FittingWithOutlierRemoval says "niter" is optional but crashes if niter is None

Description

The docstring for FittingWithOutlierRemoval says that the parameter “niter” (the maximum number of iterations) is optional, but calling with iter=None causes a crash because the __call__() method expects it to be an integer.

I see three solutions, in increasing order of (I think) desirability, so I’m not submitting a PR in case whichever solution I code is not approved.

1. Change the docstring so niter must be an integer 😃
2. Check if niter is None in the __init__() or __call__() and, if so, set it to a large number instead. Since #10642 stops iterating once convergence has been reached, one can check whether this number of iterations occurred, in which some sort of “did not converge” flag could be added to the fit_info.
3. Allow the loop to iterate infinitely, in principle, but repeatedly check that the solution is converging and break out if it doesn’t. I’m not sure this is possible though since FittingWithOutlierRemoval is a wrapper for other fitters, whose behaviour may not be well-prescribed.

Expected behavior

I expected it to iterate to convergence.

Actual behavior

It crashed.

Steps to Reproduce

>>> import numpy as np
>>> from astropy import models, fitting
>>> from astropy.stats import sigma_clip
>>> x=np.arange(100)
>>> y = x
>>> m = models.Polynomial1D(degree=1)
>>> f = fitting.FittingWithOutlierRemoval(fitting.LinearLSQFitter(), niter=None, outlier_func=sigma_clip)
>>> m_final = f(m, x, y)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/opt/anaconda3/envs/dev/lib/python3.7/site-packages/astropy/modeling/fitting.py", line 870, in __call__
    for n in range(1, self.niter + 1):
TypeError: unsupported operand type(s) for +: 'NoneType' and 'int'

System Details

Python 3.7.7 (default, May 6 2020, 04:59:01) [Clang 4.0.1 (tags/RELEASE_401/final)] Numpy 1.19.1 astropy 4.2.dev735+ga83f25b96 Scipy 1.5.2 Matplotlib 3.3.1

[pllim]

No worries, @chris-simpson ! I wasn’t aware there exists alternative definition of “optional,” so TIL as well.

[chris-simpson]

Fair enough, I’ve been working with a package where “optional” is defined as None is an acceptable value, whereas I do see that astropy is consistently using “optional” to mean there exists a default value. Sorry for wasting everyone’s time; it’s been a long year!