Example:
implicit none
real :: a,b
if(a .nonex. b) stop
end
Expected:
Error: Unknown user operator 'nonex' at (1)
Current gfortran result:
Error: Operands of user operator 'nonex' at (1) are REAL(4)/REAL(4)
NAG f95:
Error: x.f90, line 3: Unknown operator .NONEX.
ifort:
fortcom: Error: x.f90, line 3: Dotted string neither a defined operator nor a
structure component [NONEX]
sunf95:
"x.f90", Line = 3, Column = 8: ERROR: The operator "nonex" has not been
declared as a user defined operator.
I though one could simply change resolve.c's resolve_operator such as e.g.
Index: resolve.c
===================================================================
--- resolve.c (revision 123843)
+++ resolve.c (working copy)
@@ -2242,6 +2242,14 @@
goto bad_op;
case INTRINSIC_USER:
+ if (NULL == gfc_find_symtree (gfc_current_ns->uop_root,
+ e->value.op.uop->name))
+ {
+ sprintf (msg, _("Unknown operator '%s' at %%L"),
+ e->value.op.uop->name);
+ }
+ else
+ {
if (op2 == NULL)
sprintf (msg, _("Operand of user operator '%s' at %%L is %s"),
e->value.op.uop->name, gfc_typename (&op1->ts));
@@ -2249,6 +2257,7 @@
sprintf (msg, _("Operands of user operator '%s' at %%L are %s/%s"),
e->value.op.uop->name, gfc_typename (&op1->ts),
gfc_typename (&op2->ts));
+ }
goto bad_op;
However, this fails as "gfc_get_uop" automatically creates the symbol if it
cannot find it. And putting it into matchexp.c's gfc_match_defined_op_name is
of cause way too early.
--
Summary: Better error message for not-found operator
Product: gcc
Version: 4.3.0
Status: UNCONFIRMED
Keywords: diagnostic
Severity: normal
Priority: P3
Component: fortran
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: burnus at gcc dot gnu dot org
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31580
