This is the mail archive of the
mailing list for the GCC project.
Re: make: *** wait: No child processes during make -j8 check
On Tue, Jan 17, 2017 at 11:59 AM, Martin Sebor <firstname.lastname@example.org> wrote:
> On 01/17/2017 08:30 AM, Jeff Law wrote:
>> On 01/16/2017 05:37 PM, Martin Sebor wrote:
>>> I've run into this failure during make check in the past with
>>> a very large make -j value (such as -j128), but today I've had
>>> two consecutive make check runs fail with -j12 and -j8 on my 8
>>> core laptop with no much else going on. The last thing running
>>> was the go test suite. Has something changed recently that
>>> could be behind it? (My user process limit is 62863.)
>> What version of make are you using? There was a bug where make could
>> lose track of children a while back. Ask Patsy -- she may remember the
>> RHEL BZ and from that we can probably extract the version #s that were
>> affected and see if they match the version you're running.
> I have make 4.0 and I'm running Fedora 23. I haven't changed
> anything on the machine in a while and used to be able to run
> make -j16. I still can bootstrap with that but make check
> has been failing recently. Three times yesterday (-12, -j8,
> and -j12 again), but then the last one with -j12 completed.
> The frustrating thing is that after the failure I have to
> restart make check from scratch, and so it can waste hours.
I haven't seen any catastrophic failures, but have noticed some
inconsistencies with parallel make check. Sometimes it seems to
substitute a different filename. It seems like make check -j harness
assumes some unique names for temporary / intermediate files that are
not guaranteed by the scripts.