[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Involving free-threaded Python in Debian (Was: Free-threaded python3.13t packaging in Debian?)




Thanks for the invitation. I would like to briefly summarize the results of my previous communication with Stefano and the current status of my work.

Here are the repositories I have been working on. I have completed the build and testing of python3.14t and python3t. An experimental repository is also under my account:
https://salsa.debian.org/ben0i0d/python3t
https://salsa.debian.org/ben0i0d/python3t-defaults

In my previous work, I enabled nogil and the t flag to ensure that python3t is completely isolated from the existing Python line. That is, python3t/3.14t follows the “B” technical route. I isolated their dist-packages. (However, note that since I haven’t adapted pip packages yet, the current venv does not work properly, though manual pip installation works fine.)

When I communicated with Stefano via email, he agreed with the idea of using a suffix and nogil, but suggested two corrections:

  1. We cannot split out a separate python3t source package. The builds must produce two parallel interpreters from the existing CPython team packages (of course, they are isolated, including /usr/lib, i.e., the standard library is isolated).

  2. dist-packages should not be isolated. This way, we only need to focus on the porting, and if issues arise, venv can be used as a transition.

I agree with this view, and this is what I am currently doing—choosing option (a).

Regarding GIL vs. nogil, pure Python modules are usually not a problem, e.g., pip. The challenge lies in Python packages with C extensions.

From my perspective, under option (a), the main issue only occurs when mixing GIL and nogil. Perhaps we should first introduce the nogil interpreter into Debian as a base, like Fedora, which could also bring nogil support to Ubuntu 26.04.

As for adapting modules for nogil, I don’t have a good solution yet. This problem will need to be discussed by the community, and I have not conducted tests or developed feasible approaches for mixed usage so far.

Ideally, I plan to complete option (a) within two weeks, i.e., finish the rules modifications and submit them.

At that time, I will carry out this part of the investigation.

Best regards,
Xu Chen (ben0i0d)


在 2025/11/4 15:28, Chen Shengqi 写道:
Xu Chen mentioned that he had some previous email with Stefanor, so I'm looping
him into this thread.

Reply to: