On 15/04/2024 13:18, Rainer Orth wrote:
Hi Jørgen,
the new gcc.misc-tests/gcov-22.c test loops on SPARC (both Solaris and
Linux). I've filed PR gcov-profile/114720 for this, but couldn't find
any bugzilla account of yours to Cc:
Rainer
Rainer,
Could you please try this patch? I don't have a sparc nor non-glibc build
(and getting a virtual one up will take a while). I suppose the problem is
given that the test also FAILs on Linux/sparc64, it's not a
glibc-vs-non-glibc issue.
Sure, I just figured that a different libc setjmp might have the same
behaviour also on x86.
FWIW, there are both Solaris/SPARC and Linux/sparc64 systems available
in the cfarm; no need to build one yourself. But fully agreed, running
a full bootstrap to verify a single testcase is a bit much if it can be
done more easily ;-)
I was not aware of the cfarm, that seems like a splendid resource, I'll
create an account.
that after longjmp the return address is to the call to setdest(), not
jump() (like is assumed), which creates the loop. If so, just guarding the
longjmp should be fine, the point of the test is to make sure that both
branches can be taken and recorded when the cond is a setjmp. If it works I
will document it and post the patch.
I've just tried the patch on both sparc-sun-solaris2.11 and
i386-pc-solaris2.11 and the test now PASSes on both.
Thanks.
Rainer
Ok, thanks, I'll type up a patch and submit again for review.
