Escape pipes when writing a CoefTable as markdown

[andreasnoack]

show(::IO, ::MIME"text/markdown", ::CoefTable) never escapes |, so any cell containing one silently destroys the table. Downstream packages hit this with the R-style Pr(>|z|) p-value header (GLM, MixedModels, and others build their coefficient tables with it), which is why it is easy to reproduce but invisible from inside StatsBase — no header or fixture here contains a pipe.

Reproduction

julia> ct = CoefTable(Any[[0.1615], [0.0499], [3.24], [0.0012]],
                      ["Coef.", "Std. Error", "z", "Pr(>|z|)"], ["Age"], 4, 3);

julia> md = sprint(show, MIME"text/markdown"(), ct);

julia> only(Markdown.parse(md).content)   # expected a Table
Markdown.Paragraph(...)

The header row carries the two extra pipes from Pr(>|z|), so it presents 10 pipe-delimited fields against the alignment row's 8. No markdown parser accepts that as a table; Julia's degrades it to a paragraph and then applies inline typography to the --- rules, turning them into en-dashes. Rendered through Documenter or any markdown viewer, the whole table appears as one run of pipes and dashes.

Fix

Escape | in row names, column names and string cells — the same rule the Markdown stdlib applies in plain(::IO, ::Markdown.Table) — before the column widths are measured, so the widths stay consistent with what is printed.

Added a test covering a pipe in a header, a row name and a string cell, asserting that every row presents the same number of cells and that the output parses as a Markdown.Table. It fails without the fix. The two existing golden-output markdown tests are unchanged, since neither contains a pipe.

Related

#1008 is an alternative that goes further: it drops the hand-rolled writer in favour of building a Markdown.Table, which removes this class of bug rather than this instance of it. That one changes the emitted spacing and adds Markdown as a dependency, so it is offered separately rather than folded in here.

🤖 Generated with Claude Code

[palday]

@andreasnoack care to patch bump and release this right away? I'll take a look at #1008 when I get some time later in the week