0.57.0rc1 NumbaDeprecationWarning: @guvectorize even with nopython keyword argument

[oyamad]

The following code (note the nopython=True argument)

from numba import guvectorize, int64

@guvectorize([(int64[:], int64, int64[:])], '(n),()->(n)', nopython=True)
def g(x, y, res):
    for i in range(x.shape[0]):
        res[i] = x[i] + y

raises NumbaDeprecationWarning:

/Applications/anaconda3/envs/py311/lib/python3.11/site-packages/numba/np/ufunc/ufuncbuilder.py:328: NumbaDeprecationWarning: The 'nopython' keyword argument was not supplied to the 'numba.jit' decorator. The implicit default value for this argument is currently False, but it will be changed to True in Numba 0.59.0. See https://numba.readthedocs.io/en/stable/reference/deprecation.html#deprecation-of-object-mode-fall-back-behaviour-when-using-jit for details. self.nb_func = jit(_target='npyufunc', cache=cache)(py_func)

This does not occur with @vectorize:

from numba import vectorize, float64

@vectorize([float64(float64, float64)], nopython=True)
def f(x, y):
    return x + y

>>> numba.__version__
'0.57.0rc1'

[guilhermeleobas]

Hi @oyamad, thanks for the report. I can confirm this "bug" in Numba.

In the snippet of code below, the call to jit is missing **targetoptions expansion.

numba/numba/np/ufunc/ufuncbuilder.py

Line 328 in be990a8

self.nb_func = jit(_target='npyufunc', cache=cache)(py_func)

Numba doesn't report the warning on vectorize because of the **targetoptions expansion, which contains nopython=True

numba/numba/np/ufunc/ufuncbuilder.py

Lines 263 to 265 in be990a8

	self.nb_func = jit(_target='npyufunc',
	cache=cache,
	**targetoptions)(py_func)

Let me know if you'd like to send a patch fixing this issue.

[guilhermeleobas]

Hi @oyamad, really appreciate you putting some effort towards fixing this issue.

We discuss this topic at today's Numba meeting and we have decided that it would be best to suppress the NumbaDeprecationWarning. Altering the call to jit could potentially result in unexpected behavior, especially if different flags are passed to the (gu)vectorize decorator.

[oyamad]

@guilhermeleobas Sure, so, in the description in https://numba.readthedocs.io/en/stable/reference/jit-compilation.html#numba.guvectorize, the keyword arguments nopython, forceobj, and locals for guvectorize (have been and) continue to be kept ignored? I guess that should be documented?