guix-devel
[Top][All Lists]
Advanced

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

Re: i686 core-updates failure.


From: Simon Tournier
Subject: Re: i686 core-updates failure.
Date: Thu, 13 Apr 2023 17:31:28 +0200

Hi jgart,

On Thu, 13 Apr 2023 at 13:23, "jgart" <jgart@dismail.de> wrote:

> Is there a reason to keep around the broken packages?

Euh, maybe you have missed what the branch core-updates is. :-)

For packages with many dependents, it is not possible to update them
directly in master, because of two main reasons: 1) the build farm would
not be able to build in timely manner so many substitutes would be
lacking after each push and 2) it would be hard if not impossible to
know before the push the impact of one change over the complete list of
dependents.

Instead, we push these changes concerning some packages with many
dependents – considered as the “core” of Guix – to the branch
core-updates.  But because each change impacts large and intensive
rebuilds, this branch is not continuously built.

Therefore, the state of this branch core-updates is often poor; meaning
that many packages are potentially broken.

At some point, when enough changes are pushed into this core-updates
branch – well, it’s more when someone takes the lead for this branch –
we set the CI to re-build all the packages of the core-updates branch.

Then, we detect broken packages, often impacted by unrelated changes as
ungrafts and co., and we try to fix them all before merging this
core-updates branch into the master branch.

Thanks to the tireless Andreas’s lead over the past months, we are at
the point where the core-updates will be merged soon to master.

Considering the architecture x86_64, the state is almost good!
Yeah! \o/  Please give a look at « A Joyous Core-Updates Week-End 🎉 »
email which provides details about how to help and test.
<https://lists.gnu.org/archive/html/guix-devel/2023-04/msg00121.html>


Considering other architectures as i686, that’s not the case.  See [1].
The state of non-x86_64 architectures is weak compared to the usual
standards for merging.

All in all, no, it is possible to drop python-numpy for example [2]. :-)

Hope that makes sense.

1: https://ci.guix.gnu.org/eval/391720/dashboard?system=i686-linux
2: https://ci.guix.gnu.org/build/712672/details

Cheers,
simon



reply via email to

[Prev in Thread] Current Thread [Next in Thread]