EofStream should be converted to StopAsyncIterator.
Tests should be added.
Documentation -- updated.
What's about iterating over chunks also (.read(), .readexactly() and .readeany() should be supported too I believe).

OK! but i don’t get the last point.

what do you mean by “supported”? we can only add one __aiter__ and one __anext__ function here.

we could of course add explicit asnc generator functions, e.g. aiter_any, which would crate an async iterator that loops data as available instead of lines.

readexactly is parameterized, so it makes no sense to make __anext__ defer to it. read and readany are basically the same when called unparameterized, so it makes no sense to distinguish.

so:

1. what should __anext__ do for the stream readers? readany or readline?
2. how to support the other? aiter_lines()/aiter_any()/aiter_exactly(n) methods? which of those?

Well, .readexactly() doesn't make sense.
My initial intention was adding a way by both iterating over fixed size chunks and any present data.

Something like

async for chunk in resp.content.iter_chunked(1024):
    ...

async for slice in resp.content.iter_any():
    ...

Does it make sense?

Default iteration for StreamReader should be line-mode, sure -- like iteration over opened text file.

OK, then we’re on the same page 😄

we now only need a way to indicate in the docs that those methods are only available on python 3.5. is there a sphinx directive for that?

maybe we should also version-gate them?

class StreamReader(...):

    def __init__(self, ...):
        ...

    if python_version > (3, 5):
        def iter_any(self):
            ...

Yes, lease use check like https://github.com/KeepSafe/aiohttp/blob/master/aiohttp/client.py#L447

I believe just versionadded directive with text like "available for Python 3.5+ only" should be enough.

We dropped Py3.3, I'm converting the doc to use async/await syntax so Py3.5 is preferable version for next release.

OK. final thing for now: do i only need to document the iter_* functions? because __aiter__ is documented in the stdlib already.

Magic methods should not be documented itself but I would see usage example in class section, e.g. http://aiohttp.readthedocs.org/en/latest/client_reference.html#aiohttp.ClientResponse

OK!

hmm, i have this AsyncStreamReaderMixin which is used by StreamReader and EmptyStreamReader, but that results in the iter_chunked and iter_any methods not appearing in the docs.

can i force autodoc to treat the methods of AsyncStreamReaderMixin as if they were in StreamReader?

/edit: got it!

I believe we should document StreamReader manually, without autodoc.
Eventually we will drop autodoc at all.

maybe after this is done?

ok, you can use autodoc for now if you want.

done. this got more complex than i thought due to readexactly leaving this in a state where 0 bytes are left but EOF isn’t actually hit.

tests pass, flake8 also (why the hell does it complain about “unused variable _”? every other tool recognizes that as placeholder)

this got more complex than i thought due to readexactly leaving this in a state where 0 bytes are left but EOF isn’t actually hit.

Hmm. Looks like a bug. I think .readexcaltly should raise EofStream after .feed_eof() call.
Anyway .iter_chunked should use .read not .readexactly I believe.

done!

@asvetlov good to merge?

@asvetlov ping

Thanks!

This thread has been automatically locked since there has not been
any recent activity after it was closed. Please open a new issue for
related bugs.

If you feel like there's important points made in this discussion,
please include those exceprts into that new issue.