Le 06/07/2013 22:40, serge Guelton a écrit :
Hi Pythraners, Numba and parakeet are using a @jit decorator on functions to trigger jit compilation of python function to modules. I first thought it would be a good idea to do the same in Pythran. After seeing https://www.youtube.com/watch?v=OrpPDkZef5I at scipy and following the cython tutorial, it appears to me that: * users are happy with the way cython waork, and pythran is very similar to it * jit typing will give a bad user experience as compilation is really slow However, distutils integration, as done in cython, seems useful any thought? @Joker: sorry, I hope you did not go to far in the dev...
IMO distutils integration in cython is really bad.Even if you just want to do a test, you have to write a specific distutils files and after some test, you have a lot of files everywhere and you don't know what is matching with which other file. Of course you can say that it is the user who should sort his files but why should we change the way pythran works to something worse?
May be I didn't understand why it is good...About JIT, we can provide both interface. The one we have now and a jit too, isn't it? I think that something more important than decorator @jit is to compile only decorated functions in a file. For example, if we have a class and many function in a file. It can be good to compile only annotated function without crashing because of (not supported) class. In this case, annotated function can be jit if they are not already compiled? (jit is only a "plus" in this case, not the main feature)