Michael McConville wrote:
> Marc Espie wrote:
> > On Wed, Mar 16, 2016 at 07:48:53PM +0100, Andreas Kusalananda Kähäri wrote:
> > > Hi,
> > >
> > > I'm building my ports with MAKE_JOBS=4, and I noticed that a few of the
> > > ports were failing.
> > >
> > > * devel/boehm-gc failed sometimes (can't find "libgc.la" towards the
> > > end of the build).
> > >
> > > * lang/ruby/1.8 failed consistently (can't find "miniruby" at
> > > beginning of the build).
> > >
> > > * editors/vim failed consistently (can't find "ex.1" and other
> > > manuals when packaging). I tried the "huge no_x11"
> > > and "no_x11" flavours.
> > >
> > > * lang/ocaml failed consistently ("Unbound module Pervasives"
> > > early in the build, but the build doesn't halt until
> > > much later).
> > >
> > > * security/nss failed inconsistently, at different stages of the
> > > build.
> > >
> > > I added PARALLEL_BUILD=No to these ports and now they build fine.
> > >
> > > Patches included.
> > >
> > > Cheers,
> >
> > Failing is not the issue. Figuring out whether the Makefile is wrong or
> > whether there's some issue in our make(1) is...
> >
> I noticed that we installing in parallel by default if we build in
> parallel ($PARALLEL_INSTALL). I know that some projects don't advise
> this - Pidgin comes to mind. At least in the case of Ocaml, though, it's
> failing before the fake target.
Also, parallel installation isn't established practice in many
communities (the canonical command is ./configure && make -jX && make
install, after all), so I suspect it doesn't get tested in many
codebases.
If it only causes problems in a few ports, though, it's probably best to
just fix and submit upstream where necessary, IMO.
