+    let drawings = [
+        anchor(1, 914_400, WPS, &shape(n, "rect", full, "C0392B", None)),
+        anchor(2, 2_286_000, WPS, &shape(n, "rect", full, "EAF2F8", Some("Strict text box"))),
+        anchor(3, 3_657_600, WPS, &shape(n, "straightConnector1", full, "", None)),
+        anchor(4, 5_029_200, WPG, &group),
+        anchor(5, 6_400_800, WPC, &canvas),
+    ];
+    let doc = format!(
+        r#"<?xml version="1.0" encoding="UTF-8" standalone="yes"?><w:document {}><w:body><w:p><w:r><w:t>Shapes</w:t></w:r>{}</w:p><w:sectPr><w:pgSz w:w="12240" w:h="15840"/></w:sectPr></w:body></w:document>"#,
+        n.ns,
+        drawings.concat()
+    );
+    zip(&[("_rels/.rels", n.rels.as_bytes()), ("word/document.xml", doc.as_bytes())])
+}
+
+fn objects(d: &Document) -> Vec<InlineObject> {
+    d.body[0].as_para().unwrap().objects.clone()
+}
+
+/// The text of each text box story, in reading order.
+fn box_texts(d: &Document) -> Vec<String> {
+    let mut ids = Vec::new();
+    d.objects_in_reading_order(&mut |o| ids.extend(o.text_boxes()));
+    ids.iter()
+        .filter_map(|id| d.parts.get(id).filter(|p| p.kind == PartKind::TextBox))
+        .map(|p| p.blocks.iter().filter_map(|b| b.as_para()).map(|p| p.text.clone()).collect::<Vec<_>>().join("\n"))
+        .collect()
+}
+
+#[test]
+fn strict_shapes_read_like_their_transitional_twins() {
+    let transitional = wordcraft_docx::read(&package(&TRANSITIONAL)).unwrap();
+    let want = objects(&transitional);
+    // The twin itself reads every drawing: a rectangle, a text box, an arrow, a group whose
+    // nested group is flattened into it, and a canvas with its background.
+    assert_eq!(want.len(), 5, "{want:#?}");
+    assert!(matches!(&want[0], InlineObject::Shape { story: None, freeform: None, .. }), "{:?}", want[0]);
+    assert!(matches!(&want[1], InlineObject::Shape { story: Some(_), .. }), "{:?}", want[1]);
+    assert!(matches!(&want[3], InlineObject::Group { children, canvas: None, .. } if children.len() == 2), "{:?}", want[3]);
+    assert!(
+        matches!(&want[4], InlineObject::Group { children, canvas: Some(c), .. } if children.len() == 2 && c.fill.is_some() && c.stroke.is_some()),
+        "{:?}",
+        want[4]
+    );
+    assert_eq!(box_texts(&transitional), ["Strict text box"]);
+    for (name, names) in [("Word's Strict names", &STRICT_WORD), ("the Strict schema's names", &STRICT_SCHEMA)] {
+        let strict = wordcraft_docx::read(&package(names)).unwrap();
+        assert_eq!(objects(&strict), want, "{name}");
+        assert_eq!(box_texts(&strict), ["Strict text box"], "{name}");
+    }
+}
+
+#[test]
+fn strict_shapes_are_saved_and_reopened() {
+    let strict = wordcraft_docx::read(&package(&STRICT_WORD)).unwrap();
+    let saved = wordcraft_docx::write(&strict).unwrap();
+    let xml = document_xml(&saved);
+    for name in ["<wps:wsp>", "<wpg:wgp>", "<wpc:wpc>", "<wpc:bg>", "<w:txbxContent>", "<wps:cNvCnPr"] {
+        assert!(xml.contains(name), "{name} missing from the saved document");
+    }
+    let reopened = wordcraft_docx::read(&saved).unwrap();
+    assert_eq!(objects(&reopened), objects(&strict));
+    assert_eq!(box_texts(&reopened), ["Strict text box"]);
+}

crates/docx/tests/it/toc_ranges.rs

@@ -228,6 +228,7 @@ fn hyperlink_field_targets_follow_the_field_code() {
     for (para, link) in [
         (complex(r#"HYPERLINK "https://example.org/a" \l "sec""#), Some("https://example.org/a#sec")),
         (complex(r#"hyperlink \o "A tip" \t "_blank" "https://example.org/""#), Some("https://example.org/")),
+        (complex(r#"HYPERLINK "x \* Upper""#), Some("x * Upper")),
         (complex(r#"HYPERLINK "C:\\Docs\\My file.docx""#), Some(r"C:\Docs\My file.docx")),
         (simple(r#"HYPERLINK \l "Sec" \m"#), Some("#Sec")),
         (simple("HYPERLINK https://example.org/b"), Some("https://example.org/b")),

crates/engine/src/cmd/caret.rs

@@ -323,6 +323,7 @@ fn select_all(s: &mut Session, v: &Value) -> CmdResult {
     }
     let st = s.sel.focus.story;
     s.sel = Selection { anchor: s.doc.start_of(st), focus: s.doc.end_of(st) };
+    s.pending = None;
     s.column = None;
     s.column_mode = false;
     sel_result(s)
@@ -332,6 +333,7 @@ fn select_range(s: &mut Session, v: &Value) -> CmdResult {
     let a = v.get("anchor").and_then(parse_pos).ok_or_else(|| CmdError::Params("bad `anchor`".into()))?;
     let b = v.get("focus").and_then(parse_pos).ok_or_else(|| CmdError::Params("bad `focus`".into()))?;
     s.sel = Selection { anchor: s.doc.clamp(&a), focus: s.doc.clamp(&b) };
+    s.pending = None;
     sel_result(s)
 }
 
@@ -392,6 +394,7 @@ fn select_unit(s: &mut Session, unit: Unit) -> CmdResult {
     if s.doc.para_at(&f).is_some() {
         let (a, b) = unit_range(s, &f, unit);
         s.sel = Selection { anchor: a, focus: b };
+        s.pending = None;
     }
     sel_result(s)
 }
@@ -486,6 +489,9 @@ fn select_text(s: &mut Session, v: &Value) -> CmdResult {
             if count == n {
                 s.sel =
                     Selection { anchor: Pos { story, path: path.clone(), off: i }, focus: Pos { story, path: path.clone(), off: i + needle.len() } };
+                // Formatting picked for the old caret doesn't carry over (as `select.word`): what
+                // replaces the selection takes its formatting.
+                s.pending = None;
                 return sel_result(s);
             }
         }

crates/engine/src/cmd/file.rs

@@ -253,27 +253,8 @@ fn open(s: &mut Session, v: &Value) -> CmdResult {
         );
     }
     let password = p::str(v, "password");
-    let (mut doc, encrypted) = if let Some(data) = p::str(v, "data") {
-        let bytes = super::insert::base64_decode(data).ok_or_else(|| CmdError::Params("bad base64".into()))?;
-        // An empty file (File Explorer's New › Microsoft Word Document makes one) opens as a new
-        // blank document that saves back to it (#1384).
-        if bytes.is_empty() {
-            (normal_document(s), false)
-        } else {
-            crate::io::open_bytes_in(path, &bytes, password, &s.region()).map_err(CmdError::Failed)?
-        }
-    } else if crate::io::is_empty_file(std::path::Path::new(path)) {
-        (normal_document(s), false)
-    } else {
-        crate::io::open_path_in(std::path::Path::new(path), password, &s.region()).map_err(CmdError::Failed)?
-    };
-    // Plain text and Markdown carry no language: they are in the editing language.
-    let ext = std::path::Path::new(path).extension().and_then(std::ffi::OsStr::to_str).unwrap_or("").to_ascii_lowercase();
-    if let Some(lang) = s.prefs.default_language.as_ref().or(s.editing_lang.as_ref())
-        && matches!(ext.as_str(), "txt" | "text" | "md" | "markdown")
-    {
-        Arc::make_mut(&mut doc.styles).default_chr.lang = Some(lang.clone());
-    }
+    // An empty file opens as a new blank document that saves back to it (#1384).
+    let (doc, encrypted) = read_document(s, path, v)?;
     match s.used_because() {
         None => s.set_document(doc),
         Some(why) => {
@@ -294,6 +275,53 @@ fn open(s: &mut Session, v: &Value) -> CmdResult {
     )
 }
 
+/// The document the file at `path` holds (or its bytes, base64 in `data`, with `path` only
+/// naming it), read with `password`, and whether it was encrypted: what `file.open` and
+/// `file.newFromTemplate` read. An empty file (File Explorer's New › Microsoft Word Document makes
+/// one) is a new blank document (#1384). Plain text and Markdown carry no language, so they are
+/// in the editing language.
+fn read_document(s: &Session, path: &str, v: &Value) -> Result<(Document, bool), CmdError> {
+    let password = p::str(v, "password");
+    let (mut doc, encrypted) = if let Some(data) = p::str(v, "data") {
+        let bytes = super::insert::base64_decode(data).ok_or_else(|| CmdError::Params("bad base64".into()))?;
+        if bytes.is_empty() {
+            (normal_document(s), false)
+        } else {
+            crate::io::open_bytes_in(path, &bytes, password, &s.region()).map_err(CmdError::Failed)?
+        }
+    } else if crate::io::is_empty_file(std::path::Path::new(path)) {
+        (normal_document(s), false)
+    } else {
+        crate::io::open_path_in(std::path::Path::new(path), password, &s.region()).map_err(CmdError::Failed)?
+    };
+    let ext = std::path::Path::new(path).extension().and_then(std::ffi::OsStr::to_str).unwrap_or("").to_ascii_lowercase();
+    if let Some(lang) = s.prefs.default_language.as_ref().or(s.editing_lang.as_ref())
+        && matches!(ext.as_str(), "txt" | "text" | "md" | "markdown")
+    {
+        Arc::make_mut(&mut doc.styles).default_chr.lang = Some(lang.clone());
+    }
+    Ok((doc, encrypted))
+}
+
+/// File › New from a template (#2852): a new, untitled document with the template's content,
+/// attached to the template as in Word, read as `file.open` reads a file (`data`, `password`).
+/// It opens beside the open documents, as `file.new` does. With `replaceBlank` an untouched blank
+/// document gives way to it, as to a file `file.open` opens: that is how a template the platform
+/// hands the app starts (#3433). The template is only read; the new document has no file, so
+/// saving it asks where to write, and a protected template's password doesn't come along.
+pub(crate) fn new_from_template(s: &mut Session, v: &Value) -> CmdResult {
+    let path = p::req_str(v, "path")?;
+    let (mut doc, _) = read_document(s, path, v)?;
+    doc.settings.attached_template = Some(path.to_string());
+    if p::bool(v, "replaceBlank") == Some(true) && s.used_because().is_none() {
+        s.set_document(doc);
+    } else {
+        s.add_document(doc);
+    }
+    s.dirty = true;
+    sel_result(s)
+}
+
 /// Close a document with no window to ask about it (agents, headless sessions).
 fn close_document(s: &mut Session, v: &Value) -> CmdResult {
     let id = match v.get("id").filter(|x| !x.is_null()) {

crates/engine/src/cmd/find_pattern.rs

@@ -198,7 +198,9 @@ fn special(query: &str, o: &Opts) -> (String, bool) {
         if (o.ignore_punct && is_punct(c)) || (o.ignore_space && is_space(c)) {
             continue;
         }
-        atoms.push(lit(c));
+        // A keyboard apostrophe also finds text produced by smart quotes. Explicit character
+        // codes above, and literal/regex/wildcard searches, keep exact character matching.
+        atoms.push(if c == '\'' { "['‘’]".into() } else { lit(c) });
     }
     let filler = match (o.ignore_punct, o.ignore_space) {
         (true, true) => r"[\p{P} \t\x{A0}]*",

crates/engine/src/cmd/insert/links.rs

@@ -63,6 +63,17 @@ pub(super) fn link(s: &mut Session, v: &Value) -> CmdResult {
         let end = s.doc.insert_text(&at, &text, &props)?;
         s.sel = Selection::caret(end);
         s.pending = Some(CharProps::default());
+    } else if let Some(text) = retyped_text(s, v) {
+        // Text to display changed: the selection is retyped with it, then linked (as in Word).
+        crate::cmd::paste_text(s, &text)?;
+        let end = s.sel.focus.clone();
+        let start = Pos { off: end.off.saturating_sub(text.len()), ..end.clone() };
+        s.doc.format_range(&start, &end, &|c| {
+            c.link = Some(url.clone());
+            c.style = style.clone();
+        })?;
+        s.sel = Selection::caret(end);
+        s.pending = Some(CharProps::default());
     } else {
         let (a, b) = s.sel.ordered();
         s.doc.format_range(&a, &b, &|c| {
@@ -73,6 +84,33 @@ pub(super) fn link(s: &mut Session, v: &Value) -> CmdResult {
     sel_result(s)
 }
 
+/// The `text` to retype a selection with before linking it: only when it differs from the selected
+/// text and both fit in one line of one paragraph without pictures or fields, outside any field's
+/// result. Otherwise (as Word shows `<<Selection in Document>>`) the selection keeps its text.
+fn retyped_text(s: &Session, v: &Value) -> Option<String> {
+    use wordcraft_doc::para::{COLUMN_BREAK, LINE_BREAK, OBJ, PAGE_BREAK};
+    /// What can't stand in one line of text: breaks, and pictures or fields.
+    const NOT_ONE_LINE: [char; 7] = [LINE_BREAK, '\r', '\u{2029}', '\u{b}', PAGE_BREAK, COLUMN_BREAK, OBJ];
+    let text = p::str(v, "text").filter(|t| !t.is_empty() && t.len() <= 4096)?;
+    if text.contains(NOT_ONE_LINE) {
+        return None;
+    }
+    let (a, b) = s.sel.ordered();
+    if a.story != b.story || a.path != b.path {
+        return None;
+    }
+    let shown = s.doc.para(a.story, &a.path)?.text.get(a.off..b.off)?;
+    if shown.contains(NOT_ONE_LINE) || shown == text {
+        return None;
+    }
+    // A selection in a field's result (between its markers, which it doesn't hold) is retyped by
+    // the next update of the field, not here.
+    if s.doc.field_ranges(a.story).iter().any(|r| r.start < b && a <= r.end) {
+        return None;
+    }
+    Some(text.to_string())
+}
+
 /// Whether `name` can name a bookmark: starts with a letter or underscore, contains only alphanumeric characters
 /// or underscores, and is at most 40 characters (#3491). Word allows leading underscores for hidden bookmarks (_Toc, _Ref).
 pub fn bookmark_name_ok(name: &str) -> bool {

crates/engine/src/cmd/lists.rs

@@ -452,6 +452,36 @@ mod tests {
         assert_eq!(numbered(&mut b), want);
     }
 
+    /// #5757: a list numbered in Thai digits shows them on the page, with the level's text around
+    /// them (`๑.`, `(๒)`), and keeps its format through a save to .docx and back.
+    #[test]
+    fn a_thai_numbered_list_shows_thai_digits_and_survives_docx() {
+        let mut s = Session::new(Document::new());
+        for (i, t) in ["หนึ่ง", "สอง", "สาม", "a", "b", "c", "d", "e", "f", "g", "h"].iter().enumerate() {
+            if i > 0 {
+                s.run("text.newParagraph", &json!({})).unwrap();
+            }
+            s.run("text.insert", &json!({"text": t})).unwrap();
+        }
+        s.run("select.all", &json!({})).unwrap();
+        let levels = json!([{"format": "thaiNumbers", "text": "%1.", "start": 1}]);
+        s.run("list.define", &json!({"levels": levels})).unwrap();
+        let page: Vec<Option<String>> = numbered(&mut s).into_iter().map(|x| x.0).collect();
+        let want: Vec<Option<String>> =
+            ["๑.", "๒.", "๓.", "๔.", "๕.", "๖.", "๗.", "๘.", "๙.", "๑๐.", "๑๑."].iter().map(|l| Some(l.to_string())).collect();
+        assert_eq!(page, want);
+        let back = wordcraft_docx::read(&wordcraft_docx::write(&s.doc).unwrap()).unwrap();
+        let mut b = Session::new(back);
+        let again: Vec<Option<String>> = numbered(&mut b).into_iter().map(|x| x.0).collect();
+        assert_eq!(again, want, "thaiNumbers survives .docx");
+        // In parentheses too.
+        let mut s = Session::new(Document::new());
+        s.run("text.insert", &json!({"text": "x"})).unwrap();
+        s.run("select.all", &json!({})).unwrap();
+        s.run("list.define", &json!({"levels": [{"format": "thaiNumbers", "text": "(%1)", "start": 2}]})).unwrap();
+        assert_eq!(numbered(&mut s)[0].0.as_deref(), Some("(๒)"));
+    }
+
     /// #1793: the other heading presets of Word's List Library, level by level.
     #[test]
     fn every_heading_preset_numbers_three_heading_levels() {
@@ -459,7 +489,8 @@ mod tests {
             ("headings", [Some("1"), Some("1.1"), Some("1.1.1")]),
             ("article", [Some("Article I."), Some("Section 1.01"), Some("(a)")]),
             ("outlineHeadings", [Some("I."), Some("A."), Some("1.")]),
-            ("chapter", [Some("Chapter 1"), None, None]),
+            // The lower levels are "none" levels with no text: an empty label, as in Word (#4985).
+            ("chapter", [Some("Chapter 1"), Some(""), Some("")]),
         ] {
             let mut s = headings("Heading3");
             s.run("list.multilevelPreset", &json!({"preset": preset})).unwrap();

crates/engine/src/cmd/objects/picture.rs

@@ -76,9 +76,11 @@ pub fn cropped_original_size(s: &Session, o: &InlineObject) -> Option<(f32, f32)
 pub(super) fn size(s: &mut Session, v: &Value) -> CmdResult {
     let (_, o) = selected(s).ok_or_else(|| CmdError::Disabled("no object".into()))?;
     let lock = p::bool(v, "lockAspect").unwrap_or(true);
-    let (mut w, mut h) = obj_size(&o);
+    let (w0, h0) = obj_size(&o);
+    let (mut w, mut h) = (w0, h0);
     let scale = p::f32(v, "scale");
-    if p::bool(v, "original").unwrap_or(false) {
+    let original = p::bool(v, "original").unwrap_or(false);
+    if original {
         let (ow, oh) =
             cropped_original_size(s, &o).ok_or_else(|| CmdError::Disabled("only a picture whose image can be read has an original size".into()))?;
         let k = scale.unwrap_or(100.0) / 100.0;
@@ -113,7 +115,16 @@ pub(super) fn size(s: &mut Session, v: &Value) -> CmdResult {
         (None, None) => true,
         _ => lock,
     };
-    let (w, h) = fit_size(w, h, keep);
+    // Sides set in proportion (from the original, or from a size whose sides both have a length)
+    // fit the limits together. Otherwise each side the call sets fits on its own, and a side it
+    // leaves as it was keeps its size, a straight line's 0 too.
+    let (w, h) = if keep && (original || (w0 > 0.0 && h0 > 0.0 && (w, h) != (w0, h0))) {
+        fit_size(w, h, true)
+    } else {
+        let (fw, fh) = fit_size(w, h, false);
+        let left = |new: f32, old: f32| !original && new == old;
+        (if left(w, w0) { w } else { fw }, if left(h, h0) { h } else { fh })
+    };
     // Any drawing: a group's members scale with its frame (its child space is kept); a chart or
     // SmartArt graphic lays itself out again at the new size.
     let theme = s.doc.settings.theme_colors.clone();

crates/engine/src/cmd/objects/specs.rs

@@ -13,7 +13,7 @@ pub fn specs() -> Vec<CommandSpec> {
     vec![
         CommandSpec::new("picture.size", "Size", "Picture Format › Size", size)
             .params(
-                r#"{"width"?: pt, "height"?: pt, "lockAspect"?: bool, "scale"?: percent, "original"?: bool} (scale: of the current size, or with original of the picture's original size, its pixels at 96 ppi less what is cropped; original alone restores that size; one side alone keeps the proportions unless lockAspect is false; 1–4000 pt: when the proportions are kept, a side past 4000 pt shrinks the other with it)"#,
+                r#"{"width"?: pt, "height"?: pt, "lockAspect"?: bool, "scale"?: percent, "original"?: bool} (scale: of the current size, or with original of the picture's original size, its pixels at 96 ppi less what is cropped; original alone restores that size; one side alone keeps the proportions unless lockAspect is false; 1–4000 pt: when the proportions are kept (from the original, or with both sides having a length), a side past 4000 pt shrinks the other with it; a side the call leaves as it was keeps its size, a straight line's 0 too)"#,
             )
             .when(has_sizable),
         CommandSpec::new("picture.crop", "Crop", "Picture Format › Size", crop)

crates/engine/src/cmd/objects/tests.rs

@@ -1034,6 +1034,66 @@ fn picture_size_past_the_limit_keeps_the_proportions() {
     }
 }
 
+/// A side the call leaves as it was keeps its size, even outside 1–4000 pt: a straight line's 0
+/// stays 0 when only its length is typed. Sides set in proportion (the original, or a size whose
+/// sides both have a length) fit the limits together; any other side the call sets, on its own.
+#[test]
+fn picture_size_keeps_the_side_it_leaves() {
+    let mut s = Session::new(wordcraft_doc::Document::new());
+    s.run("insert.shape", &json!({"kind": "straightConnector", "width": 100.0, "height": 0.0})).unwrap();
+    let size = |s: &Session| obj_size(&selected(s).unwrap().1);
+    // A side a scale sets on its own: 0 has no length to scale.
+    let scaled = |x: f32, k: f32| if x == 0.0 { 0.0 } else { (x * k).clamp(1.0, 4000.0) };
+    // Sizes with both sides follow each other: they fit 1–4000 pt by the one factor nearest 1, worked out by hand
+    // for the calls `width` locked, `height` locked, `scale` 50 and `scale` 200 with a width (in that order).
+    let states = [
+        ((100.0, 0.0), None),
+        ((0.0, 100.0), None),
+        ((5000.0, 0.0), None),
+        ((0.0, 5000.0), None),
+        ((100.0, 0.5), Some([(200.0, 1.0), (4000.0, 20.0), (200.0, 1.0), (200.0, 1.0)])),
+        ((5000.0, 50.0), Some([(200.0, 2.0), (4000.0, 40.0), (2500.0, 25.0), (200.0, 2.0)])),
+        ((100.0, 50.0), Some([(200.0, 100.0), (400.0, 200.0), (50.0, 25.0), (200.0, 100.0)])),
+    ];
+    for ((w0, h0), together) in states {
+        let follow = |i: usize, alone: (f32, f32)| together.map_or(alone, |t: [(f32, f32); 4]| t[i]);
+        let cases = [
+            (json!({"width": 200.0, "lockAspect": false}), (200.0, h0)),
+            (json!({"height": 200.0, "lockAspect": false}), (w0, 200.0)),
+            (json!({"width": 200.0}), follow(0, (200.0, h0))),
+            (json!({"height": 200.0}), follow(1, (w0, 200.0))),
+            (json!({"scale": 50}), follow(2, (scaled(w0, 0.5), scaled(h0, 0.5)))),
+            (json!({"scale": 200, "width": 200.0}), follow(3, (200.0, scaled(h0, 2.0)))),
+            (json!({"scale": 100}), (w0, h0)),
+            (json!({}), (w0, h0)),
+            // The Size dialog: both sides sent, the height as it was.
+            (json!({"width": 200.0, "height": h0, "lockAspect": false}), (200.0, h0)),
+            (json!({"width": 200.0, "height": 300.0}), (200.0, 300.0)),
+        ];
+        for (v, (ew, eh)) in cases {
+            with_obj(&mut s, |o| o.set_size(w0, h0)).unwrap();
+            s.run("picture.size", &v).unwrap();
+            let (w, h) = size(&s);
+            assert!((w - ew).abs() < 1e-3 && (h - eh).abs() < 1e-3, "{w0}×{h0} {v}: {w}×{h}, not {ew}×{eh}");
+        }
+    }
+    // Both sides of a picture's original size fit together, whatever size it had, even when the
+    // height comes out as it was: 4050×1012.5 pt is 4000×1000, not 4000×1012.5.
+    let data = super::super::insert::base64_encode(&png(400, 100));
+    s.run("insert.picture", &json!({"data": data})).unwrap();
+    for (w0, h0) in [(3000.0, 1012.5), (4050.0, 0.0), (100.0, 100.0)] {
+        with_obj(&mut s, |o| o.set_size(w0, h0)).unwrap();
+        s.run("picture.size", &json!({"original": true, "scale": 1350})).unwrap();
+        let (w, h) = size(&s);
+        assert!((w - 4000.0).abs() < 0.01 && (h - 1000.0).abs() < 0.01, "{w0}×{h0}: {w}×{h}");
+    }
+    // A height the original sets is set, though it is the one the picture had: 300×75 pt at
+    // 5400 % is 16200×4050, so with a width of 200 the height fits to 4000.
+    with_obj(&mut s, |o| o.set_size(100.0, 4050.0)).unwrap();
+    s.run("picture.size", &json!({"original": true, "scale": 5400, "width": 200.0, "lockAspect": false})).unwrap();
+    assert_eq!(size(&s), (200.0, 4000.0));
+}
+
 /// A crop handle drop: `picture.crop` with the new size is one edit, one undo step.
 #[test]
 fn picture_crop_with_size_is_one_undo_step() {

crates/engine/src/cmd/para.rs

@@ -183,9 +183,17 @@ pub fn specs() -> Vec<CommandSpec> {
             .key(if cfg!(target_os = "macos") { "Ctrl+Shift+Right" } else { "Alt+Shift+Right" }),
         CommandSpec::new("para.shading", "Shading", "Home › Paragraph", |s, v| {
             let c = p::str(v, "color").and_then(Rgb::parse);
+            // Text selected within one paragraph is shaded itself, as in Word, so No Color takes
+            // off a shading a file gave the text (#4685); a caret or paragraphs shade paragraphs.
+            // `paragraph` (Borders and Shading › Apply to: Paragraph) shades the paragraphs anyway.
+            if p::bool(v, "paragraph") != Some(true) && shades_text(s) {
+                return super::format::apply(s, &|x| x.shading = c);
+            }
             fmt(s, &|p| p.shading = c)
         })
-        .params(r#"{"color": "RRGGBB" | null}"#),
+        .params(
+            r#"{"color": "RRGGBB" | null, "paragraph"?: bool} (text selected inside one paragraph is shaded itself, as in Word; `paragraph: true` shades the paragraphs instead, as Borders and Shading › Apply to: Paragraph)"#,
+        ),
         CommandSpec::new("para.borders", "Borders", "Home › Paragraph", borders).params(
             r#"{"kind": "bottom|top|left|right|none|all|outside|inside|horizontalLine", "width"?: pt, "color"?: "RRGGBB", "style"?: "single|double|dotted|dashed|thick|triple|dotDash|wave", "sides"?: Sides} where Sides = {"top"|"left"|"bottom"|"right"|"between": {"style"?, "width"?: pt, "color"?: "RRGGBB"} | null}: exactly these sides (others off), instead of `kind`"#,
         ),
@@ -264,3 +272,21 @@ pub fn specs() -> Vec<CommandSpec> {
         }),
     ]
 }
+
+/// Whether Shading applies to the selected text rather than to paragraphs: a selection inside one
+/// paragraph, short of its mark. A selection reaching past the mark ends at the next paragraph's
+/// start, so all of a paragraph's text is still text, except in the last paragraph of its story,
+/// cell or box, where it stands for the paragraph with its mark. A selected picture, shape or text
+/// box is its object character, not text (a field's object character is).
+fn shades_text(s: &Session) -> bool {
+    let (a, b) = s.sel.ordered();
+    if a == b || a.story != b.story || a.path != b.path {
+        return false;
+    }
+    let Some(p) = s.doc.para_at(&a) else { return false };
+    if super::objects::object_selection(s).is_some() {
+        return false;
+    }
+    let last = s.doc.next_para(a.story, &a.path).is_none_or(|q| q.parent() != a.path.parent());
+    !(last && a.off == 0 && b.off >= p.len())
+}

crates/engine/src/cmd/references.rs

@@ -103,8 +103,9 @@ pub fn specs() -> Vec<CommandSpec> {
 
 /// A field instruction split into its name, positional arguments and switches (ECMA-376
 /// §17.16.1): `TOC \o "1-3" \h` → name `TOC`, switches `o` = `1-3` and `h`. Quoted text stays
-/// whole (without the quotes; `\"` and `\\` inside it are a quote and a backslash). A switch
-/// takes the word after it as its argument unless that word is another switch.
+/// whole (without the quotes; a backslash inside it makes the next character literal, as in Word:
+/// `\"` is a quote, `\\` a backslash, `\*` an asterisk). A switch takes the word after it as its
+/// argument unless that word is another switch.
 #[derive(Clone, Debug, Default, PartialEq)]
 pub struct FieldCode {
     pub name: String,
@@ -126,7 +127,7 @@ pub fn field_words(instr: &str) -> Vec<(String, bool)> {
             while let Some(c) = chars.next() {
                 match c {
                     '"' => break,
-                    '\\' if matches!(chars.peek(), Some('"' | '\\')) => w.extend(chars.next()),
+                    '\\' if chars.peek().is_some() => w.extend(chars.next()),
                     c => w.push(c),
                 }
             }

crates/engine/src/cmd/references/tests.rs

@@ -1,3 +1,4 @@
+use super::properties::{new_property_value, text_result};
 use super::*;
 
 fn fields(s: &Session) -> Vec<(String, String)> {
@@ -446,7 +447,7 @@ fn update_fields_apply_the_capitalization_switches() {
             "seventh",
             "Seventh",
             "BUDGET",
-            "new \\* Upper"
+            "new * Upper"
         ]
     );
 }
@@ -548,6 +549,85 @@ fn caption_for_an_object_in_a_cell_stays_in_the_cell() {
     assert_eq!(texts.get(1).map(String::as_str), Some("Figure 1"), "{texts:?}");
 }
 
+/// A fresh scratch directory for the file-name tests below.
+#[cfg(not(target_arch = "wasm32"))]
+fn filename_scratch_dir(tag: &str) -> std::path::PathBuf {
+    let nanos = std::time::SystemTime::now().duration_since(std::time::UNIX_EPOCH).unwrap().as_nanos();
+    let dir = std::env::temp_dir().join(format!("wordcraft-filename-field-{tag}-{}-{nanos}", std::process::id()));
+    std::fs::create_dir_all(&dir).unwrap();
+    dir
+}
+
+/// #3796: after Save As to a new name, Update Field with the caret in a header shows the new
+/// name for FILENAME and the new full path for FILENAME \p.
+#[cfg(not(target_arch = "wasm32"))]
+#[test]
+fn filename_fields_in_a_header_follow_save_as() {
+    let dir = filename_scratch_dir("save-as");
+    let (mut s, hdr) = header_fields_session(wordcraft_doc::Document::from_text("body"), &["FILENAME", r"FILENAME \p"]);
+    let first = dir.join("report_v1.docx");
+    let second = dir.join("report_v2.docx");
+    s.run("file.save", &json!({"path": first.to_string_lossy()})).unwrap();
+    s.run("references.updateFields", &json!({})).unwrap();
+    let got = story_fields(&s, StoryRef::Part(hdr));
+    assert_eq!(got[0].1, "report_v1.docx");
+    assert_eq!(got[1].1, first.to_string_lossy());
+
+    s.run("file.saveAs", &json!({"path": second.to_string_lossy()})).unwrap();
+    s.run("references.updateFields", &json!({})).unwrap();
+    let got = story_fields(&s, StoryRef::Part(hdr));
+    assert_eq!(got[0].1, "report_v2.docx");
+    assert_eq!(got[1].1, second.to_string_lossy());
+    let _ = std::fs::remove_dir_all(&dir);
+}
+
+/// #3796: a Word-style `FILENAME \* MERGEFORMAT` field whose cached text names the old file
+/// shows the new name after Save As and Update Field.
+#[cfg(not(target_arch = "wasm32"))]
+#[test]
+fn filename_mergeformat_field_replaces_its_cached_name_after_save_as() {
+    let dir = filename_scratch_dir("mergeformat");
+    let (mut s, hdr) = header_fields_session(wordcraft_doc::Document::from_text("body"), &[r"FILENAME \* MERGEFORMAT"]);
+    let first = dir.join("report_v1.docx");
+    s.run("file.save", &json!({"path": first.to_string_lossy()})).unwrap();
+    // The cached text a file written earlier would carry.
+    let path = s.doc.para_paths(StoryRef::Part(hdr)).remove(0);
+    if let Some(InlineObject::Field { result, .. }) = s.doc.para_mut(StoryRef::Part(hdr), &path).unwrap().objects.first_mut() {
+        *result = "report_v1.docx".into();
+    }
+    s.run("file.saveAs", &json!({"path": dir.join("report_v2.docx").to_string_lossy()})).unwrap();
+    s.run("references.updateFields", &json!({})).unwrap();
+    assert_eq!(story_fields(&s, StoryRef::Part(hdr))[0].1, "report_v2.docx");
+    let _ = std::fs::remove_dir_all(&dir);
+}
+
+/// #3796: renaming the document (File > Info) changes what FILENAME shows on the next Update
+/// Field; the extension stays.
+#[cfg(not(target_arch = "wasm32"))]
+#[test]
+fn filename_field_in_a_header_follows_rename() {
+    let dir = filename_scratch_dir("rename");
+    let (mut s, hdr) = header_fields_session(wordcraft_doc::Document::from_text("body"), &["FILENAME"]);
+    s.run("file.save", &json!({"path": dir.join("report_v1.docx").to_string_lossy()})).unwrap();
+    s.run("file.rename", &json!({"name": "report_v3"})).unwrap();
+    s.run("references.updateFields", &json!({})).unwrap();
+    assert_eq!(story_fields(&s, StoryRef::Part(hdr))[0].1, "report_v3.docx");
+    let _ = std::fs::remove_dir_all(&dir);
+}
+
+/// A backslash in a quoted field argument makes the next character literal, so `\*` there is an
+/// asterisk and not the `\* Upper` switch (Word shows `x * Upper` for `TITLE "x \* Upper"`).
+#[test]
+fn quoted_field_argument_backslash_escapes() {
+    for (instr, shown) in [(r#"TITLE "x \* Upper""#, "x * Upper"), (r#"TITLE "x \\* Upper""#, r"x \* Upper"), (r#"TITLE "a \"b\" c""#, r#"a "b" c"#)]
+    {
+        let value = new_property_value(instr).unwrap();
+        assert_eq!(value, shown, "{instr}");
+        assert_eq!(text_result(instr, value), shown, "{instr}");
+    }
+    assert_eq!(text_result(r#"TITLE "ab" \* Upper"#, new_property_value(r#"TITLE "ab" \* Upper"#).unwrap()), "AB");
+}
+
 /// Output (PDF, print, page images) shows the body's unlocked SEQ, STYLEREF and REF fields with
 /// their fresh results, as Word fills them in when it prints or exports even with "Update fields
 /// before printing" off; NUMWORDS and a locked field keep their stored results there, and the

crates/engine/src/cmd/table.rs

@@ -195,7 +195,7 @@ pub fn specs() -> Vec<CommandSpec> {
         })
         .params(r#"{"text"?: string, "title"?: string} (empty clears)"#)),
         t(CommandSpec::new("table.sort", "Sort", "Table Layout › Data", sort_table)
-            .params(r#"{"column"?: n, "descending"?: bool, "header"?: bool}"#)),
+            .params(r#"{"column"?: non-negative JSON integer (0-based, default 0; must name a column in the table), "descending"?: bool, "header"?: bool}"#)),
         t(CommandSpec::new("table.toText", "Convert to Text", "Table Layout › Data", to_text)
             .params(r#"{"separator"?: "tab|comma|paragraph|other", "char"?: string (with other)} (the selected rows, or the whole table)"#)),
         t(CommandSpec::new("table.formula", "Formula", "Table Layout › Data", formula).params(r#"{"formula"?: "=SUM(ABOVE)"}"#)),

crates/engine/src/cmd/table/convert.rs

@@ -8,11 +8,18 @@ use super::{
 /// Sort the table's rows by a column. Numbers are read in each cell's language (`1,5` is one and
 /// a half in Polish) and sort first, numerically; other text sorts in the column's language.
 pub(super) fn sort_table(s: &mut Session, v: &Value) -> CmdResult {
-    let col = p::u64(v, "column").unwrap_or(0) as usize;
+    let column = match v.get("column") {
+        None => 0,
+        Some(value) => value.as_u64().ok_or_else(|| CmdError::Params("`column` must be a non-negative JSON integer".into()))?,
+    };
     let desc = p::bool(v, "descending").unwrap_or(false);
     let header = p::bool(v, "header");
     let (tp, _, _) = cell(s)?;
     let t = s.doc.table(s.sel.focus.story, &tp).ok_or_else(|| CmdError::Failed("no table".into()))?;
+    let col = usize::try_from(column)
+        .ok()
+        .filter(|&col| t.rows.iter().any(|row| col < row.cells.len()))
+        .ok_or_else(|| CmdError::Params("`column` is outside the table".into()))?;
     let skip = usize::from(header.unwrap_or(t.rows.first().is_some_and(|r| r.props.header))).min(t.rows.len());
     let column = || t.rows.iter().skip(skip).map(|r| r.cells.get(col));
     let lang = column().flatten().find_map(|c| cell_lang(&s.doc, c)).or_else(|| crate::cmd::default_lang(s));

crates/engine/src/cmd/tools/actions.rs

@@ -1,5 +1,4 @@
 use super::{Arc, CmdError, CmdResult, Document, Pos, Selection, Session, StoryRef, Value, delete_selection, json, p, sel_result};
-use wordcraft_doc::CharProps;
 
 /// Options › Language › Set as default: only the saved default proofing language (what future
 /// blank documents are written in), as Word's Set as Preferred. A setting, like `file.options`: the
@@ -15,15 +14,8 @@ pub(super) fn preferred_language(s: &mut Session, v: &Value) -> CmdResult {
 /// Set Proofing Language: the selection's language and "do not check spelling or grammar",
 /// or (`default`) the document's default and the saved language for future blank documents;
 /// `detect` is the per-user detection preference.
-/// Nothing at all sets US English, as before. `scope: "document"` sets it for every story.
+/// Nothing at all sets US English, as before.
 pub(super) fn language(s: &mut Session, v: &Value) -> CmdResult {
-    // Checked first, so a bad scope changes nothing (not even the detect preference).
-    let whole_document = match v.get("scope") {
-        None | Some(Value::Null) => false,
-        Some(Value::String(t)) if t == "selection" => false,
-        Some(Value::String(t)) if t == "document" => true,
-        Some(other) => return Err(CmdError::Params(format!("`scope` must be \"selection\" or \"document\", not {other}"))),
-    };
     let lang = match p::str(v, "lang") {
         Some(t) => Some(
             wordcraft_proof::lang::normalize_tag(t)
@@ -53,75 +45,20 @@ pub(super) fn language(s: &mut Session, v: &Value) -> CmdResult {
     if lang.is_none() && no_proof.is_none() {
         return Ok(json!({"detect": s.prefs.detect_language}));
     }
-    let change = |c: &mut CharProps| {
+    let result = crate::cmd::format::apply(s, &|c| {
         if let Some(l) = &lang {
             c.lang = Some(l.clone());
         }
         if let Some(np) = no_proof {
             c.no_proof = Some(np);
         }
-    };
-    let result = if whole_document {
-        set_document_props(s, &change)?;
-        json!({"scope": "document"})
-    } else {
-        crate::cmd::format::apply(s, &change)?
-    };
+    })?;
     if p::bool(v, "default") == Some(true) {
         s.prefs.default_language = lang;
     }
     Ok(result)
 }
 
-/// Apply `f` to the character formatting of every story of the document (Language dialog, "Apply
-/// to the whole document"): text and paragraph marks of the body, headers, footers, notes,
-/// comments, text boxes (all `Document::parts`), table cells, and the glossary's building blocks.
-/// The selection is untouched; formatting waiting for typing follows.
-fn set_document_props(s: &mut Session, f: &dyn Fn(&mut CharProps)) -> Result<(), CmdError> {
-    let stories: Vec<StoryRef> = std::iter::once(StoryRef::Body).chain(s.doc.parts.keys().map(|k| StoryRef::Part(*k))).collect();
-    for story in stories {
-        for path in s.doc.para_paths(story) {
-            let para = s.doc.para_mut(story, &path)?;
-            let len = para.len();
-            para.format(0, len, f)?;
-            f(&mut para.mark);
-            para.touch();
-        }
-    }
-    for block in &mut s.doc.glossary {
-        format_blocks(&mut block.blocks, f, 0);
-    }
-    if let Some(pend) = s.pending.as_mut() {
-        f(pend);
-    }
-    Ok(())
-}
-
-/// [`set_document_props`] for a block list outside any story (depth-limited like `para_paths`).
-fn format_blocks(blocks: &mut wordcraft_doc::Blocks, f: &dyn Fn(&mut CharProps), depth: usize) {
-    if depth > wordcraft_doc::PARA_WALK_DEPTH {
-        return;
-    }
-    for blk in blocks.iter_mut() {
-        match Arc::make_mut(blk) {
-            wordcraft_doc::Block::Para(p) => {
-                let len = p.len();
-                // A range inside the paragraph can't fail; the mark is formatted regardless.
-                let _ = p.format(0, len, f);
-                f(&mut p.mark);
-                p.touch();
-            }
-            wordcraft_doc::Block::Table(t) => {
-                for row in &mut t.rows {
-                    for cell in &mut row.cells {
-                        format_blocks(&mut cell.blocks, f, depth + 1);
-                    }
-                }
-            }
-        }
-    }
-}
-
 pub(super) fn macros(s: &mut Session, v: &Value) -> CmdResult {
     if let Some(d) = v.get("define") {
         let name = d.get("name").and_then(Value::as_str).unwrap_or("Macro").to_string();

crates/engine/src/cmd/tools/specs.rs

@@ -121,17 +121,11 @@ pub fn specs() -> Vec<CommandSpec> {
         CommandSpec::new("file.recover", "Recover Unsaved Documents", "File › Info", |s, v| s.recover(v))
         .params(r#"{"restore"?: recoveryId, "discard"?: recoveryId, "refresh"?: true}; desktop only; returns {enabled,busy,documents,error}; restore loads asynchronously without a file path"#)
         .pure(),
-        CommandSpec::new("file.newFromTemplate", "New from Template", "File › New", |s, v| {
-            let path = p::req_str(v, "path")?;
-            let mut doc = crate::io::open_path(std::path::Path::new(path)).map_err(CmdError::Failed)?;
-            // As in Word, the new document is attached to the template it came from (#2852).
-            doc.settings.attached_template = Some(path.to_string());
-            s.add_document(doc);
-            s.dirty = true;
-            sel_result(s)
-        })
-        .params(r#"{"path": string} → a new, untitled document from the template, attached to it"#)
-        .pure(),
+        CommandSpec::new("file.newFromTemplate", "New from Template", "File › New", crate::cmd::file::new_from_template)
+            .params(
+                r#"{"path": string, "data"?: base64 (the template's bytes; `path` then only names it), "password"?: string (for a password-protected template), "replaceBlank"?: bool (an untouched blank document gives way to it, as to a file file.open opens)} → a new, untitled document from the template, attached to it"#,
+            )
+            .pure(),
         CommandSpec::new("file.saveTemplate", "Save as Template", "File › Save As", |s, v| {
             let path = p::req_str(v, "path")?;
             // A .dotm keeps its name (#2852); anything else becomes a .dotx.
@@ -340,7 +334,7 @@ pub fn specs() -> Vec<CommandSpec> {
         })
         .params(r#"{"format": "decimal|upperRoman|lowerLetter…", "text"?: "%1.", "start"?: n}"#),
         CommandSpec::new("review.language", "Language", "Review › Language", language).params(
-            r#"{"lang"?: "en-US|en-GB|pl-PL|fr-FR|…" (BCP 47; default en-US), "noProof"?: bool, "detect"?: bool (detect language automatically, a preference), "default"?: bool (also the document's default and the saved language for future blank documents), "scope"?: "selection" (default) | "document" (every run and paragraph mark of every story: body, headers, footers, notes, comments, text boxes, tables, glossary; one undo step, selection unchanged)}"#,
+            r#"{"lang"?: "en-US|en-GB|pl-PL|fr-FR|…" (BCP 47; default en-US), "noProof"?: bool, "detect"?: bool (detect language automatically, a preference), "default"?: bool (also the document's default and the saved language for future blank documents)}"#,
         ),
         CommandSpec::new("review.language.preferred", "Set as Default Language", "File › Options › Language", preferred_language)
             .params(r#"{"lang": "en-US|en-GB|pl-PL|nl-NL|…" (BCP 47)} — only the saved default proofing language, what future blank documents are written in, as Options › Language › Set as default does; the open document, the selection and the undo history don't change. Returns {"preferred": tag}."#)

crates/engine/src/cmd/tools/tests.rs

@@ -290,104 +290,3 @@ fn inspect_removals_work_when_the_document_does_not_track_changes() {
     assert!(!text.contains("Hidden"), "the hidden text is still there: {text}");
     assert!(text.contains("Visible text."), "{text}");
 }
-
-/// #3518: every run and paragraph mark of a story, as (story, path, lang, noProof) rows.
-fn proofing_rows(s: &Session) -> Vec<(StoryRef, Option<String>, Option<bool>)> {
-    let stories = std::iter::once(StoryRef::Body).chain(s.doc.parts.keys().map(|k| StoryRef::Part(*k)));
-    let mut out = Vec::new();
-    for st in stories {
-        for path in s.doc.para_paths(st) {
-            let Some(p) = s.doc.para(st, &path) else { continue };
-            out.push((st, p.mark.lang.clone(), p.mark.no_proof));
-            out.extend(p.runs.iter().map(|r| (st, r.props.lang.clone(), r.props.no_proof)));
-        }
-    }
-    out
-}
-
-/// A document with its text in German in the body, a table cell, a header, a footnote, a comment
-/// and a text box, as a Word DOCX keeps a `w:lang` on each run, plus a glossary entry.
-fn multi_story_doc() -> Session {
-    use wordcraft_doc::{Block, BuildingBlock, CharProps, Paragraph, PartKind, Table};
-    let de = || CharProps { lang: Some("de-DE".into()), ..Default::default() };
-    let blk = |t: &str| Arc::new(Block::Para(Paragraph::with_text(t, de())));
-    let mut d = Document::from_text("Guten Tag");
-    d.body = vec![blk("Guten Tag")];
-    let mut t = Table::new(1, 2, 200.0);
-    if let Some(c) = t.rows.first_mut().and_then(|r| r.cells.first_mut()) {
-        c.blocks = vec![blk("Zelle")];
-    }
-    d.body.push(Arc::new(Block::Table(t)));
-    d.body.push(blk("Ende"));
-    for kind in [PartKind::Header, PartKind::Footnote, PartKind::Comment, PartKind::TextBox] {
-        d.add_part(kind, vec![blk("Teil")]);
-    }
-    d.glossary.push(BuildingBlock { name: "g".into(), blocks: vec![blk("Baustein")], ..Default::default() });
-    Session::new(d)
-}
-
-#[test]
-fn language_scope_document_sets_every_story() {
-    let mut s = multi_story_doc();
-    s.run("select.text", &json!({"text": "Guten"})).unwrap();
-    let (sel, before) = (s.sel.clone(), s.doc.clone());
-    let rows = proofing_rows(&s);
-    assert!(rows.len() >= 10 && rows.iter().any(|r| r.1.as_deref() == Some("de-DE")));
-    s.run("review.language", &json!({"lang": "fr-FR", "scope": "document"})).unwrap();
-    // Runs that had their own language, the paragraph marks and the empty cell: all French.
-    let rows = proofing_rows(&s);
-    assert!(rows.iter().all(|r| r.1.as_deref() == Some("fr-FR")), "{rows:?}");
-    for kind_story in s.doc.parts.keys() {
-        assert!(rows.iter().any(|r| r.0 == StoryRef::Part(*kind_story)));
-    }
-    assert!(s.doc.plain_text(StoryRef::Body).contains("Zelle"));
-    let g = &s.doc.glossary[0].blocks;
-    assert_eq!(g[0].as_para().map(|p| p.runs[0].props.lang.clone()), Some(Some("fr-FR".into())));
-    assert_eq!(g[0].as_para().map(|p| p.mark.lang.clone()), Some(Some("fr-FR".into())));
-    // The selection stays; the document is modified.
-    assert_eq!(s.sel, sel);
-    assert!(s.doc != before);
-    // One undo restores everything.
-    s.run("edit.undo", &json!({})).unwrap();
-    assert!(s.doc == before);
-    assert_eq!(s.sel, sel);
-}
-
-#[test]
-fn language_scope_document_no_proof_and_pending() {
-    let mut s = multi_story_doc();
-    s.run("review.language", &json!({"noProof": true, "scope": "document"})).unwrap();
-    assert!(proofing_rows(&s).iter().all(|r| r.2 == Some(true)));
-    // The language itself wasn't asked for and stays.
-    assert!(proofing_rows(&s).iter().any(|r| r.1.as_deref() == Some("de-DE")));
-    // Formatting waiting for typing follows.
-    s.run("format.bold", &json!({})).unwrap();
-    s.run("review.language", &json!({"lang": "pl-PL", "scope": "document"})).unwrap();
-    assert_eq!(s.pending.as_ref().and_then(|p| p.lang.as_deref()), Some("pl-PL"));
-}
-
-#[test]
-fn language_default_scope_is_selection_only() {
-    let mut s = multi_story_doc();
-    s.run("select.text", &json!({"text": "Guten"})).unwrap();
-    s.run("review.language", &json!({"lang": "fr-FR"})).unwrap();
-    let rows = proofing_rows(&s);
-    assert!(rows.iter().any(|r| r.1.as_deref() == Some("fr-FR")));
-    assert!(rows.iter().filter(|r| r.1.as_deref() == Some("fr-FR")).count() < rows.len() / 2);
-    let mut s = multi_story_doc();
-    s.run("select.text", &json!({"text": "Guten"})).unwrap();
-    s.run("review.language", &json!({"lang": "fr-FR", "scope": "selection"})).unwrap();
-    assert_eq!(proofing_rows(&s), rows);
-}
-
-#[test]
-fn language_bad_scope_is_a_params_error() {
-    let mut s = multi_story_doc();
-    let before = s.doc.clone();
-    for bad in [json!("everything"), json!(1), json!(true), json!(["document"])] {
-        let e = s.run("review.language", &json!({"lang": "fr-FR", "detect": false, "scope": bad})).unwrap_err();
-        assert!(matches!(e, CmdError::Params(_)), "{e:?}");
-    }
-    assert!(s.doc == before);
-    assert!(s.prefs.detect_language);
-}

crates/engine/src/cmd/user_templates.rs

@@ -337,6 +337,73 @@ mod tests {
         let _ = std::fs::remove_dir_all(&d);
     }
 
+    /// #3433: a template the platform hands the app takes an untouched blank document's place
+    /// (`replaceBlank`), as a file `file.open` opens does; otherwise, and without the flag, the
+    /// new document opens beside the others, as File › New's does. Either way the template on
+    /// disk is only read: the new document has no path, so a save can't write over it.
+    #[test]
+    fn new_from_template_replaces_an_untouched_blank_document_only_when_asked() {
+        let d = scratch("replace-blank");
+        let path = d.join("Letterhead.dotx");
+        template(&path, "Letterhead text", "");
+        let bytes = std::fs::read(&path).unwrap();
+        let from = json!({"path": path.to_string_lossy(), "replaceBlank": true});
+
+        let mut s = Session::new(wordcraft_doc::Document::new());
+        s.run("file.newFromTemplate", &from).unwrap();
+        assert_eq!(s.documents().len(), 1, "the untouched blank document gave way");
+        assert_eq!(s.path, None);
+        assert!(s.dirty);
+        assert_eq!(s.doc.settings.attached_template.as_deref(), Some(path.to_string_lossy().as_ref()));
+        assert!(s.doc.plain_text(wordcraft_doc::StoryRef::Body).contains("Letterhead text"));
+        let saved = s.run("file.save", &json!({})).unwrap();
+        assert_eq!(saved["saved"], json!(false), "no file to save to: Save As asks where");
+        assert_eq!(std::fs::read(&path).unwrap(), bytes, "the template is never written over");
+
+        // The new document has unsaved changes, so a second one opens beside it.
+        s.run("file.newFromTemplate", &from).unwrap();
+        assert_eq!(s.documents().len(), 2);
+        // Without the flag, even an untouched blank document stays (File › New).
+        let mut s = Session::new(wordcraft_doc::Document::new());
+        s.run("file.newFromTemplate", &json!({"path": path.to_string_lossy()})).unwrap();
+        assert_eq!(s.documents().len(), 2);
+        assert_eq!(std::fs::read(&path).unwrap(), bytes);
+        let _ = std::fs::remove_dir_all(&d);
+    }
+
+    /// #3433: a protected template asks for its password like `file.open`, and the document made
+    /// from it isn't protected; a template's bytes can come in `data` (a file name that isn't
+    /// valid Unicode, #1440), its path then only naming it.
+    #[test]
+    #[cfg(not(target_arch = "wasm32"))]
+    fn new_from_template_reads_a_protected_template_and_bytes_like_file_open() {
+        let d = scratch("protected");
+        let path = d.join("Locked.dotx");
+        let mut t = Session::new(wordcraft_doc::Document::from_text("Locked template"));
+        t.run("file.save", &json!({"path": path.to_string_lossy(), "password": "sesame"})).unwrap();
+        let name = path.to_string_lossy();
+
+        let mut s = Session::new(wordcraft_doc::Document::new());
+        let e = s.run("file.newFromTemplate", &json!({"path": name})).unwrap_err().to_string();
+        assert!(crate::io::needs_password(&e), "{e}");
+        let e = s.run("file.newFromTemplate", &json!({"path": name, "password": "wrong"})).unwrap_err().to_string();
+        assert!(crate::io::wrong_password(&e), "{e}");
+        assert_eq!(s.documents().len(), 1, "nothing opened yet");
+        s.run("file.newFromTemplate", &json!({"path": name, "password": "sesame", "replaceBlank": true})).unwrap();
+        assert!(s.doc.plain_text(wordcraft_doc::StoryRef::Body).contains("Locked template"));
+        assert_eq!((s.path.as_ref(), s.password.is_some()), (None, false));
+
+        let plain = d.join("Plain.dotx");
+        template(&plain, "From bytes", "");
+        let data = crate::cmd::insert::base64_encode(&std::fs::read(&plain).unwrap());
+        let mut s = Session::new(wordcraft_doc::Document::new());
+        s.run("file.newFromTemplate", &json!({"path": "Pl\u{FFFD}in.dotx", "data": data, "replaceBlank": true})).unwrap();
+        assert!(s.doc.plain_text(wordcraft_doc::StoryRef::Body).contains("From bytes"));
+        assert_eq!(s.doc.settings.attached_template.as_deref(), Some("Pl\u{FFFD}in.dotx"));
+        assert_eq!(s.path, None);
+        let _ = std::fs::remove_dir_all(&d);
+    }
+
     #[test]
     fn the_templates_folder_option_is_saved_and_checked() {
         let d = scratch("option");

crates/engine/src/tests.rs

@@ -12,6 +12,7 @@ use crate::{Session, cmd};
 
 mod bidi;
 mod documents;
+mod find_apostrophes;
 mod font_sizes;
 mod fonts;
 mod ime_noop;
@@ -4681,10 +4682,8 @@ fn typed_paragraphs_are_capped_for_every_caller() {
     assert!(!body_texts(&s).iter().any(|p| p.contains('y')));
 }
 
-#[test]
-fn hostile_params_never_panic() {
-    let reg = cmd::registry();
-    let junk = [
+fn hostile_junk() -> [serde_json::Value; 8] {
+    [
         json!(null),
         json!({}),
         json!({"text": 5, "size": "x", "path": [], "pos": {"block": 9999, "off": 99999}}),
@@ -4693,11 +4692,23 @@ fn hostile_params_never_panic() {
         json!({"page": 0, "x": 1e30, "y": -1e30, "extend": 1}),
         json!({"width": 1e-9, "height": 1e30, "top": -1e9, "left": 1e9}),
         json!({"section": {"pageW": -1e308, "pageH": 0, "marginTop": 1e30, "header": -5, "columns": {"count": 1000000, "space": -1e300, "widths": [[0, 0]]}}}),
-    ];
+    ]
+}
+
+/// `file.*`, `insert.picture` and `insert.textFromFile`: `hostile_file_params_never_panic` runs
+/// them, held to a folder of its own, and `hostile_params_never_panic` every other command.
+fn reaches_files(id: &str) -> bool {
+    id.starts_with("file.") || id == "insert.picture" || id == "insert.textFromFile"
+}
+
+#[test]
+fn hostile_params_never_panic() {
+    let reg = cmd::registry();
+    let junk = hostile_junk();
     for spec in reg.all() {
-        // These reach outside the session (files). Read Aloud (`review.readAloud`, `readAloud.*`)
-        // is fuzzed too: under `cfg(test)` its backend is the silent `Hold`, asserted below.
-        if spec.id.starts_with("file.") || spec.id == "insert.picture" || spec.id == "insert.textFromFile" {
+        // Read Aloud (`review.readAloud`, `readAloud.*`) is fuzzed too: under `cfg(test)` its
+        // backend is the silent `Hold`, asserted below.
+        if reaches_files(spec.id) {
             continue;
         }
         for j in &junk {
@@ -4711,6 +4722,127 @@ fn hostile_params_never_panic() {
     }
 }
 
+/// `file.*`, `insert.picture` and `insert.textFromFile` take junk params and paths to files they
+/// can't use (#3719). They may read only `in` and write only `out`, so a save can't replace a file
+/// another run is meant to find broken, and nothing outside those folders is touched.
+#[cfg(not(target_arch = "wasm32"))]
+#[test]
+fn hostile_file_params_never_panic() {
+    /// What is directly in `dir`: each name with its bytes (`None` for a folder).
+    fn contents(dir: &std::path::Path) -> std::collections::BTreeMap<String, Option<Vec<u8>>> {
+        let entries = std::fs::read_dir(dir).unwrap().map(|e| e.unwrap().path());
+        entries.map(|p| (p.file_name().unwrap().to_string_lossy().into_owned(), std::fs::read(&p).ok())).collect()
+    }
+    /// The names whose bytes differ, so a failure lists them instead of printing every file.
+    fn changed(before: &std::collections::BTreeMap<String, Option<Vec<u8>>>, dir: &std::path::Path) -> Vec<String> {
+        let after = contents(dir);
+        let names: std::collections::BTreeSet<_> = before.keys().chain(after.keys()).collect();
+        names.into_iter().filter(|n| before.get(*n) != after.get(*n)).cloned().collect()
+    }
+    let reg = cmd::registry();
+    let base = scratch_dir("hostile-files");
+    let (input, output, outside) = (base.join("in"), base.join("out"), base.join("outside"));
+    for dir in [&outside, &input.join("folder.docx"), &output.join("folder.docx")] {
+        std::fs::create_dir_all(dir).unwrap();
+    }
+    let at = |dir: &std::path::Path, name: &str| dir.join(name).to_string_lossy().into_owned();
+
+    // Fixed bytes, so a failure comes back on the next run.
+    let noise: Vec<u8> = (0..512u32).map(|i| (i.wrapping_mul(2_654_435_761) >> 24) as u8).collect();
+    let picture_exts = ["png", "jpg", "gif", "bmp", "svg", "emf"];
+    let mut paths = vec![];
+    for ext in crate::io::OPEN_EXTS.iter().chain(&picture_exts) {
+        let name = format!("random.{ext}");
+        std::fs::write(input.join(&name), &noise).unwrap();
+        paths.push(at(&input, &name));
+    }
+    paths.extend(crate::io::SAVE_EXTS.iter().map(|ext| at(&output, &format!("new.{ext}"))));
+    std::fs::write(input.join("empty"), b"").unwrap();
+    std::fs::write(input.join("empty.docx"), b"").unwrap();
+    std::fs::write(input.join("good.txt"), "Good text").unwrap();
+    std::fs::write(input.join("good.png"), tiny_png()).unwrap();
+    paths.extend(["empty", "empty.docx", "folder.docx", "missing", "missing.docx"].map(|name| at(&input, name)));
+    paths.extend(["folder.docx", "missing/new.docx", &format!("{}.docx", "x".repeat(5000))].map(|name| at(&output, name)));
+    // Outside the roots, however it is spelled, and not paths at all.
+    paths.extend(["new.docx", "new"].map(|name| at(&outside, name)));
+    paths.push(format!("{}/../outside/new.docx", input.to_string_lossy()));
+    paths.extend(["hostile-relative.docx", "hostile-relative", "", "\0", "a\0b.docx"].map(String::from));
+
+    // Every path alone; each of these also with every other kind of junk param. The numbers that
+    // cost time (`copies`, `scale`) stay small: huge ones are #3720.
+    let mut params: Vec<serde_json::Value> = hostile_junk().into();
+    params.extend(paths.iter().map(|p| json!({"path": p})));
+    for name in [
+        "in/random.docx",
+        "in/empty.docx",
+        "in/folder.docx",
+        "in/missing.docx",
+        "in/good.txt",
+        "in/good.png",
+        "out/new.docx",
+        "out/new.pdf",
+        "out/new.png",
+    ] {
+        let path = at(&base, name);
+        params.extend([
+            json!({"path": path, "password": "x", "format": "wordml2003", "markup": "x", "confirmDamaged": 1}),
+            json!({"path": path, "password": 5, "format": 5, "markup": true}),
+            json!({"path": path, "page": 0, "scale": 0, "width": -1e308, "alt": 5}),
+            json!({"path": path, "data": "not base64!"}),
+            json!({"path": path, "data": "AAAA"}),
+            json!({"path": path, "pagesPerSheet": 7, "range": "x", "pages": "9-1,,", "parity": "x", "copies": 3, "paper": "x"}),
+            json!({"dir": path, "name": path, "template": path, "userTemplatesFolder": path, "id": 1e30, "restore": 1e30, "save": path}),
+        ]);
+    }
+
+    let before = contents(&input);
+    let working_dir = std::path::Path::new(".");
+    let working_dir_before = contents(working_dir);
+    let roots = crate::file_access::AutomationRoots::new(Some(&input), Some(&output)).unwrap();
+    let confined = |s: &mut Session, id: &str, v: &serde_json::Value| crate::file_access::confine(roots.as_ref(), || s.run(id, v));
+    for spec in reg.all() {
+        if !reaches_files(spec.id) {
+            continue;
+        }
+        for v in &params {
+            let mut s = Session::new(crate::sample::sample_document());
+            let r = confined(&mut s, spec.id, v);
+            assert!(!caught_panic(&reg, &r), "{} panicked on {v}: {r:?}", spec.id);
+            s.clamp_selection();
+            let _ = s.layout();
+        }
+    }
+
+    let nothing = Vec::<String>::new();
+    assert_eq!(changed(&before, &input), nothing, "the read root is only read");
+    assert_eq!(changed(&Default::default(), &outside), nothing, "nothing outside the write root");
+    // Every name counts: an empty path comes out as `.pdf` or `.dotx` here.
+    assert_eq!(changed(&working_dir_before, working_dir), nothing, "nothing in the working directory");
+
+    // The junk above met the files, not a closed door: usable files work through the same roots...
+    let mut s = s();
+    confined(&mut s, "file.open", &json!({"path": at(&input, "good.txt")})).unwrap();
+    assert_eq!(s.doc.plain_text(StoryRef::Body), "Good text");
+    confined(&mut s, "insert.textFromFile", &json!({"path": at(&input, "good.txt")})).unwrap();
+    confined(&mut s, "insert.picture", &json!({"path": at(&input, "good.png")})).unwrap();
+    let saved: [(&str, &str, &[u8]); 3] =
+        [("file.saveAs", "control.docx", b"PK\x03\x04"), ("file.exportPdf", "control.pdf", b"%PDF-"), ("file.exportPng", "control.png", b"\x89PNG")];
+    for (id, name, signature) in saved {
+        confined(&mut s, id, &json!({"path": at(&output, name)})).unwrap();
+        assert_eq!(std::fs::read(output.join(name)).unwrap().get(..signature.len()), Some(signature), "{name}");
+    }
+    // ...the noise is refused by the reader, and only a path outside by the roots.
+    let refused = |s: &mut Session, path: String| match confined(s, "file.open", &json!({"path": path})) {
+        Err(crate::CmdError::Failed(m)) => m,
+        r => panic!("not refused: {r:?}"),
+    };
+    let unreadable = refused(&mut s, at(&input, "random.docx"));
+    assert!(!unreadable.contains("automation path rejected"), "{unreadable}");
+    let rejected = refused(&mut s, at(&outside, "new.docx"));
+    assert!(rejected.contains("automation path rejected"), "{rejected}");
+    let _ = std::fs::remove_dir_all(&base);
+}
+
 /// A panic that `Session::run`'s guard turned into an error, in a command or one it ran: still a
 /// crash path (#3718). Only the guard's own `"{id} panicked: "` counts, not an error that echoes
 /// those words.
@@ -7812,6 +7944,97 @@ fn colour_commands_take_any_custom_hex() {
     assert_eq!(s.doc.settings.page_color, wordcraft_doc::Rgb::parse("123ABC"));
 }
 
+/// #4685: Home › Paragraph › Shading on text selected inside a paragraph shades that text, as in
+/// Word, so No Color takes off a text shading a file gave it (an orange "highlight" from a .docx is
+/// `w:shd`, which Text Highlight Color › No Color leaves, in Word too). A caret or a whole
+/// paragraph still shades the paragraph.
+#[test]
+fn shading_selected_text_shades_the_text_and_no_color_clears_it() {
+    let orange = wordcraft_doc::Rgb::parse("FFA500");
+    let mut s = s();
+    run(&mut s, "text.insert", json!({"text": "Orange text here"}));
+    run(&mut s, "select.text", json!({"text": "Orange"}));
+    run(&mut s, "para.shading", json!({"color": "FFA500"}));
+    let p = s.doc.para_at(&s.sel.focus).unwrap();
+    assert_eq!((p.props.shading, p.props_of_char(0).shading, p.props_of_char(7).shading), (None, orange, None));
+    // Highlight › No Color doesn't take it off; Shading › No Color does.
+    run(&mut s, "format.highlight", json!({"color": "none"}));
+    assert_eq!(s.doc.para_at(&s.sel.focus).unwrap().props_of_char(0).shading, orange);
+    run(&mut s, "para.shading", json!({"color": null}));
+    assert_eq!(s.doc.para_at(&s.sel.focus).unwrap().props_of_char(0).shading, None);
+    // A caret, or the whole paragraph, shades the paragraph and leaves the text alone.
+    for whole in [false, true] {
+        let mut s = self::s();
+        run(&mut s, "text.insert", json!({"text": "Orange text here"}));
+        run(&mut s, if whole { "select.all" } else { "caret.left" }, json!({}));
+        run(&mut s, "para.shading", json!({"color": "FFA500"}));
+        let p = s.doc.para_at(&s.sel.focus).unwrap();
+        assert_eq!((p.props.shading, p.props_of_char(0).shading), (orange, None), "whole: {whole}");
+    }
+    // All the text of a paragraph that isn't the last is still text: a selection with its mark
+    // would end at the next paragraph's start.
+    let mut s = self::s();
+    run(
+        &mut s,
+        "document.setText",
+        json!({"text": "Word
+More"}),
+    );
+    run(&mut s, "select.text", json!({"text": "Word"}));
+    run(&mut s, "para.shading", json!({"color": "FFA500"}));
+    let p = s.doc.para_at(&s.sel.focus).unwrap();
+    assert_eq!((p.props.shading, p.props_of_char(0).shading), (None, orange));
+    // A selected text box shades its anchor's paragraph, not every line of the box.
+    let mut s = self::s();
+    run(&mut s, "text.insert", json!({"text": "Anchor"}));
+    run(&mut s, "insert.textBox", json!({"text": "Inside"}));
+    let id = s.doc.parts.iter().find(|(_, p)| p.kind == wordcraft_doc::PartKind::TextBox).map(|(id, _)| *id).unwrap();
+    let end = s.doc.text_box_anchor(id).unwrap();
+    let start = wordcraft_doc::Pos { off: end.off - wordcraft_doc::para::OBJ.len_utf8(), ..end.clone() };
+    run(&mut s, "select.range", json!({"anchor": start, "focus": end}));
+    run(&mut s, "para.shading", json!({"color": "FFA500"}));
+    let boxed = s.doc.parts.get(&id).and_then(|p| p.blocks.first()).and_then(|b| b.as_para()).unwrap();
+    assert_eq!(boxed.props_of_char(0).shading, None, "the box's text isn't shaded");
+    assert_eq!(s.doc.para_at(&end).unwrap().props.shading, orange);
+    // A selected field is text (its object character isn't a drawing): No Color clears its shading.
+    let mut s = self::s();
+    run(&mut s, "text.insert", json!({"text": "Page "}));
+    run(&mut s, "insert.pageNumber", json!({"position": "current"}));
+    let end = s.sel.focus.clone();
+    let start = wordcraft_doc::Pos { off: end.off - wordcraft_doc::para::OBJ.len_utf8(), ..end.clone() };
+    run(&mut s, "select.range", json!({"anchor": start, "focus": end}));
+    run(&mut s, "para.shading", json!({"color": "FFA500"}));
+    let p = s.doc.para_at(&end).unwrap();
+    assert_eq!((p.props.shading, p.props_of_char(start.off).shading), (None, orange), "the field is shaded");
+    run(&mut s, "para.shading", json!({"color": null}));
+    assert_eq!(s.doc.para_at(&end).unwrap().props_of_char(start.off).shading, None);
+    // Borders and Shading › Apply to: Paragraph shades the paragraph, whatever is selected.
+    run(&mut s, "select.text", json!({"text": "Page"}));
+    run(&mut s, "para.shading", json!({"color": "FFA500", "paragraph": true}));
+    let p = s.doc.para_at(&s.sel.focus).unwrap();
+    assert_eq!((p.props.shading, p.props_of_char(0).shading), (orange, None));
+}
+
+/// #4685 review: in a cell's last paragraph all of its text stands for the paragraph with its
+/// mark (a selection can't reach the next paragraph's start there), so Shading shades the
+/// paragraph; part of the text is text.
+#[test]
+fn shading_all_of_a_cells_last_paragraph_shades_the_paragraph() {
+    let orange = wordcraft_doc::Rgb::parse("FFA500");
+    for (from, to, paragraph) in [(0, 9, true), (0, 4, false)] {
+        let mut s = s();
+        run(&mut s, "insert.table", json!({"rows": 1, "cols": 1}));
+        run(&mut s, "text.insert", json!({"text": "Cell text"}));
+        let cell = s.sel.focus.clone();
+        assert!(cell.path.cell().is_some());
+        run(&mut s, "select.range", json!({"anchor": Pos { off: from, ..cell.clone() }, "focus": Pos { off: to, ..cell.clone() }}));
+        run(&mut s, "para.shading", json!({"color": "FFA500"}));
+        let p = s.doc.para_at(&cell).unwrap();
+        let want = if paragraph { (orange, None) } else { (None, orange) };
+        assert_eq!((p.props.shading, p.props_of_char(0).shading), want, "{from}..{to}");
+    }
+}
+
 /// #2738: Home › Underline › Underline Color sets the underline's own colour and leaves its style
 /// alone; `auto` drops it again. One undo step each, kept through a DOCX round trip, and junk is
 /// a params error that changes nothing.
@@ -10167,6 +10390,73 @@ fn column_texts(s: &Session) -> Vec<String> {
     t.rows.iter().map(|r| r.cells[0].blocks.iter().filter_map(|b| b.as_para()).map(|p| p.plain_text()).collect::<String>()).collect()
 }
 
+#[test]
+fn table_sort_rejects_invalid_columns_without_changing_rows_or_history() {
+    for column in [
+        json!(-1),
+        json!(0.5),
+        json!(0.0),
+        json!("0"),
+        json!(null),
+        json!(true),
+        json!([]),
+        json!({}),
+        json!(1),
+        json!(u32::MAX as u64 + 1),
+        json!(u64::MAX),
+    ] {
+        let mut s = column_table("en-US", &["Charlie", "Alpha", "Bravo"]);
+        s.close_typing();
+        run(&mut s, "text.insert", json!({"text": "!"}));
+        let before = s.doc.clone();
+        let selection = s.sel.clone();
+        let dirty = s.dirty;
+        assert!(s.run("table.sort", &json!({"column": column, "header": false})).is_err(), "column {column}");
+        assert_eq!(s.doc, before, "column {column}");
+        assert_eq!(s.sel, selection, "column {column}");
+        assert_eq!(s.dirty, dirty, "column {column}");
+        run(&mut s, "edit.undo", json!({}));
+        assert_eq!(column_texts(&s), ["Charlie", "Alpha", "Bravo"], "column {column}: refusal must not consume Undo");
+        run(&mut s, "edit.redo", json!({}));
+        assert_eq!(s.doc, before, "column {column}: Redo must restore the preceding edit");
+    }
+}
+
+#[test]
+fn table_sort_keeps_default_and_valid_columns() {
+    let mut s = s();
+    run(&mut s, "insert.table", json!({"rows": 3, "cols": 2}));
+    for (k, value) in ["Charlie", "1", "Alpha", "3", "Bravo", "2"].into_iter().enumerate() {
+        if k > 0 {
+            run(&mut s, "text.tab", json!({}));
+        }
+        run(&mut s, "text.insert", json!({"text": value}));
+    }
+    let original = s.doc.clone();
+    for params in [json!({"header": false}), json!({"column": 0, "header": false})] {
+        run(&mut s, "table.sort", params);
+        assert_eq!(column_texts(&s), ["Alpha", "Bravo", "Charlie"]);
+        run(&mut s, "edit.undo", json!({}));
+        assert_eq!(s.doc, original);
+    }
+    run(&mut s, "table.sort", json!({"column": 1, "header": false}));
+    assert_eq!(column_texts(&s), ["Charlie", "Bravo", "Alpha"]);
+    run(&mut s, "table.sort", json!({"column": 1, "header": false, "descending": true}));
+    assert_eq!(column_texts(&s), ["Alpha", "Bravo", "Charlie"]);
+    // A column may be missing in some rows. Its present cells still sort normally.
+    let at = s.doc.body.iter().position(|b| b.as_table().is_some()).unwrap() as u32;
+    let table = s.doc.table_mut(StoryRef::Body, &wordcraft_doc::Path(vec![at])).unwrap();
+    table.rows[0].cells.remove(1);
+    run(&mut s, "table.sort", json!({"column": 1, "header": false}));
+    assert_eq!(column_texts(&s), ["Charlie", "Bravo", "Alpha"]);
+
+    let mut s = column_table("en-US", &["Title", "3", "1", "2"]);
+    let at = s.doc.body.iter().position(|b| b.as_table().is_some()).unwrap() as u32;
+    s.doc.table_mut(StoryRef::Body, &wordcraft_doc::Path(vec![at])).unwrap().rows[0].props.header = true;
+    run(&mut s, "table.sort", json!({"column": 0}));
+    assert_eq!(column_texts(&s), ["Title", "1", "2", "3"]);
+}
+
 /// #836: table sort reads numbers in the text's language: `1,5` is one and a half in Polish.
 #[test]
 fn table_sort_reads_decimal_comma_in_polish() {

crates/engine/src/tests/find_apostrophes.rs

@@ -0,0 +1,48 @@
+use super::*;
+
+/// Keyboard smart quotes stay searchable, and replacement keeps their original spelling.
+#[test]
+fn keyboard_apostrophe_finds_smart_quoted_typing() {
+    let mut s = s();
+    for ch in "Tau'va".chars() {
+        run(&mut s, "text.insert", json!({"text": ch.to_string()}));
+    }
+    assert_eq!(text(&s), "Tau’va");
+    for query in ["Tau'", "Tau'v", "Tau'va", "tau'VA"] {
+        assert_eq!(run(&mut s, "edit.find", json!({"text": query}))["count"], 1, "{query}");
+    }
+    assert_eq!(run(&mut s, "edit.find", json!({"text": "tau'va", "matchCase": true}))["count"], 0);
+    assert_eq!(run(&mut s, "edit.replaceAll", json!({"text": "Tau'va", "with": "[^&]", "matchCase": false}))["replaced"], 1);
+    assert_eq!(text(&s), "[Tau’va]");
+    run(&mut s, "edit.undo", json!({}));
+    assert_eq!(text(&s), "Tau’va");
+    run(&mut s, "edit.redo", json!({}));
+    assert_eq!(text(&s), "[Tau’va]");
+}
+
+/// Normal search uses raw UTF-8 ranges while explicit syntax and modifier letters stay exact.
+#[test]
+fn normal_apostrophe_search_keeps_exact_modes_and_original_byte_ranges() {
+    let mut s = s();
+    let original = "🙂 Tau'va Tau’va Tau‘va Tauʼva";
+    run(&mut s, "document.setText", json!({"text": original}));
+    let found = run(&mut s, "edit.find", json!({"text": "Tau'va", "wholeWord": true}));
+    assert_eq!(found["count"], 3);
+    let ranges: Vec<_> =
+        found["matches"].as_array().unwrap().iter().map(|m| (m["start"]["off"].as_u64().unwrap(), m["end"]["off"].as_u64().unwrap())).collect();
+    assert_eq!(ranges, [(5, 11), (12, 20), (21, 29)]);
+    for options in [json!({"special": false}), json!({"wildcards": true}), json!({"regex": true})] {
+        let mut exact = super::s();
+        run(&mut exact, "document.setText", json!({"text": original}));
+        let mut params = options;
+        params["text"] = json!("Tau'va");
+        assert_eq!(run(&mut exact, "edit.find", params)["count"], 1);
+    }
+    for query in ["Tau^u39va", "Tau^u8217va", "Tau’va", "Tau‘va", "Tauʼva"] {
+        assert_eq!(run(&mut s, "edit.find", json!({"text": query}))["count"], 1, "{query}");
+    }
+    assert_eq!(run(&mut s, "edit.replaceAll", json!({"text": "Tau'va", "with": "<^&>"}))["replaced"], 3);
+    assert_eq!(text(&s), "🙂 <Tau'va> <Tau’va> <Tau‘va> Tauʼva");
+    run(&mut s, "edit.undo", json!({}));
+    assert_eq!(text(&s), original);
+}

crates/engine/src/tests/typing.rs

@@ -1539,6 +1539,110 @@ fn typing_at_either_end_of_a_link_is_not_linked() {
     assert_eq!(links(&s), [("sixte".into(), false), (String::new(), false)]);
 }
 
+#[test]
+fn a_link_on_a_selection_shows_the_text_to_display() {
+    // Ctrl+K on selected text with a new Text to display: the selection is retyped with it and
+    // linked, keeping its formatting; typing after it stays outside the link (as in Word).
+    let mut s = typed("go see kiln now");
+    run(&mut s, "select.text", json!({"text": "kiln"}));
+    run(&mut s, "format.bold", json!({}));
+    run(&mut s, "insert.link", json!({"url": "https://example.org/", "text": "Oven"}));
+    typ(&mut s, "!");
+    assert_eq!(texts(&s), ["go see oven! now"]);
+    assert_eq!(links(&s), [("Oven".into(), false)]);
+    let p = s.doc.body.iter().find_map(|b| b.as_para()).unwrap();
+    assert!(p.run_ranges().any(|(r, c)| &p.text[r] == "Oven" && c.bold == Some(true)), "the new text keeps the selection's formatting");
+    // The same text, or text that can't stand in one line, leaves the selection as it is.
+    run(&mut s, "select.text", json!({"text": "see"}));
+    run(&mut s, "insert.link", json!({"url": "https://example.org/", "text": "see"}));
+    run(&mut s, "select.text", json!({"text": "now"}));
+    run(&mut s, "insert.link", json!({"url": "https://example.org/", "text": "two\nlines"}));
+    assert_eq!(texts(&s), ["go see oven! now"]);
+}
+
+#[test]
+fn a_new_selection_drops_the_formatting_left_after_a_link() {
+    // A link leaves plain formatting for what is typed after it. A selection made next doesn't
+    // take it: retyping that selection keeps its own formatting.
+    let bold_run = |s: &Session, text: &str| {
+        let p = s.doc.body.iter().find_map(|b| b.as_para()).unwrap();
+        p.run_ranges().any(|(r, c)| &p.text[r] == text && c.bold == Some(true))
+    };
+    let mut s = typed("kiln");
+    run(&mut s, "select.text", json!({"text": "kiln"}));
+    run(&mut s, "format.bold", json!({}));
+    run(&mut s, "insert.link", json!({"url": "https://example.org/", "text": "Oven"}));
+    run(&mut s, "select.all", json!({}));
+    run(&mut s, "insert.link", json!({"url": "https://example.org/", "text": "Stove"}));
+    assert!(bold_run(&s, "Stove"), "select.all");
+    run(&mut s, "select.word", json!({}));
+    run(&mut s, "insert.link", json!({"url": "https://example.org/", "text": "Range"}));
+    assert!(bold_run(&s, "Range"), "select.word");
+    let anchor = serde_json::to_value(s.doc.start_of(s.sel.focus.story)).unwrap();
+    let focus = serde_json::to_value(s.sel.focus.clone()).unwrap();
+    run(&mut s, "select.range", json!({"anchor": anchor, "focus": focus}));
+    run(&mut s, "insert.link", json!({"url": "https://example.org/", "text": "Hob"}));
+    assert!(bold_run(&s, "Hob"), "select.range");
+}
+
+#[test]
+fn a_link_over_two_paragraphs_keeps_their_text() {
+    // Word shows <<Selection in Document>> for such a selection: the text to display is ignored.
+    let mut s = typed("one\ntwo");
+    run(&mut s, "select.all", json!({}));
+    run(&mut s, "insert.link", json!({"url": "https://example.org/", "text": "both"}));
+    assert_eq!(texts(&s), ["one", "two"]);
+    assert_eq!(links(&s), [("One".into(), false), ("two".into(), false)]);
+}
+
+#[test]
+fn a_retyped_link_text_is_tracked() {
+    let mut s = typed("go see kiln now");
+    run(&mut s, "review.trackChanges", json!({"value": true}));
+    run(&mut s, "select.text", json!({"text": "kiln"}));
+    run(&mut s, "insert.link", json!({"url": "https://example.org/", "text": "Oven"}));
+    let p = s.doc.body.iter().find_map(|b| b.as_para()).unwrap();
+    assert!(p.run_ranges().any(|(r, c)| &p.text[r] == "kiln" && c.del.is_some()), "the old text stays as a deletion");
+    assert!(p.run_ranges().any(|(r, c)| &p.text[r] == "Oven" && c.ins.is_some() && c.link.is_some()));
+}
+
+#[test]
+fn a_link_keeps_a_selection_it_cannot_retype() {
+    use wordcraft_doc::props::CharProps;
+    use wordcraft_doc::{InlineObject, Path};
+    let url = "https://example.org/";
+    // A second retyped link keeps its own selection's formatting: the first one's left no
+    // formatting pending for it.
+    let mut s = typed("go see kiln and vat now");
+    run(&mut s, "select.text", json!({"text": "kiln"}));
+    run(&mut s, "insert.link", json!({"url": url, "text": "Oven"}));
+    run(&mut s, "select.text", json!({"text": "vat"}));
+    run(&mut s, "format.bold", json!({}));
+    run(&mut s, "insert.link", json!({"url": url, "text": "Tub"}));
+    let p = s.doc.body.iter().find_map(|b| b.as_para()).unwrap();
+    assert!(p.run_ranges().any(|(r, c)| &p.text[r] == "Tub" && c.bold == Some(true) && c.link.is_some()), "{:?}", p.text);
+    // A selection holding a manual line break keeps both lines.
+    let mut s = typed("one");
+    run(&mut s, "text.lineBreak", json!({}));
+    typ(&mut s, "two");
+    run(&mut s, "select.all", json!({}));
+    run(&mut s, "insert.link", json!({"url": url, "text": "both"}));
+    assert_eq!(s.doc.plain_text(StoryRef::Body), "one\ntwo");
+    // A selection in a field's result is only linked; the field's next update owns that text.
+    let mut doc = wordcraft_doc::Document::from_text("See ");
+    let c = CharProps::default();
+    let p = doc.para_mut(StoryRef::Body, &Path::top(0)).unwrap();
+    let start = p.len();
+    p.insert_object(start, InlineObject::FieldStart { instr: "ADDIN ZOTERO_ITEM CSL_CITATION {}".into(), locked: false, dirty: false }, &c).unwrap();
+    let n = p.insert_text(p.len(), "(Smith 2020)", &c).unwrap();
+    p.insert_object(start + 3 + n, InlineObject::FieldEnd, &c).unwrap();
+    let mut s = Session::new(doc);
+    run(&mut s, "select.text", json!({"text": "Smith"}));
+    run(&mut s, "insert.link", json!({"url": url, "text": "Jones"}));
+    assert_eq!(s.doc.plain_text(StoryRef::Body), "See (Smith 2020)");
+    assert_eq!(s.doc.field_ranges(StoryRef::Body).len(), 1);
+}
+
 #[test]
 fn typing_then_selecting_then_typing_are_two_undo_steps() {
     // #419: a selection that moves the insertion point ends the typing group, like a caret move.

crates/fonts/src/lib.rs

@@ -21,7 +21,7 @@ pub use fontdb::{
 pub use harfrust::Feature;
 use harfrust::{Direction, ShapeOptions, ShapePlan, ShapePlanKey, Shaper, Tag, UnicodeBuffer};
 pub use kurbo::BezPath;
-pub use script::{fallback_rank, is_emoji, needs_layout_tables, script_fallbacks};
+pub use script::{fallback_rank, is_emoji, joins_cursively, needs_layout_tables, script_fallbacks};
 use skrifa::MetadataProvider;
 use skrifa::instance::Size;
 pub use ui_fallback::{
@@ -185,13 +185,34 @@ fn direction_runs(text: &str) -> Vec<(std::ops::Range<usize>, bool)> {
 /// order: right-to-left runs are shaped right to left, then their clusters put back in text
 /// order (each cluster's glyphs keep the shaper's order); line layout reorders them visually.
 pub fn shape(face: &FontFace, text: &str, features: &[Feature], map: impl Fn(char) -> char) -> Vec<ShapedGlyph> {
+    shape_in(face, text, features, map, ShapeContext::default())
+}
+
+/// The text shown on either side of what is being shaped. Only a letter's form depends on it: the
+/// letters of an Arabic word join across a change of formatting, so each piece has to know what
+/// it continues (#4093). Nothing outside the shaped text is drawn or measured.
+#[derive(Clone, Copy, Debug, Default)]
+pub struct ShapeContext<'a> {
+    pub before: &'a str,
+    pub after: &'a str,
+}
+
+impl<'a> ShapeContext<'a> {
+    /// This context for the part `r` of `text`: what's beyond `text` where `r` reaches its edge.
+    fn at_edges_of(self, text: &str, r: &std::ops::Range<usize>) -> ShapeContext<'a> {
+        ShapeContext { before: if r.start == 0 { self.before } else { "" }, after: if r.end == text.len() { self.after } else { "" } }
+    }
+}
+
+/// [`shape`] of a piece of longer text (see [`ShapeContext`]).
+pub fn shape_in(face: &FontFace, text: &str, features: &[Feature], map: impl Fn(char) -> char, around: ShapeContext<'_>) -> Vec<ShapedGlyph> {
     if !text.chars().any(is_rtl) {
-        return shape_dir(face, text, features, &map, false);
+        return shape_dir(face, text, features, &map, false, around);
     }
     let mut out = Vec::with_capacity(text.len());
     for (r, rtl) in direction_runs(text) {
         let Some(sub) = text.get(r.clone()) else { continue };
-        let mut g = shape_run(face, sub, features, &map, rtl);
+        let mut g = shape_run_in(face, sub, features, &map, rtl, around.at_edges_of(text, &r));
         for x in &mut g {
             x.cluster += r.start;
         }
@@ -205,7 +226,19 @@ pub fn shape(face: &FontFace, text: &str, features: &[Feature], map: impl Fn(cha
 /// mark positioning. Glyphs come back in logical order like [`shape`]: clusters in text order,
 /// each cluster's glyphs in the shaper's (visual) order.
 pub fn shape_run(face: &FontFace, text: &str, features: &[Feature], map: impl Fn(char) -> char, rtl: bool) -> Vec<ShapedGlyph> {
-    let g = shape_dir(face, text, features, &map, rtl);
+    shape_run_in(face, text, features, map, rtl, ShapeContext::default())
+}
+
+/// [`shape_run`] of a piece of longer text (see [`ShapeContext`]).
+pub fn shape_run_in(
+    face: &FontFace,
+    text: &str,
+    features: &[Feature],
+    map: impl Fn(char) -> char,
+    rtl: bool,
+    around: ShapeContext<'_>,
+) -> Vec<ShapedGlyph> {
+    let g = shape_dir(face, text, features, &map, rtl, around);
     if !rtl {
         return g;
     }
@@ -221,7 +254,14 @@ pub fn shape_run(face: &FontFace, text: &str, features: &[Feature], map: impl Fn
     groups.into_iter().flatten().collect()
 }
 
-fn shape_dir(face: &FontFace, text: &str, features: &[Feature], map: &impl Fn(char) -> char, rtl: bool) -> Vec<ShapedGlyph> {
+fn shape_dir(
+    face: &FontFace,
+    text: &str,
+    features: &[Feature],
+    map: &impl Fn(char) -> char,
+    rtl: bool,
+    around: ShapeContext<'_>,
+) -> Vec<ShapedGlyph> {
     let mut out = Vec::with_capacity(text.len());
     let shaped = face.hb().map(|hb| {
         let shaper = face.shaper.shaper(&hb).instance(face.instance.as_ref()).build();
@@ -234,6 +274,10 @@ fn shape_dir(face: &FontFace, text: &str, features: &[Feature], map: &impl Fn(ch
             for (i, c) in sub.char_indices() {
                 buf.add(map(c), (r.start + i) as u32);
             }
+            // After the text: adding to a buffer clears what follows it.
+            let around = around.at_edges_of(text, &r);
+            buf.set_pre_context(around.before);
+            buf.set_post_context(around.after);
             buf.guess_segment_properties();
             buf.set_direction(if rtl { Direction::RightToLeft } else { Direction::LeftToRight });
             let plan = shape_plan(face, &shaper, &buf, features);
@@ -386,7 +430,10 @@ mod tests {
             gb.glyph_infos().iter().zip(gb.glyph_positions()).map(|(i, p)| (i.glyph_id, i.cluster, p.x_advance, p.x_offset, p.y_offset)).collect()
         };
         let kept = |face: &FontFace, text: &str, features: &[Feature], rtl: bool| -> Vec<(u32, u32, i32, i32, i32)> {
-            shape_dir(face, text, features, &|c| c, rtl).iter().map(|g| (g.gid, g.cluster as u32, g.x_advance, g.x_offset, g.y_offset)).collect()
+            shape_dir(face, text, features, &|c| c, rtl, ShapeContext::default())
+                .iter()
+                .map(|g| (g.gid, g.cluster as u32, g.x_advance, g.x_offset, g.y_offset))
+                .collect()
         };
         let db = FontDb::global();
         let faces = [db.face(DEFAULT_FAMILY, "Regular"), db.face(DEFAULT_FAMILY, "Italic"), db.face("Source Sans 3", "Regular")];
@@ -421,7 +468,8 @@ mod tests {
         for n in 0..3 * MAX_SHAPE_PLANS as u32 {
             let features = [feature_value(b"kern", n % 2), feature_value(b"liga", n)];
             let rtl = n % 3 == 0;
-            let kept: Vec<(u32, i32)> = shape_dir(&face, "office", &features, &|c| c, rtl).iter().map(|g| (g.gid, g.x_advance)).collect();
+            let kept: Vec<(u32, i32)> =
+                shape_dir(&face, "office", &features, &|c| c, rtl, ShapeContext::default()).iter().map(|g| (g.gid, g.x_advance)).collect();
             assert_eq!(kept, fresh(&features, rtl), "{n}");
             assert!(kept_shape_plans() <= MAX_SHAPE_PLANS, "{}", kept_shape_plans());
         }

crates/fonts/src/script.rs

@@ -18,6 +18,28 @@ fn real_script(c: char) -> Option<Script> {
     }
 }
 
+/// Whether `c` belongs to a script whose letters join their neighbours (Arabic, Syriac, Mongolian…),
+/// so a letter's form depends on the text on either side of it.
+pub fn joins_cursively(c: char) -> bool {
+    c >= '\u{0600}'
+        && matches!(
+            c.script(),
+            Script::Adlam
+                | Script::Arabic
+                | Script::Chorasmian
+                | Script::Hanifi_Rohingya
+                | Script::Mandaic
+                | Script::Manichaean
+                | Script::Mongolian
+                | Script::Nko
+                | Script::Old_Uyghur
+                | Script::Phags_Pa
+                | Script::Psalter_Pahlavi
+                | Script::Sogdian
+                | Script::Syriac
+        )
+}
+
 /// Maximal single-script byte ranges of `text`. Characters of no particular script join the run
 /// before them (or, at the start, the first run), so the ranges cover `text` exactly.
 pub(crate) fn script_runs(text: &str) -> Vec<std::ops::Range<usize>> {

crates/formats/src/html/entity.rs

@@ -3,6 +3,10 @@
 
 /// Decode an entity body (`amp`, `#38`, `#x26`) to its text.
 pub fn decode_entity(name: &str) -> Option<String> {
+    decode_char(name).map(String::from)
+}
+
+fn decode_char(name: &str) -> Option<char> {
     if let Some(num) = name.strip_prefix('#') {
         let (digits, radix) = match num.strip_prefix(['x', 'X']) {
             Some(h) => (h, 16),
@@ -17,7 +21,7 @@ pub fn decode_entity(name: &str) -> Option<String> {
             0x80..=0x9F => crate::txt::cp1252(v as u8),
             _ => char::from_u32(v).unwrap_or('\u{FFFD}'),
         };
-        return Some(c.to_string());
+        return Some(c);
     }
     let c = match name {
         "amp" | "AMP" => '&',
@@ -229,11 +233,54 @@ pub fn decode_entity(name: &str) -> Option<String> {
         "zwnj" => '\u{200C}',
         _ => return None,
     };
-    Some(c.to_string())
+    Some(c)
 }
 
-/// Replace entity references in `s`.
+/// Legacy names HTML5 also decodes without a `;` (only those [`decode_char`] knows), sorted by bytes.
+const BARE_NAMES: &[&str] = &[
+    "AElig", "AMP", "Aacute", "Acirc", "Agrave", "Aring", "Atilde", "Auml", "COPY", "Ccedil", "Eacute", "Ecirc", "Egrave", "Euml", "GT", "Iacute",
+    "Icirc", "Igrave", "Iuml", "LT", "Ntilde", "Oacute", "Ocirc", "Ograve", "Oslash", "Otilde", "Ouml", "QUOT", "REG", "Uacute", "Ucirc", "Ugrave",
+    "Uuml", "Yacute", "aacute", "acirc", "aelig", "agrave", "amp", "aring", "atilde", "auml", "ccedil", "cent", "copy", "curren", "deg", "divide",
+    "eacute", "ecirc", "egrave", "euml", "frac12", "frac14", "frac34", "gt", "iacute", "icirc", "iexcl", "igrave", "iquest", "iuml", "laquo", "lt",
+    "micro", "middot", "nbsp", "ntilde", "oacute", "ocirc", "ograve", "oslash", "otilde", "ouml", "para", "plusmn", "pound", "quot", "raquo", "reg",
+    "sect", "shy", "sup1", "sup2", "sup3", "szlig", "times", "uacute", "ucirc", "ugrave", "uuml", "yacute", "yen", "yuml",
+];
+
+/// A reference written without its `;` at the start of `after` (the text after `&`): a numeric one,
+/// or the longest legacy name. In attribute values a name followed by `=` or an alphanumeric stays text.
+fn bare_reference(after: &str, attr: bool) -> Option<(char, usize)> {
+    if let Some(num) = after.strip_prefix('#') {
+        let hex = num.starts_with(['x', 'X']);
+        let radix = if hex { 16 } else { 10 };
+        let skip = usize::from(hex);
+        let n = num.get(skip..)?.bytes().take_while(|b| char::from(*b).is_digit(radix)).count();
+        if n == 0 {
+            return None;
+        }
+        let len = 1 + skip + n;
+        return decode_char(after.get(..len)?).map(|c| (c, len));
+    }
+    // Legacy names are 2 to 6 ASCII alphanumerics: try the longest prefix first.
+    let run = after.bytes().take(6).take_while(u8::is_ascii_alphanumeric).count();
+    let len = (2..=run).rev().find(|&l| after.get(..l).is_some_and(|p| BARE_NAMES.binary_search(&p).is_ok()))?;
+    let next = after.as_bytes().get(len);
+    if attr && next.is_some_and(|b| b.is_ascii_alphanumeric() || *b == b'=') {
+        return None;
+    }
+    decode_char(after.get(..len)?).map(|c| (c, len))
+}
+
+/// Replace entity references in text `s`.
 pub fn unescape(s: &str) -> String {
+    unescape_in(s, false)
+}
+
+/// Replace entity references in an attribute value `s`.
+pub fn unescape_attr(s: &str) -> String {
+    unescape_in(s, true)
+}
+
+fn unescape_in(s: &str, attr: bool) -> String {
     if !s.contains('&') {
         return s.to_string();
     }
@@ -248,10 +295,10 @@ pub fn unescape(s: &str) -> String {
             cut -= 1;
         }
         let end = after.get(..cut).and_then(|w| w.find(';'));
-        match end.and_then(|e| decode_entity(&after[..e]).map(|t| (t, e))) {
-            Some((t, e)) => {
-                out.push_str(&t);
-                rest = &after[e + 1..];
+        match end.and_then(|e| decode_char(&after[..e]).map(|c| (c, e + 1))).or_else(|| bare_reference(after, attr)) {
+            Some((c, e)) => {
+                out.push(c);
+                rest = &after[e..];
             }
             None => {
                 out.push('&');
@@ -310,3 +357,14 @@ pub fn decode_as(b: &[u8], charset: Option<&str>) -> String {
         },
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn bare_names_are_sorted_and_known() {
+        assert!(BARE_NAMES.windows(2).all(|w| w[0] < w[1]));
+        assert!(BARE_NAMES.iter().all(|n| (2..=6).contains(&n.len()) && decode_char(n).is_some()));
+    }
+}

crates/formats/src/html/lexer.rs

@@ -1,4 +1,4 @@
-use super::entity::unescape;
+use super::entity::{unescape, unescape_attr};
 
 // ---------------------------------------------------------------------------------------------
 // Tokenizer
@@ -154,7 +154,7 @@ impl Lexer<'_> {
                     Some(&q) if q == b'"' || q == b'\'' => {
                         let v0 = i + 1;
                         let e = rest.get(v0..).and_then(|r| r.find(q as char)).map(|k| v0 + k).unwrap_or(rest.len());
-                        val = unescape(rest.get(v0..e).unwrap_or(""));
+                        val = unescape_attr(rest.get(v0..e).unwrap_or(""));
                         i = e + 1;
                     }
                     _ => {
@@ -162,7 +162,7 @@ impl Lexer<'_> {
                         while b.get(i).is_some_and(|c| !c.is_ascii_whitespace() && *c != b'>') {
                             i += 1;
                         }
-                        val = unescape(rest.get(v0..i).unwrap_or(""));
+                        val = unescape_attr(rest.get(v0..i).unwrap_or(""));
                     }
                 }
             }

crates/formats/src/html/tests.rs

@@ -232,6 +232,21 @@ fn entities() {
     assert_eq!(unescape("θ λ Δ ∇ ∑ √ ≠ ≤ ≥ ♥"), "θ λ Δ ∇ ∑ √ ≠ ≤ ≥ ♥");
 }
 
+/// Browsers decode numeric and legacy named references without their `;` in text,
+/// but keep `&name=` and `&nameX` intact inside attribute values (query strings).
+#[test]
+fn references_without_semicolon() {
+    assert_eq!(unescape("&copy 2024 Acme"), "© 2024 Acme");
+    assert_eq!(unescape("&#169 2024 Acme"), "© 2024 Acme");
+    assert_eq!(unescape("&#xA9 2024 &amp &lt"), "© 2024 & <");
+    assert_eq!(unescape("&bogus &# &#x &"), "&bogus &# &#x &");
+    assert_eq!(unescape("&nbsp;x"), "\u{00A0}x");
+    let f = parse("<p><a href=\"/q?a=1&copy=2&reg2=3&amp;b=&lt\">l</a> &copy 2024</p>");
+    let FBlock::Para(p) = &f.blocks[0] else { panic!() };
+    assert!(p.inlines.iter().any(|i| matches!(i, Inline::Text(t, f) if t == "l" && f.link.as_deref() == Some("/q?a=1&copy=2&reg2=3&b=<"))));
+    assert!(p.text().ends_with("© 2024"));
+}
+
 #[test]
 fn wbr_word_break_opportunity() {
     let f = parse("<p>Super<wbr>cali<wbr>fragilistic</p>");

crates/formats/src/markdown/export.rs

@@ -330,6 +330,18 @@ pub fn export(doc: &Document) -> String {
     s
 }
 
+/// A heading's text on one line, with a trailing run of '#' escaped so a reader does not take it
+/// for the optional closing sequence (`## Item #` reads back as "Item").
+fn heading_text(body: &str) -> String {
+    let mut s = body.replace("\\\n", " ");
+    let t = s.trim_end();
+    let core = t.trim_end_matches('#');
+    if core.len() != t.len() && (core.is_empty() || core.ends_with([' ', '\t'])) {
+        s.insert(core.len(), '\\');
+    }
+    s
+}
+
 pub(super) fn write_blocks(blocks: &[FBlock], out: &mut Vec<String>) {
     let mut i = 0;
     let mut counters = [0u32; 9];
@@ -410,8 +422,8 @@ pub(super) fn write_blocks(blocks: &[FBlock], out: &mut Vec<String>) {
                     None => {
                         prev_list = false;
                         let line = match p.kind {
-                            Kind::Heading(n) => format!("{} {}", "#".repeat(n.clamp(1, 6) as usize), body.replace("\\\n", " ")),
-                            Kind::Title => format!("# {}", body.replace("\\\n", " ")),
+                            Kind::Heading(n) => format!("{} {}", "#".repeat(n.clamp(1, 6) as usize), heading_text(&body)),
+                            Kind::Title => format!("# {}", heading_text(&body)),
                             Kind::Rule => "---".to_string(),
                             Kind::Quote => {
                                 let b = escape_lines(body);

crates/formats/src/markdown/tests.rs

@@ -775,3 +775,25 @@ fn exported_link_destinations_read_back_unchanged() {
     assert_eq!(md("https://example.com/a b"), "[x](<https://example.com/a b>)");
     assert_eq!(md("https://e.com/x\ny"), "[x](https://e.com/xy)");
 }
+
+#[test]
+fn heading_trailing_hash_survives_round_trip() {
+    let heading = |md: &str| {
+        let f = parse(md);
+        let Some(FBlock::Para(p)) = f.blocks.first() else { panic!("no paragraph in {md:?}") };
+        assert_eq!(p.kind, Kind::Heading(2), "{md:?}");
+        p.text()
+    };
+    for (src, text, want) in [
+        ("## Item \\#\n", "Item #", "## Item \\#\n"),
+        ("## Issue \\##\n", "Issue ##", "## Issue \\##\n"),
+        ("## \\#\n", "#", "## \\#\n"),
+        ("## C#\n", "C#", "## C#\n"),
+    ] {
+        assert_eq!(heading(src), text, "{src:?}");
+        let md = export(&import(src));
+        assert_eq!(md, want, "first export of {src:?}");
+        assert_eq!(heading(&md), text, "md was {md:?}");
+        assert_eq!(export(&import(&md)), want, "second export of {src:?}");
+    }
+}

crates/formats/src/model/from_doc.rs

@@ -1122,6 +1122,7 @@ fn flow_table(doc: &Document, t: &Table, depth: usize, tracked: FlowOpts) -> FTa
             width: t.props.width,
             width_pct: t.props.width_pct,
             heights,
+            keep_rows: t.rows.iter().map(|r| r.props.cant_split).collect(),
             fixed: t.props.fixed,
             space: [t.props.space_before, t.props.space_after],
             header_rows: t.rows.iter().take_while(|r| r.props.header).count(),

crates/formats/src/model/to_doc.rs

@@ -844,6 +844,7 @@ impl<'a> Builder<'a> {
                 row.props.height = Some(h.min(1584.0));
                 row.props.height_rule = if exact { wordcraft_doc::props::HeightRule::Exact } else { wordcraft_doc::props::HeightRule::AtLeast };
             }
+            row.props.cant_split = t.extra.as_ref().is_some_and(|x| x.keep_rows.get(ri) == Some(&true));
             let mut g = 0usize;
             for c in r.iter().take(wordcraft_doc::table::MAX_COLS) {
                 let span = (c.colspan.clamp(1, 63) as usize).min(cols.saturating_sub(g).max(1));

crates/formats/src/model/types.rs

@@ -943,6 +943,8 @@ pub struct TableExtra {
     pub width_pct: Option<f32>,
     /// Each row's height (points) and whether it is exact (else at least), by row.
     pub heights: Vec<Option<(f32, bool)>>,
+    /// Whether each row keeps from breaking across pages, by row.
+    pub keep_rows: Vec<bool>,
     /// Its columns keep their widths, never fitted to their content (Word's fixed layout,
     /// `w:tblLayout`): an ODT table's always do.
     pub fixed: bool,

crates/formats/src/odt/body.rs

@@ -79,6 +79,10 @@ pub(super) struct Body<'a> {
     /// drawings were waiting before it.
     loose_open: Option<(usize, usize, usize)>,
     fmt: Vec<Fmt>,
+    /// For each paragraph being read: the size its heading, title or code style gives its text.
+    para_sizes: Vec<(Option<f32>, bool)>,
+    /// That size for the paragraph being finished.
+    para_size: Option<f32>,
     para: Option<Para>,
     /// The text read last ended in a space collapsed from XML whitespace (not a `text:s`).
     xml_space_end: bool,
@@ -128,6 +132,9 @@ pub(super) struct Body<'a> {
     paras_done: usize,
     /// `(width, height, picture bytes, alt text, equation read from an embedded `draw:object`)`.
     frame: Option<Frame>,
+    /// The first picture the open frame offers that WordCraft can't show: an object coming after it
+    /// keeps it as its replacement.
+    frame_unread: Option<Vec<u8>>,
     in_desc: bool,
     /// The master pages' page setups.
     pages: &'a Pages,
@@ -237,6 +244,8 @@ impl<'a> Body<'a> {
             xml_space_end: false,
             spaces_left: MAX_EXTRA_SPACES,
             para_cont: false,
+            para_sizes: Vec::new(),
+            para_size: None,
             lists: Vec::new(),
             lists_saved: Vec::new(),
             item_used: Vec::new(),
@@ -259,6 +268,7 @@ impl<'a> Body<'a> {
             pending_master: None,
             paras_done: 0,
             frame: None,
+            frame_unread: None,
             in_desc: false,
             pages,
             sec: page.clone(),
@@ -952,6 +962,8 @@ impl Body<'_> {
         } else {
             self.para_cont = false;
         }
+        let kind_size = matches!(kind, Kind::Code | Kind::Heading(_) | Kind::Title).then(|| style.and_then(|s| self.styles.named_size(s))).flatten();
+        self.para_sizes.push((kind_size, style.is_some_and(|s| self.styles.states_size(s))));
         self.fmt.push(base);
         // Drawings waiting outside a paragraph come first in it, when it is the text's own (not
         // a text box's, a note's or a cell's).
@@ -970,6 +982,7 @@ impl Body<'_> {
         self.index_marks.clear();
         let Some(mut p) = self.para.take() else { return };
         self.glue_connectors(&mut p);
+        self.para_size = self.para_sizes.pop().and_then(|(size, stated)| size.filter(|_| stated));
         // A paragraph without text has its own text formatting as its mark's: its size is its
         // height, and its drawings' runs have it.
         if let Some(base) = self.fmt.pop()
@@ -1020,13 +1033,18 @@ impl Body<'_> {
             p.kind = Kind::Normal;
         }
         if matches!(p.kind, Kind::Code | Kind::Heading(_) | Kind::Title) {
+            // The style says the size of its text, stated again or not: text of another size keeps
+            // its own.
+            let style_size = self.para_size;
             for i in &mut p.inlines {
                 if let Inline::Text(_, f) = i {
                     if p.kind == Kind::Code {
                         f.code = false;
                         f.font = None;
                     }
-                    f.size = None;
+                    if f.size == style_size {
+                        f.size = None;
+                    }
                 }
             }
         }
@@ -1323,6 +1341,7 @@ impl Body<'_> {
                 let w = get(a, "svg:width").and_then(length);
                 let h = get(a, "svg:height").and_then(length);
                 self.frame = Some((w, h, None, String::new(), None, None, None));
+                self.frame_unread = None;
                 let g = get(a, "draw:style-name").map(|s| self.styles.graphic(s)).unwrap_or_default();
                 self.frame_float = Some(super::frames::place(a, w.unwrap_or(0.0), h.unwrap_or(0.0), &g, self.text_size()));
                 self.member_start(a, w.unwrap_or(0.0), h.unwrap_or(0.0), None);
@@ -1372,19 +1391,23 @@ impl Body<'_> {
         // LibreOffice's tables keep their columns' widths, whatever their content (and however
         // far past the margins): Word's fixed layout.
         let space = get(a, "table:style-name").map(|s| self.styles.table_space(s)).unwrap_or_default();
-        let extra = model::TableExtra { align, width: None, width_pct, heights: Vec::new(), fixed: true, space, header_rows: 0 };
+        let extra =
+            model::TableExtra { align, width: None, width_pct, heights: Vec::new(), keep_rows: Vec::new(), fixed: true, space, header_rows: 0 };
         self.tables.push(TableB { rows: Vec::new(), row: Vec::new(), hcover: 0, widths: Vec::new(), extra });
     }
 
     /// A table row (`table:table-row`) begins: its height, kept as rows are (see the row's end).
     fn start_row(&mut self, a: &Attrs) {
         let height = get(a, "table:style-name").and_then(|s| self.styles.row_height(s));
+        let keep = self.styles.row_keep(get(a, "table:style-name"));
         if let Some(t) = self.tables.last_mut() {
             t.row.clear();
             t.hcover = 0;
             if t.extra.heights.len() < wordcraft_doc::table::MAX_ROWS {
                 t.extra.heights.truncate(t.rows.len());
                 t.extra.heights.push(height);
+                t.extra.keep_rows.truncate(t.rows.len());
+                t.extra.keep_rows.push(keep);
             }
         }
     }
@@ -1429,12 +1452,25 @@ impl Body<'_> {
                     && fr.2.is_none()
                     && fr.4.is_none()
                     && fr.5.is_none()
-                    && fr.6.as_ref().is_none_or(|o| o.replacement.is_none())
                 {
                     let path = href.trim_start_matches("./");
-                    fr.2 = (self.zip)(path);
-                    if let Some(o) = &mut fr.6 {
-                        o.replacement = fr.2.clone().map(std::sync::Arc::new);
+                    // A frame may offer one picture in several formats, in order of preference
+                    // (LibreOffice writes its own `.svm` metafile before a PNG): the first
+                    // WordCraft can read. An object's replacement is the first, kept whatever it
+                    // is, also when the object is still to come.
+                    if let Some(d) = (self.zip)(path) {
+                        let readable = drawing::readable_picture(&d);
+                        match &mut fr.6 {
+                            Some(o) if o.replacement.is_none() => o.replacement = Some(std::sync::Arc::new(d.clone())),
+                            Some(_) => {}
+                            None if !readable => {
+                                self.frame_unread.get_or_insert_with(|| d.clone());
+                            }
+                            None => {}
+                        }
+                        if readable {
+                            fr.2 = Some(d);
+                        }
                     }
                 }
             }
@@ -1478,9 +1514,9 @@ impl Body<'_> {
             // A chart is kept whole too, to be written back as it was while it is unchanged.
             if fr.4.is_none() {
                 fr.6 = super::objects::gather(href, ole, self.entries, self.manifest, &mut *self.zip);
-                // A replacement picture read before the object.
+                // A replacement picture read before the object: the first the frame offered.
                 if let Some(o) = &mut fr.6 {
-                    o.replacement = fr.2.clone().map(std::sync::Arc::new);
+                    o.replacement = self.frame_unread.take().or_else(|| fr.2.clone()).map(std::sync::Arc::new);
                 }
             }
         }
@@ -1526,7 +1562,20 @@ impl Body<'_> {
         match local(q) {
             "span" => {
                 let base = self.cur_fmt();
-                let fmt = get(a, "text:style-name").map_or_else(|| base.clone(), |s| self.styles.span_fmt(s, &base));
+                // A size in percent is of the size it is in: its heading's, where nothing nearer
+                // states one.
+                let sized = Fmt { size: base.size.or(self.para_sizes.last().and_then(|(size, _)| *size)), ..base.clone() };
+                let fmt = get(a, "text:style-name").map_or_else(
+                    || base.clone(),
+                    |s| {
+                        let mut fmt = self.styles.span_fmt(s, &sized);
+                        // That size is only for working out a percent: saying nothing of size, it has none.
+                        if base.size.is_none() && self.styles.span_fmt(s, &base).size.is_none() {
+                            fmt.size = None;
+                        }
+                        fmt
+                    },
+                );
                 self.fmt.push(fmt);
             }
             "a" => {

crates/formats/src/odt/body/drawing.rs

@@ -385,6 +385,11 @@ impl Body<'_> {
     }
 }
 
+/// Whether WordCraft can show these bytes as a picture (see the function below).
+pub(in crate::odt) fn readable_picture(data: &[u8]) -> bool {
+    wordcraft_metafile::is_metafile(data) || model::sniff_image(data).is_some()
+}
+
 /// A picture from a package's bytes: a raster image, or a Windows metafile (WMF, EMF: LibreOffice
 /// keeps clip art and logos so), sized `w` × `h` (points) when the frame says.
 pub(in crate::odt) fn picture(data: Vec<u8>, w: Option<f32>, h: Option<f32>, alt: &str) -> Option<model::Img> {

crates/formats/src/odt/body/tables.rs

@@ -87,7 +87,11 @@ impl TableB {
         let stated = rows.iter().flatten().filter(|c| !c.covered).all(|c| c.borders.is_some());
         let borderless = stated && rows.iter().flatten().all(|c| !c.borders.is_some_and(|b| b.any_visible()));
         let extra = self.extra;
-        let kept = extra.fixed || extra.align.is_some() || extra.width_pct.is_some() || extra.heights.iter().any(Option::is_some);
+        let kept = extra.fixed
+            || extra.align.is_some()
+            || extra.width_pct.is_some()
+            || extra.heights.iter().any(Option::is_some)
+            || extra.keep_rows.contains(&true);
         FTable { rows, widths, borderless, extra: kept.then(|| Box::new(extra)), kept: None, wrap: None }
     }
 }

crates/formats/src/odt/kept.rs

@@ -27,6 +27,7 @@ pub(super) fn defs(xml: &[u8], content: bool) -> Vec<(String, Arc<str>)> {
     let mut path: Vec<String> = Vec::new();
     let mut open: Option<(String, usize, usize)> = None;
     let mut before_1_3 = false;
+    let mut prefixes = std::collections::BTreeSet::new();
     events_at(xml, |ev, at| match ev {
         Ev::Start(q, a) => {
             path.push(q.clone());
@@ -39,11 +40,23 @@ pub(super) fn defs(xml: &[u8], content: bool) -> Vec<(String, Arc<str>)> {
                     if let Some(p) = k.strip_prefix("xmlns:").filter(|p| !p.is_empty() && p.len() <= 64)
                         && out.len() < MAX_DEFS
                     {
+                        prefixes.insert(p.to_string());
                         out.push((format!("ns:{p}"), Arc::from(v.as_str())));
                     }
                 }
                 return;
             }
+            // Those a part's container declares too: what is kept of it leaves the container behind.
+            if depth == 2 {
+                for (k, v) in &a {
+                    if let Some(p) = k.strip_prefix("xmlns:").filter(|p| !p.is_empty() && p.len() <= 64)
+                        && out.len() < MAX_DEFS
+                        && prefixes.insert(p.to_string())
+                    {
+                        out.push((format!("ns:{p}"), Arc::from(v.as_str())));
+                    }
+                }
+            }
             // A reference mark (§6.6.2), which `text:reference-ref` names: WordCraft reads it as a
             // bookmark, and writes it as both.
             if content
@@ -183,20 +196,41 @@ fn space(q: &str, family: &str) -> String {
 pub(super) fn defined(xml: &str, def: &str) -> bool {
     let head = def.split('>').next().unwrap_or("");
     let el = head.trim_start_matches('<').split([' ', '/', '\t', '\n', '\r']).next().unwrap_or("");
-    let name = ["style:name=\"", "draw:name=\""].into_iter().find_map(|a| {
-        let at = head.find(a)? + a.len();
-        let len = head.get(at..)?.find('"')?;
-        head.get(at - a.len()..at + len + 1)
-    });
-    let (Some(name), false) = (name, el.is_empty()) else { return false };
+    let name = ["style:name", "draw:name"].into_iter().find_map(|a| Some((a, raw_attr(head, a)?)));
+    let (Some((attr, name)), false) = (name, el.is_empty()) else { return false };
     let open = format!("<{el} ");
-    xml.match_indices(&open).any(|(i, _)| xml.get(i..).and_then(|t| t.split('>').next()).is_some_and(|tag| tag.contains(name)))
+    xml.match_indices(&open).any(|(i, _)| {
+        xml.get(i..).and_then(|t| t.split('>').next()).is_some_and(|tag| raw_attr(tag, attr).is_some_and(|n| attr_text(n) == attr_text(name)))
+    })
+}
+
+/// An attribute's value as it reads, however XML spells it (`O'Brien`, `O&apos;Brien`).
+pub(super) fn attr_text(raw: &str) -> std::borrow::Cow<'_, str> {
+    quick_xml::escape::unescape(raw).unwrap_or(std::borrow::Cow::Borrowed(raw))
+}
+
+/// Attribute `name`'s value as `tag` (a start tag) writes it, spaced and quoted however XML allows.
+pub(super) fn raw_attr<'a>(tag: &'a str, name: &str) -> Option<&'a str> {
+    let past_space = |at: usize| Some(at + tag.get(at..)?.find(|c: char| !c.is_ascii_whitespace())?);
+    let mut at = tag.find(|c: char| c.is_ascii_whitespace())?;
+    loop {
+        let name_at = past_space(at)?;
+        let eq = name_at + tag.get(name_at..)?.find('=')?;
+        let value_at = past_space(eq + 1)?;
+        let rest = tag.get(value_at..)?;
+        let quote = rest.chars().next().filter(|c| matches!(c, '"' | '\''))?;
+        let len = rest.get(1..)?.find(quote)?;
+        if tag.get(name_at..eq)?.trim_end() == name {
+            return rest.get(1..1 + len);
+        }
+        at = value_at + 1 + len + 1;
+    }
 }
 
 /// The children of `container` (a child of the root) in `xml` (a part WordCraft wrote): each
 /// one's key in `group` (as [`defs`] gives it) and its markup's byte range; and the range of
 /// the container's content (`None` when there is no such container).
-fn children(xml: &str, container: &str, group: &str) -> (Vec<(String, (usize, usize))>, Option<(usize, usize)>) {
+pub(super) fn children(xml: &str, container: &str, group: &str) -> (Vec<(String, (usize, usize))>, Option<(usize, usize)>) {
     let mut out = Vec::new();
     let (mut depth, mut inside) = (0usize, false);
     let mut open: Option<(String, usize)> = None;

crates/formats/src/odt/package.rs

@@ -900,6 +900,11 @@ fn with_kept_defs(
     };
     add(&mut content, "</office:automatic-styles>", auto);
     add(&mut styles, "</office:styles>", common);
+    // A font the file declares is declared as the file does: a reader without the font picks its
+    // substitute by this, and a different one would reflow every page.
+    let faces: Vec<&str> = defs.iter().filter(|(k, _)| k.starts_with("font:")).map(|(_, d)| &**d).collect();
+    with_file_faces(&mut content, &faces);
+    with_file_faces(&mut styles, &faces);
     for (xml, before) in [(&mut content, "<office:automatic-styles"), (&mut styles, "<office:styles")] {
         // No declarations yet: the element they go in, where it belongs (before the styles).
         if !fonts.is_empty()
@@ -913,6 +918,54 @@ fn with_kept_defs(
     (content, styles)
 }
 
+/// `xml` (a part WordCraft wrote) with each font it declares that `faces` declare too (the file's
+/// own `style:font-face` elements) as `faces` has it: under that name, or else of that family (a
+/// face named `F1` of the family `Mystery`, which WordCraft names by its family), then keeping
+/// WordCraft's name for it.
+fn with_file_faces(xml: &mut String, faces: &[&str]) {
+    use super::kept::{attr_text, raw_attr};
+    let mut named = std::collections::HashMap::new();
+    let mut of_family = std::collections::HashMap::new();
+    for face in faces {
+        let Some(head) = face.split('>').next() else { continue };
+        if let Some(name) = raw_attr(head, "style:name") {
+            named.entry(attr_text(name).into_owned()).or_insert(*face);
+        }
+        if let Some(family) = raw_attr(head, "svg:font-family") {
+            let family = attr_text(family);
+            let first = family.split(',').next().unwrap_or("").trim().trim_matches(['\'', '"']);
+            if let Some(font) = model::kept_font(first) {
+                of_family.entry(font).or_insert(*face);
+            }
+        }
+    }
+    // WordCraft's own declarations are empty elements.
+    let (kids, _) = super::kept::children(xml, "office:font-face-decls", "font");
+    let mut out = String::with_capacity(xml.len());
+    let mut done = 0;
+    for (_, (from, to)) in kids {
+        let (Some(before), Some(tag)) = (xml.get(done..from), xml.get(from..to)) else { continue };
+        out.push_str(before);
+        let name = raw_attr(tag, "style:name").map(attr_text);
+        let same = name.as_deref().and_then(|n| named.get(n)).map(|face| face.to_string());
+        let family = || name.as_deref().and_then(|n| of_family.get(n).and_then(|face| face_named(face, n)));
+        let empty = tag.find('>').is_some_and(|end| end + 1 == tag.len());
+        out.push_str(same.or_else(family).filter(|_| empty).as_deref().unwrap_or(tag));
+        done = to;
+    }
+    out.push_str(xml.get(done..).unwrap_or(""));
+    *xml = out;
+}
+
+/// `face` (a `style:font-face` element) with the name `name`.
+fn face_named(face: &str, name: &str) -> Option<String> {
+    let head = face.split('>').next()?;
+    let was = super::kept::raw_attr(head, "style:name")?;
+    // `was` is a piece of `head`, which starts `face`.
+    let at = (was.as_ptr() as usize).checked_sub(head.as_ptr() as usize)?;
+    Some(format!("{}{}{}", face.get(..at)?, x(name), face.get(at + was.len()..)?))
+}
+
 /// An OpenDocument text document's media type, and a template's (ODF 1.3 Part 2 §3.3, Appendix C).
 pub const TEXT_MIME: &str = "application/vnd.oasis.opendocument.text";
 pub const TEMPLATE_MIME: &str = "application/vnd.oasis.opendocument.text-template";

crates/formats/src/odt/styles.rs

@@ -252,6 +252,8 @@ struct StyleInfo {
     /// A table-row style's `style:table-row-properties`: its height (`style:row-height`, exact)
     /// or least height (`style:min-row-height`), points.
     row_height: Option<(f32, bool)>,
+    /// `fo:keep-together` of a table-row style: the row doesn't break across pages.
+    row_keep: Option<bool>,
     /// `style:master-page-name` of a paragraph or table style: its paragraphs (tables) start a
     /// page of that master page. Empty: explicitly none, which stops the lookup up the chain.
     master_page: Option<String>,
@@ -623,6 +625,9 @@ fn style_child(name: &str, q: &str, a: &Attrs, s: &mut StyleInfo) -> bool {
             } else if let Some(h) = get(a, "style:min-row-height").and_then(length).filter(|h| h.is_finite() && *h > 0.0) {
                 s.row_height = Some((h.min(MAX_LEN), false));
             }
+            if let Some(k) = get(a, "fo:keep-together") {
+                s.row_keep = Some(k.trim() == "always");
+            }
         }
         "table-column-properties" => {
             if let Some(w) = get(a, "style:column-width").and_then(length) {
@@ -1043,6 +1048,8 @@ pub(super) struct Styles {
     default_color: Option<wordcraft_doc::props::Rgb>,
     /// `style:default-style` for graphics: what every drawing shape's style falls back on.
     default_graphic: super::frames::Graphic,
+    /// `style:default-style` for table rows: `fo:keep-together`, for a row whose style doesn't say.
+    default_row_keep: Option<bool>,
     /// The named dashes and gradients graphic styles use.
     pub(super) defs: super::strokes::Defs,
     /// Date and time styles (`number:date-style`, `number:time-style`) as Word pictures.
@@ -1055,6 +1062,7 @@ impl Styles {
         let mut cur: Option<(String, StyleInfo)> = None;
         let mut in_default = false;
         let mut in_graphic_default = false;
+        let mut in_row_default = false;
         let mut dates = DateStyles::default();
         let mut list: Option<(String, Vec<ListLevel>)> = None;
         // The list or outline level being read.
@@ -1083,8 +1091,12 @@ impl Styles {
                 "default-style" if q.starts_with("style:") => {
                     in_default = get(&a, "style:family") == Some("paragraph");
                     in_graphic_default = get(&a, "style:family") == Some("graphic");
+                    in_row_default = get(&a, "style:family") == Some("table-row");
                 }
                 "graphic-properties" if in_graphic_default && cur.is_none() => self.default_graphic = super::frames::Graphic::read(&a),
+                "table-row-properties" if in_row_default && cur.is_none() => {
+                    self.default_row_keep = get(&a, "fo:keep-together").map(|k| k.trim() == "always")
+                }
                 "text-properties" if in_default && cur.is_none() => {
                     self.default_font = default_font(&a, &self.faces);
                     if let Some(sz) = get(&a, "fo:font-size").and_then(length).filter(|v| v.is_finite() && *v >= 1.0) {
@@ -1151,7 +1163,7 @@ impl Styles {
             },
             Ev::End(q) => match local(&q) {
                 "automatic-styles" => in_auto = false,
-                "default-style" if q.starts_with("style:") => (in_default, in_graphic_default) = (false, false),
+                "default-style" if q.starts_with("style:") => (in_default, in_graphic_default, in_row_default) = (false, false, false),
                 "style" if q.starts_with("style:") => {
                     if let Some((n, s)) = cur.take()
                         && self.map.len() < 100_000
@@ -1263,6 +1275,24 @@ impl Styles {
     /// The formatting of a span in text style `name` over `base`: the automatic styles' over the
     /// nearest custom character style, which the span then names ([`Fmt::extra`]'s style, by ODF
     /// name until import makes it a document style id); without one, the whole chain's.
+    /// The size of text in paragraph style `name` by its named styles, not its automatic ones,
+    /// which are the paragraph's own: what they say on what the document says.
+    pub(super) fn named_size(&self, name: &str) -> Option<f32> {
+        // The default style's: Standard's counts only for a style based on it, in the chain.
+        let size = self.default_size.filter(|v| v.is_finite() && *v >= 1.0).map(|v| v.min(1638.0));
+        let mut f = Fmt { size, ..Fmt::default() };
+        for (_, s) in self.chain(name).iter().rev().filter(|(_, s)| !s.auto) {
+            s.fmt.apply(&mut f);
+        }
+        f.size
+    }
+
+    /// Whether paragraph style `name`'s named styles state a size. Those that don't leave the size
+    /// to WordCraft's style for its kind, so text of any size isn't known to be the style's.
+    pub(super) fn states_size(&self, name: &str) -> bool {
+        self.chain(name).iter().any(|(_, s)| !s.auto && (s.fmt.size.is_some() || s.fmt.size_pct.is_some()))
+    }
+
     pub(super) fn span_fmt(&self, name: &str, base: &Fmt) -> Fmt {
         let chain = self.chain(name);
         let Some(at) = chain.iter().position(|(n, _)| self.is_named_text(n)) else { return self.text_fmt(name, base) };
@@ -1337,6 +1367,12 @@ impl Styles {
         self.chain(name).iter().find_map(|(_, s)| s.row_height)
     }
 
+    /// Whether a table row doesn't break across pages: its style says (whole chain), else the
+    /// default style.
+    pub(super) fn row_keep(&self, name: Option<&str>) -> bool {
+        name.and_then(|n| self.chain(n).iter().find_map(|(_, s)| s.row_keep)).or(self.default_row_keep).unwrap_or(false)
+    }
+
     /// Background colour of a table-cell style (whole chain).
     /// A table-cell style's background picture: its package path, repeat and position.
     pub(super) fn cell_picture(&self, name: &str) -> Option<(String, String, String)> {
@@ -1462,6 +1498,10 @@ impl Styles {
         };
         let mut fmt = Fmt::default();
         s.fmt.apply(&mut fmt);
+        // A percent is of what's under it: the parent's size, or with none the default style's.
+        if s.fmt.size_pct.is_some() {
+            fmt.size = self.named_size(odt);
+        }
         // A style with no parent inherits the default style, not Standard (the document's
         // defaults here): its size, when Standard's differs.
         if s.parent.is_none() && fmt.size.is_none() && self.doc_size() != self.default_size {

crates/formats/src/odt/tests_more/chars.rs

@@ -284,3 +284,126 @@ fn objects_in_hidden_text_are_kept() {
     assert!(p.text.contains("Hidden"), "{:?}", p.text);
     assert!(p.objects.iter().any(|o| matches!(o, wordcraft_doc::para::InlineObject::BookmarkStart { name } if name == "Secret")), "{:?}", p.objects);
 }
+
+#[test]
+fn a_headings_text_of_another_size_keeps_it() {
+    // Only the size its style gives goes with the style: a smaller word in a heading came back
+    // from a save at the heading's size.
+    let styles = format!(
+        r#"<office:document-styles {NS_T}><office:styles><style:style style:name="Heading_20_1" style:display-name="Heading 1" style:family="paragraph" style:default-outline-level="1"><style:text-properties fo:font-size="16pt"/></style:style></office:styles></office:document-styles>"#
+    );
+    let content = format!(
+        r#"<office:document-content {NS_T}><office:automatic-styles><style:style style:name="T1" style:family="text"><style:text-properties fo:font-size="10pt"/></style:style></office:automatic-styles><office:body><office:text><text:h text:style-name="Heading_20_1" text:outline-level="1">big<text:span text:style-name="T1">small</text:span></text:h></office:text></office:body></office:document-content>"#
+    );
+    let check = |d: &Document, what: &str| {
+        let p = d.body[0].as_para().unwrap();
+        let size = |run: usize| d.styles.resolve_char(p.props.style.as_deref(), &p.runs[run].props).size;
+        assert_eq!((size(0), size(1)), (16.0, 10.0), "{what}");
+    };
+    let d = import(&package(&content, Some(&styles))).unwrap();
+    check(&d, "read");
+    assert_eq!(d.body[0].as_para().unwrap().runs[0].props.size, None, "the style's stays the style's");
+    check(&import(&crate::odt::tests::export_model(&d).unwrap()).unwrap(), "written from the model");
+}
+
+/// In a heading style with no parent, where Standard's size and the default's differ, text stating
+/// a size keeps that size.
+#[test]
+fn a_parentless_headings_text_keeps_its_size() {
+    let styles = format!(
+        r#"<office:document-styles {NS_T}><office:styles><style:default-style style:family="paragraph"><style:text-properties fo:font-size="10pt"/></style:default-style><style:style style:name="Standard" style:family="paragraph" style:class="text"><style:text-properties fo:font-size="14pt"/></style:style><style:style style:name="H" style:family="paragraph" style:default-outline-level="1"/></office:styles></office:document-styles>"#
+    );
+    for size in ["10pt", "14pt"] {
+        let content = format!(
+            r#"<office:document-content {NS_T}><office:automatic-styles><style:style style:name="T1" style:family="text"><style:text-properties fo:font-size="{size}"/></style:style><style:style style:name="T2" style:family="text"><style:text-properties fo:font-weight="bold"/></style:style></office:automatic-styles><office:body><office:text><text:h text:style-name="H" text:outline-level="1"><text:span text:style-name="T1">One</text:span><text:span text:style-name="T2">Two</text:span></text:h></office:text></office:body></office:document-content>"#
+        );
+        let want: f32 = size.trim_end_matches("pt").parse().unwrap();
+        let check = |d: &Document, what: &str| {
+            let p = d.body[0].as_para().unwrap();
+            assert_eq!(p.runs[1].props.size, None, "{size} {what}: a span stating no size has none");
+            assert_eq!(
+                d.styles.resolve_char(p.props.style.as_deref(), &p.runs[0].props).size,
+                want,
+                "{size} {what}: {:?} {:?}",
+                p.props.style,
+                p.runs[0].props.size
+            );
+        };
+        let d = import(&package(&content, Some(&styles))).unwrap();
+        check(&d, "read");
+        check(&import(&crate::odt::tests::export_model(&d).unwrap()).unwrap(), "written from the model");
+    }
+}
+
+/// A percent in a heading style with no parent is of the default style's size, not Standard's.
+#[test]
+fn a_percent_in_a_parentless_heading_is_of_the_default_size() {
+    let styles = format!(
+        r#"<office:document-styles {NS_T}><office:styles><style:default-style style:family="paragraph"><style:text-properties fo:font-size="10pt"/></style:default-style><style:style style:name="Standard" style:family="paragraph" style:class="text"><style:text-properties fo:font-size="14pt"/></style:style><style:style style:name="H" style:family="paragraph" style:default-outline-level="1"/></office:styles></office:document-styles>"#
+    );
+    let content = format!(
+        r#"<office:document-content {NS_T}><office:automatic-styles><style:style style:name="T1" style:family="text"><style:text-properties fo:font-size="50%"/></style:style></office:automatic-styles><office:body><office:text><text:h text:style-name="H" text:outline-level="1"><text:span text:style-name="T1">One</text:span></text:h></office:text></office:body></office:document-content>"#
+    );
+    let d = import(&package(&content, Some(&styles))).unwrap();
+    assert_eq!(d.body[0].as_para().unwrap().runs[0].props.size, Some(5.0));
+}
+
+/// A heading style with no parent and a percent size is that percent of the default's size, so text
+/// stating that size is at the style's size.
+#[test]
+fn a_parentless_heading_style_in_percent_is_of_the_default_size() {
+    let styles = format!(
+        r#"<office:document-styles {NS_T}><office:styles><style:default-style style:family="paragraph"><style:text-properties fo:font-size="10pt"/></style:default-style><style:style style:name="H" style:family="paragraph"><style:text-properties fo:font-size="50%"/></style:style></office:styles></office:document-styles>"#
+    );
+    let content = format!(
+        r#"<office:document-content {NS_T}><office:automatic-styles><style:style style:name="T1" style:family="text"><style:text-properties fo:font-size="5pt"/></style:style></office:automatic-styles><office:body><office:text><text:h text:style-name="H" text:outline-level="1"><text:span text:style-name="T1">One</text:span>Two</text:h></office:text></office:body></office:document-content>"#
+    );
+    let d = import(&package(&content, Some(&styles))).unwrap();
+    let p = d.body[0].as_para().unwrap();
+    for r in &p.runs {
+        assert_eq!(d.styles.resolve_char(p.props.style.as_deref(), &r.props).size, 5.0, "{:?} {:?}", p.props.style, r.props.size);
+    }
+}
+
+/// Text at its heading's size goes with the heading style, stated or not, and a size in percent
+/// is of the heading's size; text of another size keeps it, and so does a paragraph's own.
+#[test]
+fn a_headings_text_sizes_are_measured_against_the_heading() {
+    let styles = format!(
+        r#"<office:document-styles {NS_T}><office:styles><style:style style:name="Heading" style:family="paragraph"><style:text-properties fo:font-size="20pt"/></style:style><style:style style:name="Heading_20_1" style:display-name="Heading 1" style:family="paragraph" style:parent-style-name="Heading" style:default-outline-level="1"><style:text-properties fo:font-size="80%"/></style:style></office:styles></office:document-styles>"#
+    );
+    let span = |name: &str, size: &str| {
+        format!(r#"<style:style style:name="{name}" style:family="text"><style:text-properties fo:font-size="{size}"/></style:style>"#)
+    };
+    let content = format!(
+        r#"<office:document-content {NS_T}><office:automatic-styles>{}{}{}<style:style style:name="P1" style:family="paragraph" style:parent-style-name="Heading_20_1"><style:text-properties fo:font-size="30pt"/></style:style></office:automatic-styles><office:body><office:text><text:h text:style-name="Heading_20_1" text:outline-level="1">plain <text:span text:style-name="Same">same</text:span> <text:span text:style-name="Pct">pct</text:span> <text:span text:style-name="Other">other</text:span></text:h><text:h text:style-name="P1" text:outline-level="1">own</text:h></office:text></office:body></office:document-content>"#,
+        span("Same", "16pt"),
+        span("Pct", "50%"),
+        span("Other", "10pt"),
+    );
+    let mut d = import(&package(&content, Some(&styles))).unwrap();
+    let sizes = |d: &Document| {
+        let p = d.body[0].as_para().unwrap();
+        let of = |word: &str| {
+            let at = p.text.find(word).unwrap();
+            let mut from = 0;
+            let run = p.runs.iter().find(|r| {
+                from += r.len;
+                at < from
+            });
+            d.styles.resolve_char(p.props.style.as_deref(), &run.unwrap().props).size
+        };
+        [of("plain"), of("same"), of("pct"), of("other")]
+    };
+    let want = |t: f32| [t, t, 8.0, 10.0];
+    let plain = |d: &Document| d.body[0].as_para().unwrap().runs.iter().filter(|r| r.props.size.is_some()).count();
+    assert_eq!(sizes(&d), want(16.0));
+    assert_eq!(plain(&d), 2, "only the two of another size state one");
+    // The paragraph's own size stays with its text.
+    let own = d.body[1].as_para().unwrap();
+    assert_eq!(d.styles.resolve_char(own.props.style.as_deref(), &own.runs[0].props).size, 30.0);
+    // A change to the style reaches what goes with it.
+    let id = d.body[0].as_para().unwrap().props.style.clone().unwrap();
+    std::sync::Arc::make_mut(&mut d.styles).get_mut(&id).unwrap().chr.size = Some(24.0);
+    assert_eq!(sizes(&d), want(24.0));
+}

crates/formats/src/odt/tests_more/floats.rs

@@ -367,3 +367,33 @@ fn a_frame_keeps_its_place_in_the_stacking_order() {
     assert_eq!(float(&d).z, 0);
     assert!(crate::odt::tests::entry(&crate::odt::tests::export_model(&d).unwrap(), "content.xml").contains(r#"draw:z-index="0""#));
 }
+
+#[test]
+fn a_frame_offering_several_formats_shows_the_first_it_can_read() {
+    // LibreOffice's letter templates: its own metafile, then a PNG. Taking the first left nothing
+    // to show, and the picture was gone once the paragraph was written from the model.
+    let png = [b"\x89PNG\r\n\x1a\n".as_slice(), &[7u8; 64]].concat();
+    let other = [b"\x89PNG\r\n\x1a\n".as_slice(), &[9u8; 64]].concat();
+    let svm = b"VCLMTF a metafile WordCraft can't draw".to_vec();
+    let frame = |images: &[&str]| {
+        let images: String = images.iter().map(|h| format!(r#"<draw:image xlink:href="{h}"/>"#)).collect();
+        format!(
+            r#"<office:document-content {NS_T} {NS_D}><office:body><office:text><text:p><draw:frame svg:width="2in" svg:height="1in" text:anchor-type="as-char">{images}</draw:frame></text:p></office:text></office:body></office:document-content>"#
+        )
+    };
+    let pictures = |d: &Document| d.body[0].as_para().unwrap().objects.iter().filter(|o| matches!(o, InlineObject::Image { .. })).count();
+    let both = frame(&["Pictures/a.svm", "Pictures/a.png"]);
+    let d = import(&zip_of(&[("content.xml", both.as_bytes()), ("Pictures/a.svm", &svm), ("Pictures/a.png", &png)])).unwrap();
+    assert_eq!(pictures(&d), 1, "the PNG");
+    let back = import(&crate::odt::tests::export_model(&d).unwrap()).unwrap();
+    assert_eq!(pictures(&back), 1, "written from the model");
+    // The first it can read, not the last.
+    let order = frame(&["Pictures/a.png", "Pictures/b.png"]);
+    let d = import(&zip_of(&[("content.xml", order.as_bytes()), ("Pictures/a.png", &png), ("Pictures/b.png", &other)])).unwrap();
+    assert_eq!(pictures(&d), 1);
+    assert_eq!(d.media.values().next().and_then(|m| m.get(8).copied()), Some(7), "the first");
+    // Nothing it can read: no picture, and no failure.
+    let only = frame(&["Pictures/a.svm"]);
+    let d = import(&zip_of(&[("content.xml", only.as_bytes()), ("Pictures/a.svm", &svm)])).unwrap();
+    assert_eq!(pictures(&d), 0);
+}

crates/formats/src/odt/tests_more/flow_breaks.rs

@@ -159,3 +159,75 @@ fn a_table_starting_a_page_starts_it() {
     let c = content_of(&crate::odt::tests::export_model(&d).unwrap());
     assert_eq!(c.matches(r#"fo:break-before="page""#).count(), 2, "{c}");
 }
+
+#[test]
+fn a_row_that_keeps_together_says_so() {
+    // Table › Text Flow: a row not to break across pages. It came back from a save free to break.
+    let styles = r#"<style:style style:name="R" style:family="table-row"><style:table-row-properties fo:keep-together="always"/></style:style><style:style style:name="H" style:family="table-row"><style:table-row-properties style:min-row-height="1in" fo:keep-together="always"/></style:style>"#;
+    let row = |s: &str, t: &str| format!(r#"<table:table-row{s}><table:table-cell><text:p>{t}</text:p></table:table-cell></table:table-row>"#);
+    let table = format!(
+        "<table:table>{}{}{}</table:table>",
+        row(r#" table:style-name="R""#, "kept"),
+        row("", "free"),
+        row(r#" table:style-name="H""#, "kept and tall")
+    );
+    let shape = |d: &wordcraft_doc::Document| {
+        d.body
+            .iter()
+            .find_map(|b| b.as_table())
+            .map(|t| t.rows.iter().map(|r| (r.props.cant_split, r.props.height)).collect::<Vec<_>>())
+            .unwrap_or_default()
+    };
+    let d = doc(styles, &table);
+    assert_eq!(shape(&d), [(true, None), (false, None), (true, Some(72.0))]);
+    let saved = crate::odt::tests::export_model(&d).unwrap();
+    assert!(content_of(&saved).contains(r#"fo:keep-together="always""#));
+    assert_eq!(shape(&import(&saved).unwrap()), [(true, None), (false, None), (true, Some(72.0))]);
+}
+
+/// A row whose style says nothing of it, or a row with no style, follows the default style; one
+/// that says otherwise doesn't.
+#[test]
+fn a_rows_keep_together_defaults_to_the_default_style() {
+    let styles = r#"<office:document-styles xmlns:office="urn:oasis:names:tc:opendocument:xmlns:office:1.0" xmlns:style="urn:oasis:names:tc:opendocument:xmlns:style:1.0" xmlns:fo="urn:oasis:names:tc:opendocument:xmlns:xsl-fo-compatible:1.0"><office:styles><style:default-style style:family="table-row"><style:table-row-properties fo:keep-together="always"/></style:default-style></office:styles></office:document-styles>"#;
+    let row = |s: &str| format!(r#"<table:table-row{s}><table:table-cell><text:p>x</text:p></table:table-cell></table:table-row>"#);
+    let content = format!(
+        r#"<office:document-content {NS_T}><office:automatic-styles><style:style style:name="H" style:family="table-row"><style:table-row-properties style:min-row-height="1in"/></style:style><style:style style:name="F" style:family="table-row"><style:table-row-properties fo:keep-together="auto"/></style:style></office:automatic-styles><office:body><office:text><table:table>{}{}{}</table:table></office:text></office:body></office:document-content>"#,
+        row(""),
+        row(r#" table:style-name="H""#),
+        row(r#" table:style-name="F""#)
+    );
+    let keeps = |d: &wordcraft_doc::Document| {
+        d.body.iter().find_map(|b| b.as_table()).map(|t| t.rows.iter().map(|r| r.props.cant_split).collect::<Vec<_>>()).unwrap_or_default()
+    };
+    let d = import(&package(&content, Some(styles))).unwrap();
+    assert_eq!(keeps(&d), [true, true, false]);
+    assert_eq!(keeps(&import(&crate::odt::tests::export_model(&d).unwrap()).unwrap()), [true, true, false]);
+}
+
+/// A row free to break under a default row style that keeps rows together says so when written from
+/// the model, or a save takes that away: the file's default style stays.
+#[test]
+fn a_row_free_to_break_stays_so_over_the_default_style() {
+    let styles = r#"<office:document-styles xmlns:office="urn:oasis:names:tc:opendocument:xmlns:office:1.0" xmlns:style="urn:oasis:names:tc:opendocument:xmlns:style:1.0" xmlns:fo="urn:oasis:names:tc:opendocument:xmlns:xsl-fo-compatible:1.0"><office:styles><style:default-style style:family="table-row"><style:table-row-properties fo:keep-together="always"/></style:default-style></office:styles></office:document-styles>"#;
+    let row = |s: &str| format!(r#"<table:table-row{s}><table:table-cell><text:p>x</text:p></table:table-cell></table:table-row>"#);
+    let content = format!(
+        r#"<office:document-content {NS_T}><office:automatic-styles><style:style style:name="F" style:family="table-row"><style:table-row-properties fo:keep-together="auto"/></style:style></office:automatic-styles><office:body><office:text><table:table><table:table-column/>{}{}</table:table></office:text></office:body></office:document-content>"#,
+        row(r#" table:style-name="F""#),
+        row("")
+    );
+    let mut d = import(&package(&content, Some(styles))).unwrap();
+    let keeps = |d: &wordcraft_doc::Document| {
+        d.body.iter().find_map(|b| b.as_table()).map(|t| t.rows.iter().map(|r| r.props.cant_split).collect::<Vec<_>>()).unwrap_or_default()
+    };
+    assert_eq!(keeps(&d), [false, true]);
+    if let Some(wordcraft_doc::Block::Table(t)) = d.body.get_mut(0).map(std::sync::Arc::make_mut)
+        && let Some(wordcraft_doc::Block::Para(p)) = t.rows[0].cells[0].blocks.get_mut(0).map(std::sync::Arc::make_mut)
+    {
+        p.text.push('!');
+        if let Some(r) = p.runs.last_mut() {
+            r.len += 1;
+        }
+    }
+    assert_eq!(keeps(&import(&export(&d).unwrap()).unwrap()), [false, true], "after an edit and a save");
+}

crates/formats/src/odt/tests_more/headers.rs

@@ -287,6 +287,97 @@ fn a_watermark_is_a_shape_in_the_header_and_reads_back_as_one() {
     );
 }
 
+/// A watermark added, changed or removed in WordCraft reaches a file's page definitions.
+#[test]
+fn a_watermark_change_to_a_file_is_saved() {
+    // A file's own page definitions were written back whatever happened to its watermark, so one
+    // added, changed or removed in WordCraft was gone from the saved file.
+    let ns = crate::odt::tests::NS_T;
+    let styles = format!(
+        r#"<office:document-styles {ns}><office:automatic-styles><style:page-layout style:name="pm1" style:page-usage="mirrored"><style:page-layout-properties fo:page-width="21cm" fo:page-height="29.7cm" fo:margin-top="2cm" fo:margin-bottom="2cm" fo:margin-left="2cm" fo:margin-right="2cm"/></style:page-layout></office:automatic-styles><office:master-styles><style:master-page style:name="Standard" style:page-layout-name="pm1"/></office:master-styles></office:document-styles>"#
+    );
+    let content =
+        format!("<office:document-content {ns}><office:body><office:text><text:p>x</text:p></office:text></office:body></office:document-content>");
+    let plain = crate::odt::tests::package(&content, Some(&styles));
+    let styles_of = |d: &Document| crate::odt::tests::entry(&export(d).unwrap(), "styles.xml");
+    let shape = "PowerPlusWaterMarkObject";
+
+    // Added.
+    let mut d = import(&plain).unwrap();
+    d.settings.watermark = Some(wordcraft_doc::Watermark { text: "DRAFT".into(), ..Default::default() });
+    let saved = export(&d).unwrap();
+    let s = crate::odt::tests::entry(&saved, "styles.xml");
+    assert!(s.contains(shape) && s.contains("DRAFT"), "added: {s}");
+    assert_eq!(import(&saved).unwrap().settings.watermark.map(|w| w.text).as_deref(), Some("DRAFT"));
+
+    // Saved again with nothing changed, it stays as the file has it.
+    let read = import(&saved).unwrap();
+    let again = styles_of(&read);
+    assert_eq!(again.matches(shape).count(), s.matches(shape).count(), "{again}");
+
+    // Changed.
+    let mut d = import(&saved).unwrap();
+    if let Some(w) = d.settings.watermark.as_mut() {
+        w.text = "COPY".into();
+    }
+    let s = styles_of(&d);
+    assert!(s.contains("COPY") && !s.contains("DRAFT"), "changed: {s}");
+
+    // Removed.
+    let mut d = import(&saved).unwrap();
+    d.settings.watermark = None;
+    let s = styles_of(&d);
+    assert!(!s.contains(shape) && !s.contains("DRAFT"), "removed: {s}");
+
+    // One with no text isn't written, so it's no change to the file's page definitions.
+    let mut d = import(&plain).unwrap();
+    d.settings.watermark = Some(wordcraft_doc::Watermark { text: "  ".into(), ..Default::default() });
+    let s = styles_of(&d);
+    assert!(s.contains(r#"style:page-usage="mirrored""#) && !s.contains(shape), "no text: {s}");
+}
+
+/// Different Odd & Even Pages turned on or off in WordCraft reaches a file's page definitions.
+#[test]
+fn different_odd_and_even_pages_changed_in_a_file_is_saved() {
+    // A file's own page definitions were written back whatever happened to this setting, so the
+    // left pages' header stayed after it was turned off.
+    let ns = crate::odt::tests::NS_T;
+    let file = |left: &str| {
+        let styles = format!(
+            r#"<office:document-styles {ns}><office:automatic-styles><style:page-layout style:name="pm1" style:page-usage="mirrored"><style:page-layout-properties fo:page-width="21cm" fo:page-height="29.7cm" fo:margin-top="2cm" fo:margin-bottom="2cm" fo:margin-left="2cm" fo:margin-right="2cm"/><style:header-style><style:header-footer-properties fo:min-height="1cm"/></style:header-style></style:page-layout></office:automatic-styles><office:master-styles><style:master-page style:name="Standard" style:page-layout-name="pm1"><style:header><text:p>Right head</text:p></style:header>{left}</style:master-page></office:master-styles></office:document-styles>"#
+        );
+        let content = format!(
+            "<office:document-content {ns}><office:body><office:text><text:p>x</text:p></office:text></office:body></office:document-content>"
+        );
+        crate::odt::tests::package(&content, Some(&styles))
+    };
+    let styles_of = |d: &Document| crate::odt::tests::entry(&export(d).unwrap(), "styles.xml");
+
+    // On in the file, and nothing changed: the file's own.
+    let both = file("<style:header-left><text:p>Left head</text:p></style:header-left>");
+    let d = import(&both).unwrap();
+    assert!(d.settings.even_odd_headers);
+    let s = styles_of(&d);
+    assert!(s.contains(r#"style:page-usage="mirrored""#) && s.contains("Left head"), "as read: {s}");
+
+    // Turned off.
+    let mut d = import(&both).unwrap();
+    d.settings.even_odd_headers = false;
+    let saved = export(&d).unwrap();
+    let s = crate::odt::tests::entry(&saved, "styles.xml");
+    assert!(!s.contains("<style:header-left") && s.contains("Right head"), "off: {s}");
+    assert!(!import(&saved).unwrap().settings.even_odd_headers);
+
+    // Off in the file, and turned on: left pages have their own header, an empty one so far.
+    let mut d = import(&file("")).unwrap();
+    assert!(!d.settings.even_odd_headers);
+    d.settings.even_odd_headers = true;
+    let saved = export(&d).unwrap();
+    let s = crate::odt::tests::entry(&saved, "styles.xml");
+    assert!(s.contains("<style:header-left") && s.contains("Right head"), "on: {s}");
+    assert!(import(&saved).unwrap().settings.even_odd_headers);
+}
+
 #[test]
 fn libreoffices_watermark_reads_as_the_watermark() {
     // As LibreOffice's Format › Watermark writes it (observed): turned, half see-through.

crates/formats/src/odt/tests_more/kept.rs

@@ -159,3 +159,161 @@ fn a_file_a_kept_definition_names_is_written_with_it() {
     let m = crate::odt::tests::entry(&saved, "META-INF/manifest.xml");
     assert!(m.contains(r#"manifest:full-path="Pictures/paper.png""#), "{m}");
 }
+
+#[test]
+fn the_files_font_declarations_are_kept() {
+    // A reader without a font picks its substitute by how the file declares it. WordCraft's own
+    // guess at a font's kind replaced the file's, and a document came back from a save with no
+    // edits 13 pages shorter in LibreOffice on a computer without its font.
+    let ns = NS_T;
+    let face = r#"<style:font-face style:name="Odd Roman" svg:font-family="&apos;Odd Roman&apos;" style:font-family-generic="roman"/>"#;
+    let styles = format!(
+        r#"<office:document-styles {ns}><office:font-face-decls>{face}</office:font-face-decls><office:styles><style:default-style style:family="paragraph"><style:text-properties style:font-name="Odd Roman" fo:font-size="12pt"/></style:default-style></office:styles></office:document-styles>"#
+    );
+    let content = format!(
+        r#"<office:document-content {ns}><office:font-face-decls>{face}</office:font-face-decls><office:automatic-styles><style:style style:name="P1" style:family="paragraph"><style:text-properties style:font-name="Odd Roman"/></style:style></office:automatic-styles><office:body><office:text><text:p text:style-name="P1">Text</text:p></office:text></office:body></office:document-content>"#
+    );
+    let check = |saved: &[u8], what: &str| {
+        for part in [styles_of(saved), content_of(saved)] {
+            assert!(part.contains(face), "{what}: {part}");
+            assert_eq!(part.matches(r#"<style:font-face style:name="Odd Roman""#).count(), 1, "{what}: declared once: {part}");
+        }
+    };
+    let mut d = import(&package(&content, Some(&styles))).unwrap();
+    check(&export(&d).unwrap(), "no edits");
+    // Text edited, written from the model: still the file's declaration.
+    if let Some(wordcraft_doc::Block::Para(p)) = d.body.get_mut(0).map(std::sync::Arc::make_mut) {
+        p.text.push('!');
+        if let Some(r) = p.runs.last_mut() {
+            r.len += 1;
+        }
+    }
+    check(&export(&d).unwrap(), "an edit");
+}
+
+/// A font's name spelled another way in XML (`O'Brien`, `O&#39;Brien`) is the same font: its
+/// declaration is the file's, once.
+#[test]
+fn a_fonts_name_spelled_another_way_is_declared_once_as_the_file_has_it() {
+    for spelled in ["O'Brien", "O&#39;Brien"] {
+        let face = format!(r#"<style:font-face style:name="{spelled}" svg:font-family="O&apos;Brien" style:font-family-generic="roman"/>"#);
+        let styles = format!(
+            r#"<office:document-styles {NS_T}><office:font-face-decls>{face}</office:font-face-decls><office:styles><style:default-style style:family="paragraph"><style:text-properties style:font-name="{spelled}" fo:font-size="12pt"/></style:default-style></office:styles></office:document-styles>"#
+        );
+        let content = format!(
+            r#"<office:document-content {NS_T}><office:font-face-decls>{face}</office:font-face-decls><office:automatic-styles><style:style style:name="T1" style:family="text"><style:text-properties style:font-name="{spelled}"/></style:style></office:automatic-styles><office:body><office:text><text:p><text:span text:style-name="T1">One</text:span></text:p></office:text></office:body></office:document-content>"#
+        );
+        let mut d = import(&package(&content, Some(&styles))).unwrap();
+        // Text edited, written from the model.
+        if let Some(wordcraft_doc::Block::Para(p)) = d.body.get_mut(0).map(std::sync::Arc::make_mut) {
+            p.text.push('!');
+            if let Some(r) = p.runs.last_mut() {
+                r.len += 1;
+            }
+        }
+        let saved = export(&d).unwrap();
+        for part in [styles_of(&saved), content_of(&saved)] {
+            assert!(part.contains(&face), "{spelled}: the file's: {part}");
+            let is_it = |t: &&str| {
+                let name = t.split('>').next().and_then(|tag| tag.split("style:name=\"").nth(1)).and_then(|v| v.split('"').next());
+                name.is_some_and(|n| n == "O'Brien" || n == "O&#39;Brien" || n == "O&apos;Brien")
+            };
+            assert_eq!(part.split("<style:font-face ").skip(1).filter(is_it).count(), 1, "{spelled}: declared once: {part}");
+        }
+    }
+}
+
+/// A face named other than its family (`F1`, `Mystery`): text written from the model names the
+/// family, and that declaration says what the file's does, not WordCraft's guess at the font's kind.
+#[test]
+fn a_face_named_other_than_its_family_gives_the_family_the_files_declaration() {
+    let face =
+        r#"<style:font-face style:name="F1" svg:font-family="&apos;Mystery&apos;" style:font-family-generic="roman" style:font-pitch="variable"/>"#;
+    let styles = format!(
+        r#"<office:document-styles {NS_T}><office:font-face-decls>{face}</office:font-face-decls><office:styles><style:default-style style:family="paragraph"><style:text-properties fo:font-size="12pt"/></style:default-style></office:styles></office:document-styles>"#
+    );
+    let content = format!(
+        r#"<office:document-content {NS_T}><office:font-face-decls>{face}</office:font-face-decls><office:automatic-styles><style:style style:name="T1" style:family="text"><style:text-properties style:font-name="F1"/></style:style></office:automatic-styles><office:body><office:text><text:p><text:span text:style-name="T1">One</text:span></text:p></office:text></office:body></office:document-content>"#
+    );
+    let mut d = import(&package(&content, Some(&styles))).unwrap();
+    if let Some(wordcraft_doc::Block::Para(p)) = d.body.get_mut(0).map(std::sync::Arc::make_mut) {
+        p.text.push('!');
+        if let Some(r) = p.runs.last_mut() {
+            r.len += 1;
+        }
+    }
+    let saved = export(&d).unwrap();
+    let part = content_of(&saved);
+    assert!(part.contains(r#"style:font-name="Mystery""#), "{part}");
+    let mystery = part.split("<style:font-face ").find(|t| t.starts_with(r#"style:name="Mystery""#)).and_then(|t| t.split("/>").next());
+    let mystery = mystery.unwrap_or_else(|| panic!("declared: {part}"));
+    assert!(mystery.contains(r#"style:font-family-generic="roman""#), "{mystery}");
+}
+
+/// Only declarations are declarations: not a script's text that reads like one. And a face's name
+/// is its own attribute, not another with `style:name` in its name.
+#[test]
+fn a_face_is_found_and_named_by_its_own_markup() {
+    let script =
+        r#"<office:scripts><office:script script:language="x"><![CDATA[<style:font-face style:name="Mystery" />]]></office:script></office:scripts>"#;
+    let face = r#"<style:font-face xstyle:name="F1" style:name="F1" svg:font-family="&apos;Mystery&apos;" style:font-family-generic="roman"/>"#;
+    let styles = format!(
+        r#"<office:document-styles {NS_T}><office:font-face-decls>{face}</office:font-face-decls><office:styles><style:default-style style:family="paragraph"><style:text-properties fo:font-size="12pt"/></style:default-style></office:styles></office:document-styles>"#
+    );
+    let content = format!(
+        r#"<office:document-content {NS_T}>{script}<office:font-face-decls>{face}</office:font-face-decls><office:automatic-styles><style:style style:name="T1" style:family="text"><style:text-properties style:font-name="F1"/></style:style></office:automatic-styles><office:body><office:text><text:p><text:span text:style-name="T1">One</text:span></text:p></office:text></office:body></office:document-content>"#
+    );
+    let mut d = import(&package(&content, Some(&styles))).unwrap();
+    if let Some(wordcraft_doc::Block::Para(p)) = d.body.get_mut(0).map(std::sync::Arc::make_mut) {
+        p.text.push('!');
+        if let Some(r) = p.runs.last_mut() {
+            r.len += 1;
+        }
+    }
+    let saved = export(&d).unwrap();
+    let part = content_of(&saved);
+    assert!(part.contains(script), "the script as it was: {part}");
+    let decls = part.split("<office:font-face-decls>").last().and_then(|t| t.split("</office:font-face-decls>").next()).unwrap();
+    assert!(decls.contains(r#"xstyle:name="F1" style:name="Mystery""#), "{decls}");
+    assert!(import(&saved).is_ok());
+}
+
+/// A font declaration in single quotes is the file's too: it replaces WordCraft's, and isn't
+/// declared a second time.
+#[test]
+fn a_font_declaration_in_single_quotes_is_kept() {
+    let ns = format!(r#"{NS_T} xmlns:svg="urn:oasis:names:tc:opendocument:xmlns:svg-compatible:1.0""#);
+    let face = r#"<style:font-face style:name='Odd Roman' svg:font-family='Odd Roman' style:font-family-generic='roman'/>"#;
+    let styles = format!(
+        r#"<office:document-styles {ns}><office:font-face-decls>{face}</office:font-face-decls><office:styles><style:default-style style:family="paragraph"><style:text-properties style:font-name="Odd Roman" fo:font-size="12pt"/></style:default-style></office:styles></office:document-styles>"#
+    );
+    let content = format!(
+        r#"<office:document-content {ns}><office:font-face-decls>{face}</office:font-face-decls><office:automatic-styles><style:style style:name="P1" style:family="paragraph"><style:text-properties style:font-name="Odd Roman"/></style:style></office:automatic-styles><office:body><office:text><text:p text:style-name="P1">x</text:p></office:text></office:body></office:document-content>"#
+    );
+    let saved = export(&import(&package(&content, Some(&styles))).unwrap()).unwrap();
+    for part in [styles_of(&saved), content_of(&saved)] {
+        assert!(part.contains(face), "{part}");
+        assert_eq!(part.matches("<style:font-face").count(), part.matches("<style:font-face style:name=").count());
+        assert_eq!(part.matches(r#"style:name="Odd Roman""#).count(), 0, "declared once: {part}");
+        assert_eq!(part.matches("style:name='Odd Roman'").count(), 1, "declared once: {part}");
+    }
+}
+
+/// A name space only the declarations' container states goes with the declaration: without it the
+/// saved part isn't well-formed in its name spaces.
+#[test]
+fn a_name_space_on_the_font_declarations_goes_with_them() {
+    let ns = format!(r#"{NS_T} xmlns:svg="urn:oasis:names:tc:opendocument:xmlns:svg-compatible:1.0""#);
+    let decls = r#"<office:font-face-decls xmlns:x="urn:example:x"><style:font-face style:name="Odd Roman" svg:font-family="Odd Roman" x:odd="1"/></office:font-face-decls>"#;
+    let styles = format!(
+        r#"<office:document-styles {ns}>{decls}<office:styles><style:default-style style:family="paragraph"><style:text-properties style:font-name="Odd Roman" fo:font-size="12pt"/></style:default-style></office:styles></office:document-styles>"#
+    );
+    let content = format!(
+        r#"<office:document-content {ns}>{decls}<office:automatic-styles><style:style style:name="P1" style:family="paragraph"><style:text-properties style:font-name="Odd Roman"/></style:style></office:automatic-styles><office:body><office:text><text:p text:style-name="P1">x</text:p></office:text></office:body></office:document-content>"#
+    );
+    let saved = export(&import(&package(&content, Some(&styles))).unwrap()).unwrap();
+    for part in [styles_of(&saved), content_of(&saved)] {
+        assert!(part.contains(r#"x:odd="1""#), "{part}");
+        assert!(part.contains(r#"xmlns:x="urn:example:x""#), "{part}");
+    }
+}

crates/formats/src/odt/tests_more/objects.rs

@@ -167,3 +167,84 @@ fn an_object_in_a_header_is_kept_too() {
     assert!(z.file_names().any(|n| n.ends_with("/content.xml") && n.starts_with("Object M")), "{:?}", z.file_names().collect::<Vec<_>>());
     assert!(entry(&saved, "styles.xml").contains("<draw:object xlink:href=\"./Object M"));
 }
+
+/// A replacement picture may come before its object. It is kept whatever it is, though WordCraft
+/// can't show it, and a save from the model writes it back.
+#[test]
+fn a_replacement_picture_before_its_object_is_kept() {
+    let content = format!(
+        r#"<office:document-content {NS_T} {NS_D}><office:body><office:text><text:p>Sheet <draw:frame draw:name="Object1" text:anchor-type="as-char" svg:width="3in" svg:height="1in"><draw:image xlink:href="./ObjectReplacements/Object 1" xlink:type="simple"/><draw:object xlink:href="./Object 1" xlink:type="simple"/></draw:frame></text:p></office:text></office:body></office:document-content>"#
+    );
+    let bytes = zip_of(&[
+        ("META-INF/manifest.xml", MANIFEST.as_bytes()),
+        ("content.xml", content.as_bytes()),
+        ("Object 1/content.xml", SHEET.as_bytes()),
+        ("Object 1/styles.xml", b"<office:document-styles/>"),
+        ("ObjectReplacements/Object 1", METAFILE),
+    ]);
+    let replacement = |bytes: &[u8]| {
+        let flow = parse(bytes).unwrap();
+        paras(&flow)
+            .into_iter()
+            .flat_map(|p| &p.inlines)
+            .find_map(|i| if let Inline::Image(img) = i { img.object.clone() } else { None })
+            .and_then(|o| o.replacement.clone())
+    };
+    assert_eq!(replacement(&bytes).as_deref().map(Vec::as_slice), Some(METAFILE));
+    let saved = crate::odt::tests::export_model(&import(&bytes).unwrap()).unwrap();
+    assert_eq!(replacement(&saved).as_deref().map(Vec::as_slice), Some(METAFILE), "written from the model");
+}
+
+/// A readable picture after an unreadable replacement is what WordCraft shows, wherever the object
+/// comes between them, and a save from the model keeps both.
+#[test]
+fn a_readable_picture_after_an_unreadable_replacement_is_shown() {
+    let mut png = std::io::Cursor::new(Vec::new());
+    image::RgbImage::new(4, 4).write_to(&mut png, image::ImageFormat::Png).unwrap();
+    let png = png.into_inner();
+    let object = r#"<draw:object xlink:href="./Object 1"/>"#;
+    let unread = r#"<draw:image xlink:href="./ObjectReplacements/Object 1"/>"#;
+    let readable = r#"<draw:image xlink:href="./Pictures/a.png"/>"#;
+    for frame in [format!("{unread}{object}{readable}"), format!("{object}{unread}{readable}")] {
+        let content = format!(
+            r#"<office:document-content {NS_T} {NS_D}><office:body><office:text><text:p>Sheet <draw:frame draw:name="Object1" text:anchor-type="as-char" svg:width="3in" svg:height="1in">{frame}</draw:frame></text:p></office:text></office:body></office:document-content>"#
+        );
+        let bytes = zip_of(&[
+            ("META-INF/manifest.xml", MANIFEST.as_bytes()),
+            ("content.xml", content.as_bytes()),
+            ("Object 1/content.xml", SHEET.as_bytes()),
+            ("Object 1/styles.xml", b"<office:document-styles/>"),
+            ("ObjectReplacements/Object 1", METAFILE),
+            ("Pictures/a.png", &png),
+        ]);
+        let check = |bytes: &[u8], what: &str| {
+            let flow = parse(bytes).unwrap();
+            let img = paras(&flow)
+                .into_iter()
+                .flat_map(|p| &p.inlines)
+                .find_map(|i| if let Inline::Image(img) = i { Some(img.clone()) } else { None })
+                .unwrap();
+            assert_eq!(img.data.as_slice(), png.as_slice(), "{what}: shows the picture it can read");
+            assert_eq!(img.object.unwrap().replacement.as_deref().map(Vec::as_slice), Some(METAFILE), "{what}: keeps the replacement");
+        };
+        check(&bytes, "read");
+        check(&crate::odt::tests::export_model(&import(&bytes).unwrap()).unwrap(), "written from the model");
+    }
+}
+
+/// With only an unreadable replacement, the placeholder drawn for it isn't written as a picture.
+#[test]
+fn the_placeholder_for_an_object_is_not_saved_as_its_picture() {
+    let content = format!(
+        r#"<office:document-content {NS_T} {NS_D}><office:body><office:text><text:p>Sheet <draw:frame draw:name="Object1" text:anchor-type="as-char" svg:width="3in" svg:height="1in"><draw:object xlink:href="./Object 1"/><draw:image xlink:href="./ObjectReplacements/Object 1"/></draw:frame></text:p></office:text></office:body></office:document-content>"#
+    );
+    let bytes = zip_of(&[
+        ("META-INF/manifest.xml", MANIFEST.as_bytes()),
+        ("content.xml", content.as_bytes()),
+        ("Object 1/content.xml", SHEET.as_bytes()),
+        ("Object 1/styles.xml", b"<office:document-styles/>"),
+        ("ObjectReplacements/Object 1", METAFILE),
+    ]);
+    let saved = crate::odt::tests::export_model(&import(&bytes).unwrap()).unwrap();
+    assert_eq!(entry(&saved, "content.xml").matches("<draw:image ").count(), 1);
+}

crates/formats/src/odt/writer/drawing.rs

@@ -244,11 +244,19 @@ impl Writer {
             None => self.pictures.push((replacement.clone(), img.data.clone(), mime_of(&img.ext).to_string())),
         }
         let el = if o.ole { "draw:object-ole" } else { "draw:object" };
-        format!(
+        let mut inner = format!(
             "<{el} xlink:href=\"./{}\" xlink:type=\"simple\" xlink:show=\"embed\" xlink:actuate=\"onLoad\"/><draw:image xlink:href=\"./{}\" xlink:type=\"simple\" xlink:show=\"embed\" xlink:actuate=\"onLoad\"/>",
             x(&name),
             x(&replacement)
-        )
+        );
+        // A picture shown that isn't the replacement (one WordCraft can't read) follows it; a
+        // placeholder is WordCraft's own, not the file's.
+        let other = o.replacement.as_ref().is_some_and(|r| r.as_slice() != img.data.as_slice());
+        if other && crate::odt::objects::placeholder(&o.mime).as_deref() != Some(img.data.as_slice()) {
+            let href = self.picture(img);
+            inner.push_str(&format!("<draw:image xlink:href=\"{href}\" xlink:type=\"simple\" xlink:show=\"embed\" xlink:actuate=\"onLoad\"/>"));
+        }
+        inner
     }
 
     /// Object `o` (a chart as read) in the package under the next object name, with its

crates/formats/src/odt/writer/tables.rs

@@ -161,19 +161,33 @@ impl Writer {
                 out.push_str("<table:table-header-rows>");
             }
             // A row of a set height: exact (`style:row-height`) or at least (`style:min-row-height`).
-            match t.extra.as_ref().and_then(|x| x.heights.get(ri).copied().flatten()) {
-                Some((h, exact)) => {
-                    let props = if exact { format!(" style:row-height=\"{}\"", pt(h)) } else { format!(" style:min-row-height=\"{}\"", pt(h)) };
-                    let k = match self.row_styles.iter().position(|s| *s == props) {
-                        Some(k) => k,
-                        None => {
-                            self.row_styles.push(props);
-                            self.row_styles.len() - 1
-                        }
-                    };
-                    out.push_str(&format!("<table:table-row table:style-name=\"{}Row{}\">", self.prefix, k + 1));
-                }
-                None => out.push_str("<table:table-row>"),
+            let height = match t.extra.as_ref().and_then(|x| x.heights.get(ri).copied().flatten()) {
+                Some((h, true)) => format!(" style:row-height=\"{}\"", pt(h)),
+                Some((h, false)) => format!(" style:min-row-height=\"{}\"", pt(h)),
+                None => String::new(),
+            };
+            // A row that doesn't break across pages (§20.247 on a row). One that may says so where
+            // the file's default row style, which stays, would keep it together.
+            let keep = if t.extra.as_ref().is_some_and(|x| x.keep_rows.get(ri) == Some(&true)) {
+                " fo:keep-together=\"always\""
+            } else if self.kept_defs.get("common~style:default-style~table-row").is_some_and(|d| d.contains("keep-together") && d.contains("always"))
+            {
+                " fo:keep-together=\"auto\""
+            } else {
+                ""
+            };
+            let props = format!("{height}{keep}");
+            if props.is_empty() {
+                out.push_str("<table:table-row>");
+            } else {
+                let k = match self.row_styles.iter().position(|s| *s == props) {
+                    Some(k) => k,
+                    None => {
+                        self.row_styles.push(props);
+                        self.row_styles.len() - 1
+                    }
+                };
+                out.push_str(&format!("<table:table-row table:style-name=\"{}Row{}\">", self.prefix, k + 1));
             }
             let mut g = 0usize;
             for c in row {

crates/formats/src/rtf/parse.rs

@@ -105,22 +105,27 @@ fn parse_full(bytes: &[u8]) -> Result<Parsed, String> {
                         i = end;
                         continue;
                     }
-                    if word == "info" {
-                        r.info_depth = Some(r.depth());
-                    }
                     if r.skip_chars > 0
-                        && matches!(r.st.dest, Dest::Body | Dest::Info(_) | Dest::FldInst | Dest::Bookmark | Dest::FontTbl)
-                        && !matches!(word, "par" | "pard" | "cell" | "row")
+                        && matches!(r.st.dest, Dest::Body | Dest::Info(_) | Dest::FldInst | Dest::Bookmark | Dest::FontTbl | Dest::RevTbl)
                     {
                         r.skip_chars -= 1;
                         first_in_group = false;
                         continue;
                     }
+                    if word == "info" {
+                        r.info_depth = Some(r.depth());
+                    }
                     r.word(word, param, starred, first_in_group);
                     first_in_group = false;
                     starred = false;
                     continue;
                 }
+                if r.skip_chars > 0 && n != b'\'' {
+                    r.skip_chars -= 1;
+                    i += 2;
+                    first_in_group = false;
+                    continue;
+                }
                 match n {
                     b'\'' => {
                         let h = bytes.get(i + 2..i + 4).and_then(|h| std::str::from_utf8(h).ok()).and_then(|h| u8::from_str_radix(h, 16).ok());

crates/formats/src/rtf/tests.rs

@@ -37,6 +37,62 @@ fn basic_reader() {
     assert!(matches!(&l.inlines[0], Inline::Text(_, f) if f.link.as_deref() == Some("http://a.b")));
 }
 
+#[test]
+fn unicode_fallback_uses_the_declared_count() {
+    for (rtf, expected) in [
+        (br"{\rtf1\uc18\u8804less than or equal rest}".as_slice(), "≤ rest"),
+        (br"{\rtf1\uc10\u233\'30\'31\'32\'33\'34\'35\'36\'37\'38\'39 rest}", "é rest"),
+        (br"{\rtf1\uc2\u8804\:ab rest}", "≤b rest"),
+        (br"{\rtf1\uc18\u8804\:12345678901234567 rest}", "≤ rest"),
+        (br"{\rtf1\uc1\u8804\:ab rest}", "≤ab rest"),
+        (br"{\rtf1\uc0\u8804\:ab rest}", "≤ab rest"),
+        (br"{\rtf1\uc9\u233abcdefgh\par VISIBLE}", "éVISIBLE"),
+        (br"{\rtf1\uc1\u233\pard VISIBLE}", "éVISIBLE"),
+        (br"{\rtf1\uc1\u233\cell VISIBLE}", "éVISIBLE"),
+        (br"{\rtf1\uc1\u233\row VISIBLE}", "éVISIBLE"),
+        (br"{\rtf1\uc0\u233\par VISIBLE}", "é\nVISIBLE"),
+        (b"{\\rtf1\\uc9\\u233abcdefgh\\\nVISIBLE}", "éVISIBLE"),
+        (b"{\\rtf1\\uc9\\u233abcdefgh\\\rVISIBLE}", "éVISIBLE"),
+        (br"{\rtf1\uc9\u233abcdefgh\*VISIBLE}", "éVISIBLE"),
+        (b"{\\rtf1\\uc0\\u233\\\nVISIBLE}", "é\nVISIBLE"),
+        (b"{\\rtf1\\uc0\\u233\\\rVISIBLE}", "é\nVISIBLE"),
+        (br"{\rtf1\uc0\u233\*VISIBLE}", "éVISIBLE"),
+        (br"{\rtf1\uc0\u233X}", "éX"),
+        (br"{\rtf1\uc1\u233?X}", "éX"),
+        (br"{\rtf1{\uc2147483647\u233?}X}", "éX"),
+        (br"{\rtf1\uc-1\u233X}", "éX"),
+    ] {
+        let doc = import(rtf).unwrap();
+        assert_eq!(doc.plain_text(wordcraft_doc::StoryRef::Body).trim(), expected, "{rtf:?}");
+    }
+}
+
+#[test]
+fn unicode_fallback_info_words_preserve_following_metadata() {
+    for (rtf, title, body) in [
+        (br"{\rtf1{\info{\title\uc9\u233abcdefgh\info X}{\author Alice}} Text}".as_slice(), "éX", "Text"),
+        (br"{\rtf1{\info{\title\uc1\u233\info X}{\author Alice}} Text}", "éX", "Text"),
+        (br"{\rtf1{\info{\title\uc0\u233X}{\author Alice}} Text}", "éX", "Text"),
+        (br"{\rtf1\uc0\u233{\info{\title Plain}{\author Alice}}Text}", "Plain", "éText"),
+    ] {
+        let doc = import(rtf).unwrap();
+        assert_eq!(doc.core.title, title, "{rtf:?}");
+        assert_eq!(doc.core.creator, "Alice", "{rtf:?}");
+        assert_eq!(doc.plain_text(wordcraft_doc::StoryRef::Body).trim(), body, "{rtf:?}");
+    }
+}
+
+#[test]
+fn unicode_fallback_in_revision_authors_counts_control_words() {
+    for (rtf, expected) in [
+        (br"{\rtf1{\*\revtbl{Unknown;}{\uc9\u233abcdefgh\par Zed;}}{\revised\revauth1 text}}".as_slice(), "éZed"),
+        (br"{\rtf1{\*\revtbl{Unknown;}{\uc0\u233\par Zed;}}{\revised\revauth1 text}}", "éZed"),
+    ] {
+        let doc = import(rtf).unwrap();
+        assert_eq!(doc.revisions.first().map(|r| r.author.as_str()), Some(expected), "{rtf:?}");
+    }
+}
+
 #[test]
 fn paragraph_borders() {
     // A heading with a grey rule under it, as hh.ru résumés write their section titles.

crates/formats/src/rtf/word.rs

@@ -89,7 +89,7 @@ impl Reader {
                 }
                 self.skip_chars = self.st.uc;
             }
-            "uc" => self.st.uc = pv.clamp(0, 8) as usize,
+            "uc" => self.st.uc = pv.max(0) as usize,
             "plain" => {
                 self.st.fmt = Fmt { link: self.st.fmt.link.clone(), ..Default::default() };
                 self.st.rev = RevSt::default();

crates/formats/src/sheet.rs

@@ -427,6 +427,27 @@ mod tests {
 
     const RELS: &str = r#"<Relationships xmlns="http://schemas.openxmlformats.org/package/2006/relationships"><Relationship Id="rId1" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/officeDocument" Target="xl/workbook.xml"/></Relationships>"#;
 
+    #[test]
+    fn workbook_sheets_find_their_relationships_by_index() {
+        // Every sheet searched every relationship; 4,096 sheets over 20,000 relationships.
+        let sheets: String = (0..MAX_SHEETS).map(|i| format!(r#"<sheet name="S{i}" sheetId="{i}" r:id="rId{i}"/>"#)).collect();
+        let wb = format!(
+            r#"<workbook xmlns="http://schemas.openxmlformats.org/spreadsheetml/2006/main" xmlns:r="http://schemas.openxmlformats.org/officeDocument/2006/relationships"><sheets>{sheets}</sheets></workbook>"#
+        );
+        let mut rels = String::from(r#"<Relationships xmlns="http://schemas.openxmlformats.org/package/2006/relationships">"#);
+        // A chart sheet sharing a worksheet's id doesn't hide it.
+        rels.push_str(r#"<Relationship Id="rId0" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/chartsheet" Target="chartsheets/c.xml"/>"#);
+        for i in 0..20_000 {
+            rels.push_str(&format!(
+                r#"<Relationship Id="rId{i}" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/worksheet" Target="worksheets/sheet{i}.xml"/>"#
+            ));
+        }
+        rels.push_str("</Relationships>");
+        let names = sheet_names(&zip(&[("_rels/.rels", RELS), ("xl/workbook.xml", &wb), ("xl/_rels/workbook.xml.rels", &rels)])).unwrap();
+        assert_eq!(names.len(), MAX_SHEETS);
+        assert_eq!(names.first().map(String::as_str), Some("S0"));
+    }
+
     fn xlsx(sheet1: &str) -> Vec<u8> {
         let wb = r#"<workbook xmlns="http://schemas.openxmlformats.org/spreadsheetml/2006/main" xmlns:r="http://schemas.openxmlformats.org/officeDocument/2006/relationships"><workbookPr/><sheets><sheet name="People" sheetId="1" r:id="rId1"/><sheet name="Chart" sheetId="2" r:id="rId9"/><sheet name="Notes" sheetId="3" r:id="rId2"/></sheets></workbook>"#;
         let wb_rels = r#"<Relationships xmlns="http://schemas.openxmlformats.org/package/2006/relationships"><Relationship Id="rId1" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/worksheet" Target="worksheets/sheet1.xml"/><Relationship Id="rId2" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/worksheet" Target="/xl/worksheets/sheet2.xml"/><Relationship Id="rId9" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/chartsheet" Target="chartsheets/sheet1.xml"/><Relationship Id="rId3" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/sharedStrings" Target="sharedStrings.xml"/><Relationship Id="rId4" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/styles" Target="styles.xml"/></Relationships>"#;

crates/formats/src/sheet/xlsx.rs

@@ -83,14 +83,13 @@ fn workbook(pkg: &mut Package<'_>) -> Result<Workbook, String> {
             }
         }
     });
-    let sheets = listed
-        .into_iter()
-        .filter_map(|(name, id)| {
-            // Worksheets only: chart and dialog sheets hold no cells.
-            let r = rels.iter().find(|r| r.id == id && r.kind.ends_with("/worksheet"))?;
-            Some(SheetRef { name, path: r.target.clone() })
-        })
-        .collect();
+    // Worksheets only: chart and dialog sheets hold no cells. Indexed by id once (the first
+    // relationship with an id wins), not searched for every sheet.
+    let mut worksheets: HashMap<&str, &str> = HashMap::new();
+    for r in rels.iter().filter(|r| r.kind.ends_with("/worksheet")) {
+        worksheets.entry(r.id.as_str()).or_insert(r.target.as_str());
+    }
+    let sheets = listed.into_iter().filter_map(|(name, id)| Some(SheetRef { name, path: worksheets.get(id.as_str())?.to_string() })).collect();
     let part = |suffix: &str| rels.iter().find(|r| r.kind.ends_with(suffix)).map(|r| r.target.clone());
     Ok(Workbook { sheets, date1904, shared: part("/sharedStrings"), styles: part("/styles") })
 }

crates/layout/src/ctx.rs

@@ -59,9 +59,9 @@ pub(crate) struct Ctx<'a> {
     pub(crate) pin: Option<Pin>,
     pub(crate) footer_wraps: HashMap<(usize, u32, bool), Vec<WrapArea>>,
     pub(crate) header_tables: HashMap<(usize, u32, bool), Vec<Rect>>,
-    /// Heights of text boxes that resize to fit their text, by (story, width), worked out once
-    /// per layout ([`textbox::fitted_height`]).
-    pub(crate) fitted: HashMap<(u32, u32), Option<f32>>,
+    /// Heights of text boxes that resize to fit their text, by story, width and page fields,
+    /// worked out once per matching layout context ([`textbox::fitted_height`]).
+    pub(crate) fitted: HashMap<(u32, u32, (u32, u32, u32, u32, u64, wordcraft_doc::section::NumFormat)), Option<f32>>,
     /// Each body float's latest placement: on which page and with what wrap area, when it is
     /// above its anchor paragraph with text before it ([`EarlyFloats`]).
     pub(crate) floats_above: HashMap<(usize, usize), Option<(usize, WrapArea)>>,
@@ -344,7 +344,7 @@ impl Ctx<'_> {
 
     /// The height of text box `id`, `w` wide, when it resizes to fit its text (once per layout).
     fn fitted_height(&mut self, id: u32, w: f32) -> Option<f32> {
-        let key = (id, w.to_bits());
+        let key = (id, w.to_bits(), self.fields.page_key());
         if let Some(h) = self.fitted.get(&key) {
             return *h;
         }
@@ -455,3 +455,85 @@ pub(crate) fn with_pct_base<T>(ctx: &mut Ctx, base: Option<f32>, f: impl FnOnce(
     ctx.pct_base = before;
     out
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use wordcraft_doc::para::{Float, ShapeKind, Wrap};
+    use wordcraft_doc::{PartKind, TextBody, para_block};
+
+    #[test]
+    fn fitted_text_box_cache_separates_page_field_contexts() {
+        let mut doc = Document::from_text("Parent");
+        let mut story_para = Paragraph::new();
+        story_para
+            .insert_object(
+                0,
+                InlineObject::Field { instr: "PAGE".into(), result: "1".into(), locked: false, dirty: false, code: false, form: String::new() },
+                &Default::default(),
+            )
+            .unwrap();
+        let id = doc.add_part(PartKind::TextBox, vec![para_block(story_para)]);
+        if let Some(part) = doc.parts.get_mut(&id) {
+            part.body = TextBody { auto_fit: true, ..Default::default() };
+        }
+        let mut p = Paragraph::new();
+        p.insert_object(
+            0,
+            InlineObject::Shape {
+                kind: ShapeKind::TextBox,
+                w: 100.0,
+                h: 20.0,
+                fill: None,
+                stroke: None,
+                stroke_width: 0.0,
+                float: Float { wrap: Wrap::Inline, ..Default::default() },
+                story: Some(id),
+                freeform: None,
+                effects: Default::default(),
+                extra: Default::default(),
+            },
+            &Default::default(),
+        )
+        .unwrap();
+        let opts = LayoutOptions::default();
+        let mut cache = LayoutCache::new();
+        let mut ctx = Ctx {
+            doc: &doc,
+            cache: &mut cache,
+            opts: &opts,
+            counters: Default::default(),
+            fields: FieldCtx::default(),
+            notes_hash: 0,
+            numbers: Default::default(),
+            eq_count: 0,
+            boxes: Default::default(),
+            chains: Default::default(),
+            grid: Default::default(),
+            hide_deleted: opts.hides_deleted(StoryRef::Body),
+            uses_totals: false,
+            track_totals: false,
+            pct_base: None,
+            quiet_marks: Vec::new(),
+            stacked_marks: Vec::new(),
+            mirror: false,
+            table_keys: Vec::new(),
+            pin: None,
+            footer_wraps: Default::default(),
+            header_tables: Default::default(),
+            fitted: Default::default(),
+            floats_above: Default::default(),
+            hf: Arc::new(HfLinks::new(doc.sections())),
+            alphas: Default::default(),
+            header_repeats: 0,
+        };
+
+        ctx.fitted_boxes(&p);
+        assert_eq!(ctx.fitted.len(), 1);
+        ctx.fields.page_format = wordcraft_doc::section::NumFormat::CardinalText;
+        ctx.fitted_boxes(&p);
+        assert_eq!(ctx.fitted.len(), 2, "the same box with a different PAGE format needs its own height");
+        ctx.fitted_boxes(&p);
+        assert_eq!(ctx.fitted.len(), 2, "matching field context reuses the fitted height");
+    }
+}

crates/layout/src/display/lines.rs

@@ -656,7 +656,7 @@ fn inline_objects(ctx: &LineContext<'_, '_, '_>, out: &mut Vec<Draw>) {
             Some(o @ InlineObject::Field { .. }) => {
                 if let Some(wordcraft_doc::legacy_form::LegacyForm::CheckBox { checked, .. }) = o.legacy_form() {
                     let color = pl.styles.get(c.style as usize).map_or(Rgb::BLACK, |st| text_color(&st.rc.color, None));
-                    let frame = Rect::new(cx, base - c.obj_h, c.adv, c.obj_h + c.obj_d);
+                    let frame = Rect::new(cx, base - c.obj_h.max(0.0), c.adv, c.obj_h + c.obj_d);
                     check_box(frame, checked, color, alpha, out);
                 }
             }

crates/layout/src/display/objects.rs

@@ -491,7 +491,8 @@ pub(crate) fn effects_in(effects: wordcraft_doc::effects::ShapeEffects, spin: Sp
 pub(crate) fn inline_rect(obj: Option<&InlineObject>, cx: f32, base: f32, adv: f32, (obj_h, obj_d): (f32, f32), line: f32) -> Rect {
     let frame = obj.and_then(InlineObject::frame);
     let [l, t, r, b] = frame.map_or([0.0; 4], |(_, _, float)| float.effect_extent());
-    let room = Rect::new(cx + l, base - obj_h + t, (adv - l - r).max(0.0), (obj_h + obj_d - t - b).max(0.0));
+    // Lowered further than its height, it is drawn with its top on the baseline (Word).
+    let room = Rect::new(cx + l, base - obj_h.max(0.0) + t, (adv - l - r).max(0.0), (obj_h + obj_d - t - b).max(0.0));
     // A horizontal rule is as wide as its share of the paragraph's width, whatever narrows its
     // line, and is placed by its alignment in what the line has left: it runs past the margin
     // beside a float or a first-line indent, as in Word (see `ParaLayout::rule_room`).

crates/layout/src/floats.rs

@@ -463,6 +463,8 @@ fn anchor_floats_but(
     late: &[usize],
 ) -> (HashMap<usize, Rect>, Arc<ParaLayout>, Vec<para::Exclusion>) {
     let FloatAnchor { p, block, y0, before, excl_mark, label, place } = anchor;
+    ctx.fields.page = pb.number;
+    ctx.fields.page_format = pb.page_format;
     let (col_x, width) = (pb.col_x(), pb.col_w());
     let mut rects = HashMap::new();
     if !pb.web && place {

crates/layout/src/headers.rs

@@ -204,6 +204,7 @@ pub(crate) fn header_table_areas(ctx: &mut Ctx, pb: &PageBuilder) -> Vec<Rect> {
     let fields = ctx.fields.clone();
     let grid = std::mem::take(&mut ctx.grid);
     let track = std::mem::replace(&mut ctx.track_totals, false);
+    ctx.fields.page_format = sect.page_num_format;
     ctx.fields.page = page.number;
     let mut areas = again(ctx, 0, |ctx| {
         with_pct_base(ctx, base, |ctx| layout_box(ctx, BoxInput::other(StoryRef::Part(id), &blocks, sect.text_width(), 0, Some(frame)))).table_areas

crates/layout/src/page.rs

@@ -9,6 +9,8 @@ use super::{Block, Border, FloatAlign, Page, ParaLayout, Rect, SectionProps, Sto
 pub(crate) struct PageBuilder<'a> {
     pub(crate) pages: Vec<Page>,
     pub(crate) sect: &'a SectionProps,
+    /// The current document section's numbering format, even when reflow uses first-section page geometry.
+    pub(crate) page_format: wordcraft_doc::section::NumFormat,
     pub(crate) sect_idx: usize,
     pub(crate) col: usize,
     pub(crate) cols: Vec<(f32, f32)>,

crates/layout/src/para/builder.rs

@@ -6,6 +6,7 @@ use super::chars::{
 use super::grid::east_asian;
 use super::model::{ClKind, Cluster, Enclosed, Glyph, ParaEnv, RubyLayout, StyleRun, SymbolRun};
 use super::{Arc, Document, FaceRef, NumeralMode, ResolvedChar, RubyAlign, SOFT_HYPHEN};
+use wordcraft_fonts::ShapeContext;
 
 pub(super) struct Builder<'a> {
     pub(super) env: &'a ParaEnv<'a>,
@@ -226,6 +227,19 @@ impl<'a> Builder<'a> {
 
     /// `dir`: the bidi levels `text` is shaped by (see [`Dir`]).
     pub(super) fn shape_with(&mut self, text: &str, base: usize, rc: &Arc<ResolvedChar>, kind_override: Option<ClKind>, dir: Dir<'_>) {
+        self.shape_around(text, base, rc, kind_override, dir, ShapeContext::default());
+    }
+
+    /// `around`: the text shown beside `text`, which its first and last letters may join.
+    pub(super) fn shape_around(
+        &mut self,
+        text: &str,
+        base: usize,
+        rc: &Arc<ResolvedChar>,
+        kind_override: Option<ClKind>,
+        dir: Dir<'_>,
+        around: ShapeContext<'_>,
+    ) {
         if text.is_empty() {
             return;
         }
@@ -374,10 +388,15 @@ impl<'a> Builder<'a> {
             let from = usize::from(rc.kerned());
             let to = if ligatures.is_some() { all.len() } else { 1 };
             let features: &[wordcraft_fonts::Feature] = all.get(from..to).unwrap_or_default();
+            // Letters join across a change of font or direction inside the run too.
+            let around = ShapeContext {
+                before: if a == 0 { around.before } else { text.get(..a).unwrap_or("") },
+                after: if b == text.len() { around.after } else { text.get(b..).unwrap_or("") },
+            };
             let shaped = if use_levels {
-                wordcraft_fonts::shape_run(&st.face, shape_text, features, map, key.rtl)
+                wordcraft_fonts::shape_run_in(&st.face, shape_text, features, map, key.rtl, around)
             } else {
-                wordcraft_fonts::shape(&st.face, shape_text, features, map)
+                wordcraft_fonts::shape_in(&st.face, shape_text, features, map, around)
             };
             let k = st.size / st.face.upem.max(1.0) as f32;
             let hscale = rc.scale / 100.0;

crates/layout/src/para/lines.rs

@@ -297,6 +297,9 @@ impl<'e> Breaker<'e> {
         };
         let end = lab.x + lab.width;
         match self.level.map(|l| l.suffix).unwrap_or(LevelSuffix::Tab) {
+            // An empty label (a "none" level with no text) that ends right at the left indent takes
+            // no tab: the text starts there, as in Word (#4985), and not at the next stop.
+            LevelSuffix::Tab if lab.width <= 0.01 && (end - self.rp.indent_left).abs() <= 0.01 => end,
             LevelSuffix::Tab => {
                 let t = next_tab(end, &self.rp.tabs, self.default_tab, self.hanging_at, self.tab_origin).pos;
                 // The level's own tab stop after the number, when it comes first.
@@ -438,8 +441,10 @@ impl<'e> Breaker<'e> {
                     if matches!(rp.line_spacing, LineSpacing::Exactly(_)) { (st.ascent, st.descent) } else { (c.obj_h, c.obj_d) }
                 } else {
                     ex.obj_asc = ex.obj_asc.max(c.obj_h);
-                    // A lowered picture reaches below the baseline.
-                    ex.obj_desc = ex.obj_desc.max(c.obj_d);
+                    // A lowered picture reaches below the baseline, with exact spacing no further
+                    // than its height (the line's box is fixed; its baseline stays where it was).
+                    let reach = if matches!(rp.line_spacing, LineSpacing::Exactly(_)) { c.obj_d.min(c.obj_h + c.obj_d) } else { c.obj_d };
+                    ex.obj_desc = ex.obj_desc.max(reach);
                     (st.ascent, st.descent)
                 }
             } else {

crates/layout/src/para/paragraph.rs

@@ -6,6 +6,8 @@ use super::justify::hyphenation_points;
 use super::lines::break_lines;
 use super::model::{ClKind, Cluster, Enclosed, Glyph, Label, LabelRun, ParaEnv, RubyLayout, undrawable_bullet};
 use super::proof::{ProofRun, is_list_item, proof_issues};
+use wordcraft_fonts::ShapeContext;
+
 use super::{
     Arc, COLUMN_BREAK, CharProps, Document, FIELD_SHADING, InlineObject, LINE_BREAK, NoteKind, OBJ, PAGE_BREAK, PTabAlign, PTabBase, ParaLayout,
     Paragraph, ResolvedChar, ResolvedPara, TabLeader, field_code_text, field_text_on, reference_link,
@@ -33,6 +35,8 @@ struct Cx<'a> {
     math_props: wordcraft_doc::math::MathProps,
     /// The first character, if it is to be a drop cap.
     drop_first: Option<char>,
+    /// The text on the page, when its letters can join across runs (see [`shown_text`]).
+    shown: Option<String>,
 }
 
 /// Where a cluster for an object or a break goes in the paragraph: its text, style and first glyph.
@@ -83,6 +87,10 @@ struct Shaped<'a> {
     bare: Vec<std::ops::Range<usize>>,
     /// Each run's proofing language and whether it is checked (spelling, grammar, hyphenation).
     runs: Vec<ProofRun>,
+    /// How far into the shown text ([`Cx::shown`]) the runs so far have come.
+    shown_at: usize,
+    /// Where in the shown text a letter's context may begin: past the drop cap, once one is drawn.
+    shown_floor: usize,
 }
 
 fn is_special(c: char) -> bool {
@@ -186,6 +194,8 @@ impl<'a> Shaped<'a> {
             left: Vec::new(),
             bare: Vec::new(),
             runs: Vec::new(),
+            shown_at: 0,
+            shown_floor: 0,
         }
     }
 
@@ -236,24 +246,44 @@ impl<'a> Shaped<'a> {
 
     /// A run shown: its text, with the breaks and objects in it as clusters of their own.
     fn visible_run(&mut self, cx: &Cx, (rc, props): (&Arc<ResolvedChar>, &CharProps), range: &std::ops::Range<usize>, text: &str) {
+        let shown_start = self.shown_at;
+        let around = self.next_shown(cx, text.len());
         let mut seg = self.drop_cap_letter(cx, rc, text, range.start);
+        if seg > 0 {
+            // A drop cap is drawn apart from its word: nothing after it joins it, in this run or
+            // in the runs after it (the letter can be formatted on its own).
+            self.shown_floor = shown_start.saturating_add(seg);
+        }
+        // The first piece continues what came before the run. After a drop cap or a break nothing
+        // joins: that letter is drawn apart, and a break is no letter. The same goes for the end.
+        let mut before = if seg == 0 { around.before } else { "" };
         for (i, c) in text.char_indices() {
             if i < seg || !is_special(c) {
                 continue;
             }
             if let Some(s) = text.get(seg..i) {
-                self.b.shape(s, range.start + seg, rc, None);
+                self.b.shape_around(s, range.start + seg, rc, None, Dir::Para, ShapeContext { before, after: "" });
             }
+            before = "";
             seg = i + c.len_utf8();
             let style = self.b.style(rc, None, false);
             let slot = Slot { start: range.start + i, end: range.start + i + c.len_utf8(), style, g: self.b.glyphs.len() as u32 };
             self.special(cx, c, slot, (rc, props));
         }
         if let Some(s) = text.get(seg..) {
-            self.b.shape(s, range.start + seg, rc, None);
+            self.b.shape_around(s, range.start + seg, rc, None, Dir::Para, ShapeContext { before, after: around.after });
         }
     }
 
+    /// The text shown on either side of the next `len` bytes shown, and moves past them. What
+    /// comes before starts after a drop cap ([`Shaped::shown_floor`]).
+    fn next_shown<'c>(&mut self, cx: &'c Cx, len: usize) -> ShapeContext<'c> {
+        let Some(shown) = &cx.shown else { return ShapeContext::default() };
+        let start = self.shown_at;
+        self.shown_at = start.saturating_add(len);
+        ShapeContext { before: shown.get(self.shown_floor.min(start)..start).unwrap_or(""), after: shown.get(self.shown_at..).unwrap_or("") }
+    }
+
     /// Drop cap: the first character, sized so its capital spans `lines` lines. Returns how much
     /// of the run's text it took.
     fn drop_cap_letter(&mut self, cx: &Cx, rc: &Arc<ResolvedChar>, text: &str, start: usize) -> usize {
@@ -338,7 +368,8 @@ impl<'a> Shaped<'a> {
     /// A picture, shape or text box: inline, as wide and high as it is; a floating one takes no room.
     /// An inline drawing: it stands on the baseline, raised or lowered by its run's position
     /// (`raise`, points, as Word moves an inline picture with `w:position`). Its cluster reaches
-    /// `obj_h` above the baseline and `obj_d` below it (negative when raised clear of it).
+    /// `obj_h` above the baseline and `obj_d` below it (negative when raised clear of it; `obj_h`
+    /// negative when lowered further than its height, drawn with its top on the baseline).
     fn drawing(&mut self, cx: &Cx, (k, o): (usize, &InlineObject), slot: Slot, raise: f32) {
         let env = cx.env;
         // A hidden one takes no room, and with no height its line doesn't draw it.
@@ -368,10 +399,10 @@ impl<'a> Shaped<'a> {
             let [el, et, er, eb] = float.effect_extent();
             let (px, py) = float.spin_pad(w * s, h * s);
             let tall = h * s + et + eb + 2.0 * py;
-            // Lowered at most until its top is on the baseline. The bound is made finite and
-            // not above `clamp`'s top whatever the extents give (`clamp` panics otherwise).
-            let floor = if tall.is_finite() { -tall.max(0.0) } else { 0.0 };
-            let raise = if raise.is_finite() { raise.clamp(floor, 1584.0) } else { 0.0 };
+            // Lowered further than its height, its top stays on the baseline while the line
+            // reaches as far down as the lowering (Word, measured: a 6 pt picture lowered 30 pt
+            // makes a 30 pt line); `obj_h` is then negative. Word's positions run to 1584 pt.
+            let raise = if raise.is_finite() { raise.clamp(-1584.0, 1584.0) } else { 0.0 };
             self.b.clusters.push(slot.cluster(ClKind::Object(k), w * s + el + er + 2.0 * px, (tall + raise, -raise)));
         } else {
             self.push(slot, ClKind::Object(k), 0.0, 0.0);
@@ -592,6 +623,26 @@ fn joined_cluster_starts(p: &Paragraph, clusters: &[Cluster]) -> Vec<usize> {
     joined
 }
 
+/// The paragraph's text without the runs left out of the layout, if it has letters that join their
+/// neighbours: a letter joins the next one on the page, across a change of formatting and past
+/// text that isn't shown (#4093). `None` for every other paragraph, which don't need it.
+fn shown_text(cx: &Cx) -> Option<String> {
+    let text = &cx.p.text;
+    if text.is_ascii() || !text.chars().any(wordcraft_fonts::joins_cursively) {
+        return None;
+    }
+    let mut shown = String::with_capacity(text.len());
+    for (range, props) in cx.p.run_ranges() {
+        let rc = cx.res.char(props);
+        if !left_out(cx.env.doc, &rc, cx.env.show_hidden, cx.env.hide_deleted)
+            && let Some(run) = text.get(range)
+        {
+            shown.push_str(run);
+        }
+    }
+    Some(shown)
+}
+
 /// Lay out one paragraph.
 pub fn layout_para(p: &Paragraph, env: &ParaEnv) -> ParaLayout {
     let doc = env.doc;
@@ -608,7 +659,9 @@ pub fn layout_para(p: &Paragraph, env: &ParaEnv) -> ParaLayout {
         shown: Vec::new(),
         substituted: Vec::new(),
     };
-    let cx = Cx { p, env, rp: &rp, res, math_props: doc.settings.math.clone().unwrap_or_default(), drop_first: drop_cap_char(p, &rp) };
+    let mut cx =
+        Cx { p, env, rp: &rp, res, math_props: doc.settings.math.clone().unwrap_or_default(), drop_first: drop_cap_char(p, &rp), shown: None };
+    cx.shown = shown_text(&cx);
     let mut sh = Shaped::new(b, env.eq_number);
     let mark_rc = cx.res.char(&p.mark);
     let mark_style = sh.b.style(&mark_rc, None, false);

crates/layout/src/pass.rs

@@ -43,6 +43,7 @@ struct Tops {
 struct Sec<'p> {
     si: usize,
     sect: &'p SectionProps,
+    page_format: wordcraft_doc::section::NumFormat,
     start: SectionStart,
     tops: Tops,
 }
@@ -158,10 +159,17 @@ fn reflow_sect(first: &SectionProps, opts: &LayoutOptions) -> SectionProps {
     s
 }
 
-fn new_page_builder<'p>(sect: &'p SectionProps, early: &'p EarlyFloats, web: bool, settle: bool) -> PageBuilder<'p> {
+fn new_page_builder<'p>(
+    sect: &'p SectionProps,
+    page_format: wordcraft_doc::section::NumFormat,
+    early: &'p EarlyFloats,
+    web: bool,
+    settle: bool,
+) -> PageBuilder<'p> {
     PageBuilder {
         pages: Vec::new(),
         sect,
+        page_format,
         sect_idx: 0,
         col: 0,
         cols: Vec::new(),
@@ -221,6 +229,7 @@ fn turned_sections(ctx: &mut Ctx, sections: &[(usize, &SectionProps)], web: bool
         .iter()
         .map(|(_, s)| {
             (!web && s.text_direction.is_turned()).then(|| {
+                ctx.fields.page_format = s.page_num_format;
                 let top = body_top_for(ctx, s, s.headers.default);
                 vertical::unturned_sect(s, top)
             })
@@ -232,6 +241,7 @@ fn turned_sections(ctx: &mut Ctx, sections: &[(usize, &SectionProps)], web: bool
 /// previous section's, and take its room. `fixed`: the margins are the room (continuous views and
 /// turned sections).
 fn section_tops(ctx: &mut Ctx, hf: &HfLinks, (si, sect): (usize, &SectionProps), fixed: bool) -> Tops {
+    ctx.fields.page_format = sect.page_num_format;
     let (heads, feet) = (hf.heads.set(si), hf.feet.set(si));
     let hb = |i: Option<u32>| hf.defining_width(false, si, i);
     let fb = |i: Option<u32>| hf.defining_width(true, si, i);
@@ -246,6 +256,8 @@ fn section_tops(ctx: &mut Ctx, hf: &HfLinks, (si, sect): (usize, &SectionProps),
 fn begin_section_state<'p>(ctx: &mut Ctx, pb: &mut PageBuilder<'p>, sec: &Sec<'p>, totals: &Totals, web: bool) {
     let (doc, sect, si) = (ctx.doc, sec.sect, sec.si);
     pb.sect = sect;
+    pb.page_format = sec.page_format;
+    ctx.fields.page_format = sec.page_format;
     pb.sect_idx = si;
     pb.sect_beneath = doc.note_options(sect, false).pos == wordcraft_doc::section::NotePos::BeneathText;
     if !web && !doc.footnote_separators.is_empty() {
@@ -263,6 +275,8 @@ fn begin_section_state<'p>(ctx: &mut Ctx, pb: &mut PageBuilder<'p>, sec: &Sec<'p
 fn end_previous_section<'p>(ctx: &mut Ctx, pb: &mut PageBuilder<'p>, hf: &HfLinks<'p>, si: usize, block: usize) {
     let Some((_, prev)) = si.checked_sub(1).and_then(|k| hf.sections.get(k)) else { return };
     pb.sect = prev;
+    pb.page_format = prev.page_num_format;
+    ctx.fields.page_format = prev.page_num_format;
     pb.sect_idx = si - 1;
     let (ph, pf) = (hf.heads.set(si - 1), hf.feet.set(si - 1));
     let top = body_top_for(ctx, prev, ph.default);
@@ -281,6 +295,8 @@ fn start_section_page<'p>(ctx: &mut Ctx, pb: &mut PageBuilder<'p>, hf: &HfLinks<
     if filler && !mode.web {
         end_previous_section(ctx, pb, hf, sec.si, block);
         pb.sect = sec.sect;
+        pb.page_format = sec.page_format;
+        ctx.fields.page_format = sec.page_format;
         pb.sect_idx = sec.si;
     }
     if restart.is_some() {
@@ -492,7 +508,7 @@ pub(crate) fn layout_pass(
     let first_sect = sections.first().map_or(&default_sect, |(_, s)| *s);
     let web_sect = mode.reflow.then(|| reflow_sect(first_sect, opts));
     let sect_ref = web_sect.as_ref().unwrap_or(first_sect);
-    let mut pb = new_page_builder(sect_ref, early, mode.web, !may_rewrap && !mode.reflow);
+    let mut pb = new_page_builder(sect_ref, first_sect.page_num_format, early, mode.web, !may_rewrap && !mode.reflow);
     let mut endnotes = endnote_queue(&ctx);
     // Text boxes laid out in balancing attempts that were thrown away.
     let mut boxes_discarded = 0usize;
@@ -500,15 +516,17 @@ pub(crate) fn layout_pass(
     let turned = turned_sections(&mut ctx, sections, mode.web);
     let mut prev_turn = TextDirection::Horizontal;
     for (si, (end, sect)) in sections.iter().enumerate() {
+        let page_format = sect.page_num_format;
         let turn = if mode.reflow { TextDirection::Horizontal } else { sect.text_direction };
         let sect: &SectionProps = match turned.get(si) {
             Some(Some(t)) => t,
             _ if mode.reflow => sect_ref,
             _ => sect,
         };
+        ctx.fields.page_format = page_format;
         let tops = section_tops(&mut ctx, &hf, (si, sect), mode.reflow || turn.is_turned());
         let start = if si == 0 { SectionStart::NextPage } else { sect.start };
-        let sec = Sec { si, sect, start, tops };
+        let sec = Sec { si, sect, page_format, start, tops };
         if !mode.web && !pb.pages.is_empty() && (start != SectionStart::Continuous || turn != prev_turn) {
             // The page the section before ends on is finished with its own margins and footnote
             // separators, before this section's replace them: the footnotes pending on it

crates/layout/src/styleref.rs

@@ -125,9 +125,11 @@ struct Entry {
     runs: Vec<(String, String)>,
     /// The list label as shown (`Chapter 2.`).
     label: Option<String>,
-    /// Each level's value up to the paragraph's, formatted; and whether the label's own
-    /// template shows that level.
-    levels: Vec<(String, bool)>,
+    /// Each level up to the paragraph's that has a number (not a bullet, nor a "none" level):
+    /// its index, its value formatted, and whether the label's own template shows it.
+    levels: Vec<(u8, String, bool)>,
+    /// The paragraph's own list level.
+    level: u8,
     /// How many paragraphs of its style there are up to this one.
     ordinal: u32,
 }
@@ -180,7 +182,7 @@ impl StyleIndex {
             *ord = ord.saturating_add(1);
             let ordinal = *ord;
             let num = p.props.numbering.or_else(|| doc.styles.resolve_para(&p.props).numbering).filter(|n| n.num != 0);
-            let (mut label, mut levels) = (None, Vec::new());
+            let (mut label, mut levels, mut level) = (None, Vec::new(), 0u8);
             if let Some(n) = num
                 && let Some((l, def)) = counters.next_label(&doc.numbering, n.num, n.level)
                 && def.format != NumFormat::Bullet
@@ -193,9 +195,10 @@ impl StyleIndex {
                     }
                     let v = if *v == 0 { lv.start } else { *v };
                     let fmt = if def.legal && lv.format != NumFormat::DecimalZero { NumFormat::Decimal } else { lv.format };
-                    levels.push((fmt.format(v), def.text.contains(&format!("%{}", k + 1))));
+                    levels.push((k as u8, fmt.format(v), def.text.contains(&format!("%{}", k + 1))));
                 }
                 label = Some(l);
+                level = n.level.min(8);
             }
             let mut runs: Vec<(String, String)> = Vec::new();
             let mut off = 0usize;
@@ -221,7 +224,7 @@ impl StyleIndex {
                     list.push(i);
                 }
             }
-            entries.push(Entry { path, style, text: clean(&p.plain_text()), runs, label, levels, ordinal });
+            entries.push(Entry { path, style, text: clean(&p.plain_text()), runs, label, levels, level, ordinal });
         }
         StyleIndex { entries, by_path, styles, para_entries, char_entries }
     }
@@ -323,18 +326,32 @@ impl StyleIndex {
         let mut out = match code.number {
             Number::Text if code.position => String::new(),
             Number::Text => text,
-            Number::Chapter => e.label.as_deref().map(chapter_number).filter(|s| !s.is_empty()).unwrap_or_else(|| e.ordinal.to_string()),
+            // The last number the label shows (`3` of `3 Response`, or of a "none" level's `%1`);
+            // a label of the paragraph's own number that shows none recognisably, its last word;
+            // else, as for an unnumbered heading or a "none" level's `Note`, the heading's count.
+            Number::Chapter => {
+                let own = e.levels.iter().any(|(k, ..)| *k == e.level);
+                e.label
+                    .as_deref()
+                    .and_then(|l| {
+                        let shown = l.split_whitespace().rev().find_map(|w| number_span(w, &e.levels));
+                        shown.map(|n| n.trim_matches(|c: char| !c.is_alphanumeric()).to_string()).or_else(|| own.then(|| chapter_number(l)))
+                    })
+                    .filter(|s| !s.is_empty())
+                    .unwrap_or_else(|| e.ordinal.to_string())
+            }
             Number::Plain | Number::Full => {
                 let own = match &e.label {
-                    Some(l) if code.suppress => suppress(l, &e.levels),
+                    Some(l) if code.suppress => suppress(l, &e.levels, e.level),
                     Some(l) => l.trim().trim_end_matches('.').to_string(),
                     None => e.ordinal.to_string(),
                 };
                 if code.number == Number::Full && e.label.is_some() {
-                    // Levels above that the label's own template doesn't show.
-                    let above: Vec<&str> =
-                        e.levels.iter().take(e.levels.len().saturating_sub(1)).filter(|(_, shown)| !shown).map(|(v, _)| v.as_str()).collect();
-                    if above.is_empty() { own } else { format!("{}.{own}", above.join(".")) }
+                    // Levels above the paragraph's that the label's own template doesn't show. By
+                    // level, not position: a "none" level has no entry, its own (#4985) included.
+                    let above: Vec<&str> = e.levels.iter().filter(|(k, _, shown)| *k < e.level && !shown).map(|(_, v, _)| v.as_str()).collect();
+                    // An empty label of its own (a "none" level with no text) adds no part: `3`, not `3.`.
+                    above.into_iter().chain(Some(own.as_str()).filter(|o| !o.is_empty())).collect::<Vec<_>>().join(".")
                 } else {
                     own
                 }
@@ -356,13 +373,46 @@ fn chapter_number(label: &str) -> String {
     word.trim_matches(|c: char| !c.is_alphanumeric()).to_string()
 }
 
-/// `\t`: the label without its words that hold no number (`Chapter 2.` → `2`).
-fn suppress(label: &str, levels: &[(String, bool)]) -> String {
-    let kept: Vec<&str> = label.split_whitespace().filter(|w| levels.iter().any(|(v, _)| !v.is_empty() && w.contains(v.as_str()))).collect();
-    let s = if kept.is_empty() { label.trim().to_string() } else { kept.join(" ") };
+/// `\t`: the label without its words that hold no number (`Chapter 2.` → `2`). A "none" level's
+/// label (`level` has no number of its own) holding none is left with nothing (`Note`, #4985);
+/// any other label that holds none is kept whole.
+fn suppress(label: &str, levels: &[(u8, String, bool)], level: u8) -> String {
+    // Each word's numbers, without the text attached to them (`Section3` → `3`).
+    let kept: Vec<&str> = label.split_whitespace().filter_map(|w| number_span(w, levels)).collect();
+    let none = !levels.iter().any(|(k, ..)| *k == level);
+    let s = match kept.is_empty() {
+        true if none => String::new(),
+        true => label.trim().to_string(),
+        false => kept.join(" "),
+    };
     s.trim_end_matches('.').to_string()
 }
 
+/// The part of `word` that holds the numbers the label's template shows (`levels` flagged shown),
+/// with the delimiters around it (`(3)`, `1.1`, `2.`), without the text attached (`Section3` →
+/// `3`); `None` when it holds none. A number written in letters (`a`, `iv`) counts only standing
+/// apart from other letters: the `a` of `Chapter` isn't one.
+fn number_span<'w>(word: &'w str, levels: &[(u8, String, bool)]) -> Option<&'w str> {
+    let alpha = |c: Option<char>| c.is_some_and(char::is_alphabetic);
+    let mut span: Option<(usize, usize)> = None;
+    for (_, v, _) in levels.iter().filter(|(_, v, shown)| *shown && !v.is_empty()) {
+        let letters = v.chars().any(char::is_alphabetic);
+        for (i, _) in word.match_indices(v.as_str()) {
+            let end = i.saturating_add(v.len());
+            let before = word.get(..i).and_then(|s| s.chars().next_back());
+            let after = word.get(end..).and_then(|s| s.chars().next());
+            if letters && (alpha(before) || alpha(after)) {
+                continue;
+            }
+            span = Some(span.map_or((i, end), |(a, b)| (a.min(i), b.max(end))));
+        }
+    }
+    let (a, b) = span?;
+    let lead: usize = word.get(..a)?.chars().rev().take_while(|c| !c.is_alphanumeric()).map(char::len_utf8).sum();
+    let trail: usize = word.get(b..)?.chars().take_while(|c| !c.is_alphanumeric()).map(char::len_utf8).sum();
+    word.get(a.saturating_sub(lead)..b.saturating_add(trail))
+}
+
 /// Does `p` hold a STYLEREF field?
 pub fn has_styleref(p: &wordcraft_doc::Paragraph) -> bool {
     p.objects.iter().any(is_styleref)
@@ -441,7 +491,14 @@ mod tests {
 
     #[test]
     fn suppress_keeps_the_number() {
-        assert_eq!(suppress("Chapter 2.", &[("2".into(), true)]), "2");
+        assert_eq!(suppress("Chapter 2.", &[(0, "2".into(), true)], 0), "2");
+        assert_eq!(suppress("Chapter", &[(0, "a".into(), false)], 1), "", "an ancestor's number the label doesn't show");
+        assert_eq!(suppress("Section3", &[(0, "3".into(), true)], 1), "3", "the text attached to the number");
+        assert_eq!(suppress("Chapter a.", &[(0, "a".into(), true)], 0), "a", "a letter number standing apart, not Chapter's a");
+        assert_eq!(suppress("(iv)", &[(0, "iv".into(), true)], 0), "(iv)", "delimiters stay");
+        assert_eq!(suppress("1.2", &[(0, "1".into(), true), (1, "2".into(), true)], 1), "1.2");
+        assert_eq!(suppress("Note", &[(0, "3".into(), false)], 1), "", "a none level's label");
+        assert_eq!(suppress("Appendix", &[(0, "3".into(), false)], 0), "Appendix", "kept whole when the level has a number");
         assert_eq!(chapter_number("Chapter IV."), "IV");
     }
 }

crates/layout/src/tests.rs

@@ -1820,6 +1820,30 @@ fn lists_get_labels_and_indent() {
     assert!((x - 108.0).abs() < 0.5, "{x}");
 }
 
+#[test]
+fn a_none_format_label_shows_its_text_and_tabs_to_the_indent() {
+    // #4985: Word draws a "none" level's literal text ("RESPONSE:"), and an empty one still tabs
+    // the text to a hanging indent, as a numbered item's. An empty label already at the left indent
+    // takes no tab (Word's Chapter 1 list: Heading 2 starts at the margin, not at 0.5").
+    use wordcraft_doc::section::NumFormat;
+    for (text, indent, hanging, want) in [("RESPONSE:", 72.0, 72.0, 144.0), ("", 72.0, 72.0, 144.0), ("", 0.0, 0.0, 72.0), ("", 50.0, 0.0, 122.0)] {
+        let mut d = Document::from_text("X");
+        let id = std::sync::Arc::make_mut(&mut d.numbering).add_list(wordcraft_doc::ListKind::Numbered);
+        for a in &mut std::sync::Arc::make_mut(&mut d.numbering).abstracts {
+            if let Some(l) = a.levels.first_mut() {
+                l.format = NumFormat::None;
+                l.text = text.into();
+                (l.indent, l.hanging) = (indent, hanging);
+            }
+        }
+        d.format_paragraphs(&Pos::body(0, 0), &Pos::body(0, 1), &|p| p.numbering = Some(wordcraft_doc::props::NumRef { num: id, level: 0 })).unwrap();
+        let l = lay(&d);
+        assert_eq!(labels(&l), [text]);
+        let x = l.caret(&Pos::body(0, 0)).unwrap().x;
+        assert!((x - want).abs() < 0.5, "{text:?} at {indent}/{hanging}: {x}, not {want}");
+    }
+}
+
 #[test]
 fn break_only_list_has_no_label() {
     use wordcraft_doc::ListKind;
@@ -5598,8 +5622,13 @@ fn an_inline_picture_moves_with_its_runs_position() {
     assert!(line.top + line.height - line.baseline >= 30.0 - 0.01, "{line:?}");
     let r = display::inline_rect(Some(&picture(40.0, 30.0, Default::default())), 0.0, 100.0, c.adv, (c.obj_h, c.obj_d), c.adv);
     assert_eq!((r.y, r.h), (100.0, 30.0));
-    // Lowered further than that: still only to its top.
-    assert_eq!(line_of(-500.0).0.obj_d, 30.0);
+    // Lowered further than that: the line reaches as far down, the picture's top staying on the
+    // baseline (Word).
+    let (c, line, _) = line_of(-500.0);
+    assert_eq!((c.obj_h, c.obj_d), (-470.0, 500.0));
+    assert!(line.top + line.height - line.baseline >= 500.0 - 0.01, "{line:?}");
+    let r = display::inline_rect(Some(&picture(40.0, 30.0, Default::default())), 0.0, 100.0, c.adv, (c.obj_h, c.obj_d), c.adv);
+    assert_eq!((r.y, r.h), (100.0, 30.0));
     // Raised: it stands clear of the baseline, as tall as before.
     let (c, _, _) = line_of(6.0);
     assert_eq!((c.obj_h, c.obj_d), (36.0, -6.0));
@@ -13694,7 +13723,7 @@ fn lowered_pictures_and_check_boxes_take_words_line_heights() {
         let Some(Placed::Lines { para, .. }) = l.pages[0].items.iter().find(|i| matches!(i, Placed::Lines { .. })) else { panic!() };
         let c = para.clusters.iter().find(|c| matches!(c.kind, para::ClKind::Object(_))).unwrap();
         let (line, st) = (&para.lines[0], &para.styles[c.style as usize]);
-        (line.height, line.baseline - c.obj_h - line.top, st.ascent + st.descent)
+        (line.height, line.baseline - c.obj_h.max(0.0) - line.top, st.ascent + st.descent)
     };
     for m in [1.0, 1.15, 1.5] {
         let share = |font: f32| (m - 1.0) * font;
@@ -13711,10 +13740,36 @@ fn lowered_pictures_and_check_boxes_take_words_line_heights() {
             let (h, top, font) = line_of("", m, 6.0, raise);
             assert!((h - font * m).abs() < 0.01 && (top - (font - reach)).abs() < 0.01, "{raise} at {m}: {h}, {top}");
         }
+        // Lowered further than its height: as far down as it's lowered, the font's height at least,
+        // its top on the baseline (Word: a 6 pt picture lowered 10 pt takes the font's 11.5 pt,
+        // lowered 30 pt 30 pt; beside text, the text's ascent and the 30 pt).
+        let (h, top, font) = line_of("", m, 6.0, -10.0);
+        assert!((h - font * m).abs() < 0.01 && (top - (font - 10.0)).abs() < 0.01, "6 lowered 10 at {m}: {h}, {top}");
+        let (h, top, font) = line_of("", m, 6.0, -30.0);
+        assert!((h - 30.0 - share(font)).abs() < 0.01 && top.abs() < 0.01, "6 lowered 30 at {m}: {h}, {top}");
+        let (h, top, font) = line_of("x", m, 6.0, -30.0);
+        let asc = h - 30.0 - share(font);
+        assert!(asc > 0.0 && asc < font && (top - asc).abs() < 0.01, "6 lowered 30 beside text at {m}: {h}, {top}");
         // Beside text, lowered: the text's ascent is no matter, the picture reaching further.
         let (h, top, font) = line_of("x", m, 30.0, -10.0);
         assert!((h - 30.0 - share(font)).abs() < 0.01 && top.abs() < 0.01, "beside text at {m}: {h}, {top}");
     }
+    // Exact spacing keeps the line's box and its baseline: a picture lowered past its height sits
+    // where one lowered by its height does.
+    let exact = |raise: f32| {
+        let mut d = Document::from_text("x");
+        let props = wordcraft_doc::CharProps { position: Some(raise), ..Default::default() };
+        let p = d.para_mut(StoryRef::Body, &Path::top(0)).unwrap();
+        p.props.line_spacing = Some(LineSpacing::Exactly(20.0));
+        let at = p.len();
+        p.insert_object(at, picture(40.0, 6.0, Default::default()), &props).unwrap();
+        let l = lay(&d);
+        let Some(Placed::Lines { para, .. }) = l.pages[0].items.iter().find(|i| matches!(i, Placed::Lines { .. })) else { panic!() };
+        (para.lines[0].height, para.lines[0].baseline - para.lines[0].top)
+    };
+    let (h, base) = exact(-6.0);
+    assert!((h - 20.0).abs() < 0.01, "{h}");
+    assert_eq!(exact(-30.0), (h, base));
     // A check box sizes its line as text of its size, alone or beside text.
     let box_line = |lead: &str, m: f32, size: u32| {
         let mut d = Document::from_text(lead);

crates/layout/src/tests/bidi.rs

@@ -639,3 +639,149 @@ fn list_label_after_leading_page_break_uses_its_text_lines_alignment() {
         }
     }
 }
+
+/// The glyph ids of each letter of paragraph 0 as laid out (marks of runs left out have none).
+fn letters(l: &DocLayout) -> Vec<Vec<u32>> {
+    let (pl, _) = first_line(l, 0);
+    pl.clusters.iter().filter(|c| c.g1 > c.g0).map(|c| pl.glyphs[c.g0 as usize..c.g1 as usize].iter().map(|g| g.gid).collect()).collect()
+}
+
+/// `text` as one right-to-left paragraph with `format` applied to each `(start, end)` byte range.
+fn formatted(text: &str, splits: &[(usize, usize)], format: &dyn Fn(&mut wordcraft_doc::props::CharProps)) -> Document {
+    let mut d = doc_of(&[(text, true)]);
+    for &(start, end) in splits {
+        d.format_range(&Pos::body(0, start), &Pos::body(0, end), format).unwrap();
+    }
+    d
+}
+
+const BEH: usize = 'ب'.len_utf8();
+
+fn red(c: &mut wordcraft_doc::props::CharProps) {
+    c.color = Some(wordcraft_doc::props::TextColor::Rgb(wordcraft_doc::props::Rgb(0xC0, 0, 0)));
+}
+
+/// The reporter's symptom: after a change of formatting inside a word, the letters on either side
+/// took the forms of the end and the start of a word, which reads as a space in it (#4093).
+#[test]
+fn letters_join_across_a_change_of_formatting() {
+    if arabic_face().is_none() {
+        eprintln!("skipped: no font with Persian letters installed");
+        return;
+    }
+    let word = "بببب";
+    let whole = letters(&lay(&doc_of(&[(word, true)])));
+    assert_eq!(whole.len(), 4, "one cluster per letter: {whole:?}");
+    assert!(whole.first() != whole.get(1) && whole.get(1) != whole.last(), "initial, medial and final forms: {whole:?}");
+    // A letter of its own at each place, then the two middle letters, then every other letter.
+    let pieces: [&[(usize, usize)]; 6] =
+        [&[(0, BEH)], &[(BEH, 2 * BEH)], &[(2 * BEH, 3 * BEH)], &[(3 * BEH, 4 * BEH)], &[(BEH, 3 * BEH)], &[(0, BEH), (2 * BEH, 3 * BEH)]];
+    for runs in pieces {
+        let d = formatted(word, runs, &red);
+        assert!(d.para_at(&Pos::body(0, 0)).unwrap().runs.len() > 1, "{runs:?} split the run");
+        assert_eq!(letters(&lay(&d)), whole, "split at {runs:?}");
+    }
+}
+
+/// A letter joins what is on the page: with a tracked deletion shown, the deleted letter; with it
+/// left out, the letter past it. The same goes for hidden text.
+#[test]
+fn letters_join_past_text_left_out_of_the_layout() {
+    if arabic_face().is_none() {
+        eprintln!("skipped: no font with Persian letters installed");
+        return;
+    }
+    let (three, two) = (letters(&lay(&doc_of(&[("ببب", true)]))), letters(&lay(&doc_of(&[("بب", true)]))));
+    assert_eq!((three.len(), two.len()), (3, 2));
+    let middle = [(BEH, 2 * BEH)];
+
+    let deleted = formatted("ببب", &middle, &|c| c.del = Some(0));
+    assert_eq!(letters(&lay(&deleted)), three, "deletion shown: it joins both neighbours");
+    let mut c = LayoutCache::new();
+    let final_text = layout(&deleted, &mut c, &LayoutOptions { hide_deleted: true, ..LayoutOptions::default() });
+    assert_eq!(letters(&final_text), two, "deletion left out: its neighbours join each other");
+
+    let hidden = formatted("ببب", &middle, &|c| c.hidden = Some(true));
+    assert_eq!(letters(&lay(&hidden)), two, "hidden text left out: its neighbours join each other");
+    let mut c = LayoutCache::new();
+    let shown = layout(&hidden, &mut c, &LayoutOptions { show_hidden: true, ..LayoutOptions::default() });
+    assert_eq!(letters(&shown), three, "hidden text shown: it joins both neighbours");
+}
+
+/// A mark in a run of its own (a coloured vowel sign) doesn't stop the letters around it joining.
+#[test]
+fn letters_join_past_a_mark_in_a_run_of_its_own() {
+    if arabic_face().is_none() {
+        eprintln!("skipped: no font with Persian letters installed");
+        return;
+    }
+    let word = "بَب";
+    let whole = letters(&lay(&doc_of(&[(word, true)])));
+    let split = letters(&lay(&formatted(word, &[(BEH, BEH + 'َ'.len_utf8())], &red)));
+    assert_eq!(split.last(), whole.last(), "the letter after the mark stays joined");
+    // Together with the mark, the letter before it has more glyphs, in the shaper's order.
+    let (before, with_mark) = (split.first().unwrap(), whole.first().unwrap());
+    assert!(before.iter().all(|g| with_mark.contains(g)), "and the letter before it: {before:?} {with_mark:?}");
+}
+
+/// What can't join still doesn't: a tab between the letters, in one run or at the edge of a run, a
+/// Latin neighbour, and a letter beside a drop cap, which is drawn apart from the word.
+#[test]
+fn letters_stay_apart_where_nothing_joins_them() {
+    if arabic_face().is_none() {
+        eprintln!("skipped: no font with Persian letters installed");
+        return;
+    }
+    let alone = letters(&lay(&doc_of(&[("ب", true)])));
+    assert_eq!(alone.len(), 1);
+    let beh = alone.first();
+    let tabbed = "ب\tب";
+    for d in [doc_of(&[(tabbed, true)]), formatted(tabbed, &[(0, BEH)], &red), formatted(tabbed, &[(BEH + 1, tabbed.len())], &red)] {
+        let got = letters(&lay(&d));
+        assert_eq!((got.first(), got.last()), (beh, beh), "across a tab: {got:?}");
+    }
+    let got = letters(&lay(&formatted("xب", &[(0, 1)], &red)));
+    assert_eq!(got.last(), beh, "after a Latin letter: {got:?}");
+
+    // The drop cap in the word's run, and formatted on its own (a run of its own).
+    let two = letters(&lay(&doc_of(&[("بب", true)])));
+    for mut d in [doc_of(&[("ببب", true)]), formatted("ببب", &[(0, BEH)], &red)] {
+        let at = Pos::body(0, 0);
+        d.format_paragraphs(&at, &at, &|p: &mut ParaProps| p.drop_cap = Some(2)).unwrap();
+        let runs = d.para_at(&at).unwrap().runs.len();
+        let got = letters(&lay(&d));
+        assert_eq!(got.get(got.len().saturating_sub(2)..), two.get(..), "{runs} run(s): the letters after a drop cap start a word: {got:?} {two:?}");
+    }
+}
+
+#[test]
+fn the_shaper_takes_a_letters_form_from_the_text_around_it() {
+    use wordcraft_fonts::{ShapeContext, shape_run_in};
+    let Some(face) = arabic_face() else {
+        eprintln!("skipped: no font with Persian letters installed");
+        return;
+    };
+    let form = |around| cluster_glyphs(&shape_run_in(&face, "ب", &[], |c| c, true, around));
+    let isolated = form(ShapeContext::default());
+    let joined = cluster_glyphs(&wordcraft_fonts::shape_run(&face, "ببب", &[], |c| c, true));
+    assert_eq!(form(ShapeContext { before: "", after: "ب" }).first(), joined.first(), "initial");
+    assert_eq!(form(ShapeContext { before: "ب", after: "ب" }).first(), joined.get(1), "medial");
+    assert_eq!(form(ShapeContext { before: "ب", after: "" }).first(), joined.get(2), "final");
+    // A vowel sign is transparent, and ZWNJ and a Latin letter stop the join.
+    assert_eq!(form(ShapeContext { before: "بَ", after: "" }).first(), joined.get(2), "past a mark");
+    assert_eq!(form(ShapeContext { before: "ب\u{200C}", after: "" }), isolated, "after ZWNJ");
+    assert_eq!(form(ShapeContext { before: "a", after: "b" }), isolated, "between Latin letters");
+    // Only the letter's form depends on the context: nothing of it is drawn.
+    assert_eq!(form(ShapeContext { before: "ببب", after: "ببب" }).len(), 1);
+}
+
+#[test]
+fn scripts_that_join_their_letters_are_told_apart() {
+    use wordcraft_fonts::joins_cursively;
+    for c in ['ب', 'پ', 'ܐ', 'ᠠ', 'ߊ', '𞤀'] {
+        assert!(joins_cursively(c), "{c}");
+    }
+    for c in ['a', 'é', '’', 'я', 'א', '中', ' ', '\t', '\u{FFFC}', '\u{200C}', 'َ'] {
+        assert!(!joins_cursively(c), "{c:?}");
+    }
+}

crates/layout/src/tests/pages.rs

@@ -612,3 +612,93 @@ fn a_page_ends_when_a_multiple_spaced_lines_text_no_longer_fits() {
         assert!(past_margin > 0, "{multiple}: some shift should leave the last line's leading past the margin");
     }
 }
+
+/// A body PAGE field shows the page number in the section's number format, as Word does (a
+/// cardinal-text section shows "one"), and `\*` overrides it.
+#[test]
+fn body_page_field_uses_the_section_number_format() {
+    let mut d = Document::from_text("Page \nPage ");
+    d.last_section.page_num_format = wordcraft_doc::section::NumFormat::CardinalText;
+    for (i, instr) in ["PAGE", "PAGE \\* Upper"].into_iter().enumerate() {
+        let p = d.para_mut(StoryRef::Body, &Path::top(i)).unwrap();
+        let off = p.len();
+        let field = InlineObject::Field { instr: instr.into(), result: "1".into(), locked: false, dirty: false, code: false, form: String::new() };
+        p.insert_object(off, field, &Default::default()).unwrap();
+    }
+    let l = lay(&d);
+    let shown: Vec<String> = l.pages[0]
+        .items
+        .iter()
+        .filter_map(|it| if let Placed::Lines { para, .. } = it { Some(para.shown.iter().map(|x| x.1.clone()).collect::<String>()) } else { None })
+        .collect();
+    assert_eq!(shown, ["one", "ONE"], "{shown:?}");
+}
+
+#[test]
+fn reflow_page_field_uses_the_current_section_number_format() {
+    let mut d = Document::from_text("First section\nSecond section");
+    let first = d.last_section.clone();
+    d.para_mut(StoryRef::Body, &Path::top(0)).unwrap().section = Some(Box::new(first));
+    d.last_section.page_num_format = wordcraft_doc::section::NumFormat::CardinalText;
+    let p = d.para_mut(StoryRef::Body, &Path::top(1)).unwrap();
+    let off = p.len();
+    p.insert_object(
+        off,
+        InlineObject::Field { instr: "PAGE".into(), result: "1".into(), locked: false, dirty: false, code: false, form: String::new() },
+        &Default::default(),
+    )
+    .unwrap();
+
+    for view in [ViewMode::Web, ViewMode::Draft, ViewMode::Read, ViewMode::Outline] {
+        let mut cache = LayoutCache::new();
+        let layout = layout(&d, &mut cache, &LayoutOptions { view, ..Default::default() });
+        let shown: Vec<_> = layout
+            .pages
+            .iter()
+            .flat_map(|page| {
+                page.items.iter().filter_map(move |item| {
+                    if let Placed::Lines { para, path, .. } = item {
+                        (path == &Path::top(1)).then(|| (page.number, para.shown.iter().map(|x| x.1.as_str()).collect::<String>()))
+                    } else {
+                        None
+                    }
+                })
+            })
+            .collect();
+        assert!(!shown.is_empty(), "{view:?}: no second-section text");
+        for (number, text) in shown {
+            let expected = wordcraft_doc::section::NumFormat::CardinalText.format(number).to_lowercase();
+            assert!(text.ends_with(&expected), "{view:?}, page {number}: {text:?}, expected {expected:?}");
+        }
+    }
+}
+
+/// A PAGE field in a table that opens the body uses the section's number format too: the table
+/// is placed without the paragraph path that sets it.
+#[test]
+fn table_cell_page_field_uses_the_section_number_format() {
+    use wordcraft_doc::Table;
+    let mut d = Document::from_text("after");
+    d.last_section.page_num_format = wordcraft_doc::section::NumFormat::CardinalText;
+    let mut t = Table::new(1, 1, 200.0);
+    let mut p = wordcraft_doc::Paragraph::with_text("", Default::default());
+    let field = InlineObject::Field { instr: "PAGE".into(), result: "1".into(), locked: false, dirty: false, code: false, form: String::new() };
+    p.insert_object(0, field, &Default::default()).unwrap();
+    if let Some(c) = t.rows.first_mut().and_then(|r| r.cells.first_mut()) {
+        c.blocks = vec![para_block(p)];
+    }
+    d.insert_block(StoryRef::Body, &Path::top(0), Block::Table(t)).unwrap();
+    let l = lay(&d);
+    let shown: String = l.pages[0]
+        .items
+        .iter()
+        .filter_map(|it| {
+            if let Placed::Lines { para, path, .. } = it {
+                (path.0.len() > 1).then(|| para.shown.iter().map(|x| x.1.clone()).collect::<String>())
+            } else {
+                None
+            }
+        })
+        .collect();
+    assert_eq!(shown, "one");
+}

crates/layout/src/tests/styleref.rs

@@ -97,6 +97,49 @@ fn character_styles_and_paragraph_numbers() {
     assert_eq!(h[3], ["two", "3", "3.a", "3"]);
 }
 
+/// #4985: a heading on a "none" list level (its label is its text, `Note`) shows its parent's
+/// number in full context (`\w`): the parent is not taken for the heading's own level, which
+/// has no number to list.
+#[test]
+fn full_context_keeps_the_parent_of_a_none_level_heading() {
+    let mut d = doc(CHAPTERS, &[r#"STYLEREF "Heading 2" \w"#, r#"STYLEREF "Heading 2" \r"#]);
+    let num = Arc::make_mut(&mut d.numbering).add_list(ListKind::Numbered);
+    if let Some(n) = Arc::make_mut(&mut d.numbering).nums.iter_mut().find(|n| n.id == num) {
+        let note = wordcraft_doc::numbering::Level { format: wordcraft_doc::section::NumFormat::None, text: "Note".into(), ..Default::default() };
+        n.level_overrides = vec![(1, note)];
+    }
+    for (i, lvl) in [(0, 0u8), (2, 0), (5, 0), (6, 1)] {
+        d.para_mut(StoryRef::Body, &Path::top(i)).unwrap().props.numbering = Some(NumRef { num, level: lvl });
+    }
+    assert_eq!(headers(&d)[3], ["3.Note", "3.Note"]);
+}
+
+/// #4985 review: `\t` leaves a "none" level's label, which holds no number, with nothing, and an
+/// empty label of its own adds no part in full context: `3`, not `3.`.
+#[test]
+fn a_none_level_heading_s_number_has_no_text_and_no_trailing_period() {
+    let numbered = |text: &str, codes: &[&str]| {
+        let mut d = doc(CHAPTERS, codes);
+        let num = Arc::make_mut(&mut d.numbering).add_list(ListKind::Numbered);
+        if let Some(n) = Arc::make_mut(&mut d.numbering).nums.iter_mut().find(|n| n.id == num) {
+            let level = wordcraft_doc::numbering::Level { format: wordcraft_doc::section::NumFormat::None, text: text.into(), ..Default::default() };
+            n.level_overrides = vec![(1, level)];
+        }
+        for (i, lvl) in [(0, 0u8), (2, 0), (5, 0), (6, 1)] {
+            d.para_mut(StoryRef::Body, &Path::top(i)).unwrap().props.numbering = Some(NumRef { num, level: lvl });
+        }
+        headers(&d)[3].clone()
+    };
+    assert!(numbered("Note", &[r#"STYLEREF "Heading 2" \n \t"#]).is_empty(), "nothing is drawn");
+    // `\s` is a chapter number: a "none" label isn't one, so the heading's count stands in; a
+    // label showing its parent's number (`%1`, `%1 Response`) gives that number.
+    assert_eq!(numbered("Note", &[r#"STYLEREF "Heading 2" \s"#]), ["1"]);
+    assert_eq!(numbered("%1", &[r#"STYLEREF "Heading 2" \s"#]), ["3"]);
+    assert_eq!(numbered("%1 Response", &[r#"STYLEREF "Heading 2" \s"#, r#"STYLEREF "Heading 2" \n \t"#]), ["3", "3"]);
+    assert_eq!(numbered("Section%1", &[r#"STYLEREF "Heading 2" \n \t"#]), ["3"]);
+    assert_eq!(numbered("", &[r#"STYLEREF "Heading 2" \w"#, r#"STYLEREF "Heading 2" \r"#]), ["3", "3"]);
+}
+
 /// Editing a heading updates the headers that show it, through the paragraph cache.
 #[test]
 fn headers_follow_heading_edits() {

crates/pdf/src/metadata.rs

@@ -8,7 +8,10 @@ pub(crate) fn metadata(doc: &Document, opts: &PdfOptions) -> Metadata {
     if !title.trim().is_empty() {
         m.title = Some(title);
     }
-    let author = opts.author.clone().unwrap_or_else(|| c.creator.clone());
+    // "Remove personal information from file properties on save" covers the PDF's author too:
+    // the document's creator is left out (an author the caller names is still written).
+    let kept = if doc.settings.remove_personal_info { String::new() } else { c.creator.clone() };
+    let author = opts.author.clone().unwrap_or(kept);
     if !author.trim().is_empty() {
         m.authors = vec![author];
     }

crates/pdf/src/tests.rs

@@ -2862,3 +2862,14 @@ fn baseline_is_the_most_shared_y() {
     assert_eq!(baseline(&[(1, 0.0, 9.0), (2, 3.0, 4.0)]), Some(4.0));
     assert_eq!(baseline(&[(1, 0.0, -0.0), (2, 3.0, 0.0)]), Some(0.0));
 }
+
+#[test]
+fn the_author_is_left_out_when_personal_information_is_removed() {
+    let mut d = Document::from_text("Private");
+    d.core.creator = "Ada Writer".into();
+    let opts = PdfOptions { tagged: false, compress: false, ..Default::default() };
+    let raw = |d: &Document| String::from_utf8_lossy(&export(d, &opts).unwrap()).to_string();
+    assert!(raw(&d).contains("Ada Writer"));
+    d.settings.remove_personal_info = true;
+    assert!(!raw(&d).contains("Ada Writer"), "the creator stays out of the PDF");
+}

crates/ui-egui/src/app.rs

@@ -13,6 +13,8 @@ use super::{
 mod commands;
 mod control;
 mod files;
+#[cfg(test)]
+pub(crate) use files::suggested_stem;
 mod fonts;
 mod frame;
 mod input;
@@ -439,7 +441,11 @@ impl WordApp {
             return;
         }
         let t = theme::Tokens::get(&ctx);
-        self.escape_free = self.dialog.is_none() && !self.ui.backstage && self.ui.keytips == keytips::Phase::Off && !egui::Popup::is_any_open(&ctx);
+        self.escape_free = self.dialog.is_none()
+            && !self.ui.backstage
+            && self.ui.keytips == keytips::Phase::Off
+            && !egui::Popup::is_any_open(&ctx)
+            && !crate::chrome::title_search_open(self, &ctx);
         // Keytips are read before layout so the frame paints the new state; badges paint after
         // the ribbon so they sit on top of it.
         crate::keytips::logic(self, &ctx);

crates/ui-egui/src/app/commands.rs

@@ -318,7 +318,8 @@ impl WordApp {
                 file_dialogs::Asked::Busy => Ok(json!({"saved": false})),
             };
         }
-        let retry = (id == "file.open").then(|| params.clone());
+        // Opening a file, or a new document from a template, reads a file that may be protected.
+        let retry = matches!(id, "file.open" | "file.newFromTemplate").then(|| params.clone());
         let retry_path = retry.clone();
         let r = self.execute(id, params);
         if let (Err(e), Some(params)) = (&r, retry)
@@ -326,7 +327,8 @@ impl WordApp {
         {
             let path = params.get("path").and_then(Value::as_str).unwrap_or("");
             let name = std::path::Path::new(path).file_name().map(|n| n.to_string_lossy().to_string()).unwrap_or_else(|| path.to_string());
-            self.dialog = Some(dialogs::Dialog::Password { name, params, password: Default::default(), message: String::new() });
+            self.dialog =
+                Some(dialogs::Dialog::Password { name, then: id.to_string(), params, password: Default::default(), message: String::new() });
             return Ok(json!({"pending": "password"}));
         }
         // Whoever asked, a failed open names the file and the cause (#1103).

crates/ui-egui/src/app/files.rs

@@ -298,12 +298,16 @@ impl WordApp {
     }
 
     /// The name Save suggests: the open file's own Word format (so a .docm keeps its macros),
-    /// otherwise .docx.
+    /// otherwise .docx. A document never saved and without a title of its own is named after its
+    /// first words, as Word suggests them (#5793), instead of `Document1`.
     pub(crate) fn default_save_name(&self) -> String {
         let ext = self.session.path.as_ref().and_then(|p| p.extension()).map(|e| e.to_string_lossy().to_string());
+        let unnamed = self.session.path.is_none() && self.session.file_name.is_none() && self.session.doc.core.title.trim().is_empty();
+        let first = || self.session.doc.body.iter().filter_map(|b| b.as_para()).map(|p| p.plain_text()).find_map(|t| suggested_stem(&t));
+        let stem = unnamed.then(first).flatten().unwrap_or_else(|| self.title_stem());
         match ext {
-            Some(e) if wordcraft_engine::io::is_word_package(&e) => format!("{}.{e}", self.title_stem()),
-            _ => format!("{}.docx", self.title_stem()),
+            Some(e) if wordcraft_engine::io::is_word_package(&e) => format!("{stem}.{e}"),
+            _ => format!("{stem}.docx"),
         }
     }
 
@@ -463,3 +467,33 @@ impl WordApp {
         }
     }
 }
+
+/// The most characters a name taken from the text keeps ([`suggested_stem`]).
+const SUGGESTED_MAX: usize = 50;
+
+/// A file name from a paragraph's text (#5793): up to its first sentence or clause end
+/// (`.`, `!`, `?`, `;`, `:` and their full-width forms) or line break, without the characters a
+/// file name can't hold (`\ / : * ? " < > |` and controls), white space collapsed, at most
+/// [`SUGGESTED_MAX`] characters cut at a word's end, and no dot or space at either end (Windows
+/// drops them). `None` when nothing is left, so the next paragraph, or `Document1`, is used.
+pub(crate) fn suggested_stem(text: &str) -> Option<String> {
+    let end = |c: char| matches!(c, '.' | '!' | '?' | ';' | ':' | '\n' | '\r' | '\u{0B}' | '\u{2029}' | '。' | '！' | '？' | '；' | '：');
+    let head = text.split(end).next().unwrap_or("");
+    let kept: String = head.chars().map(|c| if c.is_control() || matches!(c, '\\' | '/' | '*' | '"' | '<' | '>' | '|') { ' ' } else { c }).collect();
+    let mut words = String::new();
+    for w in kept.split_whitespace() {
+        let next = if words.is_empty() { w.chars().count() } else { words.chars().count() + 1 + w.chars().count() };
+        if next > SUGGESTED_MAX {
+            if words.is_empty() {
+                words = w.chars().take(SUGGESTED_MAX).collect();
+            }
+            break;
+        }
+        if !words.is_empty() {
+            words.push(' ');
+        }
+        words.push_str(w);
+    }
+    let words = words.trim_matches(|c: char| c == '.' || c.is_whitespace()).to_string();
+    (!words.is_empty()).then_some(words)
+}