[Bug tree-optimization/24351] [4.1 Regression] ICE in do_simple_structure_copy with some C++ code

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

------- Comment #6 from rguenth at gcc dot gnu dot org  2005-10-24 11:01 -------
Also, while we have the same types for lhs and rhs in do_structure_copy, namely

(gdb) call debug_tree(lhstype)
 <record_type 0x40232c94 adaptsint sizes-gimplified needs-constructing type_1
type_5 HI
    size <integer_cst 0x4018d300 type <integer_type 0x4019f05c bit_size_type>
constant invariant 16>
    unit size <integer_cst 0x4018d318 type <integer_type 0x4019f000 unsigned
int> constant invariant 2>
    align 8 symtab 0 alias set -1
    fields <field_decl 0x40232f18 functor_
        type <record_type 0x40232bdc adaptor_functorint sizes-gimplified type_1
type_5 QI
            size <integer_cst 0x4018d1f8 constant invariant 8>
            unit size <integer_cst 0x4018d210 constant invariant 1>
            align 8 symtab 0 alias set -1 fields <type_decl 0x40212c30
adaptor_functorint>
            X() X(constX&) this=(X&) n_parents=1 use_template=0
interface-unknown
            pointer_to_this <pointer_type 0x402377e8> chain <type_decl
0x40212bc8 adaptor_functorint>>
        nonlocal decl_3 QI file /suse/rguenther/src/tests/pr24351.C line 8 size
<integer_cst 0x4018d1f8 8> unit size <integer_cst 0x4018d210 1>
        align 8 offset_align 128
        offset <integer_cst 0x4018d198 constant invariant 0> bit offset
<integer_cst 0x4018d1f8 8> context <record_type 0x40232c94 adaptsint>
        chain <type_decl 0x40212d68 adaptsint type <record_type 0x40232c94
adaptsint>
            nonlocal decl_4 VOID file /suse/rguenther/src/tests/pr24351.C line
6
            align 1 context <record_type 0x40232c94 adaptsint>
           >>
   needs-constructor X(constX&) this=(X&) n_parents=1 use_template=0
interface-unknown
    pointer_to_this <pointer_type 0x40232e60> chain <type_decl 0x40212d00
adaptsint>>

the full sizes from the variable infos for lhs/rhs do not match:

(gdb) print *get_varinfo(lhs.var)
$9 = {id = 6, name = 0x898aaf7 "a", decl = 0x4019a7e8, offset = 8, size = 8, 
  fullsize = 24, next = 0x899ca54, node = 6, address_taken = 1, 
  indirect_target = 0, is_artificial_var = 0, is_special_var = 0, 
  is_unknown_size_var = 0, has_union = 0, is_heap_var = 0, 
  solution = 0x89996c8, variables = 0x89996d8, complex = 0x0}
(gdb) print *get_varinfo(rhs.var)
$10 = {id = 8, name = 0x896d7d0 "D.1877", decl = 0x4019ae18, offset = 8, 
  size = 8, fullsize = 16, next = 0x0, node = 8, address_taken = 0, 
  indirect_target = 0, is_artificial_var = 0, is_special_var = 0, 
  is_unknown_size_var = 0, has_union = 0, is_heap_var = 0, 
  solution = 0x8999708, variables = 0x8999718, complex = 0x0}


This is during the assignment of

(gdb) call debug_generic_expr(t)
#   SFT.2D.1881_20 = V_MAY_DEF <SFT.2D.1881_17>;
#   VUSE <SFT.0D.1879_8>;
aD.1849.functor_D.1824.functor_D.1764 = D.1877

i.e. during the second may-alias pass.

So another workaround may be in do_structure_copy:


      /* The size only really matters insofar as we don't set more or less of
         the variable.  If we hit an unknown size var, the size should be the
         whole darn thing.  */
      if (get_varinfo (rhs.var)->is_unknown_size_var)
        rhssize = ~0;
      else
        rhssize = get_varinfo (rhs.var)->fullsize; /*TREE_INT_CST_LOW
(rhstypesize);*/

      if (get_varinfo (lhs.var)->is_unknown_size_var)
        lhssize = ~0;
      else
        lhssize = get_varinfo (lhs.var)->fullsize; /*TREE_INT_CST_LOW
(lhstypesize);*/


Oh well.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24351