fix(content)!: spell literal text so python-markdown reads it as text (0.6.0)

[baraline]

Why

from_transport passed markdownify's output on as Markdown with escaping switched off. Literal text in a GLPI body therefore changed meaning as soon as python-markdown rendered it, which is what to_transport and any consumer does.

Measured through the reader into python-markdown, with the four extensions to_transport uses:

• \\serveur\compta lost a backslash, and __init__ became bold;
• a ----- line under text made a heading;
• * point and > merci lines became a list and a quote;
• #4521 at the start of a line became a heading;
• [1]: https://… was consumed as a reference definition;
• a | inside a table cell dropped the rest of the row;
• nested lists flattened.

What changes

• The reader (GlpiContentConverter.from_transport, and so every read model's .content) escapes a character only where python-markdown would read it as syntax:
  • with a backslash where python-markdown removes one;
  • with a character reference (<, &, =, ~) where no backslash escape exists.
• What stays untouched:
  • ordinary prose comes back byte for byte, for example fichier_de_test_v2.xlsx, C:\Temp, R&D or a # in mid-sentence;
  • pasted URLs stay <https://…>;
  • code spans and code blocks are not escaped.
• Structure:
  • nested lists indent by four spaces and keep their numbering;
  • blocks inside list items stay inside them;
  • <script>, <style> and <title> bodies are dropped, where they used to leak as prose;
  • obsolete elements (font, center, strike, …) are recognised;
  • <s> and <del> keep their words without a literal ~~.
• Dependencies: markdownify>=1.2, because the subclass needs the 1.x hooks. Version 0.6.0.

Breaking for consumers

• .content now carries backslash escapes and character references wherever literal text would otherwise be misread. Render it before displaying or indexing it.
• The Markdown read from GLPI changes once for most e-mail-created bodies: new escapes, dropped CSS, and source newlines read as spaces. Stored digests of .content therefore change once.
• A write-model value that holds one real HTML element is read as HTML throughout. Its newlines become spaces, and any Markdown syntax beside the element stays literal.

Verification

• pytest -m "not integration" with coverage: 1569 passed, 17 xfailed, 97.6 % (the gate is 95 %).
• ruff, mypy (strict), unasync_build.py --check and sphinx -W pass. Sphinx was built offline, with intersphinx disabled locally. The wheel was checked for test files.
• Property tests use an HTML-parser oracle and check that to_transport(from_transport(html)) displays what html displays. 50,000 fuzzed bodies gave 0 failures.
• An independent review re-ran the gates and tried about 3 million generated bodies. Its open findings are listed below.

Still to do before this leaves draft

These come from that review.

• A table nested inside a table cell loses all its text, which is common in Outlook signatures. The bug predates this branch (0.5.0 does the same). The fix is to flatten an inner table to its cells' text.
• A body with no HTML element is still returned verbatim, so its literal text is still misread. Spell it literally on the read path only; the write-model validator must stay verbatim.
• Edge whitespace such as   defeats the line rules in three places: the start of the first block, the start of a list item, and the end of the last block.
• Two list shapes take quadratic time, for example thousands of empty items. The CHANGELOG's "every pass is linear" needs that fix to be true.
• The setext-underline rule is copied from python-markdown 3.10.3, and 3.10.2 behaves differently. Derive the rule from the installed version, or raise the floor.
• Five guards have no test that fails without them:
  • fences in separate paragraphs;
  • SRV___PROD___01;
  • leading whitespace before top-level syntax;
  • a titled pasted URL;
  • a _ b _ c.
• Spell link destinations, link titles and the ticket-context transcript literally too.
• Narrow the documented fixed-point guarantee: some e-mail shapes settle after one extra cycle, with no visible change.
• Head the CHANGELOG section "Changed (breaking)".

The EasyVista port of this converter: baraline/easyvista_python_client#6

🤖 Generated with Claude Code

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

❌ Patch coverage is 97.12230% with 8 lines in your changes missing coverage. Please review.
✅ Project coverage is 97.64%. Comparing base (f93411d) to head (524304a).

Files with missing lines	Patch %	Lines
glpi_python_client/content/conversion.py	96.91%	8 Missing ⚠️
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #39      +/-   ##
==========================================
+ Coverage   97.59%   97.64%   +0.04%     
==========================================
  Files          90       90              
  Lines        3163     3313     +150     
==========================================
+ Hits         3087     3235     +148     
- Misses         76       78       +2     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.