Re: [PATCH] Show URL in the "Getting <foo> list" http-fetch messages

From: Junio C Hamano <junkio@cox.net>
Date: Nov 14, 2005, 20:11 UTC
Message-ID: <7vveyvxa4d.fsf@assigned-by-dhcp.cox.net>
In-Reply-To: <20051114190225.GA3793@reactrix.com>

Nick Hengeveld <nickh@reactrix.com> writes:

Show 13 quoted lines

> On Sat, Nov 12, 2005 at 09:22:02AM -0800, Nick Hengeveld wrote:
>
>> This should not be an issue with index requests because they are only
>> initiated from fetch().  The previous patch to load alternates on demand
>> added alternate handling to process_curl_messages() so that a 404 for an
>> object can be handled immediately rather than waiting for the fetch()
>> call for that object to notice.
>
> Seems like it might make sense to handle pack downloads immediately when
> an object is unavailable rather than waiting for the fetch() call.  It
> could prevent attempts to download any other objects inside that pack,
> although queued requests that activate while a pack is downloading
> would have to wait to see whether the download is successful.

I think that makes sense, and it probably is preferable to make them wait than blindly go ahead. Although we are issuing requests in parallel, these simultaneous requests are asking for related things (e.g. parent commit objects or the tree object of the commit we just saw; blob objects contained in the tree we just saw) and are more likely than not to be found in the pack being requested. Losing parallelism while a pack is in transit would not be too much of a problem.