From: Jeff King <peff@peff.net>
Date: Thu, 10 May 2012 14:29:20 GMT
Subject: Re: [PATCH 0/4] git-notes ui fixes regarding non-blobs notes
Message-ID: <20120510142920.GA14711@sigill.intra.peff.net>
In-Reply-To: <1336658701-9004-1-git-send-email-pclouds@gmail.com>

On Thu, May 10, 2012 at 09:04:57PM +0700, Nguyen Thai Ngoc Duy wrote:

> Thanks Jeff and Junio for all the comments. This is basically what you
> (and I) agree in v1's thread. The first patch is a new one, to avoid
> git-notes from converting everything to blobs. I don't restrict the
> "show" case either because git-notes uses "git show", which is
> capable of displaying all kinds of objects.

Nice, this is much better than the previous round.

I was surprised by the "git show" thing you mentioned, because I checked
before that "git notes add -C HEAD^{tree}; git notes show" generated
garbage. But the problem is that "-C $tree" was more broken than we
realized. It actually copied the tree contents into a blob object,
giving it a new sha1:

  $ git rev-parse HEAD^{tree}
  f799843170cdddbbdfec446e934bbccf94e0d2a7
  $ git notes add -C HEAD^{tree}
  $ git notes list HEAD
  587521811b7a5b84ed7cb48d11f9321ddcefd337

So it really was totally useless, and nobody could possibly have been
using it to store a real tree object, experimental or otherwise. But I
still like these patches better than just disallowing it, because they
allow people to experiment with the idea.

It's a little odd that we rewrite the object in the "-C" case at all; we
should never even need to open the object. I guess it is just to make
the code paths between "-c" and "-C" simpler.

-Peff