[markbac]

py-doc draws a diagram written in the Azure DevOps form, :::name … :::, for every diagram language (diagrams/scan.py). The extension highlighted only backtick fences, so a :::mermaid block showed as plain text. A real document with these blocks showed the gap.

Changes

• :::name rules for every diagram language: syntaxes/pydoc-fences.markdown.tmLanguage.json has one beside each language's backtick-fence rule, ending at a line of only colons, as py-doc reads it. :::bpmnt uses the bpmn-text grammar. Poster blocks have no ::: form in py-doc, so none is added.
• No overlap with the general ::: rule: the fenced-div rule in pydoc.markdown.tmLanguage.json now leaves these lines alone, so ::: notes and ::: {.pagebreak} are still highlighted as divs.
• Troubleshooting: the README now covers another extension's grammar colouring ordinary words. This was the cause of the stray colours reported, likely a Mermaid extension that reads :::mermaid. The fix is to inspect a coloured word's scopes and, if none ends in .pydoc, disable the other extension for the workspace.
• Version: 2.3.1, with a CHANGELOG entry.

Testing

• New JS test: a :::mermaid block is highlighted as Mermaid, the paragraph after it is plain Markdown again, and ::: notes stays a div. It passes with the stand-in host and with VS Code 1.105's Markdown grammar.
• New Python test: the ::: rules cover exactly the diagram fences py-doc's renderers define.
• The reported document, run through the grammars with the real Markdown grammar:
  • both :::mermaid blocks come out as Mermaid;
  • no prose line picks up another language's scopes;
  • the xml block is meta.embedded.block.xml.
• Results:
  • npm run lint && npm test: 32 passed, 1 skipped without the real grammar.
  • pytest tests/test_editor_grammar.py: 3 passed.
  • ruff is clean, and the README lints clean with Vale.
  • vsce package builds the 2.3.1 .vsix.

---

Generated by Claude Code

[markbac]

Test (windows-latest) failed on 90a03fb with a pytest Timeout in tests/test_vscode_install_script.py (from #222). This PR doesn't change that script or test.

The cause is the test's time budget, not the code. The first test builds the real extension twice (npm ci, vsce package) and runs PowerShell three times. On a Windows runner that takes about two minutes, right at the suite's 120 s pytest-timeout: it passed narrowly on #222 and went over here.

Fixed in 562977a: those tests now have pytest.mark.timeout(900), matching the 600 s each PowerShell run may take. No test is skipped, and the assertions are unchanged.

---

Generated by Claude Code