Re: [PATCH] Fix for PR fortran/21375

[Steve Kargl <sgk at troutmask dot apl dot washington dot edu>]

On Sat, Jun 11, 2005 at 03:03:41PM +0200, Tobias Schl?ter wrote:
> Steve Kargl wrote:
> > OK for mainline?  Ok for 4.0.1 with Mark's approval?
> 
> This is ok with one minor nit:
> 
> > +   else
> > +     {
> > +       pstat = apstat = integer_zero_node;
>                            ^^^^^^^^^^^^^^^^^
> 
> Those should be pointers to int4.  See the top of
> gfc_build_io_library_fndecls() for how to create that type.  We should
> probably add a global variable gfc_pint4_type_node for this along the other
> type nodes, and move that code out of gfc_build_io_library_fndecls, but I'm
> fine if you put it in a local variable.
> 

Sigh.  I wish either you or Paul B would take the time to
explain how one works within the poorly commented/documented
trans-*.[ch] files when these minor little nits arise.  While
self-education is a wonderful thing, for many of us spare time
to work on gfortran is very limited.  

Fine I changed gfc_build_builtin_function_decls() to include

  tree gfc_pint4_type_node = build_pointer_type (gfc_int4_type_node);

  gfor_fndecl_deallocate =
    gfc_build_library_function_decl (get_identifier (PREFIX("deallocate")),
				     void_type_node, 2, ppvoid_type_node,
				     gfc_pint4_type_node);

Seems to work ok.

I then change gfc_trans_deallocate() to include

  tree gfc_int4_type_node = gfc_get_int_type (4);

  else
    {
      pstat = apstat = build_pointer_type (gfc_int4_type_node);

KABOOOMMMMMM.

kargl[217] gfc41 -static -o z -fdump-tree-original a.f90
a.f90: In function 'MAIN__':
a.f90:4: internal compiler error: in gimplify_expr, at gimplify.c:4305
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://gcc.gnu.org/bugs.html> for instructions.

kargl[219] cat a.f90
program a
   real, allocatable :: b(:)
   allocate(b(3))
   deallocate(b)
end program a

kargl[220] more a.f90.t02.original

  {
    real4[0:] * * ptr.0;

    ptr.0 = &b.data;
    _gfortran_deallocate (ptr.0, int4 *);
  }

This is obviously wrong.  Sigh, I guess I'll waste another hour 
of time trying to determine how to transmute the above into 

    _gfortran_deallocate (ptr.0, (int4 *) NULL);

-- 
Steve