Remove @overload_glue for NumPy allocators. - #7999

Move this to numba.np.unsafe.ndarray?

Nice! better error message

Why does it need to set parent to NULL? The only place this is used is in ary.imag for real domain array. The .parent is always NULL in that case.

I vaguely recall adding this with view of trying provide a "general" way to set the readonly bit, was wondering (here: #7999 (comment)) whether this ought to be a utility function? Am fine to remove it or move it into numba.np.unsafe.ndarray.

Let's remove it from _force_readonly. I don't think this is the way to do it. The boxer is already setting the writeable flag according to the array readonly/mutable property:

TBH, I'd rather that this didn't exist at all, but this sort of thing fails because the return type of the impl doesn't match the declared signature:

from numba import njit
from numba.core.extending import overload
import numpy as np

def foo(x):
    pass

@overload(foo)
def ol_foo(x):
    def impl(x):
        return x
    return x.copy(readonly=True)(x), impl

@njit
def call_foo(x):
    return foo(x)

tmp = np.arange(5.)
print(call_foo(tmp))

so the intrinsic is needed to fix it. I'll go fix the patch WRT the above.