Re: [Patch, Fortran] Add caf_runtime_error to libgfortran/caf/single.c

[Tobias Burnus <burnus at net-b dot de>]

On 07/14/2011 12:14 PM, Daniel Carrera wrote:
> On 07/14/2011 12:04 PM, Tobias Burnus wrote:
> > I was wondering - based on the discussion - whether one should remove
> > the "int error" argument from caf_runtime_error and simply use "exit
> > (EXIT_FAILURE)" for all exit() calls in mpi.c/single.c, cf.
> > http://gcc.gnu.org/ml/fortran/2011-07/msg00140.html
> > But one can also do so as follow up patch.
>
> You are the boss. The message I got from Nick's post is that it 
> doesn't matter much and that you could even get surprising behaviour 
> because EXIT_SUCCESS is not required to be zero and EXIT_FAILURE is 
> not required to be non-zero. But maybe I missed the point. So it's up 
> to you.

Well, for "exit(3)" one knows that one will get a surprising in Windows: 
An abort. While EXIT_FAILURE [but also exit(1)] have a very high change 
to be working (nearly) everywhere. At the end, as long as the operating 
system does not react in a completely odd way, the exit status does not 
matter - it's the problem of the caller (e.g. some script) to make use 
of the result. If EXIT_FAILURE is 0 and the script assumes that it is a 
success, it's either a fault of the script writer, or of the one putting 
0 for EXIT_FAILURE into the header, or a fault of the OS designer.

Thus, I think the chance that one avoids surprises is higher with 
exit(EXIT_FAILURE) and one avoids having some random number (for the 
exit status) in the code. Users usually do not check the exact exit 
status value but only its 0- vs non-0-ness.

Hence, I think exit(EXIT_FAILURE) is better than the current version.

Tobias