From: Jeff King <peff@peff.net>
Date: Sun, 21 Jan 2007 22:01:15 GMT
Subject: Re: [PATCH] prune: --expire=time
Message-ID: <20070121220114.GA24729@coredump.intra.peff.net>
In-Reply-To: <7vejpo39zg.fsf@assigned-by-dhcp.cox.net>

On Sun, Jan 21, 2007 at 03:17:07AM -0800, Junio C Hamano wrote:

> If we wanted to apply this grace period conservatively,
> protecting young objects is not enough.  You need to protect
> everything they refer to as well.  In the above scenario, you

That's not sufficient either. You might not _have_ the young objects
yet, think the blob is dangling, and delete it. Meanwhile, the tree that
references it arrives. IOW,
  1. blob B arrives, but already exists
  2. prune deletes unreference and old blob B
  3. tree T arrives, referencing blob B
I think this might be safe if you add objects in a top-down way (i.e., T
before B). However, that doesn't make sense for the commit operation, in
which you add blobs (with git-add), and then eventually construct a
tree.

-Peff