The problem with Numba is that it only really works in small, limited usecases. Good for e.g. Numpy arrays, but can it provide these massive speedups with custom classes? Arrays of strings? Sets? Probably not, since both Julia and Numba relies on inferring the data types of all objects, and on them being stored efficiently in memory. The Julia type system allows easy type inference, and all custom structs are stored efficiently in memory. In Python, neither of those are true.

hijacking this discussion to raise the banner of seamless interoperability between numba (quick numerical code) and pypy (quick algorithmic code + business logic).

My dream python is pypy+mypy+numba. Alas, pypy doesn't play nice with the others "(