Re: g77 testsuite failures under hpux 10.20

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

  In message <200007041753.NAA05051@hiauly1.hia.nrc.ca>you write:
  > >   >         arg 0 <function_decl 4009c700 do_lio type <function_type 4009
  > c300>
  > >   >             addressable used public common external SI file 19990218-
  > 0.f li
  > >   > ne 5
  > >   >             (mem:SI (symbol_ref/v:SI ("")) 0) chain <function_decl 40
  > 09c380
  > >   >  s_wsle>>>
  > > OK.  So things are busted at the tree level.  A good thing to know.
  > > 
  > > Next I would recommend a breakpoint in make_decl_rtl conditional on
  > > decl == 0x4009c700 (ie the FUNCTION_DECL node mentioned in arg0 above).
  > 
  > This looks ok.  make_decl_rtl is called once for decl == 0x4009c700.  The
  > result looks ok.  Note, there are two previous calls to do_lio in MAIN__,
  > and the rtl and assembly code for these calls are ok.  It is only the calls
  > in fun_ that are bad.
OK.  This sounds more and more like a garbage collection problem.  It may
be worth turning on some of the garbage collection debug/sanity checking to
see if it helps narrow down the problem.

I would start looking at the passed in DECL and verify that its tree node is
not stored in a global variable without a suitable call to register the node
as a garbage collection root.

jeff