Comparison of branches on CI

Comparison of branches on CI

[Andreas Enge]

Hello,

Leo mentioned some time ago that cuirass was missing useful functionality
from hydra, and I wonder if I have not come across one. How about comparison
of different branches and/or evaluations? I did not see how to do it, but
maybe I just overlooked something.

For instance for the core-update merge (but also later feature branches),
it would be useful to distinguish problems in core-updates from those
already present in master; otherwise said, packages that we would lose
by making the merge. If core-updates is not 100% okay, but does not present
(important) regressions with respect to master, we may as well merge it and
sort out remaining problems in master. Is there a way of making this
comparison with our current tools?

Andreas

Re: Comparison of branches on CI

[Leo Famulari]

On Thu, Mar 30, 2023, at 05:24, Andreas Enge wrote:
> Leo mentioned some time ago that cuirass was missing useful functionality
> from hydra, and I wonder if I have not come across one. How about comparison
> of different branches and/or evaluations? I did not see how to do it, but
> maybe I just overlooked something.

This is the missing feature that I think is most important.

It is possible to use data.guix.gnu.org to create these comparisons. Sometimes it takes me a few minutes to remember how to do it, so the UI could grow some hints or "affordances". I'll provide an example later today, when I am near a computer.

Re: Comparison of branches on CI

[Leo Famulari]

On Thu, Mar 30, 2023 at 07:59:49AM -0400, Leo Famulari wrote:
> It is possible to use data.guix.gnu.org to create these comparisons. Sometimes it takes me a few minutes to remember how to do it, so the UI could grow some hints or "affordances". I'll provide an example later today, when I am near a computer.

Here is the page to use:

https://data.guix.gnu.org/compare/package-derivations?base_commit=d7673b49c086c898a8e749fd042081dc9d4631b8&target_commit=cf26ee1f99f84f817f96269541073846d546026b&build_change=broken&after_name=&limit_results=40

When comparing core-updates to master, the base commit should be from
the master branch, and the target commit should be from core-updates.

However, it looks like the Data Service doesn't know about these commits
on core-updates, which is unfortunate.

My understanding is that Cuirass stopped sending information to the Data
Service after this commit was deployed:

https://git.savannah.gnu.org/cgit/guix/guix-cuirass.git/commit/?id=e598f89445b58eede595e66076405931e6bb1e55

So, in the short term, to get useful information about core-updates from
the Data Service, we'd want to build core-updates in Bordeaux.

Re: Comparison of branches on CI

[Andreas Enge]

Am Thu, Mar 30, 2023 at 07:27:55PM -0400 schrieb Leo Famulari:
> Here is the page to use:
> https://data.guix.gnu.org/compare/package-derivations?base_commit=d7673b49c086c898a8e749fd042081dc9d4631b8&target_commit=cf26ee1f99f84f817f96269541073846d546026b&build_change=broken&after_name=&limit_results=40

Thanks, Leo!

> So, in the short term, to get useful information about core-updates from
> the Data Service, we'd want to build core-updates in Bordeaux.

That would be very welcome indeed, since things are shaping up.
Maybe we can prioritise it over patch handling now.

Andreas

Re: Comparison of branches on CI

[Andreas Enge]

Am Thu, Mar 30, 2023 at 07:27:55PM -0400 schrieb Leo Famulari:
> > It is possible to use data.guix.gnu.org to create these comparisons. Sometimes it takes me a few minutes to remember how to do it, so the UI could grow some hints or "affordances". I'll provide an example later today, when I am near a computer.

And this now also works:
https://data.qa.guix.gnu.org/compare/package-derivations?base_commit=5c893787be78a79433fe1343f5b70cd647e8f667&target_commit=51b8385acd6f7de668a994fe0523b3861240cd1a&system=x86_64-linux&target=none&build_change=broken&after_name=&limit_results=400

It finds 298 packages with problems on x86_64, which is suspiciously few.
Actually unless I made a mistake, these are packages that build on master,
but are broken on core-updates. I see more broken packages on core-updates,
but it is quite possible that these are also broken on master...
Not really: kcodecs builds on master, but not on core-updates, but it
is not (yet) "broken" because it has not been scheduled yet. So expect
the 298 to rise!

Maybe it would be an idea to add "maybe broken" (was green, is blue)
and "maybe fixed" (was red, is blue) to the search? And those that
were and remain blue could go anywhere...

Many thanks to Chris for making this possible! My impression is that we
are making useful progress on the tooling; due to this website and ci
working, it is finally possible to make assertions on core-updates beyond
what I see when trying to build my own, subjective profile.

Andreas