As complained by Coverity, which was pointed out to my
by Fredering (thanks!), that patch lead to an inconsistency:
+ if (code->ext.omp_clauses)
+ resolve_omp_clauses (code, code->ext.omp_clauses, NULL);
+ if (!code->ext.omp_clauses->critical_name
Before, we had:
!$omp critical [(name) [hint()]]
but as for hint == 0 '(name)' is not required, we now have:
!$omp critical [(name)] [hint()]
Result: gfc_match_omp_clauses is now called unconditionally
and (unless MATCH_ERROR) it always allocates 'ext.omp_clauses'.
Thus, we can now remove first condition – which caused that
Coverity's scanner assumed that the second condition
could dereference a NULL pointer in the 'false' case.
Committed as obvious.
Tobias
-----------------
Mentor Graphics (Deutschland) GmbH, Arnulfstraße 201, 80634 München / Germany
Registergericht München HRB 106955, Geschäftsführer: Thomas Heurung, Alexander
Walter
commit 20d0d6527c36005a9664f7ac9eb50f1071d5670c
Author: Tobias Burnus <tob...@codesourcery.com>
Date: Mon Jul 27 09:12:15 2020 +0200
fortran/openmp.c - silence static analyzer warning
gcc/fortran/ChangeLog:
* openmp.c (gfc_resolve_omp_directive): Remove NULL check
for clauses in EXEC_OMP_CRITICAL as it no longer can be NULL.
diff --git a/gcc/fortran/openmp.c b/gcc/fortran/openmp.c
index f8f2439b6e4..4a0466f968d 100644
--- a/gcc/fortran/openmp.c
+++ b/gcc/fortran/openmp.c
@@ -6560,8 +6560,7 @@ gfc_resolve_omp_directive (gfc_code *code, gfc_namespace *ns ATTRIBUTE_UNUSED)
resolve_omp_atomic (code);
break;
case EXEC_OMP_CRITICAL:
- if (code->ext.omp_clauses)
- resolve_omp_clauses (code, code->ext.omp_clauses, NULL);
+ resolve_omp_clauses (code, code->ext.omp_clauses, NULL);
if (!code->ext.omp_clauses->critical_name
&& code->ext.omp_clauses->hint
&& code->ext.omp_clauses->hint->ts.type == BT_INTEGER
