[Cython] Exporting inline functions from a pyx.

[Robert Bradshaw]

On Sat, May 30, 2015 at 10:46 PM, Stefan Behnel <stefan_ml at behnel.de> wrote:
> Jesus-Omar Ocegueda-Gonzalez schrieb am 27.05.2015 um 04:17:
>> we recently faced an issue exporting
>> inline functions defined in a .pyx. According to this:
>> http://docs.cython.org/src/tutorial/pxd_files.html
>> the full inline definition (not just the declaration) should be in the pxd.
>> However, if we put the definition in the .pyx and just its declaration
>> header in the .pxd,  Cython declares a pointer to the inline function
>> similar to:
>>
>> static CYTHON_INLINE double (*__function_name)(...); /*proto*/
>>
>> but this causes a compilation error in some platforms (but successfully
>> compiles in others) because variables cannot be declare as inline.
>
> Thanks for the report, I agree that this looks like a bug. I'm not entirely
> sure what the best fix is. We could make it an error if an external
> function declaration in a .pxd file is declared "inline", or we could just
> ignore the modifier for external functions on the import side.
>
> The first case would be more correct, but we'd end up with different
> declarations in the .pyx and .pxd files. That might be difficult to explain
> to users that run into this problem, as they would first have to understand
> that they can still declare their function "inline" in the .pyx file, just
> not in the .pxd. Just copying the function header over in order to export
> it will make your code fail to compile.
>
> But the second case might also not be without surprises ("why doesn't my
> 'inline' declaration work here?"), and it leaks implementation details into
> the .pxd. I'm thus leaning towards the first. I guess it just depends on
> coming up with a clear error message. (Well, after detecting this case in
> the first place...)

The first also has the disadvantage that one can't create an inline
function that is exported as an imported function elsewhere...