The whole purpose of the module is not to log to syslog - it logs to a variety of destinations. If you meant the SysLogHandler class - why exactly do you need the syslogging to be done via a syscall? You have two options already - log using a hostname/port for local or remote daemons, or use e.g. address="/dev/log" to use a Unix domain socket for access to the local syslog daemon. So I'm not sure how the purpose is defeated.

The patch as it stands will not import cleanly in Windows (since there is no "syslog" module there), so it cannot be accepted as it stands. Ordinarily I would make any small amendments myself, but at the moment I don't have the time.

You should also be aware that the syslog module may not be thread-safe on all platforms - for instance, see the first Google hit I got for the query "syslog openlog":

http://www.unet.univie.ac.at/aix/libs/basetrf2/syslog.htm

Now logging does provide an I/O lock, but that's only around the emit() call. Your syslog calls have a broader scope (handler open/close) and I'm not sure whether thread-safety could be guaranteed in all cases.

If you want to work up a patch which works across platforms and can allay my worries about multi-threading, I'll certainly look at it again.