--- Begin Message ---
- To: Debian Bug Tracking System <submit@bugs.debian.org>
- Subject: transition: numpy 1.24.x
- From: Sandro Tosi <morph@debian.org>
- Date: Mon, 26 Dec 2022 18:52:50 -0500
- Message-id: <167209877091.1154612.15056499387890065579.reportbug@zion.matrix.int>
Package: release.debian.org
Severity: normal
User: release.debian.org@packages.debian.org
Usertags: transition
X-Debbugs-Cc: numpy@packages.debian.org, morph@debian.org
Control: affects -1 + src:numpy
Resuming from https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1025788#43
In the meantime, since upstream released it, i've uploaded
numpy/1.24.0 to experimental and the autopkgtest results are
https://qa.debian.org/excuses.php?experimental=1&package=numpy
now, there's a lot of red in there but almost all of the errors come
in the format of
AttributeError: module 'numpy' has no attribute 'X'
with X being [float, int, bool, object, ...].
This is because, numpy upstream in 1.24.0, finally decided to expire
https://numpy.org/doc/stable/release/1.24.0-notes.html#:~:text=The%20deprecation%20for%20the%20aliases
some deprecations introduced in 1.20.0
https://numpy.org/doc/stable/release/1.20.0-notes.html#using-the-aliases-of-builtin-types-like-np-int-is-deprecated
(released almost 2 years ago).
All of those are quite straightforward to fix, since often it's just
necessary to stop importing them from numpy and use the python native
types.
There are handful more errors in the form of:
* ValueError: setting an array element with a sequence. The
requested array has an inhomogeneous shape after 1 dimensions. The
detected shape was (2,) + inhomogeneous part.
* Too many indices for array: array is 0-dimensional, but 1 were indexed
which will need to be looked at in more details, likely by individual
projects upstream.
This additional transition seems to be comfortably in the reach for
Bullseye and will place us in a better position to get support from
upstream. I also anticipate that a few more patch releases (fixing
bugs etc) for the 1.24.x series will be published before Bullseye is
released and we would like to update numpy to them in Debian if
reasonable.
Please let me know if i can proceed with a numpy upload to unstable.
With ginggs' reply https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1025788#48
Hi Sandro
I should have closed this bug once numpy 1:1.23.5-2 migrated. Doing
so now. Please file a new bug for discussing 1.24.0.
On Mon, 26 Dec 2022 at 00:12, Sandro Tosi <morph@debian.org> wrote:
> In the meantime, since upstream released it, i've uploaded
> numpy/1.24.0 to experimental and the autopkgtest results are
> https://qa.debian.org/excuses.php?experimental=1&package=numpy
>
> now, there's a lot of red in there but almost all of the errors come
> in the format of
>
> AttributeError: module 'numpy' has no attribute 'X'
>
> with X being [float, int, bool, object, ...].
>
> This is because, numpy upstream in 1.24.0, finally decided to expire
> https://numpy.org/doc/stable/release/1.24.0-notes.html#:~:text=The%20deprecation%20for%20the%20aliases
> some deprecations introduced in 1.20.0
> https://numpy.org/doc/stable/release/1.20.0-notes.html#using-the-aliases-of-builtin-types-like-np-int-is-deprecated
> (released almost 2 years ago).
>
> All of those are quite straightforward to fix, since often it's just
> necessary to stop importing them from numpy and use the python native
> types.
That's a lot of breakage, but good that the fix is straightforward.
> There are handful more errors in the form of:
>
> * ValueError: setting an array element with a sequence. The
> requested array has an inhomogeneous shape after 1 dimensions. The
> detected shape was (2,) + inhomogeneous part.
> * Too many indices for array: array is 0-dimensional, but 1 were indexed
>
> which will need to be looked at in more details, likely by individual
> projects upstream.
The sooner these bugs are filed, the better.
> This additional transition seems to be comfortably in the reach for
> Bullseye and will place us in a better position to get support from
> upstream. I also anticipate that a few more patch releases (fixing
> bugs etc) for the 1.24.x series will be published before Bullseye is
> released and we would like to update numpy to them in Debian if
> reasonable.
s/Bullseye/bookworm but agreed.
> Please let me know if i can proceed with a numpy upload to unstable.
Please let's wait for a bit on this one. There's still a matplotlib
transition in flight (#1026119) that needs bugs filed and autopkgtests
fixed before it can migrate. Let's aim to do this once the initial
rebuilds for Python 3.11 as default (#1026825) have been done. In the
meantime, it would help if bugs were filed against the packages that
need updating for numpy 1.24.x.
--- End Message ---