Special operator's methods (like __add__) don't work in jitclass

[polkovnikov]

Reporting a bug

Seems that jitclass works incorrectly with all special methods like __add__, __mul__, etc. Minimal reproducible example:

import numba

@numba.experimental.jitclass([('x', numba.int64)])
class C:
    def __init__(self, x):
        self.x = x
    def __add__(self, b):
        return self.x + b.x

@numba.njit
def f():
    #C(3) + C(5)
    return C(13).__add__(C(17))
print(f())

if 1:
    @numba.jit(forceobj = True)
    def g():
        #return C(3) + C(5)
        return C(13).__add__(C(17))
    print(g())

print(C(7).__add__(C(11)))
print(C(3) + C(5))

Call of njitted-function f() works and prints correct answer. Call of jitted-function g() (in object mode) with same call of __add__ prints exception AttributeError: 'C' object has no attribute '__add__'. Same exception is printed for the following not-jitted code C(7).__add__(C(11)). Seems that this __add__ method is somehow removed from class for jit/no-jit case.

And finally code C(3) + C(5) works nowhere, not in njit, not in jit, not in unjitted code. It prints error TypeError: unsupported operand type(s) for +: 'C' and 'C' (for jit/no-jit) and long error dump for njit. It means python doesn't detect __add__ method as special method that implements + operator.

Regular methods of jitclass (without special __ ... __ underscores) work correctly in njit/jit/no-jit cases.

Probably it is not a bug but unimplemented feature of yet experimental jitclass.

[gmarkall]

I believe this is implemented in #5877.
Similar requests for other methods: #1606, #5690.