Bernhard R. Link wrote: > I'd rather say "doing it right" means to properly test it's build to be > robust. Only ever testing in an artifical environment has a certain > outcome: certain failure.
That depends. How closely does the artificial environment mirror the avarage system as measured by popcon? How closely does the environment in which I test build my package mirror the average system? It actually seems likely that my environment is more diverged from the average user's environment than is a buildd. I have a random set of build dependencies installed. Nobody else on the planet is likely to have the same set. I pick weird settings for alternatives, etc. We're all weirdos. It seems strange to expect that checking that an arbitrary package builds in a single arbitrary weird environment will be much benefit to a more normal user, or even to another weirdo. Whatever benefit there is has already been diluted a lot by the DDs who always build their packages in clean environments already. Something that has already become enough of a best practice that it's *embarrasing* to say I don't. Probably that feeling means that more than 50% of you do it, so we've already lost 50% of the benefit of ad-hoc building. That was ok because we gained other, larger benefits. So I don't think that we need grand schemes of test building every package in purposefully weird systems. We could probably do as well as we do now by instead stracing builds of some fraction, checking that all files they try to accress come from build dependencies. dpkg-depcheck can nearly do that now -- just modify it to report accesses to non-installed files. -- see shy jo
signature.asc
Description: Digital signature
