Build error with uv on Oracle Linux 7

[demidenm]

Unsure what this bug(?) classifies under, but adding here in the event it is helpful for someone else. I have been using uv environment management for python. My workflow worked fine 3-4 weeks ago. Today, I would get the error:

Using CPython 3.13.2
Creating virtual environment at: .venv
...
× Failed to build pyzmq==27.0.0
  ├─▶ The build backend returned an error
  ╰─▶ Call to scikit_build_core.build.build_wheel failed (exit status: 1)
...
/tmp/tmp476q9wkj/build/_src/_zmq.c:15:6: error: #error Cython requires Python 3.8+.
         15 |     #error Cython requires Python 3.8+.
            |      ^~~~~
      ninja: build stopped: subcommand failed.

      [stderr]

Forcing uv to use python 3.12 didnt resolve the issue. Tried several other things to no avail. Updating my pyproject.toml file by reverting to v.26.4.0 worked:
dependencies = [
"pyzmq==26.4.0", # avoids build error with 27.0.0
....
]

[minrk]

Can you share the os and arch and the exact command used? A full repro example would really help, too, if that's possible.

[demidenm]

Added a light repo which I use to setup an env: https://github.com/demidenm/bidsstatsmod_groupmods . If you remove/comment out this line in the .toml file, it continuously caused the error.

Running on:

uname -a
Linux sh04-04n21.int 3.10.0-1160.119.1.el7.tuxcare.els17.x86_64 #1 SMP Mon Feb 24 04:58:14 UTC 2025 x86_64 x86_64 x86_64 GNU/Linux

The error occurs when I run the setup file:

bash setup_uv.sh

[minrk]

Thanks! I've cloned your repo and run those scripts and it successfully created an env with pyzmq 27.

What linux distro are you using? It looks like maybe it's Tux Care's post-End-of-Life Oracle Linux 7?

What is the output of

ldd --version

Unless you have a super old glibc, you should get a wheel, but Oracle Linux 7 is more than 10 years old, so maybe it's old enough to be a problem. One change from 26 to 27 is that the oldest glibc with a wheel for Python 3.10 is 2.28 and greater, where it was 2.17 for pyzmq 26.

It would also help if you could run without uv:

python3 -m pip wheel -vvv 'pyzmq==27.*'

and share the output. The error message kinda sounds like it might be invoking Cython with old Python not from the isolated build env, but that would be weird. Maybe a bug in our CMakeLists, or maybe a bug in uv? Do you perhaps have a which cython that resolves to something in the host env hooked up to older Python?

[demidenm]

Sorry for delay in responding. Got back logged in emails. You are right, it looks like the clusters on which I run things on use an older el7 linux which comes with the ldd (GNU libc) 2.17 . Which is likely the culprit, as you note.