Re: reload_in_progress || reload_completed

[Jeffrey A Law <law at cygnus dot com>]

  In message <199811251009.KAA12810@sun52.NIS.cambridge>you write:
  > Probably because nothing is said about this in the documentation (that I 
  > can find).  This is certainly something I didn't know.
As I said, it's a common mistake.  It doesn't effect most ports since few
actually want to create new pseudos via the gen_move_insn call in regclass.

And since we didn't actually have a variable which indicated the exact point
after which no new pseudos could be created, most folks have used the
approximation of "no new pseudos after reload starts".  It may be wise to
actually review all code which checks reload_in_progress or reload_completed
to see if it has the potential to lose.

  > I think the comment in reload1.c should also be fixed:
  > 
  >   /* From now on, we need to emit any moves without making new pseudos.  */
  >   reload_in_progress = 1;
Agreed.

Thanks,
jeff