[Bug lto/48246] ICE in lto_wpa_write_files

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48246

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2011.03.23 10:00:36
                 CC|                            |hubicka at gcc dot gnu.org
     Ever Confirmed|0                           |1

--- Comment #1 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-03-23 10:00:36 UTC ---
Confirmed.

An intermediate partition is completely empty (ltrans_partitions[2], after
lto_balanced_map (), out of 7 partitions in total).  Sorting moves that
empty partition last.

We do seem to allow a single empty partition (probably for an empty CU):

      gcc_assert (cgraph_node_set_nonempty_p (set)
                  || varpool_node_set_nonempty_p (vset) || !i);

so we could change !i for i == n_sets-1, but I suppose we could arrive
with multiple empty partitions here as well.

Honza, why do we even care to assert the above if we handle empty
partitions just fine (in case of !i)?  It looks like some partitioning
sanity check to me, not something we should assert on.