gc problem with nested functions

[Jeffrey A Law]

  In message < Pine.LNX.4.10.9909111117470.6144-100000@biriani.cygnus.co.uk >you 
write:
  > or whether the amount of memory allocated for them is so small that it
  > doesn't matter if we never free them (i.e. because we have one global
  > obstack).  Comments?
I suspect we do not want constants allocated per-function.  That would increase
TOC space on most targets (and for the ppc targets, we're already using way
too much TOC space).

I've also always found it awful annoying that I get multiple copies of floating
point constants because the const_doubles are per-function constants.  

Instead of looking to output constants on a per-function basis, we should be
looking to defer them until the end of compilation and emit them all at once.
One instance of each constant only please :-)


jeff