Recently I've resumed Fortran and probably there's something I'm missing but this behaviour looks very odd
When I run the following code (compiled with ifort), that just declares an array and sets one of his elements
PROGRAM sol_kernel2
IMPLICIT NONE
INTEGER, PARAMETER :: jpi=5,jpj=5
REAL, DIMENSION(jpi,jpj) :: sshn
PRINT *,jpi,jpj
sshn(10,10) = 150.0
PRINT *,sshn(10,10)
END PROGRAM sol_kernel2
I expect to get an ERROR Statement, such as SEGMENTATION FAULT, since I'm trying to set the sshn variable using indexes supposed out of memory. I get an error free output like this instead
5 5
150.0000
If I try to set in the code
sshn(100,100) = 150.0
I get this
5 5
0.0000000E+00
The compiler doesn't complain anyway, but this time 150.0 failed to be set. Have you got any hints about this? I can't figure out myself what I'm doing wrong. Thanks
The compiler is not required to diagnose your out of bound indexing. When you do something like this, what happens is unspecified by the language standard. It is also known as undefined behaviour.
You can't expect anything specific to happen.
If you wish your error to be diagnosed, compile with error checking:
ifort -g -traceback -warn -check yourcode.f90
What actually happened is that you wrote something to some random place in memory. Who knows what would happened in a more complex code. It can then crash at random occasions or give wrong results.
The compiled code does not normally check for bounds and out-of bounds access. You can expect the operating system killing the process if it tries to access memory outside of what is allocated to the process, but otherwise it will let it go. It is difficult to tell when this happens.
In your case you set and accessed memory outside of the array bounds - it is actually a memory address so possible to instruct your code to do that - but it must have been within the memory allocated to the process, and as you found out, the result is unpredictable and undefined.
You can enforce strict bound checking with ifort using the
-check bounds
compiler option. This will actually catch this error compile time for the ifort version I used. Gfortran catches the error even without any additional compiler flags.