[Bug c++/32639] ptrmem1.C now ICE's on mainline

["rguenth at gcc dot gnu dot org" <gcc-bugzilla at gcc dot gnu dot org>]

------- Comment #3 from rguenth at gcc dot gnu dot org  2007-07-05 19:06 -------
We compare two pointer-to-method types that "look" the same:

Breakpoint 3, useless_type_conversion_p (outer_type=0xb7d869b4, 
    inner_type=0xb7d866c0) at /home/richard/src/trunk2/gcc/tree-ssa.c:917
917       inner_type = TYPE_MAIN_VARIANT (inner_type);
QWidget:: *
QWidget:: *
(gdb) call debug_tree (outer_type)
 <pointer_type 0xb7d869b4
    type <method_type 0xb7d86360
        type <void_type 0xb7cc15e8 void VOID
            align 8 symtab 0 alias set -1 canonical type 0xb7cc15e8
            pointer_to_this <pointer_type 0xb7cc1654>>
        QI
        size <integer_cst 0xb7ca84b4 constant invariant 8>
        unit size <integer_cst 0xb7ca84d0 constant invariant 1>
        align 8 symtab 0 alias set -1 canonical type 0xb7d86438 method basetype
<record_type 0xb7d860d8 QWidget>
        arg-types <tree_list 0xb7d32dac value <pointer_type 0xb7d863cc>
            chain <tree_list 0xb7d32d74
                purpose <integer_cst 0xb7ca8c08 constant invariant 1> value
<boolean_type 0xb7cba57c bool>
                chain <tree_list 0xb7ca8e70 value <void_type 0xb7cc15e8
void>>>>
        pointer_to_this <pointer_type 0xb7d869b4>>
    unsigned SI
    size <integer_cst 0xb7ca863c type <integer_type 0xb7cba06c bit_size_type>
constant invariant 32>
    unit size <integer_cst 0xb7ca8428 type <integer_type 0xb7cba000 unsigned
int> constant invariant 4>
    align 32 symtab 0 alias set -1 canonical type 0xb7d866c0>
(gdb) call debug_tree (inner_type)
 <pointer_type 0xb7d866c0
    type <method_type 0xb7d86438
        type <void_type 0xb7cc15e8 void VOID
            align 8 symtab 0 alias set -1 canonical type 0xb7cc15e8
            pointer_to_this <pointer_type 0xb7cc1654>>
        QI
        size <integer_cst 0xb7ca84b4 constant invariant 8>
        unit size <integer_cst 0xb7ca84d0 constant invariant 1>
        align 8 symtab 0 alias set -1 canonical type 0xb7d86438 method basetype
<record_type 0xb7d860d8 QWidget>
        arg-types <tree_list 0xb7d32de4 value <pointer_type 0xb7d863cc>
            chain <tree_list 0xb7d32dc8 value <boolean_type 0xb7cba57c bool>
                chain <tree_list 0xb7ca8e70 value <void_type 0xb7cc15e8
void>>>>
        pointer_to_this <pointer_type 0xb7d866c0>>
    sizes-gimplified unsigned SI
    size <integer_cst 0xb7ca863c type <integer_type 0xb7cba06c bit_size_type>
constant invariant 32>
    unit size <integer_cst 0xb7ca8428 type <integer_type 0xb7cba000 unsigned
int> constant invariant 4>
    align 32 symtab 0 alias set -1 canonical type 0xb7d866c0>

(note they have the same canonical types, for both the pointer and the
pointed-to type.  note they both have their alias set not computed.)

Now at the point we verify if those pointer types are compatible we check

      /* Do not lose casts between pointers with different
         TYPE_REF_CAN_ALIAS_ALL setting or alias sets.  */
      if ((TYPE_REF_CAN_ALIAS_ALL (inner_type)
           != TYPE_REF_CAN_ALIAS_ALL (outer_type))
          || (get_alias_set (TREE_TYPE (inner_type))
              != get_alias_set (TREE_TYPE (outer_type))))
        return false;

which calls get_alias_set on both pointed-to types which (surprisingly?)
returns 2 for one and 3 for the other one.  Because, while we specialize
FUNCTION_TYPE

  /* There are no objects of FUNCTION_TYPE, so there's no point in
     using up an alias set for them.  (There are, of course, pointers
     and references to functions, but that's different.)  */
  else if (TREE_CODE (t) == FUNCTION_TYPE)
    set = 0;

we do not so for METHOD_TYPE, which looks like an omission.  Also it looks
like in useless_type_conversion_p we can move the canonical types check
to the top and use it for all types.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32639