Support njitting partial functions

[mroeschke]

Reporting a bug

• I am using the latest released version of Numba (most recent is visible in
  the change log (https://github.com/numba/numba/blob/master/CHANGE_LOG).
• I have included below a minimal working reproducer (if you are unsure how
  to write one see http://matthewrocklin.com/blog/work/2018/02/28/minimal-bug-reports).

#2916 describes that **kwargs is not supported in njit functions.

In [10]: from numba import njit

In [11]: @njit
    ...: def f(func, args, kwargs):
    ...:     return func(*args, **kwargs)
    ...:

In [12]: @njit
    ...: def my_func(*args, **kwargs):
    ...:     return list(*args)
    ...:

In [13]: f(my_func, (1, 2, 3), {'a': 1, 'b': 2})
NotImplementedError: Failed in nopython mode pipeline (step: analyzing bytecode)
CALL_FUNCTION_EX with **kwargs not supported

In [14]: numba.__version__
Out[14]: '0.45.1'

Related to #4570, I am trying to use partial to bind the kwargs earlier.

In [23]: from functools import partial

In [24]: def my_func(*args, **kwargs):
    ...:     return list(args)
    ...:

In [25]: my_func_partial = partial(my_func, **{'a':2})

In [26]: njit(my_func_partial)

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-26-6bdaac92077c> in <module>
----> 1 njit(my_func_partial)

/anaconda3/envs/pandas-2s-dev/lib/python3.7/site-packages/numba/decorators.py in njit(*args, **kws)
    222         warnings.warn('forceobj is set for njit and is ignored', RuntimeWarning)
    223     kws.update({'nopython': True})
--> 224     return jit(*args, **kws)
    225
    226

/anaconda3/envs/pandas-2s-dev/lib/python3.7/site-packages/numba/decorators.py in jit(signature_or_function, locals, target, cache, pipeline_class, **options)
    159                    targetoptions=options, **dispatcher_args)
    160     if pyfunc is not None:
--> 161         return wrapper(pyfunc)
    162     else:
    163         return wrapper

/anaconda3/envs/pandas-2s-dev/lib/python3.7/site-packages/numba/decorators.py in wrapper(func)
    175         disp = dispatcher(py_func=func, locals=locals,
    176                           targetoptions=targetoptions,
--> 177                           **dispatcher_args)
    178         if cache:
    179             disp.enable_caching()

/anaconda3/envs/pandas-2s-dev/lib/python3.7/site-packages/numba/dispatcher.py in __init__(self, py_func, locals, targetoptions, impl_kind, pipeline_class)
    581
    582         _DispatcherBase.__init__(self, arg_count, py_func, pysig, can_fallback,
--> 583                                  exact_match_required=False)
    584
    585         functools.update_wrapper(self, py_func)

/anaconda3/envs/pandas-2s-dev/lib/python3.7/site-packages/numba/dispatcher.py in __init__(self, arg_count, py_func, pysig, can_fallback, exact_match_required)
    207
    208         argnames = tuple(pysig.parameters)
--> 209         default_values = self.py_func.__defaults__ or ()
    210         defargs = tuple(OmittedArg(val) for val in default_values)
    211         try:

AttributeError: 'functools.partial' object has no attribute '__defaults__'

Is there a better workaround for what I am trying to accomplish?

[stuartarchibald]

Thanks for the report. I'm not sure there's a way around this at present. We'll discuss at the next core developer meeting.

[mroeschke]

Thanks @stuartarchibald.

[MetinSa]

Any updates on this?

I am trying to njit a function which takes as argument a tuple of partial functions. Running numba.typeof(my_partial_func) results in the following value error: ValueError: Cannot determine Numba type of <class 'functools.partial'>. If I try to re-njit the partial function: numba.njit(my_partial_func) I get the following type error: TypeError: The decorated object is not a function (got type <class 'functools.partial'>). Is there a way to achieve this with numba as of now?

[AlonSpinner]

Here for the same feature aswell