692055 - inconsistent behaviour / pipelining for POP3 account settings auto-discovery. Pipelining to server that can't handle it causes QUIT is missed by the server.

Status: NEW

(MailNews Core :: Networking: POP, defect)

Description

[Adrien de Croy]

Attachment: tb8pop3discovery.pcap

User Agent: Mozilla/5.0 (Windows NT 5.1) AppleWebKit/535.1 (KHTML, like Gecko) Chrome/14.0.835.187 Safari/535.1

Steps to reproduce:

this is while creating an account in TB 8.0.  Choosing manual configuration.

Choose POP3, set server, port and automatic for SSL/TLS and auth.

Different behaviour on initial retest than subsequent ones.  Also it's using pipelined POP3 commands CAPA and QUIT which is a BAD idea, as many servers / proxies don't support pipelining of POP3 commands. 


Actual results:

initial retest:  Makes 3 connections, sends CAPA + QUIT in 1 buffer.
subsequent retest: makes 1 connection, sends CAPA + QUIT in several buffers, but prior to receiving any response from the server, so still effectively pipelined


Expected results:

make only 1 connection, wait for +OK welcome, send CAPA, wait for response, send QUIT.

Comment 1

[Adrien de Croy]

Just analyzed a bit more with wireshark, and uploaded capture.  This shows 3 connections in initial test, and 1 in subsequent (some time later).

The 3rd of the initial connections is of concern, since it connects and spews garbage at the server.  Is this maybe a SSL client hello checking for SSL on the port?  It should know already based on the +OK it received that it's not SSL.

So maybe the behaviour is different because after the first test, the setting for SSL is changed to None (so maybe therefore doesn't re-test for it).

In any case, because the 2 commands are pipelined, the QUIT is missed by the server.  Pipelining should only be used if the server advertises it.

Comment 3

[Wayne Mery (:wsmwk)]

Do you still see this when using a current version?

Comment 4

[Wayne Mery (:wsmwk)]

(adrien appears to be gone)

Ben, would you expect this to still be an issue?

Comment 5

[Ben Bucksch (:BenB)]

The guessconfig.js logic hasn't changed much, so I expect this bug to still be there.

The reporter seems to be competent, so I think it's worthwhile to keep this open.

Comment 6

[Wayne Mery (:wsmwk)]

Thanks