Gnus PATCH: nnimap operations with literals fail when using GSSAPI

Gnus v5.8.8 XEmacs 21.4 (patch 4) "Artificial Intelligence" [Lucid] (i686-pc-linux, Mule) of Wed Aug 29 2001 on two-places-at-once.mit.edu 200 senator-bedfellow.mit.edu NNRP Service Ready - usenet(a)mit.edu (posting ok). Any IMAP command containing a literal that contains newlines will fail when using nnimap with an nnimap-stream of "gssapi". (APPEND is a good example of a command that will fail.) This is because the literal length is miscomputed. First, the fix; make imap-gssapi-open match imap-kerberos4-open in this way: --- imap.el 2002/04/03 07:42:18 1.1 +++ imap.el 2002/04/03 07:45:13 ＠＠ -520,7 +520,8 ＠＠ response) (when process (with-current-buffer buffer - (setq imap-client-eol "\n") + (setq imap-client-eol "\n" + imap-calculate-literal-size-first t) (while (and (memq (process-status process) '(open run)) (goto-char (point-min)) ;; cyrus 1.6.x (13? < x <= 22) queries capabilities Now, the background. "imtest" is a horrible, evil, unholy program that hates me. It calls up my mother at 3am and swears at her. It murdered my cat. It's *that* bad. IMAP specifies that newlines are canonicalized to \r\n. imtest requires that its input lines be terminated with \n, because it uses fgets(). (I suppose this is actually platform specific and Gnus and imtest would lose in more interesting ways on non-Unix platforms, if they ran there.) imtest then does a \n to \r\n translation on what it's read. This means that any program that uses imtest and sends literals needs to send any newlines as \n but count them as 2 characters when computing the literal length. Gnus likes to construct proper IMAP messages (with \r\n) but is smart enough to know that imtest wants \n, and translates the buffer before sending it to imtest. (That's the (setq imap-client-eol "\n") bit.) However, unless you tell it (setq imap-calculate-literal-size-first t), it gets the length wrong because it calculates the length after translating, forgetting that imtest will translate back. This obviously bit someone long ago on kerberos4 auth, because that code does the right thing. But that right thing never got copied to the GSSAPI code. Oh, by the way, remember how I said that imtest uses fgets()? It uses a 2k buffer there. Which means that if you ever try to send a line longer than 2K, imtest will insert spurious newlines in it, and you're just fucked. Isn't this fun? --Jered