Trying to catch use of uninitialized variables (revisited)

[albert]

I'm running some tests on the "Static Verifier" with the Intel Fortran 10.1 compiler (ifort (IFORT) 10.1 20080112).
When using the following program:
PROGRAM tst
implicit none
integer c
write(*,*) 'Anything'
write(*,*) c
end
subroutine tt(c)
integer c
write(*,*) c
end
it reports (as expected):
uninit.f90(6): error #12143: [SV] "C" is uninitialized
uninit.f90(8): warning #12222: [SV] routine "tt" is never
called from the MAIN program unit
command line:
ifort -diag-enable sv3 -free uninit.f90 -o uninit


When altering the program to:
PROGRAM tst
implicit none
integer, allocatable :: a(:)
integer c
write(*,*) 'Anything'
allocate(a(10))
write(*,*) c
write(*,*) a(5)
end
subroutine tt(c)
integer c
write(*,*) c
end

Again the correct result, although not catching, as expected, the uninitialized a(5):
uninit.f90(7): error #12143: [SV] "C" is uninitialized
uninit.f90(10): warning #12222: [SV] routine "tt" is never
called from the MAIN program unit


Changing the command line to:
ifort -diag-enable sv3 -g -traceback -free -check all uninit.f90 -o uninit

I get the messages:
uninit.f90(7): error #12143: [SV] "C" is uninitialized
uninit.f90(8): warning #12137: [SV] definition of
procedure "emit_diagnostic" is not found

This last message I don't understand, I would have expected a message about routine "tt"

Another question is how an I check the routines inside my own library (i.e. objects I compiled with -diag-enable sv3 and placed in a library) ?

Albert

[Steven_L_Intel1]

The message about emit_diagnostic is a bug. There is no way to use SV on a library - you would have to add the library sources to a project and build with SV.

[albert]

Thanks for the answer.

- Is there already a bug filed for the "emit_diagnostic" ?
- pitty a library cannot be used, is this a principle problem ? I will use all objects on the command line.

Albert

[Steven_L_Intel1]

I believe that the emit_diagnostic issue has been reported. Theoretically, SV could work with libraries the same way that IPO does, but the current implementation does not support it. We are working on a significantly revised implementation of SV for the future and I will ask that libraries be considered.