* fix(engine): support nested quotations and opening delimiters in smart quotes
* docs: typing-parity line for smart quotes after guillemets, low quotes and nesting
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(engine): resolve section and section pages context when unlinking fields
* cargo fmt
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(engine): support switches and quoted bookmarks when updating REF and PAGEREF fields
* cargo fmt
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Backspace and Delete remove a section break, imported or not (fixes#2127)
* Delete removes a section break on the last paragraph too (#2127)
A continuous break on the mark of the document's last paragraph, with the
last section holding no paragraph (as in the file attached to #2127), had
nothing for Delete to join, so it stayed. Delete at the end now removes it.
The text takes the last section's setup, as in Word, and the headers and
footers that section showed by linking to the removed one stay, here and
when the break goes by joining two paragraphs. Undo brings it back.
* Deleting over several section breaks keeps the headers they linked to (#2127)
A range across two or more breaks dropped the paragraphs in between with
their sections, so only the first section's headers and footers were carried
over. The joined section now takes those it leaves out from every removed
section, nearest first, as "link to previous" resolved them before the delete.
The typing-parity row says that Delete with Track Changes on still leaves a
break on the last paragraph alone.
---------
Co-authored-by: Brandon Thomas <echelon@gmail.com>
* fix(engine): use table_formula parse_number for table sorting
* contributors: drop the duplicate entry (already listed)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* cargo fmt
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* engine: filter note_refs in nextFootnote to skip endnotes
* Collapse the nested if (clippy collapsible_if)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
With text selected, Layout > Columns used to change the whole section. As in
Word (observed black-box), a continuous section break now goes where the
selection starts and one where it ends, splitting a paragraph there, unless that
end already meets a section boundary (the document's start or end, or an
existing break); the columns go on the section between, in one undo step.
`layout.columns` gains `apply: "selection"`, its default when text is selected
in the main text. More Columns... offers "Selected text" and picks it then.
Co-authored-by: Brandon Thomas <echelon@gmail.com>
With Different First Page on, page 1 shows the first-page header, but
Insert > Header/Footer/Page Number always wrote the default one (which
page 1 no longer shows), Edit Header without a page went to the default
too, and a double-click in page 1's header area was ignored while
another header was being edited (the usual case: edit the header, tick
Different First Page, double-click page 1). Now hf_part and edit_hf use
the slot of the caret's page, and a double-click in a header/footer
zone switches to the one that page shows even from another header.
Fixes#2826Fixes#2811
Claude-Session: https://claude.ai/code/session_013kKPnx8r1vcYumjyL3kcuz
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
LibreOffice puts a paragraph indented on both sides on the clipboard as
<blockquote>, its margins in the page's stylesheet. Paste mapped it to the
centred, italic Quote style. On paste only, it now becomes a Normal
paragraph indented by the blockquote's stylesheet margins (else 0.5 in each
side). Opening .html files still maps blockquote to Quote.
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
A copy or cut of one equation on its own now also puts its MathML in the
platform's MathML clipboard formats, beside the HTML (with MathML, #816) and
the linear text: "MathML" and "MathML Presentation" (UTF-16LE) on Windows,
public.mathml.presentation and public.mathml on macOS. These are the names
W3C MathML 3 section 6.2.4 registers, and Microsoft's "MathML Support in
Microsoft 365" doc says Office checks them first when it pastes math. No
public clipboard format carries OMML. A copy with text around the equation
doesn't offer them, so Word keeps the text.
Not yet verified in Word: the parity docs say MathML clipboard formats, not
an editable paste.
Claude-Session: https://claude.ai/code/session_01P8veLveuUsJJm6UEcQPEuW
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Tab already follows reading order (next logical cell) and right-to-left
tables (w:bidiVisual) already lay the first cell out at the right. The
bug was that insert.table ignored the caret paragraph's direction, so a
table made while writing RTL text was a left-to-right table and Tab went
rightwards. Like Word, insert.table now makes a right-to-left table when
the caret's paragraph reads right to left.
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Found while testing crates/doc/src/table.rs:129 (Table::delete_row), called from crates/engine/src/cmd/table.rs:526 (del_row)
Co-authored-by: Claude Haiku 5.5 <noreply@anthropic.com>
Indonesian spelling reads the installed Hunspell id_ID dictionary at runtime
(Debian/Ubuntu hunspell-id, LibreOffice dict-id, the Dictionaries folder or
WORDCRAFT_DICTIONARIES), like Czech, Polish and Portuguese: nothing is bundled,
and without a dictionary Indonesian text gets no spelling marks.
- ProofLang::Id for id, id-ID, ind and the legacy in tag
- suggestions reuse the dictionary-driven Czech ranking (MAP/REP/TRY/KEY of
the loaded dictionary), nothing Indonesian hard-coded
- hyphenated reduplication accepted whole or by parts (kupu-kupu, anak-anak);
hyphenated clitics after capitals/abbreviations (KTP-nya, se-Indonesia)
- generic grammar only: repeated word (offers anak-anak first), space before
punctuation, double space, with Indonesian messages
- Review > Language, status bar, Set Proofing Language, File > Options >
Proofing list Indonesian; name translated in every catalog
- tests use a hand-written synthetic dictionary; engine end-to-end test
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
On macOS, Cmd+Left/Right were bound to Word Left/Right (same as Option+arrows).
Bind them to Line Start/End there, as in Word for Mac and every Mac text field;
Option+arrows still move by word. Windows/Linux keys are unchanged.
Co-authored-by: justinjohn0306 <babyjjbabyjj549@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
* Tables: formulas with cell references, functions, number formats and F9
Table Layout › Data › Formula was a fixed set of string checks: any
unknown function was silently summed, cell references weren't read, and
Update Field never recalculated a formula field.
Formulas are now parsed and evaluated (crates/engine/src/cmd/table_formula.rs):
- cell references and ranges by grid column (=B2+B3, =SUM(B2:C3)),
+ - * / ^, parentheses and the comparisons = <> < <= > >=
- SUM, AVERAGE, COUNT, MAX, MIN, PRODUCT, ABS, INT, SIGN, ROUND, MOD,
IF (only the branch taken is worked out), AND, OR, NOT, TRUE, FALSE,
DEFINED
- ABOVE/LEFT/RIGHT/BELOW sum the first run of numbers next to the cell,
as observed in Word; `=SUM()` is a syntax error, as in Word; a bare
`=SUM` takes ABOVE or LEFT from the cell
- the \# numeric picture (`#,##0.00`, `0%`, `;` sections)
- an unknown function or a syntax error is an error and inserts nothing;
input is bounded (formula length, nesting depth)
Update Field (F9) re-evaluates formula fields from the cells they read.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Review fixes for #907: clamp formula ranges to the table, cell budget, depth for ^ and unary +
- A range reads only the rows and grid columns the table has, so SUM(A1:BK32767) over a small
table no longer walks two million grid positions.
- One evaluation reads at most 100,000 cells; past it the formula fails with "Formula too
complex" (F9 shows !Formula too complex) instead of hanging on a hostile document.
- Each ^ in a chain and each unary + counts toward the nesting limit of 32, like unary -.
- Regression tests for all three.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* M2.1: locale-aware number parsing and formatting in table formulas
* contributors: credit rbr48
* Merge main; rustfmt; one contributor entry for rbr48 (#906 had added one)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Drop a stray merge marker from ROADMAP
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: rbr48 <rajbrur48@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
part that inherits its lines sends it, and the picker only shows No Border checked for it.
- CellProps::valign is Option<VAlign>: a cell's own Top (Align Top, w:vAlign="top") beats a
table style part's vertical alignment, and round-trips through .docx.
- tableStyle edit rejects Part fields at the top level without "part".
- "font": null or "" makes a part's font inherited; clearing the dialog's font box sends it.
- Document ui.dialog's optional "part" in docs/control-protocol.md.
Co-authored-by: Dmitrii Khlapov <19991457+dkhlapov@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
new_draft_restarts_restores_unsaved_and_explicitly_discards failed in a
workspace run: the reopened session found no recovery record, because a
process another test was starting held the record's lock for a moment (a
forked child shares it until exec) and scan() skips locked records. The
same can happen to a user when the app starts a helper process. lock() now
tries again up to ten times, 10 ms apart, before treating the record as
owned by a live instance.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* #714: table style editor formats every table style part
Modify/New Table Style: "Apply formatting to" any of the 13 ECMA-376
§17.7.6 parts (whole table, header/total row, first/last column,
odd/even row and column bands, four corner cells), each with font,
size, bold, italic, underline, colour, fill, borders (line style,
width, colour; all/none/outside/inside/each edge) and the 9-way cell
alignment, over a live preview.
- doc: TablePart, PartFormat, PartSet; TableStyleParts::part/set_part
(new parts and per-part jc/vAlign in `regions`)
- layout: generic part resolution incl. band2, corner cells, alignment
- docx: every tblStylePr type read/written with pPr jc and tcPr vAlign
- engine: table.newStyle/modifyStyle take ECMA part names, `part`,
`parts`, per-edge `borders`, `underline`, `align`/`vAlign`/`cellAlign`
- ui: dialog moved to dialogs_table_style.rs; ui.dialog takes `part`
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* i18n: Hungarian and French strings for this change
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Review fixes for #887: Bulgarian strings for the table style parts
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* #714: Vietnamese strings for the table style parts (main's new vi catalog)
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* hr: match the catalog's word for banded
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Dmitrii Khlapov <19991457+dkhlapov@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
A `DATE` or `TIME` field showed whatever was cached the last time the file was
saved, so a letter written from a template went out carrying the date the
template was saved. Word refreshes those two whenever it shows, prints or
exports a document; WordCraft only did so on an explicit Update Field(s), on
insert, or when printing with "Update fields before printing" on. Headless
`convert` to PDF was the worst of it: nothing hints the date is old.
Refresh them where a document is read instead, in a new `io::opened` beside
`ensure_nonempty` -- the two things a freshly-read document needs -- so `text`,
PDF, rendering and the window all show the same, current value. The model's
stored result is the one source they share; patching each output path would
have needed a clock in two places and still missed the next one.
Narrow on purpose. Only `DATE` and `TIME`: `CREATEDATE`, `SAVEDATE` and
`PRINTDATE` record when something happened and are not "now", and `REF`, `TOC`
and citations stay on explicit update, as in Word. A field carrying `w:fldLock`
keeps its stored result, which is what the lock is for. Headers, footers and
text boxes are covered too, not only the body.
The field code is untouched, so saving writes the field back as it was, with a
current cached result -- as Word writes one.
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Same pattern as Portuguese (#797) and Polish (#346): the cs_CZ
dictionary is read from the system at runtime (e.g. Debian/Ubuntu
hunspell-cs), nothing is bundled. Suggestions use the dictionary's own
MAP/REP/TRY/KEY; grammar stays generic (repeated word, spacing), with
digit ratios like "3 : 1" exempt. The Czech interface now writes new
documents (and opened .txt/.md) in cs-CZ.
Engine fixes the Czech dictionary needs, also used by pl/pt:
- PFX/SFX header lines may carry a trailing comment (cs_CZ has
"SFX y Y 80 #…"); rule lines are left literal;
- NOSUGGEST stems are accepted when typed but never offered;
- KEY is parsed for neighbour-key suggestions.
Tests use a hand-written synthetic Czech dictionary only; an
end-to-end engine test mirrors the Portuguese one.
Co-authored-by: Losminek <301412321+Losminek@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
* Ruler tab stops: corner selector, click to set, drag to move, drag off to clear
The ruler's markers were read-only (double-click opened the Tabs dialog,
click sets (Left, Center, Right, Decimal, Bar; right-click cycles back),
clicking the ruler sets one on the half-division grid (Alt places
freely), dragging moves it with a dashed guide down the page and a
position tooltip, and dragging it off the ruler clears it. Markers draw
per kind, mirrored for right-to-left paragraphs. Each gesture runs
para.tabs once on release, so it is one undo step and inherited-stop
clearing keeps working. Double-click still opens the dialog.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Progress docs: ruler tab stops (ui-parity Rulers row and revision history, ROADMAP progress log)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Tabs dialog: OK applies what the fields show; double-click a stop to edit its position
Selecting a stop, changing its alignment or leader and pressing OK now
applies the change without Set, and a new position typed while a stop
is selected (the dialog preselects the first) is set on OK, so adding a
stop never requires leaving the dialog for the ruler. Double-clicking a
stop in the list edits its position in place: Enter or clicking away
commits (the interface unit by default, any unit spelled out, either
decimal separator), Escape cancels, and the commit moves the stop in
one clear + set, keeping its alignment and leader. Clear still wins
over a re-typed cleared position, and junk input changes nothing.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Tabs dialog: New button, delete from the list, and the list fills its frame
New stages a fresh stop - deselects, proposes the next free multiple of
the default interval after the last stop (skipping cleared positions),
resets alignment and leader, and focuses the position field - and Set
or OK adds it. Every row gets a small x that clears its stop, and the
Delete key clears the selected one when nothing is being typed; the
selection and an in-progress inline edit follow when a row above them
goes. The list is a fixed 150 pt wide and the rows and scrollbar fill
it, so labels stay left-aligned, the x sits at the frame's edge clear
of the scrollbar, and the scrollbar docks at the edge instead of
floating mid-list.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Tabs dialog: the stop list is an inline-edit table
Each row edits its stop in place: a unit-aware position field (rows
resort into position order once the edit ends, never mid-typing; two
stops landing on one position merge, the edited row winning), an
alignment dropdown and a leader dropdown (the closed box shows just the
fill), and the x that clears the row. The double-click editor and
move_stop go: the table subsumes them. The selected row keeps an accent
outline, its edits sync the fields below, and the frame widens to fit
the three columns.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Tabs dialog: the table is the whole editor
New adds a row straight into the table at the next free multiple of
the default interval, selected and ready to edit in place, and the
position field, the Alignment and Leader radio groups and the Set and
Clear buttons go: every one of their jobs is a table cell or the row's
x. What stays is New, Clear All, the default interval, the cleared
list and OK/Cancel, with OK applying the staged diff as one para.tabs
call. The staged-field methods (set, clear, on_ok) go with them.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Review fixes: safe ruler gestures, one tab tolerance, dialog cleanups
Ruler gestures are primary-button only, a click on an indent marker
places nothing (the markers absorb their clicks), Escape cancels a drag
like a table-border drag (#217), and a drag is stamped with its
paragraph: the caret leaving it mid-drag, or a release the ruler never
saw, abandons the gesture instead of applying it elsewhere or late. The
dragged marker no longer double-draws on its first frame. One
TAB_POS_EPS in the doc layer replaces four hardcoded 0.5s across the
engine, the dialog and the ruler. The dialog's normalize is a no-op
unless a position edit happened, its combos share the ruler's KINDS and
one LEADERS table, dead code and a stale doc comment go, and the
ui-parity revision row moves to the table's top. Regression tests cover
each gesture rule.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Selecting one cell (Table Layout > Select > Select Cell, or a click along
its left edge) and dragging its left or right border moved the border in
the whole column: one selected cell wasn't told apart from text selected
inside it. `ResizeTarget::limit_to_selection` now takes the selected cells
from `Session::cell_range` (#637) when they're in the dragged table, so one
selected cell's border moves in its row only, as in Word, like the border
of cells selected across (#720). Text selected within a cell still moves
the whole column.
Tests: engine `a_selected_cells_border_moves_only_in_its_row`; kittest
`a_selected_cells_border_drags_in_its_row_only` (a click along the cell's
edge, then a real drag). Progress docs: ui-parity, ROADMAP.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
A field begin marker flagged w:dirty by a producer that couldn't compute
its result (pandoc-style TOC placeholders) lost the flag on save, so Word
no longer refreshed the field on open. Mirror w:fldLock end to end:
FieldState, InlineObject::Field/FieldStart, the reader (fldChar and
fldSimple) and the writer (buffered, range and TOC-spanning begins).
Update Field/Update Table clear it on the fields they compute, like
Word's F9; untouched fields keep it. Round-trip and engine regression
tests, incl. a pin for #878 (imported TOC placeholder replaced in place,
fixed earlier by #617).
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Backspace at the start of the paragraph after a table nested in a cell
moved the caret into the last cell of the table around it: the search for
the table's last paragraph compared only the first index of the path, which
a nested table shares with its outer table. It now matches the table's
whole path. Regression test: backspace_after_a_nested_table_stays_in_its_cell.
Also adds my contributor entry (rbr48, GitHub profile name).
Co-authored-by: rbr48 <rajbrur48@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
discovery_reclaims_a_crashed_first_write_but_keeps_valid_records failed in
a full workspace run: load() found the record locked for a moment by a
process another test was starting (the race #1939 handled for scan()).
The three tests that reopen a store and load now retry for up to a second.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* feat(web): add option to load system installed fonts in dropdown
Add explicit 'Load System Installed Fonts…' button in font dropdowns for web (Wasm).
- Dispatches font.loadSystem via transient user activation on click
- Calls window.queryLocalFonts() to enumerate and load installed fonts into FontDb
- Invalidates font family cache in FontPreviews upon new font installation
- Complete localization across all 18 languages
- Credit GhostMitra in contributors/people.toml
* fix(ui, engine): canvas focus after chrome click and checked file access in sniff
* Revert "fix(ui, engine): canvas focus after chrome click and checked file access in sniff"
This reverts commit b0c925bf79.
---------
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
failed_write_keeps_last_valid_generation_and_discovery_falls_back hit
the same race as #1849's test (a record locked for a moment by a child
another test spawns is skipped, so scan() is Ok). Both now share a
helper that reopens until the scan reports the unreadable record.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* engine: store replacement images under their true format extension
* rustfmt
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Update Table on a TOC read from a file set the page-number tab at the text width of the
caret's section (after Select All, the last one: a landscape appendix pushed level-one
leaders past the margin), kept the style's number tab stops so a short entry or the last
line of a wrapped one stopped there without its leader, and left the paragraph that held the
field's start empty under the TOC heading. The width now comes from the TOC's own section,
entries without a tab of their own clear the style's left/centre/decimal stops before the
page-number stop, and a range that filled its own paragraphs under a TOC heading puts the
field at the end of that heading, where Table of Contents puts it.
Co-authored-by: dfgodvan <180286157+dfgodvan@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Tables: text running past an exact-height row is cut off
A row with an exact height (w:trHeight w:hRule="exact") keeps that
height whatever its cells hold. Word cuts the text that doesn't fit at
the cell's edge; WordCraft drew it over the rows below, so a certificate's
identification block came out as overlapping lines.
The cell is marked as clipped in layout (Placed::Cell clip), and display
draws its lines inside that cell's clip, on screen and in PDF. Hit
testing is unchanged: the lines stay where they are in the layout.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Tables: keep row clipping compatible with Rust 1.90
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
* Engine: the Word 2003 format check reads through io, so commands still never touch files themselves
is_word_2003_file (#1081) opened the file with std::fs, which the file-access guard test
(commands_do_not_touch_the_file_system_themselves) forbids outside io.rs; main failed it.
A new io::read_file_head reads the first bytes inside the read root.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* UI: a chrome click counts the page as focused at once, so the next keystroke isn't dropped; the ribbon-focus test passes again
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix: DOCのDIB画像変換でバイト上限を守る
* Drop the duplicate read_file_head left by the merge with main
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Dmitrii Khlapov <19991457+dkhlapov@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: haranishi <223786623+haranishi@users.noreply.github.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
review.deleteComment removed anchors from the body only, so a comment
anchored in a header, footer, note or text box left its CommentStart and
CommentEnd behind, pointing at a deleted definition. It also collected only
direct replies, so a reply to a reply survived with its parent gone.
Anchors are now removed from every story, and the deleted set is the whole
thread below the comment. The thread is collected from a parent-to-replies
map built once, with a seen set, so a parent cycle from a hostile file ends
and a long reply chain doesn't take quadratic time. Undo restores everything.
Regression tests: a header comment (anchors checked in the header story), a
three-level reply chain (with its text parts and Undo), a parent cycle, and a
50,000-reply chain.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
The model's paragraph walk (each_para / each_para_mut) stopped 16 tables
down, while a .docx opens with tables nested up to 24 deep and pasting can
nest them further. Text box pruning, which runs after every edit, took a
box shown only from deeper for one shown nowhere and dropped its story,
so an unrelated edit lost the box's text. The same walk decides which
boxes a copy takes along and which pasted shapes get stories of their
own, and serves reading order, the next drawing id and note conversion.
The walk now keeps its own stack instead of recursing, so it reaches
every paragraph however deep tables nest, in the same document order and
with no call stack to run out of; a paragraph on its own still needs no
allocation. The signature is unchanged (the depth argument is now
unused), so every caller builds as before.
Tests: a box 17 tables down in an opened .docx survives an unrelated
edit and a save, while a box shown nowhere still goes; boxes 17 and 40
tables down, in a box and in a header are kept, counted in reading order,
and new drawing ids go above the deep shape's; a copied deep table
pastes into another document showing its own box's text.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
divyanshu2074