------- Comment #2 from mikael dot morin at tele2 dot fr 2008-11-05 11:43
-------
There is a reset of the expression to NULL_EXPR in case we encounter c_null_ptr
or c_null_funptr.
However, the if conditions for this relies on the is_iso_c attribute, which is
the case of c_loc.
This patch changes this behavior by relying on the symtree names instead.
Index: trans-expr.c
===================================================================
--- trans-expr.c (révision 141583)
+++ trans-expr.c (copie de travail)
@@ -3689,15 +3682,16 @@ gfc_trans_structure_assign (tree dest, gfc_expr *
{
/* Skip absent members in default initializers. */
if (!c->expr)
- continue;
+ continue;
/* Update the type/kind of the expression if it represents either
C_NULL_PTR or C_NULL_FUNPTR. This is done here because this may
be the first place reached for initializing output variables that
have components of type C_PTR/C_FUNPTR that are initialized. */
- if (c->expr->ts.type == BT_DERIVED && c->expr->ts.derived
- && c->expr->ts.derived->attr.is_iso_c)
- {
+ if (c->expr->symtree && c->expr->symtree->name
+ && (!strcmp ("c_null_ptr", c->expr->symtree->name)
+ || !strcmp ("c_null_funptr", c->expr->symtree->name)))
+ {
c->expr->expr_type = EXPR_NULL;
c->expr->ts.type = c->expr->ts.derived->ts.type;
c->expr->ts.f90_type = c->expr->ts.derived->ts.f90_type;
With this patch alone, I still get wrong-code though (pointer uninitialized
instead of nullified).
I've had to revert this patch
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35150#c7
to have it working.
There is only one regression, c_loc_tests_12.f03.
That one is expected, and we are back at PR35150.
I looked at that one too, but I must admit I'm at least as confused as FX when
he fixed it.
If some middle-end gurus could take a look/give some tips there...
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35983
