Re: [RFH] GSoC 2015 application

From: Duy Nguyen <pclouds@gmail.com>
To: Jeff King <peff@peff.net>
Cc: Git Mailing List <git@vger.kernel.org>
Subject: Re: [RFH] GSoC 2015 application
Date: Thu, 26 Feb 2015 20:10:38 +0700	[thread overview]
Message-ID: <CACsJy8D6nAt3fMw=oh5mwUwfAJKViN1pk9pf9gZSBGyquU_WBQ@mail.gmail.com> (raw)
In-Reply-To: <20150218191417.GA7767@peff.net>

On Thu, Feb 19, 2015 at 2:14 AM, Jeff King <peff@peff.net> wrote:
> Where I really need help now is in the "ideas" page:
>
>   http://git.github.io/SoC-2015-Ideas.html

Is this too ambitious for a summer? I suspect the answer is yes, but anyway..

Due to http limitations and stateless decision, a lot of data is sent
back and forth during have/want negotiation for smart-http. I wonder
if we could implement the "long polling" scheme in a CGI program. The
program terminates HTTP requests and recreates a full duplex
connection for upload-pack to talk to the client. upload-pack falls
back to the normal mode, used by git:// and ssh://.

An example of this is BOSH [1]. From a quick glance it does not seem
to require any special thing, so it's unlikely to cause problems with
firewalls, proxies.. If this is implemented as cgi (instead of http
server), we'll need to save session infos somewhere. I suppose shm
with proper locking is enough.

[1] http://xmpp.org/extensions/xep-0124.html
-- 
Duy