github-featured / unstructured-io--unstructured

发布

0 发布 249 Git 标签

RSS 订阅源
  • 0.22.23 879e1269b5
    比较

    fix: first table chunk preserve col/row span (#4343)

    frostbyte_neo 发布于 2026-04-24 17:22:26 +00:00

    Summary

    When a Table element was split into multiple TableChunks, the
    first chunk lost colspan / rowspan on its header
    cells while continuation chunks kept them — producing inconsistent
    merged-cell structure across a single split table.

    Root cause: HtmlTable.from_html_text calls e.attrib.clear() on every
    descendant during compactification, intending to drop
    cosmetic attributes (border, class, style, …). That also wiped
    colspan/rowspan. Continuation chunks escaped the bug
    because their repeated headers come from source_row_htmls captured
    before the strip; the first chunk's rows flow through
    the compactified tree directly.

    Fix

    Preserve colspan and rowspan through the attrib.clear() step in
    unstructured/common/html_table.py. They're structural,
    not cosmetic, and the rest of the chunking pipeline then carries them
    through unchanged.

    Tests

    test_html_table.py::but_it_preserves_colspan_and_rowspan_as_structural_cell_attributes
    — unit guarantee that
    compactification keeps the two span attributes and still strips
    class/style/id/data-*.

    test_base.py::and_it_preserves_colspan_and_rowspan_in_the_first_chunk_header_rows
    — end-to-end regression: the first
    TableChunk of a split table keeps colspan/rowspan on header cells,
    matching continuation chunks.

    Both tests fail on main and pass with this change.

    Note

    Low Risk
    Low risk, localized change to table HTML normalization that only
    preserves two structural attributes; main impact is on downstream
    TableChunk HTML fidelity for merged headers.

    Overview
    Fixes a regression where HtmlTable.from_html_text() compactification
    stripped structural colspan/rowspan, causing the first chunk of a
    split table to lose merged-header layout while continuation chunks
    retained it.

    Updates the attribute-stripping step to preserve colspan and
    rowspan, adds unit + end-to-end chunking regression tests to assert
    consistent header spans across chunks, and bumps the release to
    0.22.23 with a changelog entry.

    Reviewed by Cursor Bugbot for commit
    5463e776ce02a269db84713431280375cc092cc8. Bugbot is set up for automated
    code reviews on this repo. Configure
    here.

    下载附件