From: Junio C Hamano <gitster@pobox.com>
Date: Mon, 27 Sep 2010 03:52:15 GMT
Subject: Re: What's cooking in git.git (Sep 2010, #05; Wed, 15)
Message-ID: <7vocbjn8r4.fsf@alter.siamese.dyndns.org>
In-Reply-To: <20100925143008.14dfbfa6@laptop.localhost>

Alexander Sulfrian <alexander@sulfrian.net> writes:

> that's not true. The last version (v4) logs for every listen an error
> if no socket could be created. And dies if not a single socket is there
> after processing all --listen options.

Hmm, the copy in 'pu' has at least these issues.

 - The error behaviour is different between with and without IPV6
   compilation.  A failure from getaddrinfo() on one named address results
   in die, but a failure from corresponding inet_pton() in NO_IPV6
   codepath does not.  I think I've recommended both not to die and let
   the caller do the diagnosing and dying?

 - The style of API, which passes (int **list_p, int cnt) and returns the
   updated cnt, was unwieldy; I thought I recommended to use a struct that
   can be more easily used with ALLOC_GROW() API?

Perhaps there was a v4 that came during the freeze period that I didn't
look at?