From: Junio C Hamano <gitster@pobox.com>
Date: Fri, 11 Sep 2009 08:54:55 GMT
Subject: Re: Issue 323 in msysgit: Can't clone over http
Message-ID: <7veiqdzwrk.fsf@alter.siamese.dyndns.org>
In-Reply-To: <20090909203350.e3d7e5dc.rctay89@gmail.com>

Tay Ray Chuan <rctay89@gmail.com> writes:

>> I am in favour or removing this check, not just due to its
>> unreliability, but for the sake of consistency (we very rarely send a
>> HEAD request to poll data before doing a GET).

I think this is sensible.  The only case I think it could be a problem is
this scenario:

 - You clone from remote.

 - Remote works more and has two more packs, A and B. 

 - You fetch, but this does not need both of them (perhaps you fetched
   only one branch).  You get *.idx files for both, but *.pack file only
   for A.

 - Remote works a bit more and then repacks everything into a single pack
   C.

 - You fetch, and the walker walks the loose objects, and then finds one
   object that cannot be obtained as a loose object.  It tries to look up
   in the *.idx file and finds it in B.

   But the packfile B is long gone.

I didn't follow the codepath that uses http_get_info_packs() and then uses
repo->packs list to see what it does, but as long as the above does not
happen we should be Ok.

Thanks.