Author martin.panter
Recipients Yuri.Bochkarev, agriffis, alanjds, amak, cananian, demian.brecht, icordasc, jcea, jhylton, martin.panter, mhammond, orsenthil, r.david.murray, rbcollins
Date 2015-02-19.23:27:27
SpamBayes Score -1.0
Marked as misclassified Yes
Message-id <1424388448.52.0.238896285549.issue3566@psf.upfronthosting.co.za>
In-reply-to
Content
Posting RemoteDisconnected.v4.patch with these changes:

* Renamed ConnectionClosed → RemoteDisconnected. Hopefully avoids confusion with shutting down the local end of a connection, or closing a socket’s file descriptor.

* Dropped the HTTPConnection.closed attribute

* Dropped special distinction of ECONNRESET at start versus in middle of response. It certainly makes the code and tests simpler again, and I realize that distinction is not the most important problem to solve right now, if ever. Also avoids relying on the poorly defined BufferedReader.peek() method.

I would like to retain the backwards compatibility with BadStatusLine if that is okay though.

Requests and “urllib3”: I’m not overly familiar with the internals of these packages (Requests uses “urllib3”). I cannot find any reference to BadStatusLine handling in “urllib3”, and I suspect it might just rely on detecting a dropped connection before sending a request; see <https://github.com/shazow/urllib3/blob/c8e7ea5/urllib3/connectionpool.py#L236>. In my opinion this is a race condition, but it is helpful and works most of the time. So I suspect “urllib3” would not be affected by any changes made relating to BadStatusLine.
History
Date User Action Args
2015-02-19 23:27:29martin.pantersetrecipients: + martin.panter, jhylton, mhammond, jcea, orsenthil, amak, rbcollins, cananian, r.david.murray, alanjds, agriffis, icordasc, demian.brecht, Yuri.Bochkarev
2015-02-19 23:27:28martin.pantersetmessageid: <1424388448.52.0.238896285549.issue3566@psf.upfronthosting.co.za>
2015-02-19 23:27:28martin.panterlinkissue3566 messages
2015-02-19 23:27:28martin.pantercreate