Python 3.13 addition as a supported Python version started

[Matthias Klose]

python3-defaults in unstable now adds Python 3.13 as a supported Python
3.13 version. You might see some additional build failures, until the
binNMUs for this addition are done [1]. This might take some days for
some architectures. We will most likely also see some more issues once
the lower levels of this addition are done.

Matthias

[1] https://release.debian.org/transitions/html/python3.13-add.html

[PICCA Frederic-Emmanuel]

do we know how long we will have to fix all the FTBFS and autopkgtest before the freeze ?

I am a bit worrying for the scientific stack , will we have enough time to work with our upstream in order to fix all these FTBFS. In the scientific stack, things are going slowly....

We are not 100% of our time dedicated to Debian work... so I hope that it will not ruine the effort of the trixie cycle for scientific softwares.

moving to Python 3.12 was not that simple...

Cheers

Frédéric

[Matthias Klose]

Post by PICCA Frederic-Emmanuel
do we know how long we will have to fix all the FTBFS and autopkgtest before the freeze ?

no. the freeze date is not yet announced.

Post by PICCA Frederic-Emmanuel
I am a bit worrying for the scientific stack , will we have enough time to work with our upstream in order to fix all these FTBFS. In the scientific stack, things are going slowly....
We are not 100% of our time dedicated to Debian work... so I hope that it will not ruine the effort of the trixie cycle for scientific softwares.
moving to Python 3.12 was not that simple...

this is the same as we did for the Python 3.12 transition. Please note
that we don't enable any of the experimental features in Python 3.12 (no
GIL, JIT compilation), so assuming there are currently no other RC
issues in your packages, there should plenty of time to fix any 3.13
related issues.

Matthias

[PICCA Frederic-Emmanuel]

Post by Matthias Klose
this is the same as we did for the Python 3.12 transition. Please note
that we don't enable any of the experimental features in Python 3.12 (no
GIL, JIT compilation), so assuming there are currently no other RC
issues in your packages, there should plenty of time to fix any 3.13
related issues.

the plenty of time is not only my time or Debian time but also upstream time.

I just wanted to express my concern because we rely at work on the scientific stack.

So we try hard to maintain our packages in testing, and it it always a deception to see them (part of) expelled from testing due to an FTBFS with a new Python or a failing autopkgtest.

amicalement,

Fred

[Charles Plessy]

Post by PICCA Frederic-Emmanuel
So we try hard to maintain our packages in testing, and it it always a
deception to see them (part of) expelled from testing due to an FTBFS
with a new Python or a failing autopkgtest.

On days where my thoughts are dark, I sometimes imagine some smart
people planning the future in a meeting room and saying "and the great
thing with open source is that if you push your change in a major
distribution, they will propagate patches upstream for free!"…

It is time that Python, GCC, etc. leverage AI to send patches upstream
by themselves and free us from that burden.

Bonne soirée,

[Stefano Rivera]

Hi debian-python (2024.11.13_15:01:31_+0000)

Hi PICCA (2024.11.13_10:04:26_+0000)

Post by PICCA Frederic-Emmanuel
I am a bit worrying for the scientific stack , will we have enough
time to work with our upstream in order to fix all these FTBFS. In the
scientific stack, things are going slowly....

The reality here is that Python has a 6-month release cycle, these days.

I mean 12-month, of course.

Stefano

[Stefano Rivera]

Hi PICCA (2024.11.13_10:04:26_+0000)

Post by PICCA Frederic-Emmanuel
I am a bit worrying for the scientific stack , will we have enough
time to work with our upstream in order to fix all these FTBFS. In the
scientific stack, things are going slowly....

The reality here is that Python has a 6-month release cycle, these days.

If upstreams can't stay on top of new Python releases, we are stuck with
doing the porting work or dropping them from Debian. We can't fix them
all in 6 months. There are still a lot of open 3.11 and 3.12 bugs, for
example.

If we don't have the latest stable version of Python in our stable
release, I think a large number of our users will be very disappointed.
It would certainly cement the view that Debian ships ancient software.
I don't think the users who would be upset would have any motivation to
help improve the situation (working on old scientific packages).

If we have to drop large numbers of scientific packages in our stable
releases, I imagine a small number of users would be disappointed, and
hopefully able to see how they can help avoid this situation in the
future. Sorry, but I see that as the less bad outcome. I'm not saying
I want it, but I think it's the approach we have to take, in the face of
unmaintained software.

The alternative would be to carry multiple Python releases in a Debian
stable release, which is something we haven't wanted to do.

We try to start the detection process as early as possible.

I have been doing archive wide rebuilds (as much as I could, on arm64)
since 3.13 rc2. I announced it, and our planned migration to 3.13 in
trixie, in:

https://lists.debian.org/msgid-search/***@satie.tumbleweed.org.za

I'm hoping to have even better tooling for this kind of rebuild in the
future.

Stefano

[Colin Watson]

python3-defaults in unstable now adds Python 3.13 as a supported Python 3.13
version. You might see some additional build failures, until the binNMUs
for this addition are done [1]. This might take some days for some
architectures. We will most likely also see some more issues once the lower
levels of this addition are done.

[...]

[1] https://release.debian.org/transitions/html/python3.13-add.html

While there are a few bits of that transition tracker still red, the
current target is to work on the list of autopkgtest failures shown on
https://tracker.debian.org/pkg/python3-defaults in order to get the
addition of 3.13 as a supported version into testing. As usual, this
page can be a little hard to interpret because it shows test failures of
the versions of those packages in testing, and you have to click through
to each corresponding package (sometimes through multiple levels of
failures) to see whether it's been fixed in unstable. But with ~35
packages left there, it's getting easier to wade through and we're
getting pretty close.

Here's the current state, with my comments:

* audioread: #1082047; apparently needs packaging of a couple of pieces
removed from the standard library. Reverse-dependencies are eartag,
puddletag, and python3-acoustid.

* dask/dask.distributed: #1088234 and #1088286, but also #1085947 in
sphinx-book-theme. I sank a bunch of time into trying to fix this
last month and didn't really get anywhere very satisfying. Can
anyone with more experience with these packages figure this out?

* datalad-next: #1088038. Probably not too hard if you can figure out
how that test is supposed to work.

* deepdiff: #1088239, blocked by orderly-set in NEW. I poked
#debian-ftp.

* hyperkitty: #1088312. Should be fairly easy.

* ipykernel: Fixed in unstable.

* ironic-python-agent: #1089531. Should be fairly easy; zigo said on
IRC that this is a leaf package and doesn't need to block migration.

* jinja2: Fixed in unstable.

* jupyter-server: Fixed in unstable.

* mdp: Fixed in unstable.

* ovn-octavia-provider: #1088762. zigo said on IRC that this is a leaf
package and doesn't need to block migration.

* pocketsphinx-python: #1088764. Apparently difficult.

* pytest-jupyter: Fixed by new ipykernel in unstable.

* python-attrs: Fixed in unstable; blocked on python-cattrs.

* python-beartype: #1089017. Apparently fixed upstream, though I don't
know exactly where.

* python-cattrs: #1073406/#1086614.

* python-omegaconf: #1089049.

* python-oslo.messaging: I believe this is fixed in unstable
(#1089050) and waiting for python-eventlet to migrate to testing.

* python-pure-python-adb: #1082251/#1084618; apparently just needs a
dependency on python3-zombie-telnetlib?

* python-voip-utils: #1088827 fixed in unstable, but has an autopkgtest
regression on s390x (#1089826).

* rich: #1082290; seems to be fixed upstream.

* smart-open: #1089053; upstream fix in progress.

* spyder: #1088068/#1089054.

* twisted: Fixed in unstable, just waiting for matrix-synapse to
migrate first (which should be soon).

There are also a number of architecture-specific failures showing up
there. Some might go away with a few more retries I guess, but we'll
likely need to work out what to do about the rest. I haven't looked at
these in any depth.

Can anyone help with any of the remaining problems here? This would be
especially useful for packages that aren't maintained by the Python
team.

[Julian Gilbey]

Post by Colin Watson
[...]
* spyder: #1088068/#1089054.

I'm struggling with this one; I've asked at
https://github.com/spyder-ide/spyder/issues/23074
for help, but nothing so far. I've just pushed my current work to
salsa (***@salsa.debian.org:science-team/spyder.git), and if anyone
has time to look into this, I'd really appreciate it.

Here's the little I've found so far:

* I'm trying to update to spyder 6.0.2

* In an lxc container with Debian unstable and all the test-time
packages installed, the tests fail in the way described, both with
Python 3.12 and Python 3.13.

* In the same lxc container, if I build a virtual Python 3.12 or 3.13
environment following the upstream testing scripts (in
.github/workflows) and test the package there, all works smoothly.

Perhaps it's an updated version of python3-pytestqt or something like
that?

Best wishes,

Julian

[Colin Watson]

Post by Julian Gilbey

Post by Colin Watson
[...]
* spyder: #1088068/#1089054.

I'm struggling with this one; I've asked at
https://github.com/spyder-ide/spyder/issues/23074
for help, but nothing so far. I've just pushed my current work to
has time to look into this, I'd really appreciate it.

I poked around a bit in pdb. I think the problem is that one plugin is
calling the icon machinery at class creation time, before a QApplication
or equivalent has been set up, so font loading doesn't happen. Since
pytest goes through and loads all the Python files to look for tests,
this causes it problems.

This patch helps (borrowed from similar code in a different plugin);
feel free to send it upstream if you think it makes sense:

diff --git a/spyder/plugins/preferences/widgets/configdialog.py b/spyder/plugins/preferences/widgets/configdialog.py
index c3f6bb3..d822431 100644
--- a/spyder/plugins/preferences/widgets/configdialog.py
+++ b/spyder/plugins/preferences/widgets/configdialog.py
@@ -28,11 +28,11 @@ class ConfigDialog(SidebarDialog):

# Constants
TITLE = _("Preferences")
- ICON = ima.icon('configure')
MIN_WIDTH = 940 if MAC else (875 if WIN else 920)
MIN_HEIGHT = 700 if MAC else (660 if WIN else 670)

def __init__(self, parent=None):
+ self.ICON = ima.icon('configure')
SidebarDialog.__init__(self, parent)

# Attributes

With that, the tests are able to actually start up, although there are
some other failures. It might be worth experimenting with whether
upgrading yapf to a more recent upstream version would help with the
rest? And it's a rather slow set of tests so I didn't wait for them to
finish before sending this email.

[Julian Gilbey]

Post by Colin Watson

Post by Julian Gilbey

Post by Colin Watson
[...]
* spyder: #1088068/#1089054.

I'm struggling with this one; I've asked at
https://github.com/spyder-ide/spyder/issues/23074
for help, but nothing so far. I've just pushed my current work to
has time to look into this, I'd really appreciate it.

I poked around a bit in pdb. I think the problem is that one plugin is
calling the icon machinery at class creation time, before a QApplication
or equivalent has been set up, so font loading doesn't happen. Since
pytest goes through and loads all the Python files to look for tests,
this causes it problems.
[...]

Hi Colin,

That's amazing, thank you! I'm still curious why it succeeds with the
upstream version in a virtual environment but not with our setup.
I'll try it out when I'm better and submit it upstream. (And in the
meantime, spyder has an autorm date of 2024-12-25, if I recall
correctly, so this should not hold up the Python 3.13 transition.)

Also, I'm used to having lots of new test failures with newer versions
of Spyder, so what you observed about several tests failing doesn't
surprise me!

Best wishes,

Julian

[Colin Watson]

Post by Colin Watson
While there are a few bits of that transition tracker still red, the
current target is to work on the list of autopkgtest failures shown on
https://tracker.debian.org/pkg/python3-defaults in order to get the
addition of 3.13 as a supported version into testing. As usual, this
page can be a little hard to interpret because it shows test failures of
the versions of those packages in testing, and you have to click through
to each corresponding package (sometimes through multiple levels of
failures) to see whether it's been fixed in unstable. But with ~35
packages left there, it's getting easier to wade through and we're
getting pretty close.

We've dealt with on the order of half of those one way or another, so

Post by Colin Watson
* audioread: #1082047; apparently needs packaging of a couple of pieces
removed from the standard library. Reverse-dependencies are eartag,
puddletag, and python3-acoustid.

I poked #debian-ftp about maybe getting python-deadlib through NEW.

Post by Colin Watson
* dask/dask.distributed: #1088234 and #1088286, but also #1085947 in
sphinx-book-theme. I sank a bunch of time into trying to fix this
last month and didn't really get anywhere very satisfying. Can
anyone with more experience with these packages figure this out?

Currently exchanging email with Nilson about the state of
sphinx-book-theme. It might also be worth somebody seeing if it's
practical to temporarily detach dask/dask.distributed from its
documentation theme.

Post by Colin Watson
* datalad-next: #1088038. Probably not too hard if you can figure out
how that test is supposed to work.

Fixed.

Post by Colin Watson
* deepdiff: #1088239, blocked by orderly-set in NEW. I poked
#debian-ftp.

ftpmaster processed orderly-set through NEW, so Emmanuel fixed this in
unstable. There's currently an i386 autopkgtest failure, but Fabian
fixed rust-associative-cache to build on i386, and that should sort it
out once it reaches testing.

Post by Colin Watson
* hyperkitty: #1088312. Should be fairly easy.

No progress - anyone?

Post by Colin Watson
* ironic-python-agent: #1089531. Should be fairly easy; zigo said on
IRC that this is a leaf package and doesn't need to block migration.
* ovn-octavia-provider: #1088762. zigo said on IRC that this is a leaf
package and doesn't need to block migration.

Thomas fixed these two in unstable.

Post by Colin Watson
* pocketsphinx-python: #1088764. Apparently difficult.

No progress as far as I know. (I guess we could wait for autoremoval
from testing, which is currently scheduled for 29 December.)

Post by Colin Watson
* python-attrs: Fixed in unstable; blocked on python-cattrs.
* python-beartype: #1089017. Apparently fixed upstream, though I don't
know exactly where.
* python-cattrs: #1073406/#1086614.
* python-omegaconf: #1089049.

No progress on this lot as far as I know.

Post by Colin Watson
* python-oslo.messaging: I believe this is fixed in unstable
(#1089050) and waiting for python-eventlet to migrate to testing.

Fixed.

Post by Colin Watson
* python-pure-python-adb: #1082251/#1084618; apparently just needs a
dependency on python3-zombie-telnetlib?

Emmanuel fixed this in unstable.

Post by Colin Watson
* python-voip-utils: #1088827 fixed in unstable, but has an autopkgtest
regression on s390x (#1089826).

No progress. I poked Edward on IRC.

Post by Colin Watson
* rich: #1082290; seems to be fixed upstream.

I investigated this (with help from debusine's new reverse-dependency
autopkgtest workflow) and concluded that upgrading rich and textual
should fix it; details in the bug. Waiting for feedback from Sandro.

Post by Colin Watson
* smart-open: #1089053; upstream fix in progress.

Fixed in unstable.

Post by Colin Watson
* spyder: #1088068/#1089054.

See replies to this email thread. There's clearly some more work to do
here, but it looks somewhat tractable.

Post by Colin Watson
* twisted: Fixed in unstable, just waiting for matrix-synapse to
migrate first (which should be soon).

Fixed.

Post by Colin Watson
There are also a number of architecture-specific failures showing up
there. Some might go away with a few more retries I guess, but we'll
likely need to work out what to do about the rest. I haven't looked at
these in any depth.

I chased down a few of these. dipy is fixed in unstable; fenics-basix
is #1090766; I fixed pytest-forked and pytest-mpi in unstable.

python-libtmux/s390x (e.g.
https://ci.debian.net/packages/p/python-libtmux/testing/s390x/55551127/)
looks as though it needs investigation.

statsmodels/armel (e.g.
https://ci.debian.net/packages/s/statsmodels/testing/armel/55559166/)
probably just requires skipping some slow tests, since the test takes
twice as long while we have two supported Python versions and it's not
very fast on armel to begin with. Somebody who knows the package better
than I do can probably make a guess at which tests are most reasonable
to exclude.