How to install pip for python3.13t

Python Help
1

Hello,
I’ve installed v3.13.3t with PEP703 support in an Ubuntu 24.04 system (which already included a v3.12 version). The interpreter seems to be working OK but I simply cannot install pip. If I try using the package manager ($ sudo apt-get install python3-pip), the OS will install pip for v3.12, which is not what I want. When I try a semi-manual installation

$ python -m ensurepip --upgrade
/usr/bin/python3.13t: No module named ensurepip

It doesn’t seem to have ensurepip. I’m not sure if pip would need to be installed from source or which would be the workaround this issue.
Thanks.

2

Do you absolutely need to use the distro package for 3.13t? You can use pyenv to build Python from source if you don’t mind having compilers and some system libraries installed.

3

Hi @ngoldbaum,
Thank you for the reply. The problem is that once python is installed, I need to build a long list of apps on top of it and, in the past, some dependencies have caused some issues (more particularly those needed by PyQt and QGIS). I guess that the only way to assess if everything can work together is to try a clean installation, which I’ll do first thing tomorrow.

4

I’d suggest raising a bug report with ubuntu about this.

5

Just checked on Fediora 42 and pip is not packaged there as well.

I wonder if this is because it is unclear if packages are free threading compatible?

6

I just did this an experiment on Fedora.
I set the PYTHONPATH to include the paths from python3.13 and setup a venv, then ran pip in the venv.

$ PYTHONPATH=/usr/lib/python3.13/site-packages python3.13t -m venv create tmpdir/qqq-t
...
$ /home/barry/tmpdir/qqq-t/bin/pip --version
pip 24.3.1 from /home/barry/tmpdir/qqq-t/lib64/python3.13t/site-packages/pip (python 3.13)

I don’t think any binary extensions will work. Because they will not have been compiled for free threading.

7

I’m not sure how to proceed. I’d expect modules such as ‘threading’ to work (otherwise how would I control the free threading) but, for example, ‘numpy’ wouldn’t work (or I wouldn’t even be able to build it) because there is no version compatible with free threading (I understand that numpy blocks the GIL by itself).

8

Do you absolutely need distro packages for the Python libraries? You can use pip to install binary wheels for many packages, for example.

I agree though based on what you and Barry have observed that the linux distros seem to be packaging free-threaded Python in a way that assumes you don’t need packages, which is pretty suboptimal. Given what you’ve found, I don’t think it’s tenable to rely on them to experiment with free-threading.

9

No that is not what I hope I said.

There is a lack of free-threading validated/built packages.
There is little that a distro can do about that.

10

Lots of packages with native extensions are shipping free-threaded wheels.

I’m a NumPy maintainer, we ship free-threaded wheels, I’m not sure what’s holding up distros from following suit. As far as I know, we haven’t heard from any distros about shipping free-threaded NumPy builds.

I would need to find the distro packaging discussions about this to see whether it’s about confidence in upstream, not wanting to re-do packaging work, or some other reason.

1 Like
11

@Conan_Kudo can you provide insight from a Fedora perspective?