On 2020/07/27 20:09, Alexander Bluhm wrote:
> On Sun, Jul 26, 2020 at 05:52:23PM +0100, Stuart Henderson wrote:
> > On 2020/07/26 09:10, wen heping wrote:
> > > ii) Remove unneeded BUILD_DEPENDS=${RUN_DEPENDS}
> >
> > I don't generally like this, in cases where upstream has their own
> > checks (common in Perl ports) it makes it harder to check if the
> > dependencies are correct.
>
> Some years ago we agreed to avoid BUILD_DEPENDS=${RUN_DEPENDS}. By
> reducing install and uninstall the bulk build is faster.
>
> I detect missing dependencies by comparing portgen output and
> inspecting Makefile.PL changes. Pure Perl modules have no build
> dependencies anyway. I doubt that many porters will see a missing
> run dependency when something scrols by during a ports build.
Sometimes pure Perl modules do have build dependencies - I also had to
fix p5-Rex the same day where something was added as a RUN_DEPENDS that
was required as a build dependency (and not needed at runtime).
Porters often won't spot discrepancies in p5-* ports because half of
them don't have BUILD_DEPENDS=${RUN_DEPENDS}, if it was common to do this
then it would be more obvious when something is missing. Of course things
would still be missed because we often have a wide set of p5-* packages
already installed but it would catch some. (Porters are also supposed to
check autoconf output for "hidden" dependencies - the Perl dependency
warnings are much easier to see than these!)
> Of course if it helps you maintaining your modules, keep them.
I do find it's useful for things which I maintain.
Also (as a bulk builder) I'm not too bothered about the effect on bulk
build time from installing/uninstalling dependencies. The things which
are most time-consuming for bulk builds are 1) building 4 copies of
chrome, 2) the rust compiler being so slow (not helped by the ecosystem
resulting in dozens or more libraries being rebuilt in each port), and
3) having to deal with bad commits. Of these, 3 is easily the worst
because it needs human intervention, I find anything that even slightly
reduces the likelihood of this to be quite helpful and easily worth a
few extra hours of build time.
