------- Comment #13 from jvdelisle at gcc dot gnu dot org 2007-12-22 16:28
-------
This is surprising! If I get rid of the hack that was needed before for
optional_dim_2.f90 :
Index: iresolve.c
===================================================================
--- iresolve.c (revision 131133)
+++ iresolve.c (working copy)
@@ -590,12 +590,6 @@ gfc_resolve_cshift (gfc_expr *f, gfc_exp
if (dim != NULL)
{
- if (dim->expr_type != EXPR_CONSTANT &&
dim->symtree->n.sym->attr.optional)
- {
- /* Mark this for later setting the type in gfc_conv_missing_dummy.
*/
- dim->representation.length = shift->ts.kind;
- }
- else
{
gfc_resolve_dim_arg (dim);
/* Convert dim to shift's kind to reduce variations. */
The new case passes (for cshift and I would presume eoshift) and the
optional_dim_2.f90 passes and I see no apparent regressions. Still testing.
This means something got fixed somewhere else. I also checked this on
ppc64-linux with the same result.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34540
