[issue23377] HTTPResponse may drop buffer holding next response

[Demian Brecht]

Demian Brecht added the comment:

> I think it's ok to slightly break a non-public API since it's required to fix an obvious bug.

To play devil's advocate here, /is/ this really non-public API? According to documented Python naming conventions, HTTPResponse is part of the public API. From code I've written myself, this would likely create a number of breakages. Granted, it would largely be in lower level testing code, but it would still create a headache. Is something that's lightly documented as being non-public take precedence over well known and used naming conventions? I'd argue no, it shouldn't.

That said, I'm not overly enthused about the alternative Martin describes, so at first thought, I'm pretty much square on the fence about this.

> By the way, I guess we don't support HTTP pipelining, right?

As Martin pointed out, sadly no. There's currently an internal state machine that seems to possibly be remnants of an initial implementation (just a guess, I haven't done any research into it to prove that though) that prevents pipelining. It's on my list of things to tackle as it seems to be a natural progression to HTTP/2.0 multiplexing, but unfortunately my time is a little shorter than usual at the moment (taking a new job has cut into my spare time a little).

@Martin:

I haven't had a chance to review the change in detail yet, but I think the idea's sound. I'll try to get some time over the next few days to dig into it further.

----------

_______________________________________
Python tracker <report at bugs.python.org>
<http://bugs.python.org/issue23377>
_______________________________________