From: Jonathan Nieder <jrnieder@gmail.com>
Date: Sun, 04 Jul 2010 00:35:43 GMT
Subject: Re: [WIP/PATCH] Teach fast-import to print the id of each imported commit
Message-ID: <20100704003543.GA25858@burratino>
In-Reply-To: <1278201760.9701.39.camel@denix>

Sam Vilain wrote:

> It should be very simple to implement --batch-verify as well, which is
> essentially the same thing, just only printing the header and not the
> contents.
> 
> How hard do you think it would be to implement a 'rev-parse' command?
> Perhaps you can see where this is going... :-)

Well, I think I understand what you are saying, but it’s not quite
like that.  fast-import is already doing a lot of work to keep track
of this collection of objects in a pack that is not yet accessible to
git because its index has not been written yet.

svn-fe needs those objects (well, it at least needs a few blobs when
it learns to apply a delta to the latest revision).  So what to do?

One option is to make fast-import into a library, or copy the code we
need into svn-fe.  And that would be possible!  But it would make
svn-fe much more complicated, and there would be this second
fast-import interface to maintain.

Or svn-fe could keep the full text of all files from the latest
revision in a temporary directory.  To get blobs for an _older_
revision, it would still need to ask git, but such copy operations are
uncommon enough that the time to request a checkpoint and wait for
things to settle could be tolerable.  The main downside: in a huge
repository like ASF’s, that’s a good amount of wasted space.

Or one can teach fast-import to supply exactly what we need from it,
which is the raw content of some blobs (and perhaps trees) it has
written to pack and not indexed yet.

Jonathan