[Bug tree-optimization/45857] [4.5 Regression] memory exhausted compiling some variadic templates

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45857

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2
             Status|UNCONFIRMED                 |NEW
           Keywords|                            |compile-time-hog,
                   |                            |memory-hog
   Last reconfirmed|                            |2010.10.01 11:40:17
          Component|c++                         |tree-optimization
                 CC|                            |hubicka at gcc dot gnu.org
     Ever Confirmed|0                           |1
            Summary|memory exhausted compiling  |[4.5 Regression] memory
                   |some variadic templates     |exhausted compiling some
                   |                            |variadic templates
   Target Milestone|---                         |4.5.2

--- Comment #1 from Richard Guenther <rguenth at gcc dot gnu.org> 2010-10-01 
11:40:17 UTC ---
It never terminates running ehcleanup.

0x0000000000c5c53d in cleanup_all_empty_eh ()
    at /space/rguenther/src/svn/gcc-4_5-branch/gcc/tree-eh.c:3831
3831          changed |= cleanup_empty_eh (lp);
Value returned is $5 = 1 '\001'
(gdb) finish
....

(gdb) p *cfun->eh->lp_array
$12 = {base = {num = 5903794, alloc = 6824169, vec = {0x0}}}

looks like very many landing-pads.  Huh.

<bb 2>:
  this.5_10 = (struct _Head_base *) &a.foo._M_t;
  _Head_base::_Head_base (this.5_10);
  D.2498_12 = this.5_10 + 1;
  _Head_base::_Head_base (D.2498_12);
  this.5_16 = (struct _Head_base *) &a.bar._M_t;
  [LP 5903793] _Head_base::_Head_base (this.5_16);

<bb 3>:
  D.2505_18 = this.5_16 + 1;
  [LP 5903793] _Head_base::_Head_base (D.2505_18);
  goto <bb 7>;

<L4>:

<L3>:

<L5>:

<L6>: [LP 5903793]

<L7>:
  goto <bb 8> (<L6>);


I suppose we're doing no-op transforms keeping changed true.  We seem
to be moving the LP through the loop.