When defining an OPERATOR(...) for intrinsic operators or ASSIGNMENT(=), the
interface of the provided procedure must match certain characteristics (like
one or two dummy arguments, being a FUNCTION/SUBROUTINE, respectively, and so
on).
For a wrong interface, gfortran correctly prints an error message. The source
location this is associated to is however quite inconsistent and misleading.
For instance with the example:
MODULE m
IMPLICIT NONE
TYPE t
INTEGER :: num
END TYPE t
INTERFACE OPERATOR(+)
MODULE PROCEDURE myplus
END INTERFACE OPERATOR(+)
CONTAINS
INTEGER FUNCTION myplus (a, b, c)
TYPE(t), INTENT(IN) :: a, b, c
myplus = a%num + b%num + 42
END FUNCTION myplus
END MODULE m
I get this error:
[/tmp]# gfortran-dev test.f03
test.f03:14.25:
INTEGER FUNCTION myplus (a, b, c)
1
Error: Operator interface at (1) has the wrong number of arguments
It is of course right, but I would expect at least a reference to where myplus
is actually defined as OPERATOR(+). On the other hand, for
MODULE m
IMPLICIT NONE
TYPE t
INTEGER :: num
END TYPE t
INTERFACE OPERATOR(+)
MODULE PROCEDURE myplus
END INTERFACE OPERATOR(+)
CONTAINS
SUBROUTINE myplus (a, b)
TYPE(t), INTENT(IN) :: a, b
END SUBROUTINE myplus
END MODULE m
I get:
[/tmp]# gfortran-dev test.f03
test.f03:9.27:
MODULE PROCEDURE myplus
1
Error: Intrinsic operator interface at (1) must be a FUNCTION
Which is quite better, in my opinion. I suggest we change the error location
in every case to the MODULE PROCEDURE line, which is more reasonable IMHO as
well as consistent. Alternatively, we could also reference both locations (but
should also do this in both cases to stay consistent).
--
Summary: Inconsistent error locations for wrong interfaces with
overloaded operators
Product: gcc
Version: unknown
Status: UNCONFIRMED
Keywords: diagnostic
Severity: normal
Priority: P3
Component: fortran
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: domob at gcc dot gnu dot org
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41023
