Re: [gentoo-dev] using -j1 with distcc?

[Ferris McCormick]

On Wed, 2006-09-13 at 07:52 -0700, Brian Harring wrote:
> On Wed, Sep 13, 2006 at 10:34:52AM -0400, Aron Griffis wrote:
> > From bind-9.3.2-r4.ebuild:
> >
> >     # idea from dev-libs/cyrus-sasl
> >     if has distcc ${FEATURES}; then
> >         einfo "You have \"distcc\" enabled"
> >         einfo "build with MAKEOPTS=\"-j1\""
> >         jobs="-j1"
> >     else
> >         einfo "build with MAKEOPTS=${MAKEOPTS}"
> >         jobs=""
> >     fi
> >
> >     emake ${jobs} || die "failed to compile bind"
> >
> > I think this is bogus.  If building with distcc reveals a parallel
> > build issue, then the issue exists with or without distcc, it just
> > seems to happen less often without it.  We've been down this road
> > before, maybe people have forgotten?
> >
> > bind-9.3.2-r4.ebuild failed to build for me on dual ia64.  Building
> > with -j1 works.
> >
> > Unless somebody can explain how this is valid, I'll go ahead and fix
> > the bind ebuilds (where "fix" means "use -j1 unconditionally since the
> > Makefiles aren't parallel safe").
>
> Similar trickery in app-office/openoffice, although they enable -jN if
> distcc is enabled, else -j1 ...
>
> Always wondered how that was valid, just avoid OO compiles enough it
> wasn't something I ever got around to looking into :)
> ~harring

I don't see how it can be valid, especially ferringb's example.
Enabling distcc doesn't mean the build will distribute; only that the
possibility is there.  If you happen to build when the other system(s)
is(are) unavailable for some reason, you get everything on the host, so
for ferringb, e.g., this would mean -jN on one system.

As for Aron's case, I agree with him.

Regards,
Ferris
-- 
Ferris McCormick (P44646, MI) <fmccor@g.o>
Developer, Gentoo Linux (Devrel, Sparc)