Antoine: I believe I have everything you mentioned addressed with the new patch.  That was an awesome review, thank you so much.

The only things I didn't do were parts of the last two items you bring up:

If PyTuple_New(0) fails, bail out.
If syslog_openlog fails, bail out.

syslog(3) can continue even if the openlog() fails.  It won't have the expected "ident" string, but it *WILL* log.

I believe this is the desired behavior.

NOTE: I puled the code out that does all the sys.argv handling, which I think made that whole section of code much easier to read, particularly with the new changes.  The down side is that the code to be reviewed is quite different now.

I also found a leak in the call to syslog_openlog() where I wasn't DECREFing the return.

Can you please review these changes?