feat(ooxml): pptx slide size + tables, xlsx merged cells + value types, docx table merges

[andiwand]

🤖 Generated with Claude Code

A batch of low-hanging fruits from the ooxml module AGENTS.md open-work lists, plus one latent shared-code bug they exposed.

pptx

• Real slide size: slide_page_layout reads p:presentation/p:sldSz (EMUs, resolved once in the Document ctor; ECMA-376 default 10in × 7.5in when absent) instead of a hardcoded 11.02in × 8.27in. Slide names come from p:cSld@name.
• Tables actually parse now: a:tbl sits inside p:graphicFrame/a:graphic/a:graphicData, which the parser never descended into — tables were silently absent from the element tree. Now: p:graphicFrame→frame (with p:xfrm positioning), a:tbl wired through create_table_element/append_column (a:tblGrid/a:gridCol), rows/cells from a:tr/a:tc, spans from gridSpan/rowSpan, covered cells from hMerge/vMerge, column widths and row heights from w/h EMUs. The dead ODF-name table_dimensions copy-paste is gone.

xlsx

• Merged cells: mergeCells is parsed per sheet into anchor span + is_covered flags on the SheetCell side map; sheet_cell_span/sheet_cell_is_covered report them and merged ranges render as colspan/rowspan. Ranges whose anchor cell is absent from sheetData are dropped to keep the renderer's cursor bookkeeping consistent.
• Cell value types: sheet_cell_value_type derives number-vs-string from c/@t (default "n" → float_number when <v> exists), so numeric cells get the odr-value-type-float class.

docx

• Table adapter rewritten against w: names (the methods previously queried ODF names and were dead code): dimensions from w:tblGrid/w:gridCol + w:tr; colspan from w:gridSpan; w:vMerge continuations report covered, and the restart cell computes its rowspan by walking following rows at the same grid column.
• bookmark_name reads w:name instead of ODF's text:name.

TableCursor: latent infinite loop (exposed by xlsx merges, affects ODF too)

handle_rowspan_ assumes each row's pending covered ranges are sorted by column, but add_cell appended them. A long-rowspan cell followed in a later row by new rowspan anchors at smaller columns (e.g. F4:F9 then A6:A7 — real pattern in ruski.xlsx) produced an unsorted list; the skip missed the covered position and translate_sheet's covered-cell continue never advanced the cursor → infinite loop. Fixed with sorted insertion + a tolerant skip, a TableCursor regression test, and a defensive one-column advance on the (normally unreachable) covered branch in the renderer. ruski.xlsx went from hanging forever to 0.13s.

Verification

• Hand-crafted xlsx (merged range incl. absent covered cells, shared/numeric cells) and docx (gridSpan + vMerge chain, bookmark) verified end-to-end through odr_test: correct colspan/rowspan, covered cells skipped, float classes, bookmark anchors.
• style-various-1.pptx renders its slide-8 table with real column widths; the 16:9 deck renders 10in × 5.625in slides.
• Suites pass: pptx, docx, xlsx, odt/ods/odp/odg, xls, csv, TableCursor.*.
• Note: CI reference-output comparison will show expected diffs (slide sizes, new pptx tables, xlsx merges/value classes) — reference submodules need regenerating.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f0599f62f1

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".