Remember the session highlighter choice and toggle it when no explicit colour is given. Keep explicit palette/automation setters and No Color semantics, and preserve undoable formatting.
Two command/key regressions and four headless UI screenshots passed. Full cargo xtask ci passed all 8 steps with 3962 passed, 0 failed, 4 existing ignored using RUST_TEST_THREADS=1. Initial parallel run hit an unchanged Russian suggestion timing assertion; isolated unchanged binary passed at13ms. Native Word comparison pending.
With Track Changes on, the border-line AutoFormat (`---` and the others)
and the `+---+` table AutoFormat removed the typed characters outright, so
Reject All could not bring them back. At the top of a story the border
was not recorded as a formatting change either.
Both AutoFormats now delete the characters through the tracked deletion
that Backspace uses: characters the same author inserted are removed, and
other characters are marked deleted. A border at the top of a story is
recorded as a tracked formatting change, and the caret goes after any
characters marked deleted. The table is inserted untracked, as Insert >
Table inserts one. A line that holds a tracked deletion no longer
triggers either AutoFormat, so Enter after the deleted characters is a
plain Enter.
AutoFormat As You Type gains "Built-in Heading styles", off by default,
as `headingStyles` on `tools.autocorrect` and a checkbox under "Apply as
you type" in the AutoCorrect dialog.
With it on, Enter in an empty paragraph right after a line of one to five
words that ends in a letter or digit makes that line Heading 1. One or two
tabs before the line make it Heading 2 or 3, and the tabs are removed. The
line must be a Normal paragraph of the main text outside a list. The caret
stays in the empty paragraph after the heading, and Undo right after
brings back the tabs and the Normal style while the Enter stays. With Track
Changes on, the tabs are marked deleted (or removed, when the same author
inserted them) and the style change is a tracked formatting change; a line
holding a tracked deletion stays as it is. A heading style locked by
formatting restrictions is not applied.
type_text with paragraphs is one text.insert now, and insert_text turns a
carriage return into a paragraph break whatever it is inside, so a
one-line plain-text control, where Enter does nothing, was split in two
(text.insert of "a\rb" had always done that; the per-line way never sent
the break because text.newParagraph refused it). Typed text now leaves
the breaks out there, as the Enter key does; pasted text keeps its own.
The #1356 test (a refused paragraph break is reported and nothing is typed
after it) is now about a refused insert: the lines are one command, so a
protected document refuses all of them at once.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
AutoFormat As You Type gains "Set left- and first-indent with tabs and
backspaces", off by default, as `tabIndent` on `tools.autocorrect` and a
checkbox under "Automatically as you type" in the AutoCorrect dialog.
With it on, Tab before the first line of a paragraph with text indents the
first line by one default tab stop, and Tab before a later line (after a
line break or where the text wraps) indents the whole paragraph. A second
Tab before an indented or hanging first line also indents the whole
paragraph. Backspace at the start of the paragraph takes the first-line
indent back, then the paragraph's own left indent one tab stop at a time.
An empty paragraph and Tab inside a line still take a tab character.
While formatting is limited to the document's styles, Tab and Backspace
work as with the option off.
It was a text.newParagraph and a text.insert per line. Each Session::run
that edits snapshots the whole document for Undo and for rollback, so a
command costs about the document's size: 10,000 lines took 2.8 s through
MCP (release build; doubling the lines cost 2.4-3.2x), and left an Undo
step per line. text.insert already takes a carriage return as a paragraph
break, so the lines go in as one command: ~10 ms for 10,000 (linear), one
Undo step.
Measured per call on a 10,000-paragraph document (release, microseconds):
text.newParagraph 188 in run (61 + 66 for the two snapshots, 53 in the
handler), text.insert of one character 85 (47 snapshot, 31 handler), and
caret.docEnd 261, all in its handler (it finds the last paragraph by
listing them all). The snapshots are the bulk, which is why a command per
line cannot be linear; this does not change them.
Behaviour: a CRLF is one break, not two (the per-line way inserted the
carriage return as a break and then added its own). The breaks no longer
run Enter's AutoCorrect: the per-line way capitalised every line but the
last (no Enter followed it) and fixed typos there, so what came out
depended on the line's place; the text now goes in as written, as the
tool's raw insert already did within each line.
Test: the paragraphs made equal those of the per-line way (in a list and
not, blank lines, leading and trailing breaks, indents, tabs) for text
AutoCorrect has nothing to change; lowercase lines and a typo stay as
written; CRLF; one Undo removes every line. Fails on the old code.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The spelling check found the start of the white-space-delimited token around
a word one byte after the last white-space character. After a no-break space
(two bytes in UTF-8) or an ideographic space (three bytes) that offset is
inside the character, so the token could not be read and was never classified
as a web or e-mail address. With Ignore web and file addresses on, the words
of such an address were still marked as misspelled.
The token now starts after the whole white-space character. A proofing test
covers a web address, a www. address and an e-mail address after the two
spaces, with the option on and off.
ROADMAP.md and docs/typing-parity.md now carry 2026-10-24 in their status
lines and a revision history row for the AutoCorrect fix. docs/gaps.md has
no entry for it.
After a `*word*` or `_word_` span is converted on a space, punctuation
or Enter, the word without its marks is AutoCorrected as before:
replacements, capitals, ordinals and links (`*teh* ` becomes bold
`The `). Undo right after still restores the marks in one step, and
after Enter the Enter stays. A marked web address becomes a link
without carrying its bold or italic into the text typed after it.
* chore: resolve SonarCloud maintainability issues
* fix: restore compiling refactors, revert behaviour changes, run cargo fmt
- fontdb::norm: keep the closure for filter (char::is_alphanumeric takes
self, not &self, so it does not fit Iterator::filter).
- story.rs: write_mark_rpr needs &mut self (rev_attrs, rpr_change).
- Revert the viewport meta change and the return 0 added to
workspace_version/sha256, which changed behaviour or masked failures.
- cargo fmt.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Import Document in odt page_parts after dropping glob re-export
* fix: apply_kind_properties binds ContentControl; cell_ref keeps char closures
* fix(engine): pass case helpers by value (clippy needless borrows)
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Dylan Taylor <dylan@craftmaker.dev>
AutoCorrect found the start of the word before the caret one byte after the
last white-space character. After a no-break space (two bytes in UTF-8) or an
ideographic space (three bytes) that offset is inside the character, so the
word could not be read and AutoCorrect left it as typed: no replacement, no
sentence capital, no ordinal superscript and no link.
The word now starts after the whole white-space character. A typing test
covers a replacement, a sentence capital and a web address after each of the
two spaces, and an ordinal after each.
AutoFormat As You Type has a new option, "*Bold* and _italic_ as real
formatting". When it is on and a word is finished by a space,
punctuation or Enter, text enclosed in asterisks becomes bold and text
enclosed in underscores becomes italic, for plain and complex-script
text, and the marks are removed. The opening mark has to start a word
and the enclosed text may not start or end with a space, so 2*3* and
file_name_ stay as typed, and so do marks around a picture, field or
content control.
Undo right after brings the marks back; after Enter, the Enter stays.
The paragraph Enter makes is formatted as after a plain Enter. With
Track Changes on, only marks the current author typed as tracked
insertions are converted, and nothing is converted while formatting is
limited to the document's styles.
The option is off by default. Scripts set it with "boldItalic" on
tools.autocorrect, and the AutoCorrect dialog shows it on the AutoFormat
As You Type tab.
The wrap areas of footers and the table areas of headers are looked up
per page, and on a cache miss the code asked Document::sections() for
the section properties — a walk of the whole body. A document whose
every paragraph starts a page (200 k one-line paragraphs, 2 MB) took
68 s to lay out, 84 % of it in those walks: the cost grew with pages x
paragraphs. Carry the section list on the Ctx instead — one walk per
pass — and let both sites read it there; the document does not change
during a pass, so the list is the same.
* #1365: PDF: pin craft-libs main, whose Krilla leaves out simple-text (no rustybuzz, ttf-parser)
craft-libs 8f5a2831 turns Krilla's default features off in the shared
workspace and keeps raster-images, so simple-text (the only path to the
unmaintained rustybuzz and ttf-parser) is no longer built. WordCraft
shapes text itself and hands Krilla positioned glyphs. The vendored Krilla
and craft-pdf's writer are unchanged between the two revisions.
Cargo.lock loses rustybuzz, ttf-parser, core_maths,
unicode-bidi-mirroring, unicode-ccc and unicode-properties.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* #1365: docs/shared-pdf.md: rewrap the backend features paragraph
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* PDF: pin craft-libs main, which bounds decoded PNG images in Krilla
* PDF: pin craft-libs 9f75d3dc, with the PDF input bounds of craft-libs#254
---------
Co-authored-by: rabis <340973947+rabis7575@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Each of these languages accepted every word in capitals of up to six letters and every word
with a digit on its own, so turning off Ignore words in UPPERCASE or Ignore words that contain
numbers had no effect on them for short words. The check now takes the options, as the Spanish
one does; the direct is_correct/is_correct_with keep the defaults.
Co-authored-by: Federico Olivares <FedeOlivares@users.noreply.github.com>
Co-authored-by: aiongg <67543387+aiongg@users.noreply.github.com>
Use the pre-spacing origin only for paragraph-relative box floats. Expand
regression coverage for line anchors and frameless page/margin fallbacks,
and document the cap conversion helpers and synthetic regression contracts.
Merge upstream main through 478fd445 and reconcile the roadmap progress log.
Native CI stages passed; the WASM gate passed on a targeted rerun after a
local sandbox linker adjustment. Optional catalog/font inputs were absent.
Co-Authored-By: GPT-6 <codex@openai.com>
The Comments pane showed the file's UTC time and the balloons showed no time. Both now
show each comment's and reply's time under its author in local time, as Word does. The
file's UTC time (Word's w16cex:dateUtc) wins over w:date, a time naming its zone is
converted and one without a zone shows as written. The desktop app reads the zone with
chrono, the web app asks the browser.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Maintainer review on #3562: opening_marks reverse-scans the characters before a
question or exclamation mark, and ends_clause asked in_address for each one,
which scanned every address range, so a paragraph of thousands of addresses
followed by a question took quadratic time (about 99 KB: hundreds of ms).
address_at() finds the range a byte is in with a binary search over the
sorted, disjoint ranges, and also gives ends_clause the range's end without
another scan. Behaviour is unchanged (the URL and punctuation regressions
still pass).
- address_lookup_finds_the_range_a_byte_is_in: boundaries of the lookup
- many_addresses_before_a_question_stay_linear: 99 KB of addresses before a
question or exclamation (the review's reproducer and two variants), checked
for the right issue range and for taking under 75 ms (about 1 to 5 ms now,
about 300 ms with a scan per character)
Carry flat, square and round outline caps through ShapeExtra, DrawingML and VML
import, DOCX export, display items, raster strokes and PDF strokes. Keep older
JSON documents default-flat and preserve WordCraft ink's tool-defined caps.
Add JSON, DOCX/VML round-trip, raster-pixel and PDF-operator regressions. Record
the verified coverage and remaining theme/native Word checks in the parity docs.
Co-Authored-By: GPT-6 <codex@openai.com>
Use the paragraph origin before effective space-before when placing floats in
headers, footers, table cells and text boxes. Keep text and wrap exclusions
relative to the text top, matching the existing body layout convention.
Cover all four box contexts with synthetic layout regressions and update the
gap, parity and progress records. Native Word comparison remains pending.
Co-Authored-By: GPT-6 <codex@openai.com>
Upstream merged #3177 (dictionary hints for every Hunspell language) and #3480.
Resolve the conflicts in crates/proof/src/lib.rs by taking #3177's
dictionary_loaded/dictionary_needed structure and adding the four Spanish
varieties to it, and in docs/spelling-dictionaries.md keeping both sides.
Catalog rows from #3480 and #3177 move to craft-i18n (storytold/craft-i18n sync/wordcraft-topup); the pin moves to 04600cb.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>