skia-python: add build-skia-python.yml for riscv64 wheels - #1697
Open
luhenry wants to merge 8 commits into
Open
Conversation
luhenry
added a commit
that referenced
this pull request
Sep 10, 2026
…t a display Rocky 10 (the riscv64 manylinux image's base) carries no virtual X server at all -- xorg-x11-server-Xvfb, tigervnc-server and weston are all absent from BaseOS/AppStream/CRB/Extras, on every architecture, not just riscv64. Drops the dnf install and the xvfb-run wrapper, and carries a patch so the 'gpu'-parametrized tests skip gracefully instead of erroring when glfw can't reach a display.
build/gen.py's GenerateLastCommitPosition runs 'git describe HEAD --match initial-commit' unconditionally whenever .git exists, to generate a cosmetic version header; a depth-1 clone has none of the history back to that root tag, so it fails outright with 'No names found, cannot describe anything.' gn's repo is 42MB full, ~4s to clone -- shallow isn't worth the fragility here.
build/gen.py's WriteGNNinja defaults cxx to os.environ.get('CXX',
'clang++') on any non-MSVC/AIX/Windows host, unconditionally --
verified against a real clone of gn's own source, and against the
live manylinux_2_39_riscv64 image manifest (quay.io), which has no
layer installing clang anywhere; only GCC is provisioned. The
/opt/clang/bin PATH entry baked into the image's own ENV is a dead
path on this architecture. Skia's own build already assumes GCC
(upstream's gn args set no is_clang), so this just makes gn's
bootstrap use the same toolchain the rest of the build already relies
on.
GCC's -Wcomment fires on src/gn/header_checker.h's ASCII-art dependency-graph diagram (a // line ending in a backslash, e.g. '// / \'), and gen.py sets -Werror unconditionally unless told otherwise. gn's own gen.py ships an --allow-warnings flag for exactly this (source only gets -Werror coverage against clang upstream); using it is cleaner than patching a third-party bootstrap tool's own source for an ASCII-art comment. gn itself never ships in the wheel, so this carries no risk to the built extension.
gn links itself with -static-libstdc++ -pthread; the riscv64 manylinux image doesn't ship libstdc++.a by default (gotcha 77), only via the CRB libstdc++-static package, already enabled for the other -devel packages on this same line.
…wnload tools/git-sync-deps calls bin/fetch-gn unconditionally, with no skip flag (unlike its --skip-emsdk sibling). fetch-gn's CPU map has no riscv64 entry and Google's CIPD has no prebuilt gn for riscv64 to add one for, so it KeyErrors outright. We already built and installed our own gn to PATH before this step runs; nothing downstream reads skia/bin/gn or third_party/gn/gn directly, so skipping the download when a gn is already on PATH is a correct no-op here.
BUILD.gn's 'skia.h' action target hardcodes
args = [rebase_path("//bin/gn")] + ... to invoke gn/find_headers.py,
which runs 'gn desc' to enumerate public include dirs for the
generated skia.h umbrella header -- this is evaluated at ninja build
time (not gn gen time), and reads //bin/gn directly rather than
resolving through $PATH or GN's own GetExePath()-based
self-regeneration mechanism. Confirmed against BUILD.gn and
gn/find_headers.py at the pinned skia commit before pushing, since
this failure only surfaces ~2.5h into the ninja build. Our patched
bin/fetch-gn correctly no-ops (gotcha: no riscv64 CIPD binary) but
never populates this path since it never runs its download logic;
copy our self-built gn there explicitly, same binary already at
/usr/local/bin/gn.
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
1 participant
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.
skia-python144.0.post2Compiles Google's Skia 2D graphics library from source (GN/Ninja) and links it into a pybind11 extension. Upstream publishes no riscv64 wheel.
Mirrors upstream's
ci.ymlbuild_wheelsjob, grouping all four interpreters into oneCIBW_BUILDso the Skia compile inCIBW_BEFORE_ALL_LINUXruns once, not per interpreter.Differs from upstream
scripts/build_Linux.sh'syum install ninja-build gnreimplemented inline - both packages come from EPEL, which the riscv64 image doesn't carry.xvfb-runwrapper andxorg-x11-server-Xvfbinstall - Rocky 10 ships no virtual X server at all (Xvfb, Xvnc, weston all absent, every arch).Testing
moderngl(no riscv64 wheel): deselectstest_GrDirectContext_MakeGL_EGL, the one test that imports it.License: OK
Patches
0001-Skip-GPU-context-tests-when-glfw.init-fails-not-just.patch- Inappropriate [Rocky 10 ships no Xvfb/Xvnc/weston]. Without it, everygpu-parametrized test errors, not skips, with no display. Riscv64-only.Built on cp312/cp313/cp314/cp314t; pending CI results.