Re: jobs safety of gcc8 package

[David Holland <dholland-pkgtech%netbsd.org@localhost>]

On Sun, Oct 04, 2020 at 05:36:51PM +0100, Jonathan Perkin wrote:
 > > Yesterday I was building gcc8 so I could update my fortran stuff, and
 > > with MAKE_JOBS set to 6 it failed, the symptoms being first doing two
 > > concurrent ar updates of libbackend.a and then shortly after crashing
 > > out because libbackend.a was corrupted. Restarting the build at this
 > > point after deleting all files namedlibbackend.a repeated the problem
 > > consistently (tried several times). After restarting with 1 job the
 > > build (eventualy) completed.
 > > 
 > > Should we tag it not MAKE_JOBS_SAFE? Does anyone have data about how
 > > many jobs it can handle?
 > 
 > I skipped gcc8, so only have data about lang/gcc49, lang/gcc7, and
 > joyent/gcc9, all of which work perfectly with MAKE_JOBS=48.  I don't
 > know if that's helpful...

It makes it odd that only gcc8 would be broken. But not necessarily
surprising given that it's all autotools.

Anyway, maybe I'll run some test builds this weekend.

 > > And did anyone ever implement per-package jobs limits?
 > 
 > Yeh, MAKE_JOBS.${PKGPATH}=

I was thinking of the other side of the matter, something like
MAKE_JOBS_SAFE_LIMIT=4. Forcing it to 1 if it can safely be 2-3 isn'
good.

-- 
David A. Holland
dholland%netbsd.org@localhost