[anticipate218]

The C SGR parser applies colon-separated colors immediately, then applies all ordinary parameters afterward. A reset followed by a color, such as \x1b[0;38:2::255:0:0m, therefore loses its color when propagate_sgr() restores it on the next line or clip() restores it at the clipping boundary. The Python implementation already preserves this order.

This extracts the existing integer-parameter application loop and flushes each integer run before applying the next colon color. The integer-only path and color serialization are unchanged. Tests cover resets, default/basic colors, and replacement between the two color syntaxes through both public functions and the C library.

The scope is valid SGR parameter sequences. Incomplete color parameters interrupted by a colon tuple can also behave differently; this does not try to define or normalize malformed input.

Validation on Windows x64, CPython 3.13, MSVC:

• The new public-API tests fail on the original C extension: 10 failed, 4 passed. The new C-library test also fails on the original implementation.
• Full suite with the fixed C extension: 1,934 passed, 11 skipped.
• Full suite in Python mode: 1,897 passed, 48 skipped.
• All 13 CTest programs pass, including allocation-failure coverage.
• Changed files pass clang-format, flake8, isort, docformatter, pylint, and git diff --check.
• The changed files pass codespell. The sdist and Windows C-extension wheel build successfully and pass twine check --strict; the sdist contains the exact tested C source.
• An independent local comparison of 2,190 valid parameter combinations, including integer runs across the stack-allocation boundary, agrees with the Python implementation. Allocation-failure probes found no retained allocations.

This was found while reading the SGR parser, rather than from a reported production incident. No real terminal rendering or performance claim is made. AI assistance: OpenAI Codex generated the patch and ran the checks above.

[jquast]

Superseded by #327

[anticipate218]

Thanks for reviewing this and consolidating it in #327. I appreciate the xterm/kitty investigation and the simpler allocation-free implementation. Thanks for taking this forward and merging the consolidated fix.