GMP and Rational's Purify

[Kevin Ryde]

"Krzysztof Kozminski" <Krzysztof.A.Kozminski@nsc.com> writes:
>
> Has anyone figured out whether it is possible to build GMP 4.1 or later under
> Solaris so that it does not get broken once it is instrumented with Purify?

Kent had a go a while back I think.  It turned up some uninitialized
memory accesses, which were mostly bogus if I remember (limbs fetched
unconditionally then ignored according to what the size field says).

> It seems that the instrumented code overflows somewhere.

Something evil in mpz_sizeinbase presumably.  Let us know if you can
find what's wrong.

For checking we claim to work with valgrind and gcc checker, both of
which are free software.  They don't get a run as often as they could
though :-).