# HG changeset patch
# User Mark Doliner <mark@kingant.net>
# Date 1220551469 0
# Node ID 85bed17fe5c134f93b2d73ca1aea36a5966df7ab
# Parent  a0381a68ceef7102793a10b3eb9d58a3c32df400
The variable we use to keep track of the watcher of the ssl connection
should be unsigned.  This isn't really a problem in Pidgin, where we
use glib's mainloop and GIOChannels because glib starts assigning the
handle IDs sequentially starting from 1.

But if an eventloop implementation ever returns a handle ID greater
than the largest possible signed integer (2,147,483,647) then we
won't be able to remove the watcher because purple_ssl_close() in
sslconn.c only removes it if inpa > 0, and since it interprets inpa
as a signed value then handles over 2,147,483,647 appear as negative
numbers.

I stumbled upon this when playing around with libevent, which can
use epoll.  My implementation generated a random handle ID which
was sometimes greater than 2,147,483,647.

I don't believe this breaks binary compatibility.  And I don't think
it breaks source compatibility, but I guess it might depend on what
compiler you're using.

diff -r a0381a68ceef -r 85bed17fe5c1 libpurple/sslconn.h
--- a/libpurple/sslconn.h	Thu Sep 04 04:05:54 2008 +0000
+++ b/libpurple/sslconn.h	Thu Sep 04 18:04:29 2008 +0000
@@ -67,7 +67,7 @@
 	int fd;
 	/** Glib event source ID; used to refer to the received data callback 
 	 * in the glib eventloop */
-	int inpa;
+	guint inpa;
 	/** Data related to the underlying TCP connection */
 	PurpleProxyConnectData *connect_data;