fix(python): complete filenames for script arguments

[akinomyoga]

Ref. ohmybash/oh-my-bash#462

When [-c command | -m module | file] is already specified before cword, the current implementation of the python completer tries to complete filenames as arguments to the command/module/script. However, this doesn't work after an option of the form -?. The following is a case reported in ohmybash/oh-my-bash#462:

python some_script.py -p [TAB]

The current implementation seems to try to detect the situation by checking the previous and the second previous arguments, but we should explicitly test if [-c command | -m module | file] is already specified.

Note: Removing an old case pattern that excludes python and -?

I think this is actually related to the following line that we once discussed in #952 (see replies [1,2]):

bash-completion/completions/python

Lines 77 to 81 in 8b4db1c

	!(?(*/)?(micro)python*([0-9.])|?(*/)py@(py|ston)*([0-9.])|-?))
	if [[ $cword -lt 2 || ${words[cword - 2]} != -[QWX] ]]; then
	_comp_compgen_filedir
	return
	fi

• [1] refactor: { => _comp_compgen}_filedir and _comp_compgen [opts] NAME #952 (comment)
• [2] refactor: { => _comp_compgen}_filedir and _comp_compgen [opts] NAME #952 (comment)

After looking at this code again, I now believe this piece of code tried to detect whether the current word is an argument just after [-c command | -m module | file] or not. The case pattern and the if statement exclude the case where

1. the previous argument is the command name (?(*/)?(micro)python*([0-9.])|?(*/)py@(py|ston)*([0-9.])),
2. the previous argument is an option (-?), or
3. the second previous and previous arguments form a pair of an option plus an option argument (${words[cword - 2]} != -[QWX]).

The remaining situation is likely to be only the case where the previous argument is a command/module/script of [-c command | -m module | file]. Since we now explicitly check [-c command | -m module | file], we should remove this heuristic detection of the first argument.

[scop]

Makes sense. This is a bit of a corner case maybe, could we add some test cases to cover/guard the intended behavior?

[akinomyoga]

I added small adjustments and tests.

[scop]

Thanks!