Re: [patch, fortran] PR32124 - Don't give a run-time error if stat= has been specified for ALLOCATE

[Brooks Moses <brooks dot moses at codesourcery dot com>]

At 10:35 AM 5/30/2007, Tobias Burnus wrote:
> Brooks Moses wrote:
> >> Any ideas? (I knew why I did not want to have a test case when I
> >> originally submitted the patch.)
> > It would be nice if the runtime library could error out on when the
> > size overflows, without needing to rely on the result ending up negative.
> I think there are two ways this can happen: (a) at compile time - and
> this is what happens here - and (b) at run time. At compile time, we
> might be able to create a check whether we are overflowing (this is was
> happens here). [I don't know whether there is a simple way of doing so,
> maybe a test if negative could be build in? Or is the size argument not
> easily accessible in the front end? It is in the dumped original tree.
> GMP does not seem to supported.]
>
> For run time, I think there is no easy way. I don't have seen may
> systems which offer trapping (SIGFPE) of integer overflows. (There are
> some which do, as googling shows.)

I'll have to think about this.  There are certainly ways to check and see 
if a given multiply is going to overflow; yes, they take a couple more 
operations, but a memory allocation operation is already slow enough that 
the extra time should be negligible.

For instance, something along the lines of
  if (shape[n] > INT_MAX / stride[n])
should check whether shape[n] * stride[n] will overflow or not.

- Brooks