------- Comment #3 from burnus at gcc dot gnu dot org 2009-07-01 13:40 -------
But of cause also the run-time test fails; for absent arguments a NULL is
passed.
We need to check:
a) optional pointer actual to non-optional dummy
b) optional pointer actual to optional dummy
The -fcheck=pointer specific check would be to check for
"*actual == NULL"
though that needs to be guarded by "actual != NULL".
In general, passing a not-present optional to a non-optional dummy should be
checked somewhere; the question is with which option (-fcheck=call?).
Patch below.
(Note: One has to use TRUTH_ANDIF_EXPR and not TRUTH_AND_EXPR!)
--- trans-expr.c (revision 149129)
+++ trans-expr.c (working copy)
@@ -2778 +2778 @@ gfc_conv_procedure_call (gfc_se * se, gf
- if (gfc_option.rtcheck & GFC_RTCHECK_POINTER)
+ if (gfc_option.rtcheck & GFC_RTCHECK_POINTER && e != NULL)
@@ -2806 +2806,15 @@ gfc_conv_procedure_call (gfc_se * se, gf
- cond = fold_build2 (EQ_EXPR, boolean_type_node, parmse.expr,
+ if (sym->attr.optional)
+ {
+ tree present, nullptr, type;
+ type = TREE_TYPE (parmse.expr);
+ present = fold_build2 (NE_EXPR, boolean_type_node, parmse.expr,
+ fold_convert (type, null_pointer_node));
+ type = TREE_TYPE (type);
+ nullptr = fold_build2 (EQ_EXPR, boolean_type_node,
+ build1 (INDIRECT_REF, type, parmse.expr),
+ fold_convert (type, null_pointer_node));
+ cond = fold_build2 (TRUTH_ANDIF_EXPR, boolean_type_node,
+ present, nullptr);
+ }
+ else
+ cond = fold_build2 (EQ_EXPR, boolean_type_node, parmse.expr,
--
burnus at gcc dot gnu dot org changed:
What |Removed |Added
----------------------------------------------------------------------------
AssignedTo|unassigned at gcc dot gnu |burnus at gcc dot gnu dot
|dot org |org
Status|NEW |ASSIGNED
Last reconfirmed|2009-07-01 13:16:43 |2009-07-01 13:40:51
date| |
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40605
