[Committed] PR fortran/44491 -- Select non-NULL locus

Message ID 20180612180008.GA73720@troutmask.apl.washington.edu
State New
Headers show
  • [Committed] PR fortran/44491 -- Select non-NULL locus
Related show

Commit Message

Steve Kargl June 12, 2018, 6 p.m.
The attached patch has been committed to trunk after
regression testing.  Two loci are available for the
error messages.  If the first locus is NULL, then it
uses the second one to report the error.

In the audit trail of the PR manu@ suggested removing
the assert() that caused the ICE.  His patch tested
for a NULL locus, and simply reported UNKNOWN LOCATION
in the error message.  I find the ICE to be an aid
in locating where gfortran goes sideways, so I've left
the assert() in place.



Index: gcc/fortran/expr.c
--- gcc/fortran/expr.c	(revision 261376)
+++ gcc/fortran/expr.c	(working copy)
@@ -3414,6 +3414,8 @@  gfc_check_assign (gfc_expr *lvalue, gfc_expr *rvalue, 
   /* Only DATA Statements come here.  */
   if (!conform)
+      locus *where;
       /* Numeric can be converted to any other numeric. And Hollerith can be
 	 converted to any other type.  */
       if ((gfc_numeric_ts (&lvalue->ts) && gfc_numeric_ts (&rvalue->ts))
@@ -3423,8 +3425,9 @@  gfc_check_assign (gfc_expr *lvalue, gfc_expr *rvalue, 
       if (lvalue->ts.type == BT_LOGICAL && rvalue->ts.type == BT_LOGICAL)
 	return true;
+      where = lvalue->where.lb ? &lvalue->where : &rvalue->where;
       gfc_error ("Incompatible types in DATA statement at %L; attempted "
-		 "conversion of %s to %s", &lvalue->where,
+		 "conversion of %s to %s", where,
 		 gfc_typename (&rvalue->ts), gfc_typename (&lvalue->ts));
       return false;
Index: gcc/testsuite/gfortran.dg/pr44491.f90
--- gcc/testsuite/gfortran.dg/pr44491.f90	(nonexistent)
+++ gcc/testsuite/gfortran.dg/pr44491.f90	(working copy)
@@ -0,0 +1,4 @@ 
+! { dg-do compile }
+! PR fortran/44491
+      character*2 escape /z'1B'/  ! { dg-error "Incompatible types in DATA" }
+      end