fix(xlsx): seat top-aligned fixed-row text on Excel's whole-point top seat

[developer0hye]

File submission policy

Before attaching or committing files, read the submission policy.
Complete a security review and obtain your organization's internal approval for
public sharing where applicable. You are responsible for lawful disclosure;
the project and maintainers disclaim related liability to the extent permitted
by applicable law.

• Any submitted sample files or attachments satisfy the submission policy, or none are submitted.

Summary

A top-aligned worksheet cell in a fixed-height row rested its line box on the cell's top inset, so its first baseline sat at the inset plus the face's continuous ascent. Excel seats that line a whole number of sheet points below the row's top boundary: the face's hhea ascent plus line gap at the declared size, offset by a constant and rounded together, evaluated in sheet space and then printed through the fit-to-page scale. On the public #982 workbook the nine-line Segoe UI 14 instruction block therefore matched native by coincidence at 100% (1pt inset + 15.107pt ascent ≈ 16) and sat 0.79pt low on the 0.82-fitted original; every other top-aligned fixed row was off by the rounding remainder of its own face and size.

Key changes:

• sheet_cell_top_baseline_from_track_top_pt (typst_gen_text.rs) states the seat: round((ascent + line gap) x declared size + 1.17) sheet points below the track top, minus the shared fitted-sheet lift, times the print scale. The 1.17pt constant is the window [1.12, 1.23) the XLSX: every baseline on a fixed-track sheet page seats 0.6-1.5pt above the native export's #1063 Arial probes admit; no composition of the centred seat's separately rounded ascent and gap reproduces the series (documented on the constant).
• word_cell_line_box gains a top-alignment arm that redistributes the box around that baseline (the seat less the cell's top inset above, the rest below), so the box height, the row advance and the XLSX: a wrapped cell's lines advance on the face's bare hhea box, 1.95pt short of the native export's per line #1163 wrapped-line leading are unchanged. generate_table_cell no longer excludes single-row top-aligned cells from cell_sheet_seat; top-aligned multi-row merges stay outside the measured regime.
• The XLSX: landscape sheet cell baselines share a 0.70pt downward seat #1496 fitted lift moves into fitted_sheet_lift_pt, shared by the centred and top seats and documented as the stand-in for the XLSX: fitted worksheet text keeps the physical margin origin while its paint layer snaps to Excel's sheet-space origin #1719 text-origin snap so both seats are re-derived together when that lands.
• Tests: top_aligned_sheet_cell_seat_reproduces_the_native_excel_probe pins the thirteen-size Arial series of the XLSX: every baseline on a fixed-track sheet page seats 0.6-1.5pt above the native export's #1063 probes (probe 1 size and track sweeps, probe 2 bordered/unbordered pairing), Segoe UI 14 from the two native Gift Budget and Tracker1.xlsx misc issues #982 exports and Malgun Gothic 14 from the XLSX: whether a sheet row's East Asian line box keys on the face or the text is unmeasured #1060 probe's native export; scaled_top_aligned_sheet_cell_seat_prints_the_lifted_sheet_point pins the 0.82 case; top_aligned_fixed_track_sheet_cell_starts_on_the_native_seat compiles a Libertinus control at both scales and reads the first baseline and the wrapped advance back off the page. spreadsheet_rows_share_one_line_box_whatever_script now asserts the seated top edge instead of the bare ascent; its script-invariance assertion is unchanged.

Related issue

Fixes #1606
Related: #1063, #1496, #1719, #1659

Testing

• cargo test --locked -p office2pdf --lib -- top_aligned_ fails on main (the helper does not exist; on the compiled control the first baseline sits at the inset plus the continuous ascent) and passes with the change.
• cargo test --locked --workspace --no-fail-fast — see the run recorded below.
• cargo +1.97 clippy --locked --workspace --all-targets -- -D warnings and cargo fmt --all -- --check clean.
• Native GTs: the Excel for Mac 16.112 exports of the unchanged public workbook (2 pages, SHA-256 2aa03cafa6a156d553a2f1e03e0458257319ff6416a188f1132ce7817ecaeae1) and of the issue's unscaled control (fitToPage=false, 3 pages, SHA-256 0935ca38e497a657d2f972fb7ddfc3163cf69e8edfe6d79d230202b5db5d7bfa); before binary is main at d79360d; both conversions with --font-path for the Office cloud-font cache (Segoe UI) and Excel's DFonts bundle.
• mutool draw -F trace, page 2 of the fitted original: B4's nine baselines were 154.948/172.168/…/292.708pt (native 154.16/171.38/…/291.92, +0.79pt each); they are now 154.040/171.260/…/291.800pt (-0.12pt each, pitch 17.22pt on both sides). Unscaled control: 177.107 → 177.000pt against native 177.000, all nine lines exact on the 21pt pitch. The XLSX: whether a sheet row's East Asian line box keys on the face or the text is unmeasured #1060 Malgun probe's eight top-aligned ht=36 rows now sit 16pt below their tracks with 0.00pt relative deviation from its native export.
• python3 scripts/compare_layout.py GT.pdf after.pdf --audit --fine-shift 0.5 --noise-floor 0.5 --json (page 2: 34 matched, 0 fine shifts; before the fix 8 fine shifts of +0.79pt and one re-grouped line); python3 scripts/compare_render.py GT.pdf after.pdf --page 2 --dpi 300 --fine-shift 0.5 --artifacts-dir … --cluster-report … --cluster-dispositions … --strict-clusters (34/34 clusters dispositioned, strict PASS); python3 scripts/compare_text_layer.py (the GT carries two more spaces from native's own word fragmentation, as in the XLSX: category-axis stroke overpaints a coincident line series #1604 audit). Page 1 is byte-identical to the before PDF's page 1 in the trace; a 5% fuzz comparison of the before and after page-2 renders with the sidebar block masked out differs on 0 pixels.

Visual impact

• No rendered PDF change
• Rendered PDF change or visual evidence added
• Reason: N/A

Visual audit

• Issue: XLSX: top-aligned wrapped worksheet text starts above the native baseline #1606
• Fixture: tests/fixtures/xlsx/issue_1603_gift_budget.xlsx (the unchanged public Gift Budget and Tracker1.xlsx from Gift Budget and Tracker1.xlsx misc issues #982, SHA-256 25f5dc75dab19ea12042979a61842314ddc226e3e45d447e36b2a2a104112613)
• Page(s): 2
• Renderer and DPI: pdftoppm, 300 DPI
• Evidence mode: fix
• Layout audit report: assets/bugfixes/issue-1606/layout-audit.json
• Render cluster reports: assets/bugfixes/issue-1606/render-clusters-page-2.json
• Reference exporter differences: None
• Fine-detail threshold: 0.5pt
• Layout audit page count: Pass
• Layout audit text flow: Pass
• Layout audit visible fills: Pass
• Layout audit rectangle geometry: Pass
• Layout audit large shifts: Pass
• Layout audit fine shifts: Pass
• New follow-up issues found in this audit: None
• Model vision findings: Inspected the full native and output page 2 side by side, the 5% diff page, a GT/output/diff stack of the whole sidebar instruction block, a GT/before/after stack and the before-vs-GT and after-vs-GT diffs of its first two lines, four 4x crops of the largest remaining sidebar clusters (differ / ily a / s with / v yea) and the before-vs-after diff of the block. Before the fix the block's nine lines sit visibly below native in the first-lines stack and their diff is horizontal bands along every glyph's top and bottom, the signature of a vertical shift; after the fix the lines share native's baselines and the remaining diff is thin vertical-stroke edges that grow toward each line's end, the output glyphs sitting a fraction of a stroke to the right of native on the same baseline (XLSX: fitted worksheet text keeps the physical margin origin while its paint layer snaps to Excel's sheet-space origin #1719 for the fitted text origin, XLSX: Excel for Mac advances each sheet glyph a whole point, so long strings drift up to 1.3pt from our fractional advances #1659 for the whole-point advance drift along the string). The before-vs-after diff is red only on the nine sidebar lines; the title, chart, gridlines, bars, markers, legend, table and footer are on the same pixels before and after. No bold, italic or underlined run is lost; the only hairlines are the 0.60pt chart gridlines and axis, present on both sides at the same x extent, y and width, with no dash pattern on either side; the two zero-value marker rings and the title's glyph edges are the renderer-only differences carried over from the XLSX: category-axis stroke overpaints a coincident line series #1604 audit.
• GT: assets/bugfixes/issue-1606/gt.jpg
• Before: assets/bugfixes/issue-1606/before.jpg
• After: assets/bugfixes/issue-1606/after.jpg
• Native: None
• Compare: None

Visual comparison

GT	Before	After

Required inspection

• Rendered all evidence at 150 DPI or higher
• Stored progressive JPEG quality 86 assets with metadata stripped
• Used Codex/Claude vision to inspect the full GT/output pages, diff, and matched crops
• Inspected matched region crops at full resolution
• Ran compare_layout.py --audit --fine-shift PT and dispositioned every fine/large text-instance shift, rectangle geometry deviation, painted-text visibility mismatch, and visible-fill occlusion
• Ran compare_render.py --cluster-report PATH --strict-clusters and dispositioned every material 5% fuzz diff cluster by explicit ID
• Inventoried hairlines and border dash styles
• Inventoried font weight, italic, and underline emphasis

Deviation audit

Check	Result
Page count/order	Matches GT
Element presence	Matches GT
Position/size	Remaining: #1719, #1659
Rotation/flip	No deviation observed
Fill	Matches GT
Stroke/border	Matches GT
Shape outline geometry	No deviation observed
Text content	Matches GT
Font family/weight/style	Matches GT
Text color	Matches GT
Alignment	Fixed: the top-aligned instruction block starts on Excel's whole-point seat at both print scales
Line/paragraph spacing	Matches GT
Clipping/overflow	Matches GT

Checklist

• Commits include a Signed-off-by line
• PR scope contains one root cause
• Remaining converter or harness deviations each reference an open issue

🤖 Generated with Claude Code