This fixes PR54898, we can't disregard comparing or hashing
TYPE_MAIN_VARIANT when we rely on intact variant relationships.
LTO bootstrapped on x86_64-unknown-linux-gnu, testing in progress.
Richard.
2012-10-12 Richard Biener <rguent...@suse.de>
PR lto/54898
* lto.c (gimple_types_compatible_p_1): Also compare
TYPE_MAIN_VARIANT.
(iterative_hash_gimple_type): Also hash TYPE_MAIN_VARIANT.
Index: gcc/lto/lto.c
===================================================================
*** gcc/lto/lto.c (revision 192392)
--- gcc/lto/lto.c (working copy)
*************** gimple_types_compatible_p_1 (tree t1, tr
*** 581,586 ****
--- 581,595 ----
if (!compare_type_names_p (t1, t2))
goto different_types;
+ /* The main variant of both types should compare equal. */
+ if (TYPE_MAIN_VARIANT (t1) != t1
+ || TYPE_MAIN_VARIANT (t2) != t2)
+ {
+ if (!gtc_visit (TYPE_MAIN_VARIANT (t1), TYPE_MAIN_VARIANT (t2),
+ state, sccstack, sccstate, sccstate_obstack))
+ goto different_types;
+ }
+
/* We may not merge typedef types to the same type in different
contexts. */
if (TYPE_NAME (t1)
*************** iterative_hash_gimple_type (tree type, h
*** 1101,1106 ****
--- 1110,1121 ----
&& TYPE_P (DECL_CONTEXT (TYPE_NAME (type))))
v = visit (DECL_CONTEXT (TYPE_NAME (type)), state, v,
sccstack, sccstate, sccstate_obstack);
+
+ /* Factor in the variant structure. */
+ if (TYPE_MAIN_VARIANT (type) != type)
+ v = visit (TYPE_MAIN_VARIANT (type), state, v,
+ sccstack, sccstate, sccstate_obstack);
+
v = iterative_hash_hashval_t (TREE_CODE (type), v);
v = iterative_hash_hashval_t (TYPE_QUALS (type), v);
v = iterative_hash_hashval_t (TREE_ADDRESSABLE (type), v);
