Mainline failed to bootstrap on Linux/x86

[Richard Kenner]

    I reverted this hunk, this allows bootstrap (well stage2
    still running right now), but unlikely to be the proper fix.

I don't have a test environment just yet, but my guess is that this will work,
meaning that the original change exposed a latent bug, which is actually good
to get rid of.  Can you verify?  (I mean to leave the old hunk there and
apply this.)

*** stor-layout.c	29 Apr 2003 22:48:55 -0000	1.153
--- stor-layout.c	30 Apr 2003 19:40:09 -0000
*************** mode_for_size_tree (size, class, limit)
*** 244,247 ****
--- 244,248 ----
  {
    if (TREE_CODE (size) != INTEGER_CST
+       || TREE_OVERFLOW (size)
        /* What we really want to say here is that the size can fit in a
  	 host integer, but we know there's no way we'd find a mode for