This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Fix !$omp end .... symbols handling
Quoting Jakub Jelinek <jakub@redhat.com>:
> On Thu, Feb 16, 2006 at 05:38:51PM +0100,
> Tobias.Schlueter@Physik.Uni-Muenchen.DE wrote:
> > Quoting Jakub Jelinek <jakub@redhat.com>:
> > > As shown on the attached testcase, if variable list was on the
> > > !$omp end directive and contained a reference to a symbol that
> > > hasn't been referenced before at all, following reject_statement
> > > would free the symbols, eventhough they were listed in omp_clauses,
> > > so we would then dereference freed memory (or worse reused for something
> > > else). The *_OMP_END_* statements aren't accepted, because omp_clauses
> > > are always attached to the matching starting omp directive.
> > > Tested on x86_64-linux.
> >
> > Why not mark the symbol as referenced when parsing the OMP directive?
>
> I do that
> m = gfc_match_symbol (&sym, 1);
> switch (m)
> {
> case MATCH_YES:
> gfc_set_sym_referenced (sym);
> but that is apparently not enough. All symbols need to be either committed
> or undone. Or do you think I should call gfc_commit_symbol (sym); on each
> symbol as well? As the starting omp directives are known to call
> gfc_commit_symbols (); at the end if they are parsed properly, that
> sounds like an overkill. Plus, if parsing of the starting or end omp
> directives fails, I guess the symbols can be safely undone, their addresses
> aren't stored anywhere.
Ok, yes, I was expressing myself unclearly. I had meant to ask: why don't you
commit the symbols after every OMP statement? I was afraid that you might have
to add this code in several places.
- Tobi