On Sun, 2014-03-30 at 08:14 -0700, Joeish W wrote:
I was just wondering why cffi doesn't do automatic conversions.
One reason may be that the FFI of the underlying Lisps (some? all?) don't do automatic conversions. They signal an error instead. If some library always wraps values with additional calls like (COERCE ,X 'FLOAT), it'll cost performance in general. Advanced type-inferencing compilers like cmucl/sbcl might be able to eliminate that extra call, depending on context. But there's no reason to let CFFI impose an extra performance burden on other implementations.
Of course, that implies that mismatching int and float in "portable CFFI code" is unspecified, because the behaviour of the underlying FFI is -- presumably -- not the same across all Lisps. Some definitely signal an error, perhaps some will silently coerce.
Regards, Jörg Höhle