[Bug middle-end/64922] runtime error: member call on misaligned address for type 'struct _Rep'

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64922

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2015-02-03
                 CC|                            |hubicka at gcc dot gnu.org
   Target Milestone|---                         |5.0
     Ever confirmed|0                           |1

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Goes away with -fno-devirtualize or -fno-devirtualize-speculatively.
Looking at the dumps, after IPA we have:
  [pol6.C:39:3] Head::parse (_33, _30);
call (looks fine), but instead of the 3 expected
  [pol6.C:39:3] Code_Block::parse (_36, _30);
  [pol6.C:39:3] Code_Block::parse (_36, _30);
  [pol6.C:39:3] Code_Block::parse (_36, _30);
calls that show e.g. with -fno-devirtualize-speculatively we instead have:
  [pol6.C:39:3] __builtin_unreachable (_36, _30);
  [pol6.C:39:3] __builtin_unreachable (_36, _30);
  [pol6.C:39:3] __builtin_unreachable (_36, _30);
and no wonder everything goes wrong afterwards.
I don't see why IPA devirtualization thinks something is wrong on the
Code_Block::parse calls.  And, because for these __builtin_unreachable calls
IPA doesn't strip the undesirable arguments, it isn't transformed even with
-fsanitize=undefined into __ubsan_handle_unreachable (), which would make it
far easier to figure this out.

Honza, can you please have a look?