David Laight <david@l8s.co.uk> writes:

> On Thu, Jun 07, 2007 at 08:24:59PM -0400, Greg Troxel wrote:
>> Normally I just commit adding MAKE_JOBS_SAFE=no when something doesn't
>> build but does build with MAKE_JOBS_SAFE=no.  But this time I tried to
>> figure it out because I thought upstream might care.
>
>>  cc -c -DHAVE_CONFIG_H -I. -I../include -I/home/n0/ANONCVS/NetBSD-current/pkgsrc/graphics/sane-backends/work/.buildlink/include -D_REENTRANT -DPATH_SANE_CONFIG_DIR=/usr/pkg/etc/sane.d -DPATH_SANE_DATA_DIR=/usr/pkg/share -DPATH_SANE_LOCK_DIR=/usr/pkg/var/lock/sane -DV_MAJOR=1 -DV_MINOR=0 -O2 -W -Wall sanei_access.c -o sanei_access.o >/dev/null 2>&1
>
> Maybe the compilation failed ?
> The compiler output might be useful!

That was the whole output (had done 'pkg_rolling-replace -uv &>
RR.000').  I reran make in the package, and it worked.  I started a new
build, and encountered a similar failure on a different library and .o
(again with no compile failure evident).  And, I find that the .o is
present after the failure, and that make succeeds just doing the
ar/ranlib.  So I'm pretty sure this is a make synchronization issue.

I'm doing all of this on netbsd-4 from April on a vanilla
single-processor i386:
  cpu0: "Intel(R) Pentium(R) 4 CPU 3.40GHz"
with MAKE_JOBS=2.

If I use BSD make in the sane-backends directory, I can't get it to fail
(but it only builds static and not shared).  With GNU make, it fails
pretty reguarly.  So I'll follow up with the sane or GNU make folks -
apparently this is not a pkgsrc bug.