gh-87414: add musl support to platform.libc_ver

[cmaureir]

This includes an implementation from the packaging module to detect the musl version on the system,
allowing libc_ver to properly return the version rather than an empty tuple.

• Issue: platform.libc_ver() returns no information on Alpine Linux #87414

[humitos]

We QAed this patch locally and we got it working. However, we are not 100% sure if this is a good implementation and, in the case it is, we are not sure if the place were we copied the functions and put the new file is OK, so:

• Is it fine adding the _elffile.py in the Lib/ directory?
• Are tests for this class required? Where we should add them? Should we add a binary file to the repository so execute the tests over? 🤔
• We copy some of the functions from packaging._musllinux.py into Lib/platform.py. Is that OK or we should create a Lib/_musllinux.py instead?
• We are checking for b'musl' in binary inside the while loop and the opening the same binary again to be parsed by ELFFile. The idea behind this was to not open the executable twice for all the cases, but only for musl. We are not convinced about this either, but we are not sure what could be better.

cc @pradyunsg@malemburg

[malemburg]

A few comments:

• Please add tests for this (possibly mocking things using dumps of actual musl files)
• Please don't add type annotations to the platform stdlib code. If you want to add those, please post them to typesched.
• Adding the helper module to Lib/ is fine, just please add a notice to the module that this is an internal module and not meant for public consumption.
• Before adding the helper module, please make sure you get proper license information for it from the original source and add this to the module doc-string.
• Function doc-strings always go first in a function definition, not after some imports (see _get_musl_version()). Please adjust the code.
• Please replace the NamedTuple _MuslVersion with a simple Python tuple. No need to over-engineer the code.

I have no way to test the code, so can only do source level review. Please find someone who can test this in real life.

Thanks.

[cmaureir]

Thank you Marc 🎉

I addressed the following points:

Please don't add type annotations to the platform stdlib code. If you want to add those, please post them to typesched.

Adding the helper module to Lib/ is fine, just please add a notice to the module that this is an internal module and not meant for public consumption.

Function doc-strings always go first in a function definition, not after some imports (see _get_musl_version()). Please adjust the code.

Please replace the NamedTuple _MuslVersion with a simple Python tuple. No need to over-engineer the code.

Before adding the helper module, please make sure you get proper license information for it from the original source and add this to the module doc-string.

I naively assumed that due to the code being used inside the packaging module it should have been straightforward, but I totally understand, will be double check :)

We will work on this soon with @humitos

Please add tests for this (possibly mocking things using dumps of actual musl files)

I have no way to test the code, so can only do source level review. Please find someone who can test this in real life.

Understood, thanks! maybe the person who reported it can give us a hand.

[humitos]

I have no way to test the code, so can only do source level review. Please find someone who can test this in real life.

This is how we are testing this code. I hope it makes sense:

1. Pull the Alpine Docker image and run it

docker pull alpine docker run -it -u root -v ~/Source/cpython:/mnt/cpython sh 

2. Compile Python inside Alpine

export PYTHON_BUILD_PACKAGES="\ bzip2-dev \ coreutils \ dpkg \ dpkg-dev \ expat-dev \ findutils \ gcc \ gdbm-dev \ libc-dev \ libffi-dev \ libnsl-dev \ libtirpc-dev \ linux-headers \ make \ ncurses-dev \ libressl-dev \ pax-utils \ readline-dev \ sqlite-dev \ tcl-dev \ tk \ tk-dev \ util-linux-dev \ xz-dev \ zlib-dev \ " apk add --no-cache --virtual .build-deps $PYTHON_BUILD_PACKAGES cd /mnt/cpython ./configure make 

3. Execute Python and check the response

Python 3.12.0a7+ (main, Apr 25 2023, 16:58:19) [GCC 12.2.1 20220924] on linux Type "help", "copyright", "credits" or "license" for more information. >>> import platform >>> platform.libc_ver() ('musl', '1.2') >>> 

Note: based on https://github.com/jfloff/alpine-python/blob/master/3.8/Dockerfile#L33

[malemburg]

Thanks for the instructions, @humitos. I forgot that Docker containers are an easy way to access Alpine builds :-)

BTW: I thought a bit more about adding a new modules to Lib/ and would like to revert my earlier comment. For the small extra logic that is needed, it will create more maintenance burden than is worth it. Could you please try to add the required code directly the platform module ? Perhaps it can be refactored and shortened, since you are not really using all of the parsing logic.

Thanks.

[humitos]

I updated the PR to live only in the platform.py module. I also trimmed the code to remove the things that are not strictly required for this purpose.

I wrote some small tests for ELFFile class and two tests for the functions that uses it. These are under a @support.requires_musl decorator that I created to avoid running these tests in Windows, Mac and Linux versions that don't have musl installed.

Please, let us know if there are more changes required 😄

[malemburg]

Still a few minor things to fix, but overall, this looks good.

Please also add a NEWS entry.

Thanks.

[bedevere-bot]

A Python core developer has requested some changes be made to your pull request before we can consider merging it. If you could please address their requests along with any other requests in other reviews from core developers that would be appreciated.

Once you have made the requested changes, please leave a comment on this pull request containing the phrase I have made the requested changes; please review again. I will then notify any core developers who have left a review that you're ready for them to take another look at this pull request.

[malemburg]

The test fails on Macs due to a missing "ldd". On Macs, you typically use "otool" instead.

It would be better to simply return False in requires_musl() or use "otool" instead on "ldd" on Macs.

[mara004]

@humitos Thanks for working on this PR. May I ask how this is going on? Seems to be stalled with requested changes since May...

[blubberdiblub]

@cmaureir this pull request seems nearly ready for merging, only 3 small issues (comment for the musl in binary check in platform.py, fixing the requires_musl() function to work gracefully on MacOS and replacing the web links into pypa/packaging sources with permalinks pinned to a specific commit/tag) remain which should be easy to address.

Would be really nice if this could be merged soon

[cmaureir]

Super sorry @malemburg@blubberdiblub I thought this was merged long time ago 😢
I will keep an eye on this issue until it gets in 👍 thanks for your time.

[encukou]

Will this always be 1.2?

[encukou]

@malemburg, if you want to merge this, please go ahaed. I'll write about what would make me comfortable with merging it.

---

This is quite a lot of code for an unsupported platform.
By unsupported I mean both that:

• musl is not only not officially listed in PEP 11, and
• we don't have musl in CI; while we have a buildbot I doubt anyone's checking it; there's no person to contact with musl-related issues.

Getting a platform supported is a bit of catch-22: you need a core dev, and it's hard to become one if you're interested in an unsupported platform. To help solve that, I'd like to figure out how community support for a platform could work.

Is anyone here interested in maintaining CPython on musl? Ideally someone interested in eventually becoming a core dev and bringing musl to official Tier 3 support? If so, please add yourself to the experts index with a PR here.

The indication that someone wants to help keep CPython working on musl, with Marc-Andre's “this looks good” comment, would convince me to merge this.

[malemburg]

This is quite a lot of code for an unsupported platform. By unsupported I mean both that:

This is really only a side effect of adding ELF header parsing instead of using the file output. We may be able to make good use of it on other platforms as well.

Regarding supporting the musl platform: I think we simply have to accept that CPython will run on unsupported platforms as well 🙂

Alpine is a very widely used platform for running CPython on Docker/Podman, whether we have official tier-X support for it or not. Simply rejecting patches for such a common platform, just because we can't check all the boxes for official tier-X support feels somewhat unreasonable -- at least to me.

Of course, happy to learn if someone wants to help check those boxes for Alpine platforms.

[mara004]

This is quite a lot of code for an unsupported platform.

Given that there has been a successful PEP for musllinux platform tags (656), "unsupported" sounds a bit hyperbolic to me. 😅 Python is definitely being used on musl systems.

there's no person to contact with musl-related issues

Not in CPython itself perhaps, but there are still downstream maintainers -- Alpine packagers would probably be happy to help.

[encukou]

This is really only a side effect of adding ELF header parsing instead of using the file output.

Sure. Then again, the first thing I'd like to ask a musl maintainer is whether ELF parsing is the proper way to get the libc version.
(If it breaks somewhere in CPython's 5-year support period, it won't be easy to fix. Maybe users would be better off using a third-party platform identification library, which would be easier to update?)

Not in CPython itself perhaps, but there are still downstream maintainers -- Alpine packagers would probably be happy to help.

Yup. We don't test it here, so we can't really take responsibility for fixing it when it breaks. And we don't have musl experts on board. That's an issue to be fixed, and while I can't fix it myself, I can help on the organizational side of things.
I suppose Alpine packagers do run the test suite and carry necessary patches. I'm all for inviting them to do some of that work in this repository, if they want to sign up for it.

[zware]

As near as I can tell, this has mostly been superseded by GH-131313. Is there anything here that can/should improve on what was done there?

[malemburg]

Indeed and the solution there is much leaner than this one. I'm going to close it for now.

Thanks @cmaureir for investing the time. Please reopen the issue if you think there are still some parts which would be useful to have in the platform module.

[mcepl]

docker pull alpine docker run -it -u root -v ~/Source/cpython:/mnt/cpython sh export PYTHON_BUILD_PACKAGES="\ bzip2-dev \ coreutils \ dpkg \ dpkg-dev \ expat-dev \ findutils \ gcc \ gdbm-dev \ libc-dev \ libffi-dev \ libnsl-dev \ libtirpc-dev \ linux-headers \ make \ ncurses-dev \ libressl-dev \ pax-utils \ readline-dev \ sqlite-dev \ tcl-dev \ tk \ tk-dev \ util-linux-dev \ xz-dev \ zlib-dev \ " apk add --no-cache --virtual .build-deps $PYTHON_BUILD_PACKAGES cd /mnt/cpython ./configure make 

For anybody who tries that, let me add that Python officially doesn’t support LibreSSL anymore (e.g., #89532), replace libressl-dev with openssl-dev.

And yes, platform.libc_ver() works:

/mnt/cpython # ./python Python 3.14.0a7+ (main, Apr 29 2025, 18:48:02) [GCC 14.2.0] on linux Type "help", "copyright", "credits" or "license" for more information. >>> import platform >>> platform.freedesktop_os_release(){'NAME': 'Alpine Linux', 'ID': 'alpine', 'PRETTY_NAME': 'Alpine Linux v3.21', 'VERSION_ID': '3.21.3', 'HOME_URL': 'https://alpinelinux.org/', 'BUG_REPORT_URL': 'https://gitlab.alpinelinux.org/alpine/aports/-/issues'} >>> platform.libc_ver() ('musl', '1.2.5') >>> /mnt/cpython # 

[cmaureir]

Hey! sorry for being late to the party :P

I'm happy this was done anyways, independently if this patch was not the final solution. I know many Alpine Linux (and other musl users) will be happier

A big thanks to @humitos for all the time we spend discussing this, testing, and pinging each other in order to have one-more-push to finish it. It was a nice experience.

And thanks to everyone that took the time to review what we were doing 🤓

See you all in a future PR!