Re: 01/01: gnu: Add rclone.

[ludo@gnu.org (Ludovic Courtès)]

Hello,

Nicolas Goaziou <mail@nicolasgoaziou.fr> skribis:

> Mark H Weaver <mhw@netris.org> writes:
>
>> Did you test this?
>
> Yes, of course I tested it. I was using it for my own needs before
> sending the patch for review. I obviously fumbled somehow when I sent
> the mail, but I still do not get when or why.

Yeah, these things happen, and evidently I overlooked it as well when
reviewing the patch.  Mark reverted it quickly enough that this was
mostly harmless.

>> There's a syntax error in the code, such that it not only fails to
>> build, but it fails even to produce a _derivation_ describing the build.
>> Unfortunately, this kind of error in any package causes problems for our
>> build farm, which currently assumes that this kind of error will not
>> occur.  Such errors not only cause the individual build to fail, but
>> moreover for the entire process that generates a new CI "evaluation" to
>> fail, which effectively prevents subsequent package updates on that
>> branch from being built until the problem is corrected.
>
> I suggest to add a check, if possible, for that kind of mistake — e.g.,
> check if `arguments' value is a valid property list — because forgetting
> a line in a leaf package should not cause the whole build farm to die.

Longer-term I think we should all push to a dedicated branch and have
server-side automation apply to ‘master’ patches that pass a few tests
(commit signed by an authorized key, ‘guix lint’, things like that.)

We’ll get there but for now the test is when CI gets around to
evaluating the given commit.

Ludo’.