Conversation
tomli ships a pure-Python py3-none-any wheel plus mypyc-compiled binary wheels for common platforms, but none for riscv64. This builds the compiled riscv64 wheels (cp312/cp313/cp314/cp314t). The compiled build is opt-in: upstream's pyproject.toml uses the flit_core backend (pure-Python only), and setup.py invokes mypyc only when TOMLI_USE_MYPYC=1. Mirror upstream's binary-wheels job: run their own scripts/use_setuptools.py to swap the backend to setuptools + mypy[mypyc] on the host, then cibuildwheel with TOMLI_USE_MYPYC forwarded into the container. Tests run upstream's exact wheel test command (python -m unittest discover); the toml-test corpus is committed at the tag, so a plain checkout is self-contained. Validated on aarch64 under QEMU: cp312 full build+test cycle green (17 tests), cp314t build green. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
threexc
approved these changes
Aug 24, 2026
threexc
left a comment
threexc
left a comment
Collaborator
There was a problem hiding this comment.
The only thing I think should be changed is removal of the comments in the body about the matrix, checkout step, python install, and build backend switch - these are the LLM explaining things in a lot of extra detail, and someone else trying to use our workflows as a reference might think we did more to customize things than is the case.
LGTM otherwise.
Address PR review: remove the verbose narration on the matrix, checkout, Python install, and backend-switch steps. They over-explained and could mislead someone using our workflows as a reference into thinking we customized more than we did — this is a faithful reproduction of upstream's binary-wheels job. Keep concise one-line "why" notes only. Comment-only change; build logic is unchanged. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
luhenry
added a commit
that referenced
this pull request
Aug 24, 2026
Per repo guidance (PR #308 review): workflows are read as reference, so reserve comments for the non-obvious and cut per-step narration. The full rationale lives in the commit message and CLAUDE.md gotchas 20-22. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
luhenry
added a commit
that referenced
this pull request
Aug 24, 2026
These workflows are read as reference; verbose per-step narration makes it look like we customized more than we did (PR #308 review). Cut the matrix/step paragraphs, keep one why-line each for the non-obvious bits (Cython pin, the easy_install preinstall workaround, the upstream test command). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
luhenry
added a commit
that referenced
this pull request
Aug 24, 2026
Per PR #308 review guidance: reserve comments for the non-obvious (setuptools-rust abi3 flag, no-cp313t, musl skip), one line each; drop the per-step paragraphs that narrated standard steps. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
luhenry
added a commit
that referenced
this pull request
Aug 24, 2026
These workflows are mined as copy-paste reference, so verbose per-step narration makes it look like we customized more than we did (PR #308 review). Cut the multi-line paragraphs and step narration down to single "why" lines, keeping only the non-obvious: BUNDLED-not-vcpkg, the OpenSSL-shared and UNITY_BUILD-off riscv deviations, registry-only deps, and the pandas test fallback. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
luhenry
added a commit
that referenced
this pull request
Aug 24, 2026
* fonttools: add build-fonttools.yml for riscv64 manywheel builds Build compiled riscv64 wheels for fonttools 4.63.0 and publish them to pypi.riseproject.dev. fonttools is a setuptools project whose six Cython extensions (cu2qu, qu2cu, bezierTools, momentsPen, varLib.iup, feaLib.lexer) are compiled from pure-python sources; FONTTOOLS_WITH_CYTHON=1 forces the compile and makes a failure hard (setup.py otherwise falls back to a pure wheel). Build-from-checkout with cibuildwheel over the repo-standard cp312/cp313/cp314/cp314t interpreter set, mirroring upstream's wheels.yml - including its py-cy-noextra tox test command, which asserts the extensions compiled and runs the full pytest suite without optional extras. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * fonttools: pin cython<3.3.0 to fix qu2cu strict-annotation failures The first CI run built compiled wheels on all four interpreters but 9 qu2cu tests failed identically with `TypeError: Expected list, got tuple` at qu2cu.py:237. Root cause is build-tool version drift, not a riscv64 issue: fonttools pins Cython nowhere (setup_requires=["cython"], no [build-system]), so the build pulled the newest Cython, 3.3.0 (released 2026-08-22). 3.3.0 enforces the `quads: List[List[Point]]` argument annotation as a strict runtime type check; the caller passes a list-of-tuples, so the compiled extension rejects it. Upstream's 4.63.0 wheels (sdist 2026-05-14) were built with Cython 3.2.x, which ignores the annotation. Reproduced natively on x86 (not riscv64-specific): Cython 3.3.0 -> 9 failed; 3.2.4 and 3.2.9 -> all pass. Fix pins cython<3.3.0 and preinstalls it with build isolation disabled, because setup.py's setup_requires fetch is an easy_install that ignores pip specifiers - the pinned Cython must already be importable for its has_cython path to use it (gotcha-17 preinstall shape, applied to a build tool). Verified under QEMU: the pin resolves to 3.2.9, all six extensions cythonize, and the wheel builds + auditwheel-repairs cleanly. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * fonttools: rename matrix.build to matrix.python Match the build-fastuuid.yml convention: matrix entries are bare interpreter tags (cp312, ...) and the -manylinux_riscv64 suffix is appended at each use site (job name, cibuildwheel only:, artifact name). No behavior change. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * fonttools: trim workflow comments to single why-lines These workflows are read as reference; verbose per-step narration makes it look like we customized more than we did (PR #308 review). Cut the matrix/step paragraphs, keep one why-line each for the non-obvious bits (Cython pin, the easy_install preinstall workaround, the upstream test command). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
luhenry
added a commit
that referenced
this pull request
Aug 24, 2026
* workflows: add build-bcrypt.yml for riscv64 wheels
bcrypt 5.0.0 is a PyO3/Rust extension built with setuptools-rust (not
maturin). The sdist bundles a committed Cargo.lock + tests/ and a plain
`python -m build --sdist` reproduces the PyPI sdist, so this uses the
sdist -> bdist shape: build the sdist once on ubuntu-latest, then feed
the tarball to cibuildwheel on the riscv64 runner.
Two build jobs mirror the two wheel shapes upstream ships:
* build_abi3: one cp312-abi3 wheel (RISE's min Python floor), built
once and reused+tested on cp312/cp313/cp314. setuptools-rust only
emits abi3 when bdist_wheel gets --py-limited-api, which cibuildwheel
does not set itself, so it is passed via CIBW_CONFIG_SETTINGS.
* build_freethreaded: a per-interpreter cp314t wheel (abi3 has no
free-threaded stable ABI; pyo3 auto-disables it under
Py_GIL_DISABLED). The manylinux_2_39_riscv64 image has no cp313t, so
cp314t is the only free-threaded target.
Rust is installed in-container via CIBW_BEFORE_ALL_LINUX (rustup); musl
is skipped (no riscv64 musl host toolchain). Each wheel runs bcrypt's
bundled pytest suite. Validated end to end under QEMU on aarch64: both
wheels build for riscv64 and all 147 tests pass (abi3 wheel loaded on
cp314, free-threaded wheel on cp314t).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* workflows: trim build-bcrypt.yml comments to single why-lines
Per PR #308 review guidance: reserve comments for the non-obvious
(setuptools-rust abi3 flag, no-cp313t, musl skip), one line each; drop
the per-step paragraphs that narrated standard steps.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Remove unused variable
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Adds
.github/workflows/build-tomli.ymlto build riscv64 wheels for tomli 2.4.1 and publish them topypi.riseproject.dev.tomli ships a pure-Python
py3-none-anywheel plus mypyc-compiled binary wheels for common platforms (cp311–cp314/cp314t, x86_64/aarch64 manylinux+musllinux, Windows, macOS) — but none for riscv64. This builds the compiled riscv64 wheels so riscv64 users get the same accelerated parser everyone else does.Ref: #94
How
Mirrors upstream's own
binary-wheelsjob intests.yaml:pyproject.tomldeclares theflit_corebackend (pure-Python only).setup.pyinvokesmypyc.build.mypycify()only whenTOMLI_USE_MYPYC=1. So the workflow runs upstream's ownscripts/use_setuptools.pyon the host (rewrites[build-system]→setuptools>=77.0.3 + mypy[mypyc]>=1.15), then runs cibuildwheel withTOMLI_USE_MYPYC=1forwarded into the manylinux container viaCIBW_ENVIRONMENT_PASS_LINUX.[cp312, cp313, cp314, cp314t]— mypyc emits per-interpreter (non-abi3) wheels; PyPI confirms separatecpXYwheels with no-abi3-tag. (cp311 dropped per the repo's 3.12+ policy.)python -m unittest discover --start-directory {project}. The full toml-test corpus is committed directly at the tag (no submodules), so a plain checkout is self-contained. tomli has zero runtime and zero test deps, so no registry index env is needed.Validation (local, aarch64 under QEMU)
cp312-manylinux_riscv64full cibuildwheel build + test cycle green — 17 tests OK; cibuildwheel ran the suite from a temp cwd soimport tomlicorrectly resolved to the installed compiled.so(not./src).cp314t-manylinux_riscv64build green (free-threaded + mypyc)._parserloads from.cpython-312-riscv64-linux-gnu.so), auditwheel-repairs tomanylinux_2_31_riscv64.manylinux_2_39_riscv64, ~286 kB, no vendored external libs.python -c "import yaml; yaml.safe_load(...)"andactionlintclean (only the expectedubuntu-24.04-riscvunknown-label warning).The
publishjob dry-runs on this PR branch (offmain), so it's safe here.🤖 Generated with Claude Code