External procedures with the same name as intrinsic procedures

[Arjen_Markus]

Hello,

we had a rather awkward problem with a routine that we called IIDIM. This is the same name as one
of the specific routines/functions forthe DIMintrinsic function. It turned out that the intrinsic function was
called, not our function - completely according to the Fortran standard of course.

My question is: is there a compiler option in Intel Fortran that flags routine names identical to the names
of intrinsic procedures? I could not find any in the help information.

Another thing: does the set of intrinsic procedures depend on the use of modules like IFPORT? From the
documentation I do get that impression, but I would like to make sure.

Regards.

Arjen

[jimdempseyatthecove]

Arjen,

Do you have (and use) and explicit interface to your routine? If not, try adding one.

Note, IDEM works as a generic function for integer(1), integer(2), integer(4) and integer(8) so
if your IIDEM takes two integer arguments .and. you are relying on the compiler for promotion/demotion then there may be some ambiguity which results in the incorrect (non-intended) function being called. In this situation you may need to make a generic interface to your IIDEM using permutations of acceptible argument types.

Jim Dempsey

[Arjen_Markus]

Hi Jim,

this problem turned up more or less out of the blue, as our function has beenaround
for a long while (at least as far as I know). The followingcode gives two warnings with gfortran -Wall,
but ifort -warn:all does not complain:

! intrinsic.f90 --

! Define a function with the same name as an intrinsic

! Any messages?

!

real function sin( x )

real :: x

sin = x + 1.0

end function sin

real function dim( x, y )

real :: x, y

dim = x + y

end function dim

I actually do see the generated interfaces appear. Hm, I should check whether
my version is indeed used in this case.

Regards,

Arjen

[Arjen_Markus]

If I add a small test program to this code:

program test_intrinsic
real :: x, y,z
x = 1.0
y= 2.0
z = sin(x)
write(*,*) 'Sin(x) =', z
z = dim(x,y)
write(*,*) 'Dim(x,y) =', z
end program test_intrinsic

I get the values as computed by the intrinsic functions, notbythe above ones:

Sin(x) = 0.8414710
Dim(x,y) = 0.0000000E+00

I do get the generated modules:

!COMPILER-GENERATED INTERFACE MODULE: Fri May 13 14:30:45 2011

MODULE SIN__genmod

INTERFACE

FUNCTION SIN(X) RESULT(SIN_0)

REAL(KIND=4) :: X

REAL(KIND=4) :: SIN_0

END FUNCTION SIN

END INTERFACE

END MODULE SIN__genmod

If I put the two functions in a module, and use that module, then:

Sin(x) = 2.000000
Dim(x,y) = 3.000000

so in that case my versions are indeed called.

Regards,

Arjen

[Steven_L_Intel1]

There is no option to warn if you have a procedure with the same name as an intrinsic, as that is perfectly standard Fortran. The language provides a way to guard against conflicts - either an explicit interface as described above, or the EXTERNAL statement.

[jimdempseyatthecove]

Put your interface, not the compiler generated interface, in a module you supply, then USE that module name if the routines that you wish to use your IIDIM (and/or other conflicting names). Do not rely on auto-magicly generated interfaces for this purpose. You have seenthat anbiguous implicit/specified USE interfaces results in not getting what you want. IOW explicitly tell the compiler what you want.

Jim Dempsey

[Arjen_Markus]

Steve, Jim,

I understand the issues - this occurred in a large program that is slowly being converted in Fortran 90.
I keep promoting modules as a means to avoid this and other nastiness, but it takes a lot of time to
get there.

Thanks for the replies.

Regards,

Arjen