RFC: Page size on PPC/PPC64 builders
Tom Lane
tgl at redhat.com
Tue Mar 4 04:59:52 UTC 2008
Colin Walters <walters at redhat.com> writes:
> Yeah, we should have separate the build and test processes rather than
> trying to cram all testing into the build process. If you're having a
> problem, it's useful to be able to get binaries out of Koji to test them
> locally, without having to indirect all testing through Koji.
> And testing may be very resource-intensive; we don't want to block the
> builders on that.
Not sure what advantage you see here. As to the first argument: if the
Koji build fails in a way that is reproducible on your own machine, that
says that you didn't bother to do a trial build before submitting to
Koji; which is surely not a habit that we should condone or encourage.
As to the second: the only way that separate build and test steps save
any resources is if a significant percentage of build jobs don't get
tested, which again is not something that I think we should encourage or
optimize for. It appears to me that Koji has pretty sizable per-job
overhead, and so I'm not going to voluntarily break one job into two
if I plan to always run both steps anyway.
Obviously, the precise extent of testing that ought to be integrated
into a routine build will vary from one package to the next. I say this
should be left to the discretion of each package maintainer. I see no
advantage whatever from instituting a general policy that "build and
test must be separate".
regards, tom lane
More information about the fedora-devel-list
mailing list