Re: [Buildroot] [PATCH v4 2/4] package/python-gobject: fix event source registration with Python 3.13

From: Thomas Petazzoni via buildroot <buildroot@buildroot.org>
To: Fiona Klute via buildroot <buildroot@buildroot.org>
Cc: Fiona Klute <fiona.klute@gmx.de>,
	James Hilliard <james.hilliard1@gmail.com>,
	Fabrice Fontaine <fontaine.fabrice@gmail.com>
Subject: Re: [Buildroot] [PATCH v4 2/4] package/python-gobject: fix event source registration with Python 3.13
Date: Sat, 17 May 2025 17:27:14 +0200	[thread overview]
Message-ID: <20250517172714.691fef81@windsurf> (raw)
In-Reply-To: <20250414161545.2863615-2-fiona.klute@gmx.de>

Hello Fiona,

Thanks for working on this (which we know is a complex topic). However,
while the series looks good and I was about to apply it, there's one
fairly significant concern below.

On Mon, 14 Apr 2025 18:15:43 +0200
Fiona Klute via buildroot <buildroot@buildroot.org> wrote:

> From: "Fiona Klute (WIWA)" <fiona.klute@gmx.de>
> 
> Since Python 3.13 BaseSelectorEventLoop.add_reader() and
> BaseSelectorEventLoop.add_writer() use the mapping returned by
> selector.get_map() to detect if a file object is already
> registered. This fails with the implementation in gi.events._Selector
> if some calls use a file object, and others the raw file descriptor.
> 
> Full upstream bug report:
> https://gitlab.gnome.org/GNOME/pygobject/-/issues/689
> 
> This bug breaks package/python-aiomqtt, because its client object uses
> file objects in some places for the connection socket, and the file
> descriptor in others. The result is that the connection attempt times
> out because source registration fails, and the Future that marks
> successful connection never resolves.
> 
> Signed-off-by: Fiona Klute (WIWA) <fiona.klute@gmx.de>
> ---
> The bug fixed by this patch is present in 3.50.0 and 3.52.3 alike, the
> reason for applying the update first is simply that the patch doesn't
> apply with 3.50.0, and backporting would be unnecessary effort when
> the update is due anyway.

And actually that doesn't work, and we want exactly the opposite: the
fix before the bump. Indeed, the fix is needed for our 2025.02.x LTS
branch, which has python-gobject 3.50.0. So we definitely want the fix
on 3.50.0 (so that we can backport it to 2025.02.x) and *then* the bump
(which would be only on the master branch).

Do you think you could rework your series with this change?

Thanks a lot for all your work!

Thomas
-- 
Thomas Petazzoni, co-owner and CEO, Bootlin
Embedded Linux and Kernel engineering and training
https://bootlin.com
_______________________________________________
buildroot mailing list
buildroot@buildroot.org
https://lists.buildroot.org/mailman/listinfo/buildroot