On 30.4.2014, at 0.27, Michael M Slusarz <slusarz@curecanti.org> wrote:
Given this test message, with admittedly incorrect QP encoding: .. Dovecot 2.2 returns this:
C: 5 UID FETCH 4464 (BINARY.PEEK[1]) S: * 1 FETCH (UID 4464 BINARY[1] NIL) S: 5 OK Fetch completed.
Contrast with, e.g., Cyrus 2.4:
C: 6 UID FETCH 1 (BINARY.PEEK[1]) S: * 1 FETCH (UID 1 BINARY[1] {57} S: [LITERAL DATA: 57 bytes] S: ) S: 6 OK Completed (0.000 sec)
(Cyrus FETCH output strips out the spurious non-encoding '=', IIRC).
Not sure if this is an example of Cyrus' QP decoder being more robust (or lenient) than Dovecot's. Or whether this is intentional to return NIL for this kind of bad data.
It was kind of intentional. Dovecot's istream-qp-decoder aborts when it finds anything broken. I guess it could simply skip errors, but I'm not sure how good idea that is either..
Although if intentional, output should probably be a NO response with UNKNOWN-CTE response code, since this appears to be an instance of "the server does not know how to decode the section's CTE". (RFC 3516 [4.3]).
Yeah, I think that's better. Fixed: http://hg.dovecot.org/dovecot-2.2/rev/197f77f6ef0d
Also this fix more or less requires this: http://hg.dovecot.org/dovecot-2.2/rev/99f59d6fce05