chore(deps): bump the rust-deps group with 2 updates

[dependabot]

Bumps the rust-deps group with 2 updates: office_oxide and thiserror.

Updates office_oxide from 0.1.11 to 0.1.12

Release notes

Sourced from office_oxide's releases.

v0.1.12 | Read-path fidelity across all six formats, and the first real fidelity work on the legacy binaries. 141 issues closed, every dependency at its latest release, and a corpus verification that compares against independent readers, not just the previous version.

Verified

Two passes over ~/projects/office_oxide_tests/ — 10,599 real Office documents, all six formats plus their macro/template/strict/encrypted variants, 38 sources.

Pass 1 — release regression sweep, after every dependency was bumped. v0.1.11 and this branch, each over every file on four surfaces (text, markdown, html, ir) — 42,396 observations per arm, compared on status, on word multisets with split/join awareness (a \w+ tokenizer misreads restored entities as loss), and on timing, using scripts/regression-sweep/sweep_tree.py + compare_tree.py.

	v0.1.11	0.1.12
ok observations	40,202	40,716
crashes (SIGKILL/abort)	5	0
timeouts (120 s)	5	0
err	2,184	1,680

Every ok → err (20 files) replaced a 0-byte "success": 7 encrypted .pptx now say so (#296, #320), 12 .ppt files with no readable PowerPoint stream (PowerPoint 4.0/95-only files and fuzzer artefacts) now say so (#351), and POI's own duplicate-entry .xlsx is refused rather than silently resolved to one copy. Every err → ok (146 files) is a macro-enabled or template extension that used to be rejected (#282). The sweep's first run was not clean — it found the ir memory regression (#349), the TOC-entries regression (#350) and two smaller ones (#355, #356) that this release had introduced, all fixed before this section was written.

Pass 2 — quorum check against independent readers. The same corpus through Apache Tika 3.3.2 (POI), python-docx 1.2.0, pandoc, openpyxl 3.1.5, python-calamine, xlrd 2.0.2, catdoc, antiword, xls2csv and catppt (scripts/regression-sweep/refpanel.py + quorum.py). A file is flagged when at least two references agree with each other while office_oxide recovers under 70 % of their words. This is the pass a two-arm diff cannot do, and it found the defects present in v0.1.11 too: 80 .xls files at 2–58 % of every reference — one shared-string parsing bug (#353) — plus #352, #357, #358. Flagged files went from 149 to 64; the .xls odd-one-out count from 80 to 0. What remains flagged is explained and deliberate: Word 6.0/95 .doc rejected with a clear error (23), hidden text (w:vanish, including style-inherited) and tracked deletions the reference tools show and Word does not, .xlsx sheet names (#359), and fuzzer artefacts.

Pass 3 — performance and rendering, after the fixes above. The same corpus through perf_sweep.py on both arms (zero status regressions; every timing flag re-measured serially), the quorum re-run against Tika, the python libraries and the C tools (74 flagged files, every one explained: hidden text and tracked deletions the references show, formulas vs cached values, references that returned nothing, Word 6/95 and .xlsb), and a structural comparison of our markdown against pandoc's GFM over the 3,521 .docx files, which found the three renderer defects listed under Fixed.

Pass 4 — one harness, one verdict per file (#401). The passes above became scripts/regression-sweep/harness.py: both arms, the reference panel, a write→reread round trip and the markdown-structure comparison over the corpus, reduced to a per-file verdict with a stated decision model — the consensus of references that agree with each other is the expected output; a two-arm loss counts only when the consensus has the words and the new arm does not; an arm that sides with a different reference than the other arm is a policy split, not a regression; w:vanish runs and field vocabulary explain a loss on their own. It added two axes a two-arm diff and a reference quorum are both blind to: our own text, markdown and html of one file compared with each other (the dual-renderer class), and the words of the original IR against those of the IR read back from what we wrote. Those found the defects under Fixed from #381 on: every docx/headers_footers/* file lost its headers on a round trip, every picture below body level was dropped on write, formulas became constants, .doc fast-saved files had their paragraphs re-cut at piece boundaries and their tracked deletions shown on the structured surfaces only, and no markdown renderer escaped document text.

Test-coverage audit. Every one of the milestone's 162 issues was mapped to the tests its fix commits added (plus a topic search for the ones committed without a number) and judged against the cases its report enumerates. Gaps found and closed: endnote and comment bodies in plain_text()/to_markdown() (only footnotes were asserted), the DOCX side of numbered headings, paragraph-style-inherited w:vanish, FORMTEXT default values, CR-only runs' xml:space, docProps/app.xml and has_macros on PPTX and XLSX and a legacy _VBA_PROJECT storage, XLSX plain_text() sheet names, and a README-vs-clap consistency test for the CLI section (which found the undocumented replace --output). The has_macros gap turned out to be a real defect (#380).

Plus: 1,380 library and workspace tests green (--all-features), clippy -D warnings clean, rustdoc -D warnings clean, cargo fmt --check clean, cargo audit and cargo deny clean, C# (xUnit runner 4.0) and Node (koffi 3.3.1) suites green, and the house-rules gate green.

Added

• Word 6.0/95 text (#374). The family was refused outright while catdoc, antiword, LibreOffice and POI read it. Its own FIB layout, no table stream, one-byte characters in the document's code page (or two-byte units under fExtChar, the Macintosh and Far East builds), non-complex and fast-saved files, subdocuments included; 15 of the 17 openable corpus files at 0.90–1.00 recall against catdoc, antiword and Tika. Text only — no paragraph or table structure.
• Excel Binary Workbooks (.xlsb) (#375). The BIFF12 parts — sheet list, shared strings, number formats and cell formats, every cell and formula-value record, merged ranges — decoded into the same model as .xlsx, so every converter and renderer applies; 29 of the 35 openable corpus files at 1.00 recall against python-calamine, the rest differing only in value representation. Cached values only (no formula text); fonts, fills and borders not read.
• Excel 2.x–4.0 (BIFF2/3/4) worksheets (#376). The pre-OLE2 record stream — refused by name before — is read as a one-sheet workbook: the version's own cell records, CODEPAGE text, formula results via STRING; 1.00 recall against xlrd on the corpus files. Number formats are not applied.
• Legacy .doc structure. Footnotes and endnotes as one IR element each (#286); comments split per comment via PlcfandTxt/PlcfandRef with authors from GrpXstAtnOwners (#345, #298); header/footer stories split via PlcfHdd (#285); CHPX character formatting and PAP alignment/indent/spacing (#287); list identity through PlfLst/PlfLfo for start_number and ordered/unordered (#250); embedded OLE objects under ObjectPool (#284); subdocument bodies in plain_text()/to_markdown() (#248); HYPERLINK field targets with field instructions stripped (#249).
• Legacy .ppt structure. Direct character/paragraph formatting from StyleTextPropAtom (#254); placeholder formatting inherited from the master's TextMasterStyleAtom (#335); hyperlinks from InteractiveInfo/ExHyperlink (#257); grid-of-shapes tables reconstructed as Element::Table (#255); hidden slides (#297); header/footer/date text (#308); embedded and linked OLE objects (#337).
• Legacy .xls structure. Cell comments from NOTE/TXO/OBJ (#307); hyperlinks from HLINK (#306); embedded chart text via SeriesText (#246); pre-OLE2 BIFF2/3/4 files named instead of failing at the CFB layer (#227) — and, later in this release, read (#376).
• Across formats. Metadata::has_macros for all six formats (#283); SummaryInformation title/author/dates for .doc/.xls/.ppt (#244); docProps/app.xml read on open (#245); conditional formatting as a new IR type for .xlsx/.xls (#252); data validation rules (#275, #343); defined names reach to_ir() (#251); placeholder role (subtitle/date/footer/…) on PPTX and PPT shapes (#258); PPTX placeholder formatting inherited from <p:txStyles> (#291).
• Writers. Hyperlinks in XLSX and PPTX output — a total, unconditional loss before (#262); speaker-note formatting and list structure carried through read, write and every renderer (#290); table-cell character formatting written and read back from inline rich strings (#346); cell comments written as real comments (#344).
• Renderers. to_html_with() with base64 image embedding (#318); underline, super/subscript, highlight, colour, font and caps in HTML (#314); List.start_number honoured in HTML and markdown (#315); adjacent same-style lists merged (#317).
• examples/corpus_validate, the per-step JSONL harness behind the verification above.
• scripts/house-rules-check.sh and its CI job: every test function is named test_*, no allow(dead_code), no issue numbers in code or comments.

Fixed

• DOCX read. Footnote/endnote/comment bodies in plain_text()/to_markdown() (#240); footnote/endnote/comment reference markers reach to_ir(), so a note is anchored to where it was cited (#241); HYPERLINK field codes via w:fldChar/w:instrText and w:fldSimple (#267); a hyperlink carrying both r:id and w:anchor keeps its external target (#242); OMML equation text (#270); WordArt v:textpath text (#274); legacy VML v:imagedata images (#268); form-field checkbox/dropdown/default state from w:ffData (#276); embedded chart title/category/series/data text (#273); SmartArt diagram text (#271); embedded native OOXML packages (#304); OMML equation text and mc:AlternateContent (#272); w:vanish hidden runs excluded (#305); sdt-wrapped cells/rows kept, tracked-change cellDel excluded (#277, #266); text box with prstGeom no longer double-extracted (#263); interrupted numbered lists keep counting (#243); <w:startOverride> read back (#260); numbered headings emitted as Heading, not ListItem (#223); outline-level headings keep their other paragraph properties, cell <w:jc> maps to text_align, thematic breaks become Element::ThematicBreak (#215); nested-table recursion bounded (#329); the default to_markdown() renderer increments ordered-list numbers instead of repeating the start value (#316); truncated_subtrees() reports content dropped past the depth bound (#218); malformed dcterms:created/modified normalised or dropped (#217).
• DOCX write. Hyperlink relationships scoped to the part that uses them (#293); r:id + w:anchor both kept (#292); custom footnote marks round-trip (#219); table caption no longer duplicates each round trip (#311); grid width from col_span sums (#265); nested lists share one numId across levels but each nested-in-container list gets its own (#261, #339); title heading not duplicated (#338, #259); tab/CR/LF-only runs get xml:space="preserve" (#294); image and font parts get a Default content type (#295); pretty-print indentation removed — it was Θ(depth²) memory on the write path (#220); rarely-populated sub-structs boxed to cut per-element memory (#328).
• XLSX. Shared-string run-level rich text (#303); inline-string whitespace (#269); inter-run spaces and scatter/bubble points (#280, #281); formula text through the IR and renderer (#279); shared-formula follower cells keep their formula text (#278); plain_text() includes the chart text to_markdown() already had (#331); merged cells reach to_ir() (#235); threaded comments preferred over legacy boilerplate (#301); in-cell rich-value images instead of a fabricated #VALUE! (#302); date-serial conversion bounded, UTF-16 XML parts decoded (#225, #229); accounting negatives, trailing literals and scientific exponents in number formats (#234); a cell keeps its column after a row-spanning merge (#340).
• XLS. Embedded chart's nested EOF no longer closes the parent sheet (#237); DATEMODE 1904 threaded through (#233); declared CODEPAGE applied to BIFF5 text (#309); record-parsing budget raised and truncation surfaced (#236); date walk bounded and format overrides honoured (#225); merged cells (#235); hidden and very-hidden sheets are kept and flagged rather than dropped, as XLSX already did (#231).
• PPTX. Speaker notes routed to Section.speaker_notes (#238); PowerPoint Sections' p14:sldIdLst no longer overwrites the slide list (#228); embedded chart text (#239); merged-cell tables rendered as a proper grid (#289); shape click actions and non-text AutoShape alt text reach the IR (#299, #300); body/title placeholder content flows instead of being wrapped in a positioned TextBox (#342); TextBox written as its own positioned shape (#264); table cells and bullet items get real run formatting on write (#341).

... (truncated)

Changelog

Sourced from office_oxide's changelog.

[0.1.12] - 2026-09-21

Read-path fidelity across all six formats, and the first real fidelity work on the legacy binaries. 141 issues closed, every dependency at its latest release, and a corpus verification that compares against independent readers, not just the previous version.

The legacy .doc/.xls/.ppt readers went from "extracts the text" to structured: per-note footnotes, endnotes, comments and headers/footers in .doc; character and paragraph formatting from CHPX/PAPX; list identity via PlfLst/PlfLfo; .ppt direct formatting, master inheritance, hyperlinks, hidden slides, OLE objects and grid-of-shapes tables; .xls comments, hyperlinks, data validation, chart text and the 1904 date system. On the OOXML side the recurring shape was parsed-then-discarded: the parser had it, a converter or one of the four renderers dropped it — rich-text runs, merged cells, defined names, formula text, speaker-note formatting, OMML equations, SmartArt, embedded charts, shape click actions.

Verified against 10,599 real Office documents — arm-to-arm with v0.1.11 (5 crashes and 5 timeouts in the baseline, 0 here) and against ten independent readers, which is how the largest content-loss defect in the crate's history (80 .xls files at 2–58 % of every reference) was found and fixed before this shipped.

Verified

Two passes over ~/projects/office_oxide_tests/ — 10,599 real Office documents, all six formats plus their macro/template/strict/encrypted variants, 38 sources.

Pass 1 — release regression sweep, after every dependency was bumped. v0.1.11 and this branch, each over every file on four surfaces (text, markdown, html, ir) — 42,396 observations per arm, compared on status, on word multisets with split/join awareness (a \w+ tokenizer misreads restored entities as loss), and on timing, using scripts/regression-sweep/sweep_tree.py + compare_tree.py.

	v0.1.11	0.1.12
ok observations	40,202	40,716
crashes (SIGKILL/abort)	5	0
timeouts (120 s)	5	0
err	2,184	1,680

Every ok → err (20 files) replaced a 0-byte "success": 7 encrypted .pptx now say so (#296, #320), 12 .ppt files with no readable PowerPoint stream (PowerPoint 4.0/95-only files and fuzzer artefacts) now say so (#351), and POI's own duplicate-entry .xlsx is refused rather than silently resolved to one copy. Every err → ok (146 files) is a macro-enabled or template extension that used to be rejected (#282). The sweep's first run was not clean — it found the ir memory regression (#349), the TOC-entries regression (#350) and two smaller ones (#355, #356) that this release had introduced, all fixed before this section was written.

Pass 2 — quorum check against independent readers. The same corpus through Apache Tika 3.3.2 (POI), python-docx 1.2.0, pandoc, openpyxl 3.1.5, python-calamine, xlrd 2.0.2, catdoc, antiword, xls2csv and catppt (scripts/regression-sweep/refpanel.py + quorum.py). A file is flagged when at least two references agree with each other while office_oxide recovers under 70 % of their words. This is the pass a two-arm diff cannot do, and it found the defects present in v0.1.11 too: 80 .xls files at 2–58 % of every reference — one shared-string parsing bug (#353) — plus #352, #357, #358. Flagged files went from 149 to 64; the .xls odd-one-out count from 80 to 0. What remains flagged is explained and deliberate: Word 6.0/95 .doc rejected with a clear error (23), hidden text (w:vanish, including style-inherited) and tracked deletions the reference tools show and Word does not, .xlsx sheet names (#359), and fuzzer artefacts.

Pass 3 — performance and rendering, after the fixes above. The same corpus through perf_sweep.py on both arms (zero status regressions; every timing flag re-measured serially), the quorum re-run against Tika, the python libraries and the C tools (74 flagged files, every one explained: hidden text and tracked deletions the references show, formulas vs cached values, references that returned nothing, Word 6/95 and .xlsb), and a structural comparison of our markdown against pandoc's GFM over the 3,521 .docx files, which found the three renderer defects listed under Fixed.

Pass 4 — one harness, one verdict per file (#401). The passes above became scripts/regression-sweep/harness.py: both arms, the reference panel, a write→reread round trip and the markdown-structure comparison over the corpus, reduced to a per-file verdict with a stated decision model — the consensus of references that agree with each other is the expected output; a two-arm loss counts only when the consensus has the words and the new arm does not; an arm that sides with a different reference than the other arm is a policy split, not a regression; w:vanish runs and field vocabulary explain a loss on their own. It added two axes a two-arm diff and a reference quorum are both blind to: our own text, markdown and html of one file compared with each other (the dual-renderer class), and the words of the original IR against those of the IR read back from what we wrote. Those found the defects under Fixed from #381 on: every docx/headers_footers/* file lost its headers on a round trip, every picture below body level was dropped on write, formulas became constants, .doc fast-saved files had their paragraphs re-cut at piece boundaries and their tracked deletions shown on the structured surfaces only, and no markdown renderer escaped document text.

Test-coverage audit. Every one of the milestone's 162 issues was mapped to the tests its fix commits added (plus a topic search for the ones committed without a number) and judged against the cases its report enumerates. Gaps found and closed: endnote and comment bodies in plain_text()/to_markdown() (only footnotes were asserted), the DOCX side of numbered headings, paragraph-style-inherited w:vanish, FORMTEXT default values, CR-only runs' xml:space, docProps/app.xml and has_macros on PPTX and XLSX and a legacy _VBA_PROJECT storage, XLSX plain_text() sheet names, and a README-vs-clap consistency test for the CLI section (which found the undocumented replace --output). The has_macros gap turned out to be a real defect (#380).

Plus: 1,380 library and workspace tests green (--all-features), clippy -D warnings clean, rustdoc -D warnings clean, cargo fmt --check clean, cargo audit and cargo deny clean, C# (xUnit runner 4.0) and Node (koffi 3.3.1) suites green, and the house-rules gate green.

Added

• Word 6.0/95 text (#374). The family was refused outright while catdoc, antiword, LibreOffice and POI read it. Its own FIB layout, no table stream, one-byte characters in the document's code page (or two-byte units under fExtChar, the Macintosh and Far East builds), non-complex and fast-saved files, subdocuments included; 15 of the 17 openable corpus files at 0.90–1.00 recall against catdoc, antiword and Tika. Text only — no paragraph or table structure.
• Excel Binary Workbooks (.xlsb) (#375). The BIFF12 parts — sheet list, shared strings, number formats and cell formats, every cell and formula-value record, merged ranges — decoded into the same model as .xlsx, so every converter and renderer applies; 29 of the 35 openable corpus files at 1.00 recall against python-calamine, the rest differing only in value representation. Cached values only (no formula text); fonts, fills and borders not read.
• Excel 2.x–4.0 (BIFF2/3/4) worksheets (#376). The pre-OLE2 record stream — refused by name before — is read as a one-sheet workbook: the version's own cell records, CODEPAGE text, formula results via STRING; 1.00 recall against xlrd on the corpus files. Number formats are not applied.
• Legacy .doc structure. Footnotes and endnotes as one IR element each (#286); comments split per comment via PlcfandTxt/PlcfandRef with authors from GrpXstAtnOwners (#345, #298); header/footer stories split via PlcfHdd (#285); CHPX character formatting and PAP alignment/indent/spacing (#287); list identity through PlfLst/PlfLfo for start_number and ordered/unordered (#250); embedded OLE objects under ObjectPool (#284); subdocument bodies in plain_text()/to_markdown() (#248); HYPERLINK field targets with field instructions stripped (#249).
• Legacy .ppt structure. Direct character/paragraph formatting from StyleTextPropAtom (#254); placeholder formatting inherited from the master's TextMasterStyleAtom (#335); hyperlinks from InteractiveInfo/ExHyperlink (#257); grid-of-shapes tables reconstructed as Element::Table (#255); hidden slides (#297); header/footer/date text (#308); embedded and linked OLE objects (#337).
• Legacy .xls structure. Cell comments from NOTE/TXO/OBJ (#307); hyperlinks from HLINK (#306); embedded chart text via SeriesText (#246); pre-OLE2 BIFF2/3/4 files named instead of failing at the CFB layer (#227) — and, later in this release, read (#376).
• Across formats. Metadata::has_macros for all six formats (#283); SummaryInformation title/author/dates for .doc/.xls/.ppt (#244); docProps/app.xml read on open (#245); conditional formatting as a new IR type for .xlsx/.xls (#252); data validation rules (#275, #343); defined names reach to_ir() (#251); placeholder role (subtitle/date/footer/…) on PPTX and PPT shapes (#258); PPTX placeholder formatting inherited from <p:txStyles> (#291).
• Writers. Hyperlinks in XLSX and PPTX output — a total, unconditional loss before (#262); speaker-note formatting and list structure carried through read, write and every renderer (#290); table-cell character formatting written and read back from inline rich strings (#346); cell comments written as real comments (#344).
• Renderers. to_html_with() with base64 image embedding (#318); underline, super/subscript, highlight, colour, font and caps in HTML (#314); List.start_number honoured in HTML and markdown (#315); adjacent same-style lists merged (#317).
• examples/corpus_validate, the per-step JSONL harness behind the verification above.
• scripts/house-rules-check.sh and its CI job: every test function is named test_*, no allow(dead_code), no issue numbers in code or comments.

Fixed

• DOCX read. Footnote/endnote/comment bodies in plain_text()/to_markdown() (#240); footnote/endnote/comment reference markers reach to_ir(), so a note is anchored to where it was cited (#241); HYPERLINK field codes via w:fldChar/w:instrText and w:fldSimple (#267); a hyperlink carrying both r:id and w:anchor keeps its external target (#242); OMML equation text (#270); WordArt v:textpath text (#274); legacy VML v:imagedata images (#268); form-field checkbox/dropdown/default state from w:ffData (#276); embedded chart title/category/series/data text (#273); SmartArt diagram text (#271); embedded native OOXML packages (#304); OMML equation text and mc:AlternateContent (#272); w:vanish hidden runs excluded (#305); sdt-wrapped cells/rows kept, tracked-change cellDel excluded (#277, #266); text box with prstGeom no longer double-extracted (#263); interrupted numbered lists keep counting (#243); <w:startOverride> read back (#260); numbered headings emitted as Heading, not ListItem (#223); outline-level headings keep their other paragraph properties, cell <w:jc> maps to text_align, thematic breaks become Element::ThematicBreak (#215); nested-table recursion bounded (#329); the default to_markdown() renderer increments ordered-list numbers instead of repeating the start value (#316); truncated_subtrees() reports content dropped past the depth bound (#218); malformed dcterms:created/modified normalised or dropped (#217).

... (truncated)

Commits

• 642ea18 Release v0.1.12 — legacy-format structure, 183 read/write fidelity fixes, eve...
• 14ce2cf chore(ci): bump actions/attest-build-provenance from 4.1.1 to 4.2.2 (#129)
• 9147f92 chore(deps): bump koffi from 3.1.6 to 3.2.1 in /js (#198)
• 719e0da chore(deps): bump encoding_rs from 0.8.35 to 0.8.41 (#222)
• See full diff in compare view

Updates thiserror from 2.0.20 to 2.0.21

Release notes

Sourced from thiserror's releases.

2.0.21

• Fix parsing of generic unit variants in display expressions (#459)

Commits

• b1827ee Release 2.0.21
• 58037b5 Merge pull request #459 from dtolnay/turbofish
• f82a0cf Keep track of nested turbofish depth
• 72ea492 Raise required compiler to Rust 1.77
• 72eea0d Resolve io_other_error clippy lint in tests
• 07f09a2 Raise required compiler to Rust 1.74
• 2715388 Update ui test suite to nightly-2026-09-22
• 5a306c7 Update ui test suite to nightly-2026-09-05
• ef9383b Update ui test suite to nightly-2026-08-22
• 8336b84 Update ui tests for version 2.0.20
• See full diff in compare view

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.

---

Dependabot commands and options

You can trigger Dependabot actions by commenting on this PR:

• @dependabot rebase will rebase this PR
• @dependabot recreate will recreate this PR, overwriting any edits that have been made to it
• @dependabot show <dependency name> ignore conditions will show all of the ignore conditions of the specified dependency
• @dependabot ignore <dependency name> major version will close this group update PR and stop Dependabot creating any more for the specific dependency's major version (unless you unignore this specific dependency's major version or upgrade to it yourself)
• @dependabot ignore <dependency name> minor version will close this group update PR and stop Dependabot creating any more for the specific dependency's minor version (unless you unignore this specific dependency's minor version or upgrade to it yourself)
• @dependabot ignore <dependency name> will close this group update PR and stop Dependabot creating any more for the specific dependency (unless you unignore this specific dependency or upgrade to it yourself)
• @dependabot unignore <dependency name> will remove all of the ignore conditions of the specified dependency
• @dependabot unignore <dependency name> <ignore condition> will remove the ignore condition of the specified dependency and ignore conditions

[dependabot]

Labels

The following labels could not be found: dependencies, rust. Please create them before Dependabot can add them to a pull request.

Please fix the above issues or remove invalid values from dependabot.yml.