fix #659

[ponte-vecchio]

See #659 which this PR patches. Specifically, this addresses the following problem:

What seemed to be happening is that before span sizes are distributed, \__tblr_collect_column_widths_skips: populates \l__tblr_col_item_skip_size_prop using the @col-width of each column. For X columns where co > 0, this value is initially their unwrapped natural width, which can be massive. During \__tblr_set_column_widths_from_span_widths:, the routine evaluates the span distributions. Now, because the X columns are not covered by a span, the distribution logic leaves their massive natural width untouched in \l__tblr_col_item_skip_size_prop. It then populates \l__tblr_column_min_width_prop for the X columns using these untouched values. Consequently, when \__tblr_adjust_extendable_column_width: runs to assign the final target widths to the X columns, it checks against \l__tblr_column_min_width_prop. Since the recorded min width is the huge unwrapped natural width it locks the X column to be that size causing table overflow as exhibited by #659.

[Copilot]

Pull request overview

Fixes issue #659, where X columns were assigned an inflated minimum width whenever a horizontally merged (\SetCell[c=...]) cell appeared in the same table, causing the table to overflow. The root cause is that \__tblr_set_column_widths_from_span_widths: recorded each column's current @col-width (the unwrapped natural width for X columns) as its minimum, which was later used by \__tblr_adjust_extendable_column_width: and locked X columns to that huge size. The fix detects columns with a non-zero coefficient (X columns) and skips recording a minimum width for them while resetting their entry in \l__tblr_col_item_skip_size_prop to 0pt.

Changes:

• In \__tblr_set_column_widths_from_span_widths:, branch on the column coefficient: for X-style columns, zero out the col-item skip size and don't populate \l__tblr_column_min_width_prop; otherwise behave as before.
• Bump the release date in \ProvidesExplPackage to 2026-05-28.
• Add a new regression test table-018 (tex, tlg, pdftex.tlg, md5) and a corresponding entry in testfiles/README.md; refresh cell-005.md5; add .DS_Store to .gitignore.

Reviewed changes

Copilot reviewed 7 out of 10 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
tabularray-dev.sty	Skip min-width recording for X (coefficient > 0) columns and bump release date.
testfiles/table-018.tex	New regression test reproducing the merged-cell + X-column scenario from #659.
testfiles/table-018.tlg, table-018.pdftex.tlg, table-018.md5	Expected log/checksum baselines for the new test.
testfiles/cell-005.md5	Updated checksum reflecting the column-width fix's effect on existing test.
testfiles/README.md	Documents the new table-018 test entry.
.gitignore	Adds .DS_Store to ignore list.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[peaR-red]

Hmm, seems alright, other than a few stuff Copilot pointed out (was the AI review necessary though?).

What I'm curious about is why cell-005 warranted changes in light of this fix. Compiling locally there doesn't seem to be any noticeable difference.

[mariuslp]

Great fix, thanks. A tiny detail that I am not sure whether it is a bug or a feature, but calling hspan=even seems to have an unexpected (but small) effect on the column size.

\documentclass{article}
\usepackage{tabularray-dev} 
\usepackage{xcolor}

\begin{document}
\begin{tblr}{
  colspec = {lllXX},
  hlines,
  vlines,
  hspan=even,
}
    1-1 & 1-2 & \SetCell[c=2]{c} merged & very very long text in the X column & sooooooooooooo very much long it will not stop and continue \\
    2-1 & 2-2 & 2-3 & 2-4 & 2-5 \\ 
\end{tblr}

\begin{tblr}{
  colspec = {lllXX},
  hlines,
  vlines,
  hspan=default,
}
    2-1 & 2-2 & 2-3 & 2-4 & 2-5 \\ 
    1-1 & 1-2 & \SetCell[c=2]{c} merged & very very long text in the X column & sooooooooooooo very much long it will not stop and continue \\
\end{tblr}

\end{document}

I would say that intuitively, a merged cell over an X cell is not extending and therefore the effects of hspan=even should be the same as hspan=minimal or hspan=default (which is no effect), but that might be up to debate.

[ponte-vecchio]

@peaR-red

Hmm, seems alright, other than a few stuff Copilot pointed out (was the AI review necessary though?).

Big oopsie with fat fingers, ignore the clanker.

What I'm curious about is why cell-005 warranted changes in light of this fix. Compiling locally there doesn't seem to be any noticeable difference.

Visually indistinguishable, but the checksum test fails, so something must be different - in any case, I've had to change it for that reason alone.

[peaR-red]

A minor addition to \__tblr_set_column_widths_from_span_widths:...