g77 testsuite failures under hpux 10.20

[Jeffrey A Law]

  In message < 200007071743.NAA03949@hiauly1.hia.nrc.ca >you write:
  > I added a simple patch to store in gc memory if ggc_p is true.  This
  > appears to fix the problem.  However, I am uncertain as to whether
  > gc memory is suitable "permanent" labels.
It is.  The whole point is we need that string to hang around until it's
not needed anymore :-)  That's precisely what GC does.  Once the last
reference goes away, the string itself can go away too.

  > There is one other use of obstack_alloc is pa.c.  This is for deferred
  > plabels.  I am wondering if this name should also be in ggc memory when
  > ggc_p is true.
I believe we do the right thing in the deferred plabel code.

The difference is the encode case you're fixing modifies a hunk of RTL
and stuffs an obstack thingie into it.

The deferred plabels just uses the obstack as a place to hold a bunch of
strings until we don't need them anymore -- it never stomps on RTL bits.

  > In looking at ia64_encode_section_info, it referred to the v850 port.
  > On inspection, it may have the same problem as the pa port.  Thus, the
  > problem doesn't appear limited to just the pa.
Depends on the precise usage.

jeff