[patch, fortran] PR20373 - intrinsic symbols can be given the wrong type (take 3)

[Daniel Franke]

On Wednesday 20 June 2007 22:52:32 Daniel Franke wrote:
> gcc/fortran:
> 2007-06-20  Daniel Franke  <franke.daniel@gmail.com>
>
>        PR fortran/20373
>        * intrinsic.c (add_functions): Additional function types.
+        (gfc_convert_type_warn): Remove intrinsic status of
+	 __convert_%s%d_%s%d functions.
>        * resolve.c (resolve_symbol): Warn if intrinsic functions are
>        explicitely typed.
>
> gcc/testsuite:
> 2007-06-20  Daniel Franke  <franke.daniel@gmail.com>
>
>        PR fortran/20373
>        * gfortran.dg/intrinsic.f90: New test.


Yet another update of this patch:

__convert_* functions are internally used and generated to implicitely convert 
between types where possible. The symbols of these functions have 
the "intrinsic" bit set, which results in something unintended:

logical, parameter :: maybe = -1
                               1
Warning: Type specified for intrinsic function '__convert_i4_l4' at (1) is 
ignored

Applying the change below avoids the warning (gfc_convert_type_warn):

Index: intrinsic.c
===================================================================
--- intrinsic.c (revision 125720)
+++ intrinsic.c (working copy)
@@ -3562,7 +3562,6 @@
   new->symtree->n.sym->ts = *ts;
   new->symtree->n.sym->attr.flavor = FL_PROCEDURE;
   new->symtree->n.sym->attr.function = 1;
-  new->symtree->n.sym->attr.intrinsic = 1;
   new->symtree->n.sym->attr.elemental = 1;
   new->symtree->n.sym->attr.pure = 1;
   new->symtree->n.sym->attr.referenced = 1;

Is there any explicit reason why these conversion functions have to have the 
intrinsic bit set? There are no apparent problems during regression testing?!

Thanks
	Daniel