silent integer overflow

[eliosh]

The following example is taken from Nick MacLaren's slides (Cambridge Univ.)

[plain]program intoverflow
  integer, parameter :: n=1800
  integer :: n3
 
  ! real :: my_data(n*n*n) <---- Ooops here

  n3 = n*n*n
  print *, 'n^3 =', n3
  
end program intoverflow[/plain]

When compiled by ifort (v11.1.046) the overflow is not signaled and n3 becomes equal to 1537032704 instead of 5832000000.
gfortran, on the other hand, does recongize the problem and emits an error message during compilation.

[Ron_Green]

Quoting - eliosh

The following example is taken from Nick MacLaren's slides (Cambridge Univ.)

[plain]program intoverflow
  integer, parameter :: n=1800
  integer :: n3
 
  ! real :: my_data(n*n*n) <---- Ooops here

  n3 = n*n*n
  print *, 'n^3 =', n3
  
end program intoverflow[/plain]

When compiled by ifort (v11.1.046) the overflow is not signaled and n3 becomes equal to 1537032704 instead of 5832000000.
gfortran, on the other hand, does recongize the problem and emits an error message during compilation.

This question comes up from time to time. There is a feature request to enable a compile time check for just such a situation. However, there is no requirement in the Fortran standard to emit an error for this scenario. This feature request is currently on our list but there is no timeline for it's implementation.