Roberto Bagnara bagnara@cs.unipr.it writes:

I have been bitten by the following problem on a bi-arch system. By accident, I was compiling with the 32-bit version of <gmp.h> and linking with the 64-bit version of libgmp.*. The failure I got was not trivial at all (only one test out of thousands was failing), and debugging took a long time.

Do you know a defense against this kind of errors? More specifically:

- for version mismatches, the obvious thing to do is to compare the result of gmp_version() with the macros __GNU_MP_VERSION, __GNU_MP_VERSION_MINOR, __GNU_MP_VERSION_PATCHLEVEL;

- I am wondering if something similar can be done for ABI mismatches: for example <gmp.h> defines __GMP_BITS_PER_MP_LIMB: does libgmp.* "leak" this information so that a comparison can be made?

There is a documented variable mp_bits_per_limb.