fails to install in python 3.10 about swiglpk HOT 4 CLOSED

After hitting this problem quite a few times in different contexts, I want to note that this issue and closely related ones will be re-experienced by your users with every release of a new Python major version.

Here is what happens every time, as far as I understand it:

1. Users install swiglpk, thinking this gives them everything they need. Since normally there is a wheel and that wheel appears to include GLPK (as e.g. the Python:3.9 docker image lets you solve problems after installing only swiglpk from pip), they are for that moment correct.
2. A new Python major version is released. There are no wheels for this version yet and there won't be any for a couple more days to months. pip sees that there are no wheels and downloads the source instead. It executes setup.py.
3. Now either (a) setup.py does not find GLPK installed to link swiglpk against (this issue) or (b) GLPK is installed but setup.py fails to locate it as the location code is not platform-independent (#40) or (c) the user has GLPK but not SWIG or some other build-time dependency installed (#36).

I don't know a good long-term solution but I suggest to at least make the error messages produced by setup.py really explain the situation, like I tried above. This would give users running swiglpk as part of a CI/CD pipeline a heads-up warning that it will break again when they bump the Python version another time, and it would allow them to work around the issue by making sure every dependency necessary for a source build is installed (and by patching setup.py to find it, if necessary).

from swiglpk.

Okay, this time it's really fixed. Wheels for 3.10 are now provided for swiglpk 5.0.4.

from swiglpk.

I sent in a PR to support 3.10.

@Viech this is more an issue of a lack of contributors more than anything else. We have very few contributors across the COBRA ecosystem at the moment and new Python versions will always require at least a minimal amount of adjustments. We definitely welcome PRs to fix those issues though :)

Additional work-around is to use conda with a supported Python version or setting the GLPK_HEADER_PATH environment variable. swiglpk and optlang are on conda-forge and have a few more contributors.

from swiglpk.

Great, thank you @cdiener !

from swiglpk.