Re: Progress with automating testing of patches

[Christopher Baines]

Ludovic Courtès <ludo@gnu.org> writes:

> I wonder if it’s due to recent changes since I last looked, but I’m a
> bit confused by the numbers in this example:
>
>   https://qa.guix.gnu.org/issue/58186
>
> The numbers before/after patches don’t match and the lint warnings (nice
> addition!) appear to unrelated to the patch at hand.
>
> Any idea what’s going on?

I've had an initial look. One important clue is that the basis of the
comparison [1] differs between data.qa.guix.gnu.org and
data.guix.gnu.org [2]. The package number differs for example.

1: 
https://data.qa.guix.gnu.org/revision/e6777cfa5eb5e9c36eaf7810b42cac0fbcaa367c
2: https://data.guix.gnu.org/revision/e6777cfa5eb5e9c36eaf7810b42cac0fbcaa367c

That shouldn't happen, one revision of Guix should have the same number
of packages regardless of what server looked at it. This being wrong
explains the bad data on qa.guix.gnu.org, since the comparison being
done by data.qa.guix.gnu.org is based on bad data.

This reminds me of [3] where data.guix.gnu.org processed a revision and
somehow got things wrong.

3: https://lists.gnu.org/archive/html/guix-devel/2021-09/msg00192.html

I'll try investigating this further when I have more time, there should
be locking in place so that even when multiple jobs are being processed
at the same time, only one job at a time is able to call
latest-channel-instances, so I don't currently have a theory as to how
this goes wrong. I think I added more logging off the back of the
previous issue, so maybe I'll be able to get further this time.

Thanks,

Chris

signature.asc
Description: PGP signature