Python High Performance - Second Edition

Limitations in Numba

There are some instances where Numba cannot properly infer the variable types and will refuse to compile. In the following example, we define a function that takes a nested list of integers and returns the sum of the element in every sublist. In this case, Numba will raise ValueError and refuse to compile:

    a = [[0, 1, 2],          [3, 4],          [5, 6, 7, 8]]    @nb.jit    def sum_sublists(a):        result = []        for sublist in a:            result.append(sum(sublist))        return result    sum_sublists(a)    # ValueError: cannot compute fingerprint of empty list

The problem with this code is that Numba is not able to determine the type of the list and fails. A way to fix this problem is to help the compiler determine the right type by initializing the list with a ...