You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
meson detection broken (on Windows?) in latest release #108
For spin we'll want to do a similar thing as for hatch-meson, because that is the use case where we're already in the build environment. So it can continue to work without setting any environment variable.
For building with build isolation, this improves things. We already use export PKG_CONFIG_PATH there to point at a scipy_openblas in the root (rather than the isolated) environment.
I think the actual reason meson fails to detect pkgconf might be that pkgconf --version gives no output when PKG_CONFIG_PATH is set:
(pkgconf) msmith@MacBook-Pro ~ % pip listPackage Version------- -----------pip 26.1.2pkgconf 3.0.1.post0(pkgconf) msmith@MacBook-Pro ~ % echo$PKG_CONFIG_PATH(pkgconf) msmith@MacBook-Pro ~ % pkgconf --version /Users/msmith/.venv/pkgconf/bin/pkgconf:6: UserWarning: PKG_CONFIG_PATH not specified, and the system is unavailable! The 'pkgconf' executable provided by pkgconf-pypi does not support searching for system-level .pc files, you must provide a search path instead. If you want the search path to be computed from the installed Python packages, please use the 'pkgconf-pypi' executable instead, or set FORCE_PKGCONF_PYPI. sys.exit(_vanilla_entrypoint())3.0.1(pkgconf) msmith@MacBook-Pro ~ % PKG_CONFIG_PATH=whatever pkgconf --version (pkgconf) msmith@MacBook-Pro ~ % pkgconf-pypi --version 3.0.1(pkgconf) msmith@MacBook-Pro ~ % PKG_CONFIG_PATH=whatever pkgconf-pypi --version3.0.1
Yay, thanks everyone! Bumps to pkgconf 3.0.7.post0 are merged for SciPy, and I think @mhsmith's diagnosis was right.
I do have a feeling we'll be revisiting this once more at some point, but I can't quite put my finger on it right now, and the observable problem seems gone.
Activity
lucascolley commentedon Apr 2, 2026
cc @rgommers
rgommers commentedon Apr 2, 2026
I already just noticed, working on a fix. Due to #106 (comment).
virtuald commentedon Apr 2, 2026
I made a fix for hatch-meson at virtuald/hatch-meson#14 -- a fix for meson-python would likely be very similar.
rgommers commentedon Apr 2, 2026
Thanks for sharing @virtuald, that's a useful PR.
rgommers commentedon Apr 2, 2026
@lucascolley my take is that:
spinwe'll want to do a similar thing as forhatch-meson, because that is the use case where we're already in the build environment. So it can continue to work without setting any environment variable.export PKG_CONFIG_PATHthere to point at ascipy_openblasin the root (rather than the isolated) environment.lucascolley commentedon Apr 2, 2026
sounds fine
lucascolley commentedon Jun 6, 2026
any news here?
mhsmith commentedon Aug 8, 2026
I think the actual reason meson fails to detect pkgconf might be that
pkgconf --versiongives no output whenPKG_CONFIG_PATHis set:mhsmith commentedon Oct 17, 2026
The behavior in my previous comment has been fixed in pkgconf 3.0.7.post0. Should this issue be closed now?
lucascolley commentedon Oct 17, 2026
Let me see how CI looks on scipy/scipy#26433
lucascolley commentedon Oct 17, 2026
looks green, @rgommers reckon we can merge the linked PR and close this?
rgommers commentedon Oct 18, 2026
Yay, thanks everyone! Bumps to
pkgconf3.0.7.post0 are merged for SciPy, and I think @mhsmith's diagnosis was right.I do have a feeling we'll be revisiting this once more at some point, but I can't quite put my finger on it right now, and the observable problem seems gone.