Re: Divide by zero detection

On Sep 20, 6:27 am, Hans-Peter Diettrich <DrDiettri...@aol.com> wrote:
> news.club-internet.fr wrote:
> > I am looking for tools able to detect inside the source code,
> > potential divide by zero errors. ...
>
> 1) find all divisions
> 2) exclude those with a constant non-zero right hand side
> Optionally:
>
> 3) exclude those with a preceding check for a non-zero RHS
> 4) exclude those with a handled zero-divide error ...


> [You can do better than that, e.g.
>
> a = 42;;
> ...
> b = c/a;
>
> No zero divide there, either. Doing the analysis properly is doubtless
> undecidable, but with aggressive dataflow analysis you should be able to
> rule out a lot of known non-zero divisors. -John]


Yes! It's an interesting coincidence that there have been some
questions about abstract interpretation recently. This is the kind of
problem where it could apply. E.g. with a simple lattice over {-,
0,+}, you could pretty easily decide that


i = 20; while ... { a = 1.0 / i; ; ... ; i = i + 1; }


poses no danger and also that


i = -20; while ... { a = 1.0 / i; ... ; i = i + 1; }


does!

Return to the comp.compilers page.
Search the comp.compilers archives again.