Re: more GNU make 4.2.90 issues on Solaris 10

[Paul Smith]

On Tue, 2019-08-27 at 12:13 -0700, Paul Eggert wrote:
> After applying the patches I recently sent to bug-make, I still see several 
> tests failing on native Solaris 10 sparc (32-bit), built with Solaris tools 
> (including Oracle Studio 12.6) instead of with GNU tools. Please see attached 
> files, giving the relevant part of the 'make' output and the *.diff files for 
> the failing tests.
> 
> One problem seems to be that some tests assume that 'make check' runs GNU 
> Make 
> instead of /usr/ccs/bin/make, which chatters less.

I don't think (I hope!!) this is not the problem.  If any part of the
tests don't run GNU make it's a serious fail clearly.  Can you expand
upon why you think this?

For all the tests failing with the "No such file or directory" errors
missing, I think that's due to a difference in the behavior in
posix_spawn() as I mentioned to Dennis.

If you have a minute can you retry with adding --disable-posix-spawn to
the configure line and retest?

> The work/functions/wildcard test fails because on Solaris 10, the command 
> "/bin/sh -c 'kill -2 $$'" causes /bin/sh to call _exit (2000) (!),

!!!

Hm.  I can try to have the test run the command and see what the result
is before starting the test.  They do that in various places where
results are system-dependenct.

> The archives test fails because /usr/ccs/bin/ar is a bit chattier on Solaris.

Most ar's print different things, which is why the test framework runs
a sample ar and uses the output to compare against the output of make. 
I think I might have messed that up somehow since it only fails with
the long name... ah yes, I see it (dumb mistake!)

> The vpathplus test fails due to file name order. Perhaps this is because on 
> Solaris 10 GNU Make uses its own glob, not Solaris's glob.

I don't think this is related to glob, this is removing intermediate
files.  I don't know offhand why we would remove these in a different
order, it should be deterministic.  I'll check it out.

> I don't offhand see why the output-sync tests are failing.

Me either, that's super strange.

Does anyone remember if GNU make 4.2.1's output-sync tests failed this
way?