[SciPy-dev] Illegal instruction in ndimage test

[David Cournapeau]

Matthieu Brucher wrote:
> This is not very different from the Numpy API interface. You have an 
> array of void* that contains your functions, and then you define 
> macros that use this array with a cast to get the real function. Here, 
> it seems that there is no array, but the result is identical (but it 
> shouldn't redefine an already defined function :|)
Yes, sure, arrays of function pointers are nothing unusual. But for some 
time, I was really wondering how could a preprocessor symbol refer to 
itself (it did not, it was just a function). I still do not understand 
how a function pointer can generate invalid opcode (the error message 
seems to appear only with gcc 4.2, not with 4.1, which is still really 
common), though.

cheers,

David