[PATCH] Fix PR fortran/21977, nested function returning array

[Tobias.Schlueter@Physik.Uni-Muenchen.DE]

Quoting Andrew Pinski <pinskia@physics.uc.edu>:
> After talking with Tobias on IRC, him and I agreed on a slightly
> different patch.
> This patch just moves the NULLing of current_fake_result_decl to be
> right before
> the NULLing of current_function_return_label.
>
>
> OK? Bootstrapped on x86_64-linux-gnu and powerpc-darwin with no
> regressions.

Yes, but please add a testcase where both the contained function and the
container require a fake_result_decl.  I do think it would be valuable to have
an execution test, but I'll leave this to your discretion.  Finally, I thought
we had abandoned adding testcases to the torture suite, but besides XFAILing
which we shouldn't ever have to in this case, I can't think of an advantage of
the dg testsuite, so I won't oppose adding the testcase to the torture suite,
unless someone digs out a relevant mailing list thread, or brings forth a good
argument.

Thanks,
- Tobi