bpo-31722: Update codecs.IncrementalDecoder to use __subclasshook__

[corona10]

I 've updated to codecs.IncrementalDecoder to be able to get True from below code snippets.

>>> issubclass(_pyio.IncrementalNewlineDecoder, codecs.IncrementalDecoder)
True
>>> issubclass(_io.IncrementalNewlineDecoder, codecs.IncrementalDecoder)
True
>>> issubclass(io.IncrementalNewlineDecoder, codecs.IncrementalDecoder)
True
>>> class A(_io.IncrementalNewlineDecoder):
...     pass
...
>>> class B(_pyio.IncrementalNewlineDecoder):
...     pass
...
>>> issubclass(A, codecs.IncrementalDecoder)
True
>>> issubclass(B, codecs.IncrementalDecoder)
True

https://bugs.python.org/issue31722

[corona10]

@serhiy-storchaka @vstinner

For reviewers,
IMHO, The more straightforward way to solve this issue is set tp_base of _io.IncrementalNewlineDecoder to codecs.IncrementalDecoder.
But I can't find a way to this on C API level to work with pure python object type.

So my choice is to use ABC.abc.
Please let me know if I am going in the wrong way.

Thanks always!

[corona10]

Or How about switching codecs.IncrementalDecoder into the C implementation _codecs.IncrementalDecoderand then
inherit it on the _pyio. IncrementalNewlineDecoder and _io. IncrementalNewlineDecoder ?

[serhiy-storchaka]

The problem is not that issubclass() returns False, but that two classes are not truly compatible.

[serhiy-storchaka]

Implementing IncrementalDecoder in C could work, but it would add more complexity. I do not know what is the best solution of this issue.

[corona10]

@serhiy-storchaka

Yes, I tried it today and I found that it is a complex issue better than I expected.
Since this class is used in the below pattern.

class IncrementalDecoder(mbc.MultibyteIncrementalDecoder,
                                              codecs.IncrementalDecoder):

We should implement it to be able to avoid lay-out conflict.