This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
[Bug c++/18968] [4.0 regression] ICE: tree check: expected ssa_name, have addr_expr in vrp_hash
- From: "pinskia at gcc dot gnu dot org" <gcc-bugzilla at gcc dot gnu dot org>
- To: gcc-bugs at gcc dot gnu dot org
- Date: 13 Dec 2004 23:18:19 -0000
- Subject: [Bug c++/18968] [4.0 regression] ICE: tree check: expected ssa_name, have addr_expr in vrp_hash
- References: <20041213212814.18968.schwab@suse.de>
- Reply-to: gcc-bugzilla at gcc dot gnu dot org
------- Additional Comments From pinskia at gcc dot gnu dot org 2004-12-13 23:18 -------
(In reply to comment #7)
> Subject: Re: [4.0 regression] ICE: tree
> check: expected ssa_name, have addr_expr in vrp_hash
>
> So what does the condition look like? I don't think we should be
> getting into this code with anything other than an SSA_NAME as
> the first argument in a COND_EXPR.
Sorry about that, I was still working through the code of DOM.
The condition looks like &0B->a == 0 which means that the 0 is an integer but other size of the equal
expression is a pointer. This means that the types are mismatched by the front-end.
This patch is the correct fix which fixes the problem in the front-end which fixes the mismatched types:
Index: class.c
===============================================================
====
RCS file: /cvs/gcc/gcc/gcc/cp/class.c,v
retrieving revision 1.692
diff -u -p -r1.692 class.c
--- class.c 8 Dec 2004 08:35:33 -0000 1.692
+++ class.c 13 Dec 2004 23:12:29 -0000
@@ -295,8 +295,11 @@ build_base_path (enum tree_code code,
/* Now that we've saved expr, build the real null test. */
if (null_test)
- null_test = fold (build2 (NE_EXPR, boolean_type_node,
- expr, integer_zero_node));
+ {
+ tree zero = cp_convert (TREE_TYPE (expr), integer_zero_node);
+ null_test = fold (build2 (NE_EXPR, boolean_type_node,
+ expr, zero));
+ }
/* If this is a simple base reference, express it as a COMPONENT_REF. */
if (code == PLUS_EXPR && !virtual_access
I found where the problem was in the front-end by using this patch which catches mismatched types
with EQ_EXPR and NE_EXPRs (only, it should be extended for 4.1 to support more expressions):
Index: tree.c
===============================================================
====
RCS file: /cvs/gcc/gcc/gcc/tree.c,v
retrieving revision 1.456
diff -u -p -r1.456 tree.c
--- tree.c 9 Dec 2004 10:54:36 -0000 1.456
+++ tree.c 13 Dec 2004 23:16:53 -0000
@@ -2573,6 +2573,10 @@ build2_stat (enum tree_code code, tree t
t = make_node_stat (code PASS_MEM_STAT);
TREE_TYPE (t) = tt;
+ if (code == EQ_EXPR || code == NE_EXPR)
+ {
+ gcc_assert (TREE_TYPE (arg0) == TREE_TYPE (arg1));
+ }
/* Below, we automatically set TREE_SIDE_EFFECTS and TREE_READONLY for the
result based on those same flags for the arguments. But if the
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18968