[RFC] type safe trees

[Mike Stump]

On Thursday, June 24, 2004, at 09:35  PM, Geoff Keating wrote:
> Wasn't the above description sufficient?  The symptom is that the GC 
> parameters are locked at 4k, even with --disable-checking.

The problem is that the kernel says that applications should use an 
insanely small amount of memory, so gcc does clamps at 4M, the minimum. 
  This is wrong, as the kernel should not say that.  This is a darwin 
bug.

You can do:

ulimit -d unlimited

in .profile, and avoid the issue.  ulimit -d will tell you what the 
value is currently.  Anything less than the total size of your real 
memory is just plain, well, wrong.  On one of my systems, it comes up 
at 6M, which is stupid.

ggc_rlimit_bound is the routine, and you can step it and see what they 
kernel says to use.  We had an apple local patch:

   /* APPLE LOCAL begin GC heuristics */
   /* ... but not on Darwin, where rlimit is gloriously broken! */
   /* min_heap_kbytes = ggc_rlimit_bound (min_heap_kbytes); */
   /* APPLE LOCAL end GC heuristics */

to ignore the bug, but that seems to have gone missing.  This issue 
mirrors the stack limit bug, that has _now_ been fixed.

This issue has been talked about before, 
http://gcc.gnu.org/ml/gcc/2003-02/msg01671.html has the past discussion.

> No, I don't know how to fix it.  I don't even know if it's really in 
> FSF GCC, it may be a merge bug into our local tree.

A lost apple local.  It really must not be apple local.  :-(