unstructured

1c28b2cd - fix(html_table): keep a declared rowspan of zero on empty cells (#4611)

fix(html_table): keep a declared rowspan of zero on empty cells (#4611) ## Summary `HtmlCell.html` lost a declared `rowspan="0"` whenever the cell had no text, while keeping it when the same cell had text. `HtmlCell.rowspan` reports `None` for `rowspan="0"` and documents it as HTML's spelling for *"spans every remaining row in the containing row group"*, but the empty-cell branch went through `_format_td("", self.colspan, self.rowspan or 1)`, and `_format_td` only emits the attribute when `rowspan > 1`. The declared span was silently downgraded to a one-row cell: ```python HtmlCell(fragment_fromstring('<td rowspan="0">a</td>')).html # '<td rowspan="0">a</td>' HtmlCell(fragment_fromstring('<td rowspan="0"></td>')).html # '<td/>' ``` ## Fix * `_format_td()` now omits `rowspan` only when it is the implicit default `1`, so a `rowspan` of `0` is emitted (`unstructured/common/html_table.py`, `_format_td`). * `HtmlCell.html` passes `0` for the `None` that represents `rowspan="0"` instead of collapsing it with `or 1`, and the docstring of `_format_td` says so. With the fix both spellings round-trip: ```python cell = HtmlCell(fragment_fromstring('<td colspan="2" rowspan="0"></td>')) cell.html # '<td colspan="2" rowspan="0"/>' HtmlCell(...).rowspan # None again after a round-trip ``` ## Tests `test_unstructured/common/test_html_table.py`: * added `DescribeHtmlCell::and_it_preserves_a_rowspan_of_zero_on_an_empty_cell`, next to the existing empty-cell span test. Before / after on `test_unstructured/common/test_html_table.py`: ```text before: 1 failed, 51 passed in 2.18s (the new test) after: 52 passed in 2.06s ``` No other test edited; the rest of `test_unstructured/common` plus the chunking suites stay green. ## Notes The cell has to be empty to show the bug. `unstructured/chunking/base.py::own_cell_html` currently recomputes the remaining span for empty cells, so chunked output is not immediately wrong — what is lost is the serialization contract of `HtmlCell.html` (and the `len(cell.html)` bookkeeping in `BaseChunker`, plus any caller that re-parses the produced HTML). Fixes #4610