Python 3.11.0a4 is blocked
Hi everyone,
I am writing this to notify you that unfortunately the release of 3.11.0a4 is blocked as there are a bunch of release blockers (some of them affect Python 3.10):
https://bugs.python.org/issue46263 https://bugs.python.org/issue46208 https://bugs.python.org/issue46006 https://bugs.python.org/issue43683
If this was a single release blocker I would think about moving forward but unfortunately, there are several of them and one of them is that Python fails to compile FreeBSD, so I am halting the release until these are fixed.
Regards from rainy London, Pablo Galindo Salgado
Hi Pablo,
Issue 43683 should not have been marked as a release blocker, as it is a performance issue, not a bug.
The related bug (which was a release blocker) is https://bugs.python.org/issue46009, and was fixed a while ago.
Cheers, Mark.
On 04/01/2022 11:12 pm, Pablo Galindo Salgado wrote:
Hi everyone,
I am writing this to notify you that unfortunately the release of 3.11.0a4 is blocked as there are a bunch of release blockers (some of them affect Python 3.10):
https://bugs.python.org/issue46263 <https://bugs.python.org/issue46263> https://bugs.python.org/issue46208 <https://bugs.python.org/issue46208> https://bugs.python.org/issue46006 <https://bugs.python.org/issue46006> https://bugs.python.org/issue43683 <https://bugs.python.org/issue43683>
If this was a single release blocker I would think about moving forward but unfortunately, there are several of them and one of them is that Python fails to compile FreeBSD, so I am halting the release until these are fixed.
Regards from rainy London, Pablo Galindo Salgado
Python-Dev mailing list -- python-dev@python.org To unsubscribe send an email to python-dev-leave@python.org https://mail.python.org/mailman3/lists/python-dev.python.org/ Message archived at https://mail.python.org/archives/list/python-dev@python.org/message/7OWGAL4U... Code of Conduct: http://python.org/psf/codeofconduct/
https://bugs.python.org/issue46006 <https://bugs.python.org/issue46006>
This one is due to a regression made in bpo-40521 "Make free lists and unicode caches per-interpreter". It affects 3.10 as well, and hes broken a number of third-party libraries there (e.g. mod_wsgi and kodi). It's not easy to revert: I think only one person understands the changes made for bpo-40521 (and the particular wider strategy behind it). I feel powerless to help here.
As Mark Dickinson said in bpo-40521, "It's feeling as though the normal Python development process is being bypassed here." IMO, there should be a PEP for wide-reaching changes like bpo-40521, and the PEP should be actually *approved* before such changes are made.
In bpo-46006 itself, there are two proposed PRs, each corresponding to a slightly different vision of how subinterpreters should be isolated. I don't know which is right. If we had a PEP, I could read it and decide, but I can't -- the people driving this effort have different ideas of the big picture.
Victor, as the author of the breaking change, could you revert it? Or do you have a different in mind (assuming your PR for bpo-46006 has an unacceptable performance impact, as Mark Shannon pointed out)?
On Wed, Jan 5, 2022 at 7:04 AM Petr Viktorin <encukou@gmail.com> wrote:
IMO, there should be a PEP for wide-reaching changes like bpo-40521, and the PEP should be actually *approved* before such changes are made.
FWIW, I plan on posting a PEP by the end of the week regarding per-interpreter GIL and the related work.
In bpo-46006 itself, there are two proposed PRs, each corresponding to a slightly different vision of how subinterpreters should be isolated. I don't know which is right. If we had a PEP, I could read it and decide, but I can't -- the people driving this effort have different ideas of the big picture.
Victor's goal is the same as mine. The only difference here is the steps involved. (In this case, the interned strings were made per-interpreter prematurely.) I'm clarifying this in that PEP.
-eroc
Hi everyone,
An update on this. Unfortunately, we are still blocked. Some of the blockers have been fixed (thanks to everyone involved) but the following are still pending:
This issue has a PR being reviewed but the fix is still not merged.
Victor made a revert of his PR but unfortunately, we cannot easily backport it to 3.10 as it affects the ABI. It affects the interpreter state structure that although is not on the stable ABI, changing it can affect some projects so we need to discuss how we want to proceed as a project here.
One of the FreeBSD issues has been fixed (thanks Christian) but another issue has been unveiled (some problem in test_capi) and the other one still lingers (test_embed failing due to freezing modules).
Thanks for all your help,
Kind regards from cloudy London, Pablo Galindo Salgado
On Tue, 4 Jan 2022 at 23:12, Pablo Galindo Salgado <pablogsal@gmail.com> wrote:
Hi everyone,
I am writing this to notify you that unfortunately the release of 3.11.0a4 is blocked as there are a bunch of release blockers (some of them affect Python 3.10):
https://bugs.python.org/issue46263 https://bugs.python.org/issue46208 https://bugs.python.org/issue46006 https://bugs.python.org/issue43683
If this was a single release blocker I would think about moving forward but unfortunately, there are several of them and one of them is that Python fails to compile FreeBSD, so I am halting the release until these are fixed.
Regards from rainy London, Pablo Galindo Salgado
On Thu, Jan 6, 2022 at 12:33 PM Pablo Galindo Salgado <pablogsal@gmail.com> wrote:
Victor made a revert of his PR but unfortunately, we cannot easily backport it to 3.10 as it affects the ABI. It affects the interpreter state structure that although is not on the stable ABI, changing it can affect some projects so we need to discuss how we want to proceed as a project here.
Well, the regression is fixed in the main branch, so Python 3.11.0a4 should be unblocked for this issue.
I updated my backport to 3.10 to fix the ABI issue.
Victor
Night gathers, and now my watch begins. It shall not end until my death.
participants (5)
-
Eric Snow -
Mark Shannon -
Pablo Galindo Salgado -
Petr Viktorin -
Victor Stinner