http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51852
--- Comment #18 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-01-27 08:52:28 UTC --- Created attachment 26476 --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=26476 pointer_map patch local_specializations isn't GTY(()) marked at all and inserts GC TREE_LIST nodes into the hash table that aren't reachable otherwise. The obvious fix I've tried first: --- pt.c 2012-01-27 09:28:26.887327122 +0100 +++ pt.c 2012-01-27 09:45:01.787479945 +0100 @@ -79,7 +79,7 @@ static tree cur_stmt_expr; /* A map from local variable declarations in the body of the template presently being instantiated to the corresponding instantiated local variables. */ -static htab_t local_specializations; +static GTY ((param_is (union tree_node))) htab_t local_specializations; typedef struct GTY(()) spec_entry { for some reason crashed elsewhere. There is additionally the saving of the local_specializations pointer, creating a new hash table and destroying + restoring the pointer afterwards in two places. Can we ggc_collect () only when there is at most one local_specializations array around (i.e. when saved_local_specializations in whomever created the hash table is NULL)? And the second question is, are the local specialization trees also reachable from some other GC root except for local_specializations htab? If they are reachable from elsewhere, it would seem to me it would be much cheaper to use a pointer_map as done in this patch (the testcase works then).
