implements more dunder protocols

[cmacdonald]

Inspired by #494 (comment), this PR has more support for Java Collections Framework, Iterators, Iterable and Comparable

I worked from https://rszalski.github.io/magicmethods/. Various test cases included.

Ones I didn't implement:

• __hash__(self) - we couldnt guarentee the Python contract of hashCode() vs equals() wouldnt have been implemented on the Java side, as Comparable doesnt require it.
• __nonzero__(self) - java.lang.Boolean (etc) are automapped anyway
• __str__(self) - this could be mapped this to toString(), but it was still handy in Python to know the jnius representation, i.e. that its a wrapper object.

I haven't verified on Python versions other than 3.6 and Java versions other than 13, so would appreciate the github workflow being run on this PR.

[tshirtman]

Really not much to say there, it's really nice.

About __str__ i think that usage should be achieved by repr(obj) (implemented by __repr__) instead of str(obj), but it's true that it's very common to use the latter rather than the former in programs to display debug info, and it sometime accidentally makes it easier to spot issues, so i understand this decision for practical reasons.

[tshirtman]

make these docstrings instead :)

[tshirtman]

just curious why this pass-through override is needed? leftover from tests?

[cmacdonald]

yes. i developed the protocol_map in the test constructor before copying into reflect.py. do you want removed?

[tshirtman]

that would be appreciated, not a big deal, but good dead code is deleted code :P.

[cmacdonald]

feedback address, thanks!

[cmacdonald]

problems on Python 2 - see https://github.com/kivy/pyjnius/runs/579173024?check_suite_focus=true#step:9:138. Do you have a Python 2 dev environment @tshirtman ?

[tshirtman]

woopsie, yes i do, i should have tested, thanks for the heads up!

[tshirtman]

i can reproduce localy, let's try to make sense of it :]
the call to keySet() and to iterator() on it both, work, the issue seems to be when iterating on the result.

[cmacdonald]

https://github.com/kivy/pyjnius/runs/579173024?check_suite_focus=true#step:9:138
Python 2 needs next() rather than __next__()? (see https://stackoverflow.com/questions/41524531/python-2-vs-python-3-class-with-iter) The Java next() implementation wont raise StopIteration.

Not sure how to fix this, as I'm not sure we can call "super().next()"?

https://github.com/kivy/pyjnius/runs/579173024?check_suite_focus=true#step:9:181 this line of code should be removed.

[tshirtman]

yeah, it calls next() all the time, so the _iterator_next is not used, and so doesn't check for hasNext(). i think the last link is not what you wanted to paste.

[cmacdonald]

Last link was not a fix for this, just an observation that a print(dir()) had made it into a test

We would need a special if() in autoclass for Python 2, which would need to rename the classDict entry for the Java next(). As we have discovered with the fields, renaming them on the Python side is tricky.

How long do you plan to support Python 2 for?

Or we "degrade" the protocol_map support for Python 2.

[tshirtman]

How long do you plan to support Python 2 for?

This is a strong reminder that we probably want to drop it next release :D (it would be a bit abrubt to drop without forewarning, right?) i'm myself would be perfectly fine dropping it asap, normaly everybody using pyjnius for android should have migrated to python3 anyway, as this one did drop support a couple releases ago.

We would need a special if() in autoclass for Python 2, which would need to rename the classDict entry for the Java next(). As we have discovered with the fields, renaming them on the Python side is tricky.

Yeah, my ideas of how to fix this on the python side are quite gross, like

def _iterator_iter(self):
    print("starting iterator")
    if PY2:
        # in py2 the next() is called on the iterator, not __next__
        # so we need to monkey patch this one to check hasNext
        j_next = self.next
        log.debug("monkey patching next()")
        def safe_next():
            if not self.hasNext():
                raise StopIteration()
            return j_next()
        self.next = safe_next

    return self

(though it's not working yet in my tests, not sure why, but if it did, would that be fine?)

[cmacdonald]

I can live with that implementation until Python 2 expires.

[tshirtman]

so this works, and feels a bit less hacky #507