Re: Question on PR26890, possible regression.

[Jakub Jelinek <jakub at redhat dot com>]

On Fri, Apr 07, 2006 at 02:44:10PM -0700, Jerry DeLisle wrote:
> Thanks for the explanation. I will commit the fix to put pad back where it 
> belongs shortly under the obvious rule.  I will also had a note explaining 
> this connection back to the frontend.  I will check the size of the pad 
> relative to the size of the other union to see where we stand.

What you committed is not what I was looking for.
u.pad really should be equal to what trans-io.c has compiled in and u.p
should be smaller or equal to u.pad.
Ok for HEAD (and similar change for gcc-4_1-branch)?

2006-04-10  Jakub Jelinek  <jakub@redhat.com>

	* io/io.h (st_parameter_dt): Revert 2005-12-10 change to
	u.pad, fix comment.
	(check_st_parameter_dt): New compile time assert.

--- libgfortran/io/io.h.jj	2006-04-10 14:00:31.000000000 +0200
+++ libgfortran/io/io.h	2006-04-10 14:11:24.000000000 +0200
@@ -436,13 +436,19 @@ typedef struct st_parameter_dt
 	  char value[32];
 	  gfc_offset size_used;
 	} p;
-      /* This pad size must be greater than or equal to the pad_size declared in
-	 trans-io.c (gfc_build_io_library_fndecls)  */
-      char pad[16 * sizeof (char *) + 34 * sizeof (int)];
+      /* This pad size must be equal to the pad_size declared in
+	 trans-io.c (gfc_build_io_library_fndecls).  The above structure
+	 must be smaller or equal to this array.  */
+      char pad[16 * sizeof (char *) + 32 * sizeof (int)];
     } u;
 }
 st_parameter_dt;
 
+/* Ensure st_parameter_dt's u.pad is bigger or equal to u.p.  */
+extern char check_st_parameter_dt[sizeof (((st_parameter_dt *) 0)->u.pad)
+				  >= sizeof (((st_parameter_dt *) 0)->u.p)
+				  ? 1 : -1];
+
 #undef CHARACTER1
 #undef CHARACTER2
 


	Jakub