[patch, fortran] PR33609 ICE on arithmetic overflow

Jerry DeLisle jvdelisle@verizon.net
Sat Oct 6 23:40:00 GMT 2007


Hi,

The ICE is caused by a NULL pointer returned by arith.c (gfc_real2real) when 
error condition is detected:

   else if (rc != ARITH_OK)
     {
       arith_error (rc, &src->ts, &result->ts, &src->where);
       gfc_free_expr (result);
       return NULL;
     }

Execution then goes back to the caller gfc_simplify_real which then does:

   return range_check (result, "REAL");

The segfault occurs in range_check.

I will commit the following patch as obvious. Test case attached.

Regression tested on x86-64-Gnu-linux.

Jerry

Index: simplify.c
===================================================================
--- simplify.c  (revision 129029)
+++ simplify.c  (working copy)
@@ -70,6 +70,9 @@ gfc_expr gfc_bad_expr;
  static gfc_expr *
  range_check (gfc_expr *result, const char *name)
  {
+  if (result == NULL)
+    return &gfc_bad_expr;
+
    switch (gfc_range_check (result))
      {
        case ARITH_OK:

2007-10-06  Jerry DeLisle  <jvdelisle@gcc.gnu.org>

	PR fortran/33609
	* simplify.c (range_check): Return gfc_bad_expr if incoming expression
	is NULL.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: error_recovery_4.f90
Type: text/x-fortran
Size: 160 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/fortran/attachments/20071006/e1b17beb/attachment.bin>


More information about the Fortran mailing list