_lib._util.getargspec_no_self is missing KEYWORD_ONLY support

[pvanmulbregt]

_lib._util.getargspec_no_self() only checks for 3 of the 5 possible parameter kinds returned by Python's inspect.signature() : (VAR_POSITIONAL, POSITIONAL_OR_KEYWORD, VAR_KEYWORD), and ignores the other two (POSITIONAL_ONLY, KEYWORD_ONLY)

Reproducing code example:

import numpy as np, scipy, scipy._lib, scipy._lib._util, inspect                                                                                                                                                             
class rv_generic(object): 
     def _rvs(self, *args, size=None): 
         U = self._random_state.random_sample(size) 
         Y = self._ppf(U, *args) 
         return Y 

# Create an instance of rv_generic
rv_obj = rv_generic()  
# Now print the results of the 4 ways to get the arg specifications
# First SciPy's which is missing the 'size' argument
print(scipy._lib._util.getargspec_no_self(rv_obj._rvs))

# Now the 3 Python inspect approaches
# Python's inspect.signature returns the KEYWORD_ONLY arguments. 
sig = inspect.signature(rv_obj._rvs)                                                                                                                                                                    
print([[p.name, p.kind] for p in sig.parameters.values()])

# Python's inspect.getfullargspec returns information on all the arguments. 
print(inspect.getfullargspec(rv_obj._rvs))

# Python's inspect.getargspec raises an Error as there are KEYWORD_ONLY arguments, which are not handled by inspect.getargspec
print(inspect.getargspec(rv_obj._rvs))

The output is

ArgSpec(args=[], varargs='args', keywords=None, defaults=None)

[['args', <_ParameterKind.VAR_POSITIONAL: 2>],
 ['size', <_ParameterKind.KEYWORD_ONLY: 3>]]

FullArgSpec(args=['self'], varargs='args', varkw=None, defaults=None, kwonlyargs=['size'], kwonlydefaults={'size': None}, annotations={})

ValueError: Function has keyword-only parameters or annotations, use inspect.signature() API which can support them

Scipy/Numpy/Python version information:

1.5.0.dev0+4242f73 1.17.3 sys.version_info(major=3, minor=8, micro=0, releaselevel='final', serial=0)