How returning a string when using jitclass from numba?

Question

I'm trying to return a string when I call a function in a class that uses jitclass but I get an error:

numba.errors.InternalError: Failed at nopython (nopython mode backend)
cannot convert native const('Something') to Python object
[1] During: resolving callee type: BoundFunction((<class 'numba.types.misc.ClassInstanceType'>, 'get_Names') for     instance.jitclass.myclass#3f2d488<A:float64,B:float64>)
[2] During: typing of call at <string> (3)

I'm using this code to test the function:

from numba import jitclass
from numba import boolean, int32, float64,uint8

spec = [('A' ,float64),
        ('B' ,float64)]

@jitclass(spec)
class myclass:

    def __init__(self,):
        self.A = 3.25
        self.B = 22.5

    def get_Names(self):
        return "Something"



mC = myclass()
print(mC.get_Names())

Does somebody know How I could return a string?

Answer 1

You can sort of work around this by using an array of bytes to represent the string, as shown below.

That said, I think this is pretty ugly/hard to maintain. Assuming it fits the problem, I think you are better off using a plain python class with jit functions for speedups as necessary, or dropping into something like cython which has richer support for extension types.

from numba import jitclass, float64

SOMETHING = np.frombuffer(b"Something", dtype='uint8')

spec = [('A' ,float64),
        ('B' ,float64)]

def get_jitclass_str(val):
    return bytes(val).decode('utf-8')

@jitclass(spec)
class myclass:

    def __init__(self,):
        self.A = 3.25
        self.B = 22.5

    def get_Names(self):
        return SOMETHING

Usage

In [16]: mc = myclass()

In [17]: get_jitclass_str(mc.get_Names())
Out[17]: 'Something'