Re: [PATCH] git-daemon: Simplify child management and associated logging by

From: Junio C Hamano <gitster@pobox.com>
Date: Aug 12, 2008, 22:05 UTC
Message-ID: <7vzlnhq48b.fsf@gitster.siamese.dyndns.org>
In-Reply-To: <20080812212534.6871.19377.stgit@aristoteles.cuci.nl>

"Stephen R. van den Berg" <srb@cuci.nl> writes:

Show 10 quoted lines

> making the signal handler almost a no-op.
> Fix the killing code to actually be smart instead of the
> pseudo-random mess.
> Get rid of the silly fixed array of children and make
> max-connections dynamic and configurable in the process.
> Make git-daemon a proper syslogging citizen with PID-info.
> Simplify the overzealous double buffering in the logroutine,
> remove the artificial maximum logline length in the process.
>
> Signed-off-by: Stephen R. van den Berg <srb@cuci.nl>

Sorry, but this does too many things in one patch.

 - Taking advantage of poll() getting interrupted by SIGCHLD, so that you
   do not have to do anything in the signal handler, is so obvious that I
   am actually ashamed of not having to think of it the last time we
   touched this code.  Is there a poll() that does not return EINTR but
   just call the handler and restart after that as if nothing has
   happened, I have to wonder...

 - Conversion from silly fixed array to dynamic and configurable maximum
   would be a good idea, but that is independent from the above, isn't it?

 - I see you have a call to vsyslog, which is the first user of the
   function.  How portable is it (the patch coming from you, I know
   Solaris would have it, and recent 4BSD also would, but what about the
   others)?