[smoltafet]

Description

Fixes #1501.

Currently a valueless attribute in an attribute list is stored as key="key", which makes the same source render differently — with different meaning — depending on the output format:

>>> markdown.markdown('[download](file.txt){download}', extensions=['attr_list'])
'<p><a download="download" href="file.txt">download</a></p>'
>>> markdown.markdown('[download](file.txt){download}', extensions=['attr_list'], output_format='html')
'<p><a download href="file.txt">download</a></p>'

download="download" gives the downloaded file a default name of download; bare download does not. The reverse problem also existed: an explicit {download="download"} was collapsed to bare download under html.

This implements the approach sketched by @waylan in #1501 (comment):

• A list of known boolean attributes is incorporated (markdown/serializers.py: BOOLEAN_ATTRIBUTES), differing per output format — the XHTML list is the one from the XHTML 1.0 spec, the HTML list is from the HTML spec — and download is on neither list.
• At parse time (markdown/extensions/attr_list.py), a valueless attribute is represented internally as key=None. When the attribute is assigned to an element, the historic key="key" behaviour is maintained only for known boolean attributes of the active output format; any other valueless attribute keeps the value None, preserving what the user provided.
• The serializer (markdown/serializers.py) renders a None value as the bare attribute name in both formats, and renders key="key" in the minimized form under html only for known boolean attributes — so explicit values are now preserved as provided for non-boolean attributes.
• Knock-on fixes: UnescapeTreeprocessor skips None attribute values, and fenced_code's handle_attrs (which shares the parser) keeps its historic treatment of a valueless config word as a flag using the key as its own value.

Result:

>>> markdown.markdown('[download](file.txt){download}', extensions=['attr_list'])
'<p><a download href="file.txt">download</a></p>'
>>> markdown.markdown('[download](file.txt){download}', extensions=['attr_list'], output_format='html')
'<p><a download href="file.txt">download</a></p>'
>>> markdown.markdown('[download](file.txt){download="download"}', extensions=['attr_list'], output_format='html')
'<p><a download="download" href="file.txt">download</a></p>'
>>> markdown.markdown('[x](y){checked}', extensions=['attr_list'])
'<p><a checked="checked" href="y">x</a></p>'   # known boolean: historic behaviour

Backward compatibility: known boolean attributes render exactly as before in both formats. Three cases in the legacy attr_list fixture had expectations which changed, each being an instance of the bug (a valueless non-boolean attribute such as foo now renders as bare foo instead of foo="foo"); at the reviewer's request, those cases were moved out of the legacy fixture and into the new-style tests in tests/test_syntax/extensions/test_attr_list.py.

Tests: 17 new tests in tests/test_syntax/extensions/test_attr_list.py covering valueless non-boolean attributes in both formats, cross-format consistency, explicit download="download" preservation, known booleans (shared, XHTML-only compact, HTML-only open), the key=None parse representation, serializer-level rendering of None values, and the fenced-code config-word regression. Full suite (python -m unittest discover tests, the runner from tox.ini): 1154 tests OK (baseline 1137 OK, same 111 skips) — zero regressions. Note: this change was developed with AI assistance.

AI Assistance Disclosure

• No AI tools were used in preparing this PR.

• AI tools were used and I have disclosed which ones, and have fully reviewed and verified their output.

  AI tools were use to assist with the following:

  • Code
  • Code Comments
  • Documentation
  • Changelog Entry
  • Commit Message
  • The Description Above
  • Other: ______________

  Tool used: Muse, Anthropic's coding assistant. It generated the code changes, the new tests, the commit message, and the description above. I ran the full test suite locally to verify the results (1154 tests OK, zero regressions against the 1137-test baseline).

Checklist

• This PR follows the contribution guidelines.
• The code follows the Code Style Guide.
• The commit message follows the Commit Message Style Guide.
• I have added or updated relevant docs, including release notes if applicable which follow the Documentation Style Guide.
• I have added or updated relevant tests.
• I have not requested, and will not request, an automated AI review for this PR.

[waylan]

As per our Contributing Guide:

All contributions must disclose whether AI tools were used or not. Disclosures of AI use must include the specific tools used and what content was generated by the tool, including, but not limited to code, documentation, comments, commit messages, and explanations. Issue and pull request templates must be followed when provided. If a template is ignored or deleted, it will be assumed that an AI agent was used to submit the pull request or issue autonomously and the pull request or issue may be closed immediately.

You appear to have deleted the PR Template. This will be your only warning.

[smoltafet]

Sorry about that — the description now follows the PR template, and the disclosure section is filled in with the tool used and what it generated.

[smoltafet]

@waylan another look please ty

[smoltafet]

@waylan fixed — moved the three changed legacy cases (random attributes, Bad Syntax header, key without value) out of tests/extensions/attr_list.txt/.html into new-style tests in tests/test_syntax/extensions/test_attr_list.py, and added the changelog entry. Also fixed the spelling check, it was flagging the word config in two comments I added. Tests pass locally (1154 OK) and the spelling check passes too.

[smoltafet]

@waylan thanks for the review and for merging this, appreciate it