I’m not a specialist in languages and programming but I never wrote that “I don’t understand the idea that there are differences in language semantics that affect compilers”. Please do not deform my words. It’s really not nice.
I don’t understand what is so wrong in terms of performance in the semantic of Python / Numpy used for efficient numerical kernels. And of course, one should not use extensible arrays of references (Python list) for critical parts of numerical kernels, and it is not a problem of a particular language. It is also possible to write very inefficient things in Julia.
I would be very interested to understand why for these simple computations (no user-def types) the Julia language (not the interpreter) is so much better. I think I need to study examples in numerical kernels used in the real life and compare the semantics. A Pythran challenge would definitively be interesting.
I’m going to stop with this discussion now. Thank you for your answers. Bye.