feat(spreadsheet): state the cell an xlsx file spells none of

[andiwand]

🤖 Generated with Claude Code

Rebased on main now that #868 has landed. Step 2.2 of docs/design/spreadsheet-editing.md.

What it does

An .xlsx file spells a c only for a cell that holds something, so a write into an empty position refused. The page carries no lock on such a cell, so the user typed and the host got the refusal back. This was the last such hole.

insert_cell states the c in its row in column order, and the row in sheetData in row order where the file states none, then widens dimension around the new cell. Nothing is reindexed — the cell map is keyed by position, so an insert touches one entry.

A position a merge covers refuses first, because Excel ignores what a covered c holds. That check reads mergeCells again, since the parsed is_covered flags only exist for a cell the file states.

The fix that came with it

TableRange::contains was half open — it left the last row and column of the range out. The parser marks the covered cells of a big merge with contains and those of a small one with an inclusive loop, so the two disagreed: a merge too big to walk cell by cell marked one row and one column too few, and SheetCell::is_covered was wrong for those cells.

to() is read as the last position of the range everywhere else — the parser's own loop, the dimension handling — so contains is closed now too. It had exactly one caller before this pull request. TableRange.contains pins it, and the_last_cell_of_a_wide_merge_refuses_too pins what it meant for the parser; both fail against the old comparison.

Checks

• soffice --headless --convert-to csv on a grown Ordnerruecken.xlsx reads 7.5 at the new column of row 1 and grown at the new bottom-right corner.
• The reference output is byte-identical to the pins, so the contains fix moves no rendered cell. It reaches the renderer only through a backstop the cursor normally makes unreachable.

Test

OoxmlSpreadsheetWrite gains seven cases: an absent cell written, column order, row order, the widened dimension, the absent covered cell that still refuses, the last cell of a wide merge that now refuses too, and a save and reopen. The test workbook gained a sheet_prefix argument so a <dimension> can sit where the schema orders it.

https://claude.ai/code/session_01KKFKbUVCYF2VhujdmjhhPW