Re: [Patch/gfortran] Fix for PR16939

[THOMAS Paul Richard 169137 <prthomas at drfccad dot cea dot fr>]

Tobi,

>> I thought to do the same. However, I am not sure that it is clearer. More
>> importantly, I decided to ring-fence the scalar case because I think that
>> dummy argument arrays of character pointers need the same treatment. They
>> are broken elsewhere, such that I cannot even get to this point usefully.

> They will still need an array descriptor, won't they? So I don't think
this
> will be the case, as the code there is meant to do the correct thing with 
> array descriptors IIUC.

The array descriptor is what is passed but it needs dereferencing for
pointer, dummy, character arrays.  Doing so fixes PR18689, in addition
PR16939.

>> How about I make it leaner and meaner once I have a clear view of what to
do
>> with arrays?

> I don't understand how pushing this backwards

I am not pushing backwards!  Anyway, I am now happy to combine the ifs
appropriately.  I'll do it tonight and resubmit the patch as a fix for both
PRs.

It will be along the lines:

      /* Dereference non-character scalar and pointer
         dummy arguments. HOW DO I INDENT THIS?*/
      if (sym->attr.dummy
	  && ((sym->ts.type != BT_CHARACTER
	      && !sym->attr.dimension)
	    || (sym->attr.pointer
	      && sym->ts.type == BT_CHARACTER)))
	se->expr = gfc_build_indirect_ref (se->expr);

      /* Dereference pointer variables.
         TODO - sort out attr.function and attr.result
         for characters. */
      if ((sym->attr.pointer || sym->attr.allocatable)
	  && (sym->attr.dummy
	      || sym->attr.result
	      || sym->attr.function
	      || !sym->attr.dimension)
          && sym->ts.type != BT_CHARACTER)
	se->expr = gfc_build_indirect_ref (se->expr);

I believe that some of the other character PRs will be fixed by going
through the logic of the second block, as the TODO indicates. 

Paul T