Jakub Jelinek <ja...@redhat.com> writes:
>> I just noticed that I see this error, so the problem still
>> exists at r209347 (seen for cris-elf).
>>
>> (The FAILs are old and provided only for context; the ERRORs are
>> fatal.)
>
> Does say following patch make the problem go away?
>
> --- gcc/gcc/testsuite/g++.dg/dso/dlclose1.C 2014-04-11 10:33:22.552213377
> +0200
> +++ gcc/gcc/testsuite/g++.dg/dso/dlclose1.C 2014-04-14 08:27:57.311138264
> +0200
> @@ -1,7 +1,7 @@
> // PR c++/60731
> // { dg-do run { target { dlopen && disable-for-4.9 } } }
> // { dg-add-options dlopen }
> -// { dg-build-dso "dlclose1-dso.cc" }
> +// ( dg-build-dso "dlclose1-dso.cc" )
It would: I've used this instead:
> -// { dg-build-dso "dlclose1-dso.cc" }
> +// { dg- build-dso "dlclose1-dso.cc" }
dg-build-dso is evaluated even if the test isn't run. And the problem
is indeed the missing support for nested dg-test invocations. I should
have known since I provided the patch for that 3 years ago ;-)
My suggestion is to leave Jason's reversal for the 4.9 branch. For
mainline, we should require DejaGnu >= 1.5, but before reverting the
reversal, I'll try to come up with a check that avoids skipping large
parts of g++.dg with an older version.
Rainer
--
-----------------------------------------------------------------------------
Rainer Orth, Center for Biotechnology, Bielefeld University
