fix(xlsx): quantize interior horizontal bar-chart category baselines

[developer0hye]

Summary

Native Excel snaps a horizontal bar chart's interior category-label baseline to floor(sheet_frame_top_pt + row_top + row / 2) + K, with K an integer that depends only on the declared category-axis size (4 at 8/10pt, 5 at 14pt) — not on which chart, how many categories it has, or the row's own position. Measured across two worksheet bar charts, two category counts and three sizes over three prior sessions (#1763, #1764, both merged, measurement-only).

bar_category_label_baseline_pt (new, typst_gen_diagrams.rs) implements this, and generate_chart_axis applies it to every category row strictly between the plot rectangle's own top and bottom edge, gated to a worksheet-anchored chart and a category-axis size this issue actually measured against native Excel.

Why edge rows are excluded, not "not yet fixed": a regression check against all 24 previously-measured rows (assets/validation/issue-1621/measurements.json) before writing any code showed the same formula cannot be safely applied to every row. Income's top edge row already sits inside this issue's own 0.5pt gate at 10pt (+0.096pt) and 14pt (+0.324pt); the naive formula would push it to exactly -1.0 sheet points (-0.78 printed points), failing a case that passes today. Expense's edges fail the same way in the other direction. So "interior rows only" is the largest rule this data supports without regressing an already-passing row — see assets/validation/issue-1621/README.md's new "Shipped: interior rows only" section for the full arithmetic.

The remaining edge-row mechanism (why an edge deviates and in which direction is chart-dependent, for a reason not yet isolated) is unchanged from the prior sessions' findings and keeps #1621 open.

Testing

• Red/green: bar_category_label_baseline_matches_measured_native_interior_rows (15 real native-Excel data points across 3 plot geometries and 3 sizes — the pure function), bar_category_label_baseline_keeps_no_rule_for_an_unmeasured_size, a_worksheet_bar_chart_seats_only_interior_category_labels_on_the_quantized_baseline and a_worksheet_bar_chart_at_an_unmeasured_size_keeps_every_labels_centred_seat (the wiring) all fail without the fix.
• cargo test --locked -p office2pdf --lib: 3097 passed.
• cargo test --locked --workspace: all suites green.
• cargo fmt --all --check / cargo clippy --locked -p office2pdf --all-targets: clean (pre-existing unrelated warnings only).
• python3 assets/validation/issue-1621/verify_measurements.py: PASS (unchanged claims; this PR ships the interior-row half of claim 4, the edge-row exception set in claims 4/5 is still open).
• Before/after self-diff of office2pdf's own output in isolation (compare_layout.py, --noise-floor 0.01 --fine-shift 0.01): exactly the 7 interior category labels move (4 expense, 3 income), each landing on native's whole sheet point; 592 text instances compared, 0 missing/extra; rect census bit-identical (738/738, 0 geometry deviations) — nothing else on the page moves.
• Required delegated read-only documentation freshness audit: PASS (two passes; the first caught a doc-comment/behavior mismatch and a forward-referenced README claim, both fixed and re-verified).

Visual impact

• No rendered PDF change
• Rendered PDF change or visual evidence added

Visual audit

• Issue: XLSX: horizontal bar-chart category baselines deviate from native row placement #1621

• Fixture: tests/fixtures/xlsx/issue_1181_fit_to_height.xlsx (unmodified; SHA-256 2b4a2d8dceda58758593c88409875efbda05780559154c02bd13fef4f7a1c65b)

• Page(s): 2

• Renderer and DPI: Microsoft Excel for Mac native reference; pdftoppm, 300 DPI

• Evidence mode: fix

• Layout audit report: assets/bugfixes/issue-1621/layout-audit.json

• Render cluster reports: assets/bugfixes/issue-1621/render-clusters-page-2.json

• Reference exporter differences: None

• Fine-detail threshold: 0.5pt

• Layout audit page count: Pass

• Layout audit text flow: XLSX: title+header-row text on one native baseline reports as a 1-into-2 line reflow #1760

• Layout audit visible fills: Pass

• Layout audit rectangle geometry: XLSX: fitted worksheet cell fills sit one sheet point above Excel's #1632, XLSX: horizontal bar-chart value-bar fills sit ~0.544pt above native, widest bar 11pt narrow #1759

• Layout audit large shifts: Pass

• Layout audit fine shifts: XLSX: horizontal bar-chart category baselines deviate from native row placement #1621, XLSX: fitted horizontal bar-chart value labels sit about 0.61pt too high #1622, XLSX: large left-aligned Cambria title starts 1.255pt left of native #1623, XLSX: right-aligned cells on the fitted budget sheet land 0.9pt left to 2.2pt right of Excel by number format #1631, XLSX: fitted worksheet text keeps the physical margin origin while its paint layer snaps to Excel's sheet-space origin #1719, XLSX: a form-control label row sits 2.16pt above Excel where the rest of the fitted sheet sits ~0.6-0.8pt off #1757, XLSX: repeated %INC column-header text sits 0.39pt right of native #1761

• New follow-up issues found in this audit: None

• Model vision findings: Full GT/before/after page-2 crops were inspected at 300 DPI (assets/bugfixes/issue-1621/gt.jpg/before.jpg/after.jpg), plus tight crops of both bar charts' label columns. At normal viewing scale the fix is invisible — each affected label moves under 1 printed point and stays vertically centered on its own bar in both the before and after renders, matching the near-zero measured residual. The compare_render.py --strict-clusters pass (395/395 clusters dispositioned) surfaced 15 clusters not present in the last full committed baseline (assets/bugfixes/issue-1564/render-clusters-page-2.json, 380/395 reused by exact bbox match); all 15 were individually cropped and inspected this session: 9 are XLSX: horizontal bar-chart value-bar fills sit ~0.544pt above native, widest bar 11pt narrow #1759's own thin bar-fill edge strips (visually confirmed against XLSX: horizontal bar-chart value-bar fills sit ~0.544pt above native, widest bar 11pt narrow #1759's description), 3 are XLSX: repeated %INC column-header text sits 0.39pt right of native #1761's repeated "% INC" header text, 2 (the "JAN" spinner band and a small chart indicator fill) directly match XLSX: fitted worksheet cell fills sit one sheet point above Excel's #1632's own already-written disposition note text and, for the indicator fill, its exact measured -0.78pt (one sheet point) offset, and 1 is a comma-grouped amount ("$1,056") matching XLSX: right-aligned cells on the fitted budget sheet land 0.9pt left to 2.2pt right of Excel by number format #1631's number-format cell family. A before/after self-diff at the pixel-cluster level is bit-identical (395/395, 0 new/removed clusters) — this fix's effect sits under the 5%-fuzz clustering floor at 300 DPI. other expenses and room & board (expense-chart edge rows) and 8 render clusters newly attributed to XLSX: a 14-15pt bottom-aligned cell in a 19.5pt fixed row is centred as a tight row where Excel bottom-seats it #1721 (a 14-15pt bottom-aligned cell being centred as a tight row, all off the plot area) were individually cropped and confirmed against those issues' own descriptions; XLSX: title+header-row text on one native baseline reports as a 1-into-2 line reflow #1760's pre-existing 1-into-2 line reflow is unchanged by this fix.

• GT: assets/bugfixes/issue-1621/gt.jpg

  

• Before: assets/bugfixes/issue-1621/before.jpg

  

• After: assets/bugfixes/issue-1621/after.jpg

  

• Native: None

• Compare: assets/bugfixes/issue-1621/compare.jpg (filed with the original issue)

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: #1621 (2 edge rows), #1622, #1623, #1631, #1719, #1721, #1757; 7 interior category-label rows fixed by this PR.
Rotation/flip	No deviation observed
Fill	Remaining: #1632, #1759
Stroke/border	No deviation observed
Shape outline geometry	No deviation observed
Text content	No deviation observed
Font family/weight/style	Matches GT
Text color	Matches GT
Alignment	Remaining: #1621 (2 edge rows), #1631, #1719, #1721, #1761; 7 interior category-label rows fixed by this PR.
Line/paragraph spacing	Remaining: #1760
Clipping/overflow	No deviation observed

Checklist

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

Related: #1621

🤖 Generated with Claude Code