Re: [patch, fortran] PR24227 initialized symbol in COMMON: Missing checks

[Jerry DeLisle <jvdelisle at verizon dot net>]

Jerry DeLisle wrote:
> Tobias Burnus wrote:
> > Jerry DeLisle wrote:
> > > This patch adds back a simplified error message and checks to see if
> > > the symbol is in BLOCK DATA or not.  Allowed if -std=gnu.
> > > Regression tested on x86-64.
> > > OK for trunk?
> >
> > +      if (gfc_add_in_common (&sym->attr, sym->name, NULL) == FAILURE) 
> > +        goto cleanup;
> > [...]
> > +
> >        if (gfc_add_in_common (&sym->attr, sym->name, NULL) == FAILURE)
> >          goto cleanup;
>
> The double statement above was there before.  I thought it was odd also, 
> but put it back the way it was before.  I will take it out and test.
>
> > Adding in_common twice looks wrong.
> >
> >
> > Additionally, the following variant using DATA compiles without giving
> > an error:
> >
> > program main
> >  implicit none
> >  integer, parameter:: nmin = 2
> >  character(len=3) :: emname(nmin)
> >  data emname/'bar','baz'/
> >  common/dd/emname ! { dg-error "can only be COMMON in BLOCK DATA" }
> > end program main
> I will have a look later this evening.

Attached is a revised patch.  Regression tested on x86-64.  I also compiled all 
of the Polyhedron test cases, NIST, and LAPACK.

OK for trunk?

Jerry
Index: match.c
===================================================================
--- match.c	(revision 130417)
+++ match.c	(working copy)
@@ -2783,6 +2783,16 @@ gfc_match_common (void)
 	      goto cleanup;
 	    }
 
+	  if (((sym->value != NULL && sym->value->expr_type != EXPR_NULL)
+	       || sym->attr.data) && gfc_current_state () != COMP_BLOCK_DATA)
+	    {
+	      if (gfc_notify_std (GFC_STD_GNU, "Initialized symbol '%s' at %C "
+					       "can only be COMMON in "
+					       "BLOCK DATA", sym->name)
+		  == FAILURE)
+		goto cleanup;
+	    }
+
 	  if (gfc_add_in_common (&sym->attr, sym->name, NULL) == FAILURE)
 	    goto cleanup;