This is the mail archive of the gcc-bugs@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[Bug fortran/13575] [gfortran] ICE for wrong code


------- Additional Comments From tobi at gcc dot gnu dot org  2004-05-16 18:59 -------
This bug has a very simple fix:
--- resolve.c	14 May 2004 13:00:04 -0000	1.3
+++ resolve.c	15 May 2004 22:33:10 -0000
@@ -3711,7 +3711,7 @@ resolve_symbol (gfc_symbol * sym)
   if (sym->ts.type == BT_UNKNOWN)
     {
       if (sym->attr.flavor == FL_VARIABLE || sym->attr.flavor == FL_PARAMETER)
-	gfc_set_default_type (sym, 0, NULL);
+	gfc_set_default_type (sym, 1, NULL);
 
       if (sym->attr.flavor == FL_PROCEDURE && sym->attr.function)
 	{
This is obviously correct, because if at this point a variable can't be given a
type it's an error.

Unfortunately, the parser's behavior which makes it create symbols where it
shouldn't, breaks this: with this "fix" COMMON /x/ a will give an error for x,
because x has been added to the namespace of our function, but can't be given a
name at this place. Similar for SELECTED_REAL_KIND(p=5), here it will complain
about the "symbol" p. Because of this it also related to PR 13372.

I believe there are two options:
1. fix the frontend and apply my patch. The Right Thing, but definitely
involved. I looked at how g95 deals with this, and it looks like a rewrite of
symbol handling.
2. traverse the namespaces and eliminate all symbols which don't have a type
after this place. I believe this would work, but it seems like very hackish
patchery. 

If I can convince myself that the second option wouldn't break anything, I will
prepare a patch to this effect.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13575


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]