Arch Linux User Repository

I'm seriously considering bumping the pkgrel for a bunch of changes:

• --with-threads is no longer recognized, which should be removed. No questions with that.

• Enable --with-ensurepip. For now, I don't see building --without-ensurepip could yield any benefits. The ensurepip module and .whls of setuptools & pip are installed anyway even when building --without-ensurepip. It's just the bundled .whls might become out of date in the future, which doesn't meet Arch's "only run latest software" guideline.

• Running tests (either in check() or building with --enable-optimizations) seems to be causing trouble for some people. Building with --enable-optimizations claims to boost performance for 10% ~ 20% (1). Personally I don't quite care about Python's performance on my dev machine that much. Another problem is: the test suite could run for as long as 30+ minutes, which is pretty slow comparing with other AUR packages that I use. But I'm guessing that people might expect this package to be "just like python in the official repo", which is built with --enable-optimizations enabled. So I still haven't decided on this yet.

I found this in the makefile:

# The task to run while instrumented when building the profile-opt target.
# We exclude unittests with -x that take a rediculious amount of time to
# run in the instrumented training build or do not provide much value.
PROFILE_TASK=-m test.regrtest --pgo

But whoever wrote the comment must have forgotten to actually add -x to the command. So I added this to PKGBUILD in prepare():

  sed -i -Ee 's|(PROFILE_TASK=-m test.regrtest --pgo)|\1 -x|' Makefile

This avoids running the tests while still building an optimized build, and gave me a functional 3.7 build in about 10 minutes.

sorry for false-flag

@Fuzzy: Yes, building with --enable-optimizations does run tests due to PGO optimization. Maybe I'll just remove this ./configure option in the next version of PKGBUILD (but I haven't decided yet).

About the missing _ctypes module: you can try building the package in a clean chroot, which is explained in the wiki: https://wiki.archlinux.org/index.php/DeveloperWiki:Building_in_a_clean_chroot This isn't trivial to setup and it will take up a fair bit of disk space. But it should solve most problems of "package builds differently on different machines".

Hi! First, thanks for this package. I also ran into troubles with last arch upgrade... But: I don't see check() section in my PKGBUILD but somehow tests are ran. There are several failed tests but worse - 'test_socket' gets stuck. I had removed --enable-optimizations and yay built and installed it successfully. At first look it works but now I have another error: No module named '_ctypes' (pacman shows that I have libffi installed...)

I'm seriously considering bumping the pkgrel for a bunch of changes:

• --with-threads is no longer recognized, which should be removed. No questions with that.

• Enable --with-ensurepip. For now, I don't see building --without-ensurepip could yield any benefits. The ensurepip module and .whls of setuptools & pip are installed anyway even when building --without-ensurepip. It's just the bundled .whls might become out of date in the future, which doesn't meet Arch's "only run latest software" guideline.

• Running tests (either in check() or building with --enable-optimizations) seems to be causing trouble for some people. Building with --enable-optimizations claims to boost performance for 10% ~ 20% (1). Personally I don't quite care about Python's performance on my dev machine that much. Another problem is: the test suite could run for as long as 30+ minutes, which is pretty slow comparing with other AUR packages that I use. But I'm guessing that people might expect this package to be "just like python in the official repo", which is built with --enable-optimizations enabled. So I still haven't decided on this yet.

@leandro: for pip3.7 you can edit PKGBUILD to replace --without-ensurepip with --with-ensurepip and rebuild the package.

@MuffinBomber: I'm no expert in building Python. Maybe the test case you are stuck on is a different test case. Not the one removed in https://bugs.python.org/issue34587.

And yes: by removing the line of --enable-optimizations from PKGBUILD, a clean rebuild (with makepkg -Csf) no longer runs the tests on my machine.

When building the package 'test_socket' gets stuck. Is there a way to solve this or should I just remove the '--enable-optimizations' line and skip the tests completely (while also dealing with slower executing code)?

EDIT: According to https://bugs.python.org/issue34587 that test was removed from 3.7, but somehow it is still present?

Thanks for your insights. I'm no expert especially with python.

My package doesn't run with 3.8 anymore and upstream is (nearly) dead. Looks like this will be a bigger task to get it run again.

@Lucki: I'm not quite sure if this pacakges should claim provides=... in its PKGBUILD.

Python packages are, in a way, built for a specific version of Python. Most of their files are installed into /usr/lib/pythonX.Y and they won't run under a different version of Python. If a package only runs under Python 3.7, I'd suggest it claims python37 as its dependency. The Python version is hard-coded in the package name already.

Another problem is: only Arch's official python package provides /usr/bin/python. Other packages must not do that. Even if python37 does provides=python<3.8, any python scripts starting with #!/usr/bin/python won't run. So the provides=... just become a false promise. This whole situation may be resolved once pacman implements an alternatives system (which is being discussed here: https://lists.archlinux.org/pipermail/pacman-dev/2019-October/023691.html) in the future.

But still, this is my first time writing a "serious" PKGBUILD. And by writing I mean "copying from official repo and python36". What I've said above may very well be wrong. Feel free to point out any mistakes.