Am Sonntag, den 23.02.2020, 12:04 +0100 schrieb Han-Wen Nienhuys:
> On Sun, Feb 23, 2020 at 11:55 AM Jonas Hahnfeld <
> [email protected]
> > wrote:
> 
> > But a CI should test the changes in every possible way we care about -
> > exactly because "this change cannot possibly break". And here I really
> > mean CI to be integrated into tooling and running without human
> > interaction, just as Patchy is right now. And if somebody needs to
> > build a new image every few days to have an updated ccache, that's
> > something that should be clearly weighted against the benefits IMO.
> 
> If you can automatically run a CI job, you could automatically run a
> seeding job.

This sounds contradicting to me: Either you want to use the ccache for
multiple jobs or you're running it for every commit - not gaining
anything except for increased complexity.

> In practice, I think we would have to have a mechanism (git tags?) by
> which we can annotate that a certain release changed the regtest
> output. Assuming we have those often enough, that takes care of
> reseeding the cache.

And that's where the infrastructure comes into play: Assuming that all
of this is to happen automatically, it will get pretty hairy from what
I've seen so far. Sure, all of this is possible, but at what cost?

> Note that the test baseline takes as much space as the cache, so it's
> not going to make a huge difference by not having them.

It does: complexity. From what I've seen in the last months, some
things in LilyPond are overly complicated. I'm not saying that there
are easy solutions to all problems, but I don't think a more complex
setup helps in this case.

Jonas

Attachment: signature.asc
Description: This is a digitally signed message part

Reply via email to