POP3 behavior on broken tcp connection

[John Gardiner Myers <[e-mail]>]

Mark Crispin <[e-mail]> writes:
>       A related issue comes up in terms of what happens if the TCP connection
> breaks before a QUIT is received.  Are deleted messages purged?  Matters are
> complicated in two ways:
>  a) c-client/IMAP has a flag, \Deleted, that is preserved across sessions, and
>     a corresponding ``expunge'' operation removes it.
>  b) having observed that most POP3 clients are utterly unprepared to handle
>     the case of messages being deleted at startup (that is, a preserved
>     deletion status), my POP3 server makes deleted messages at startup be
>     invisible (that is, I adjust the numbers of subsequent messages downwards
>     as if an expunge had happened).  No messages are actually destroyed until
>     a QUIT (or expunge in some other c-client context), so you can end up with
>     some deleted messages which are invisible and others which are not, all
>     depending upon whether or not they were deleted in this session.
>  c) c-client also supports shared mailboxes.  This means that having deleted
>     status as a session flag isn't any good; you could have a delete, followed
>     by an undelete later in another session, then a QUIT in the first session
>     would expunge the message even though it was undeleted!
> So, this leads to:
>  5) what the hell am I supposed to do?  Somehow, it seems that the POP3
>     protocol is very much oriented towards non-shared mailbox technology with
>     no state stored in the data.  If I leave things the way they are, then a
>     broken connection makes deleted messages inaccessible to any other POP3
>     connection (although other c-client access can get at them).  Why don't
>     POP3 clients understand about messages that are deleted at startup?

Marshall Rose <[e-mail]> writes:
> I think that if the TCP connection breaks before a QUIT is received,
> that the session doesn't enter the UPDATE state and that deleted
> messages are not purged.


The following sentence in the various drafts could be interpreted to
say a session enters the UPDATE state upon a broken TCP connection:

   When the client has
   finished its transactions, the session enters the UPDATE state.

-- 
_.John G. Myers		Internet: [e-mail]
			LoseNet:  ...!seismo!ihnp4!wiscvm.wisc.edu!give!up