Classes and ids on pipe tables: probe, 2026-10-12 Question Can each tool give a Markdown pipe table (| a | b |) a class (for example table-compact) and an id, without rewriting it as a list-table? Method - A copy of concepts/tables made outside the repository (APFS clone of the working tree at 05:33; the repository was not changed). All builds, servers and checks ran in the copy. - Native syntax first: each candidate written next to a 3 x 3 pipe table and compiled with the tool's own Markdown library as installed in the sites (Markdoc and Python-Markdown directly; MDX with remark-gfm through @mdx-js/mdx evaluate; GitHub through its Markdown API, mode gfm): attribute line after the table {.table-compact #t1} the same after a blank line attribute line before the table attribute on the first header cell | A {: .table-compact #t1 } | | A {% .table-compact #t1 %} | HTML comment before the table Then an HTML wrapper around the table, built in each site. - Extension: one small change per site that reads the HTML comment in the block just above a pipe table and puts its classes and id on the . Built with each site's `make check`. - Test pages (one per site, native mode; Starlight native-plain, where its pipe tables are): pipe-attrs case 1 comment on the line above, .table-compact #probe-one case 2 two classes, blank line between comment and table case 3 an unrelated comment above the table case 4 table inside a numbered list item, comment above it case 5 no comment case 6 the comment syntax inside a fenced code block pipe-wrapper the table inside an HTML
carrying class and id The roles table's native pipe form got `` on its first line (Docusaurus native, Zensical native and native-plain, Starlight native-plain): the id the source has and native mode lost. - Wide-table check (checks/wide-tables.mjs) run against each site served on ports 4610-4612, for the roles pages, and a copy limited to the probe pages. Versions (read from node_modules and uv.lock) Docusaurus 3.10.2 (@docusaurus/mdx-loader 3.10.2), @mdx-js/mdx 3.1.1, remark-gfm 4.0.1, mdast-util-to-hast 13.2.1, @slorber/remark-comment 1.0.0 Zensical 0.0.69, Python-Markdown 3.11, pymdown-extensions 12.1 Starlight 0.42.6, Astro 7.3.8, @astrojs/markdoc 2.0.11, @markdoc/markdoc 0.5.10 Antora 3.2.1, Asciidoctor.js 2.2.9 (Asciidoctor 2.0.23) Upstream files fetched at the commits below were byte-identical to the installed copies (diff -q), except Docusaurus and Astro, which ship compiled code; their line numbers are from the TypeScript source at the release tag. Native syntax, 3 x 3 table candidate MDX+gfm Python-Markdown Markdoc GitHub {...} line after table parse error extra body row error: "Can't extra body row (acorn) (cell text) apply inline annotations to 'document'" {...} line, blank line before parse error paragraph same error paragraph {...} line before table not tried table lost: one same error visible paragraph paragraph attribute in 1st header cell - class, id on
class, id on text in the cell above table, no table, no attrs, table, no attrs, table, comment attrs, comment comment kept in comment dropped hidden dropped output {/* ... */} after (MDX only) table, no - - text attrs No tool attaches a class or id to a pipe table natively. HTML wrapper (built), class and id land on a
around the table Docusaurus
+ blank line + table + blank line +
->
Zensical
... (md_in_html) ->
Starlight
... ->
The spelling differs per tool (className in MDX; the markdown attribute in Python-Markdown). Extension: one comment syntax, three small changes Syntax: an HTML comment in the block just above the table, a blank line between them allowed: | Activity | Admin | ... `.name` adds a class (several allowed), `#name` sets the id. Docusaurus plugins/table-attrs.js (new, 30 lines, 20 of code) and one changed line in docusaurus.config.js. The parser drops comments, so the plugin reads the source line above the table's position and sets the mdast table's hProperties. Zensical table_attrs.py (new, 58 lines, 46 of code) and one line in zensical.toml. A preprocessor adds a blank line after the comment (a table in a list item must follow one); a treeprocessor finds the comment's paragraph (or its raw HTML placeholder) just before a
and sets class and id. Starlight table-attrs.mjs (new, 37 lines, 24 of code) and two changed lines in markdoc.config.mjs. A document transform finds a comment node just before a table node and sets its class and id attributes; the existing table transform (list-table.mjs) renders them. Commands (in the copy, per site) make check spot checks, all modes grep -o ']*>' node ../../checks/wide-tables.mjs 'http://localhost:461N/%m/%p/' node ../../checks/wide-probe.mjs (the same, pages = pipe-attrs) python3 cmp-tables.py (every
on every page compared) Rendered
start tags, pipe-attrs page case Docusaurus Zensical Starlight 1 class="table-compact" id="probe-one" class="table-compact" id="probe-one" id="probe-one" class="table-compact" 2 class="table-compact table-striped" the same the same id="probe-two" 3 (none) (none) (none) 4 class="table-compact" id="probe-four" the same the same 5 (none) (none) (none) 6 code block shows the comment as written in all three; no table Comment in the HTML output: dropped (Docusaurus, Starlight), kept as an HTML comment (Zensical). GitHub (Markdown API, gfm): all five tables render; the comments are hidden; case 6 shows the comment as code. Roles table (native pipe form),
start tag Docusaurus native
Zensical native
Zensical native-plain
Starlight native-plain
Antora native, native-plain
(already; AsciiDoc's [#id] on the table) Every
on every built page compared with the repository's own builds: the only differences are these roles tables, and each is identical apart from the id. Spot checks after the change Docusaurus passthrough 49, extension 49, native 49 (of 51; the two misses are the known page descriptions, as before) Zensical passthrough, extension, native, native-plain: 51/51 each Starlight passthrough, extension, native, native-plain: 51/51 each Wide tables (1280 and 390 px) roles pages and all others, after the change: Docusaurus 3 modes 90 passed; Zensical native + native-plain 60 passed; Starlight native-plain 30 passed; 0 failed pipe-attrs page: Docusaurus 10 passed; Starlight 10 passed Zensical 4 passed, 6 failed: tables 1, 2 and 4 (the classed ones) widen the page. The theme styles and wraps only tables without a class, so a classed table loses its look and its scroll box, as an HTML table does. Zensical with `.table .table-compact` (the standard class first): 10 passed. The site's table.table rules restore look and scrolling. Antora (no pipe tables) A GFM pipe table given to Asciidoctor renders as one paragraph of text. AsciiDoc's own table takes id and role in its block attributes: [.table-compact#probe-one%header] ->
Converted with the installed Asciidoctor.js; not built as an Antora page. Side finding (Docusaurus, headings) Docusaurus's heading code reads an id from a trailing comment, but in an .mdx page the comment is dropped before that code runs: `## A heading ` built as id="a-heading-". Code references (pinned) Markdoc 0.5.10 (16e1840c) annotation outside inline content is an error; inside, it goes to the inline's parent (the cell): https://github.com/markdoc/markdoc/blob/16e1840c687e17c57d2b3664eac4edfbd01e7283/src/parser.ts#L118-L125 https://github.com/markdoc/markdoc/blob/16e1840c687e17c57d2b3664eac4edfbd01e7283/src/parser.ts#L186 table node has no attributes of its own; class and id are global: https://github.com/markdoc/markdoc/blob/16e1840c687e17c57d2b3664eac4edfbd01e7283/src/schema.ts#L120-L122 https://github.com/markdoc/markdoc/blob/16e1840c687e17c57d2b3664eac4edfbd01e7283/src/transformer.ts#L17-L19 comments become comment nodes when allowComments is on: https://github.com/markdoc/markdoc/blob/16e1840c687e17c57d2b3664eac4edfbd01e7283/src/tokenizer/index.ts#L31 https://github.com/markdoc/markdoc/blob/16e1840c687e17c57d2b3664eac4edfbd01e7283/src/tokenizer/plugins/comments.ts#L48-L51 https://github.com/markdoc/markdoc/blob/16e1840c687e17c57d2b3664eac4edfbd01e7283/src/schema.ts#L246-L250 @astrojs/markdoc 2.0.11 (3571e2d2) turns allowComments on: https://github.com/withastro/astro/blob/3571e2d2e80a66764254fa9b7750704a150ca3f6/packages/integrations/markdoc/src/tokenizer.ts#L11-L15 Python-Markdown 3.11 (0ffbf00c) every line of the table block after the delimiter row is a row; the
is created with no attributes: https://github.com/Python-Markdown/markdown/blob/0ffbf00cf667971d46f43aafba825e8d25098bc5/markdown/extensions/tables.py#L87-L107 attr_list reads a list from an element's own text; cells (td, th) take it at the end of their text, a table has no text of its own: https://github.com/Python-Markdown/markdown/blob/0ffbf00cf667971d46f43aafba825e8d25098bc5/markdown/extensions/attr_list.py#L103-L154 raw HTML blocks are set aside at priority 20, as placeholders: https://github.com/Python-Markdown/markdown/blob/0ffbf00cf667971d46f43aafba825e8d25098bc5/markdown/preprocessors.py#L41 https://github.com/Python-Markdown/markdown/blob/0ffbf00cf667971d46f43aafba825e8d25098bc5/markdown/util.py#L79-L81 pymdown-extensions 12.1 (9b896a8a): fenced code set aside at priority 25: https://github.com/facelessuser/pymdown-extensions/blob/9b896a8ae4ae2c9bbf42aa525925b245432fb3d2/pymdownx/superfences.py#L325 Docusaurus 3.10.2 (f37f9035) HTML comments in MDX handled by a fork of remark-comment, on by default (mdx1Compat.comments): https://github.com/facebook/docusaurus/blob/f37f9035584917a97a260b91fc2842cba4f8b94f/packages/docusaurus-mdx-loader/src/processor.ts#L86-L87 https://github.com/facebook/docusaurus/blob/f37f9035584917a97a260b91fc2842cba4f8b94f/packages/docusaurus-mdx-loader/src/processor.ts#L152 https://github.com/facebook/docusaurus/blob/f37f9035584917a97a260b91fc2842cba4f8b94f/packages/docusaurus/src/server/configValidation.ts#L130-L131 heading id from : https://github.com/facebook/docusaurus/blob/f37f9035584917a97a260b91fc2842cba4f8b94f/packages/docusaurus-mdx-loader/src/remark/headings/index.ts#L51-L60 @slorber/remark-comment 1.0.0 (installed index.js identical to slorber/remark-comment 426dde74): a comment leaves a node in the tree only with the `ast` option, which Docusaurus does not pass: https://github.com/slorber/remark-comment/blob/426dde74a26d9d9adfe1a649d602366c46027350/index.js#L35-L60 mdast-util-to-hast 13.2.1 (174795b2): data.hProperties on an mdast node is copied onto the element, tables included: https://github.com/syntax-tree/mdast-util-to-hast/blob/174795b21f7757fffb54dd8d5fb4012f4751f791/lib/handlers/table.js#L60 https://github.com/syntax-tree/mdast-util-to-hast/blob/174795b21f7757fffb54dd8d5fb4012f4751f791/lib/state.js#L352-L382 Asciidoctor 2.0.23 (2c03d22e): table id and role on
: https://github.com/asciidoctor/asciidoctor/blob/2c03d22eb9fe83d04ee452939581d1b842f6d8b4/lib/asciidoctor/converter/html5.rb#L856-L878 GFM spec: an HTML comment is an HTML block (type 2) that ends at -->; a table ends at a blank line or another block: https://github.github.com/gfm/#html-blocks https://github.github.com/gfm/#tables-extension- Not verified - The MDX wrapper spelling with `class` instead of `className` was not tried. - The Zensical preprocessor that adds a blank line after the comment runs after fenced code is set aside at the top level (case 6); a fenced block inside a list item that holds such a comment line was not tried. - No CSS for table-compact was written; the probe shows only that the class and id reach the
. Follow-up, same day: titles, and Zensical adds `table` (supersedes the extension line counts and the Zensical wide-table result above; everything else above still holds) Changes - Title: a table's title is a bold line above it (Template conventions). The comment may sit above the title instead of directly above the table, with or without a blank line between comment and title: **Label requirements** | Part of Label | Length in characters | ... Each extension first looks at the block directly above the table; if that is not the comment but a paragraph whose whole content is one bold run, it looks at the block above that. Only one title is skipped; any other block in between stops the search. Class and id go on the table, never on the title. - Zensical: a table given a class by the comment also gets `table` (unless named already), so the site's `table.table` rules give back the theme's look and scroll box. The source names `.table-compact` alone, the same in every tool. Method - A new copy of concepts/tables, made from the repository at 40da120 (with a45d898 and 94771c6 since: Makefile screenshot targets only), the earlier patches applied, then the two changes. Baseline: a second, unpatched copy at 94771c6, built and checked the same way. - Test page pipe-attrs gained cases 7-14: 7 comment, blank line, title, table -> applies 8 comment directly above the title, table -> applies 9 comment, normal paragraph, table -> no change 10 title, table, no comment -> no change 11 comment, two titles, table -> no change 12 title, comment directly above the table -> applies to the table 13 comment, title with more than bold text -> no change 14 .table .table-compact (standard class named) -> no duplicate - metadata (Docusaurus native, Zensical native and native-plain, Starlight native-plain): `` above `**Label requirements**`. Extension size now Docusaurus plugins/table-attrs.js 44 lines (29 of code) + 1 config line The parser drops comments, so the title is found in the tree (paragraph with one strong child, ending on the line just above the table) and the comment in the source above it. Zensical table_attrs.py 79 lines (46 of code) + 1 config line The treeprocessor now runs after inline processing (priority 15, after `inline` at 20), when bold text is a element and the comment is a raw-HTML placeholder at every level (a block placeholder at the top level, an inline one in a list item). Starlight table-attrs.mjs 45 lines (28 of code) + 2 config lines Markdoc keeps the comment and the title as sibling nodes of the table. Rendered
start tags, pipe-attrs (class and id order as emitted) case Docusaurus Zensical Starlight 1 table-compact / probe-one table table-compact / probe-one table-compact / probe-one 2 table-compact table-striped table table-compact table-striped as Docusaurus 3 (none) (none) (none) 4 table-compact / probe-four table table-compact / probe-four table-compact / probe-four 5 (none) (none) (none) 7 table-compact / probe-seven table table-compact / probe-seven table-compact / probe-seven 8 table-compact / probe-eight table table-compact / probe-eight table-compact / probe-eight 9 (none) (none) (none) 10 (none) (none) (none) 11 (none) (none) (none) 12 table-compact / probe-twelve table table-compact / probe-twelve table-compact / probe-twelve 13 (none) (none) (none) 14 table table-compact / fourteen table table-compact / fourteen table table-compact / fourteen In all three: no element other than a
carries a probe id; the six bold-only titles render unchanged as

…

. GitHub (Markdown API, gfm): 13 tables, comments hidden, titles bold. metadata, Label requirements Docusaurus native, Zensical native and native-plain, Starlight native-plain:
; the bold title line is kept (spot check "title shown" passes); the other two tables unchanged. Spot checks, all modes (Docusaurus passthrough, extension, native; Zensical and Starlight passthrough-raw, passthrough, extension, native, native-plain) Docusaurus 49 / 49 / 49 Zensical 46 / 51 / 51 / 51 / 51 Starlight 41 / 51 / 51 / 51 / 51 The checks now also list the four bosh.io pages (5de50e2), which the sites do not have yet, so many FAIL lines print. The ok counts and every FAIL line are identical to the unpatched copy's (Docusaurus 129, Zensical 206, Starlight 211 FAIL lines in both). Wide tables (1280 and 390 px; checks/wide-tables.mjs with its page list taken from the environment) roles-table-host and metadata, all modes: Docusaurus 24, Zensical 32, Starlight 32 passed; 0 failed pipe-attrs: Docusaurus 26, Zensical 26, Starlight 26 passed; 0 failed (Zensical was 4 passed, 6 failed before `table` was added) roles table in Zensical native given `.table-compact` alone (tried, then reverted):
, 2 passed, 0 failed Code references added (pinned) Python-Markdown 3.11 (0ffbf00c) inline HTML (a comment inside a list item) goes to the same stash: https://github.com/Python-Markdown/markdown/blob/0ffbf00cf667971d46f43aafba825e8d25098bc5/markdown/inlinepatterns.py#L555-L561 inline processing runs at treeprocessor priority 20: https://github.com/Python-Markdown/markdown/blob/0ffbf00cf667971d46f43aafba825e8d25098bc5/markdown/treeprocessors.py#L42 Other pages Every
on every page built in both copies compared (cmp-tables.py): the only differences are the roles tables and the Label requirements tables named above (Docusaurus 2, Zensical 4, Starlight 2 pages), and each is identical to the unpatched build apart from its id.