From: Jonathan Nieder <jrnieder@uchicago.edu>
Date: Fri, 22 Aug 2008 06:38:02 GMT
Subject: Re: [PATCH 3/3] git-add --intent-to-add (-N)
Message-ID: <Pine.GSO.4.62.0808220119250.12851@harper.uchicago.edu>
In-Reply-To: <7vtzddd1z5.fsf@gitster.siamese.dyndns.org>

Hi,

Junio C Hamano wrote:

> Didn't I already tell you that you seem to have picked only one out of
> _three_ patch series?

I am using all three patches.  If you try ">a && git add -N a && git
commit" in an empty repo, you should get the same behavior (I checked
with commit 038a213^2, which is three commits ahead of master).  And
yes, I do understand where your suspicion came from.

But the reason for the behavior is that update_one in cache-tree.c
contains the test

	if (mode != S_IFGITLINK && !missing_ok && !has_sha1_file(sha1))

which fails for the empty blob in a new repo because, as I said, we
don't have that sha1 file.

I still wonder, do we want to pretend we have that object on disk
and proceed with the commit, or are the hardcoded objects only
supposed to be sufficient for in-core use?  If the former, I will
have to make some tests to be comfortable: are the objects properly
transfered to older clients without the hardcoded objects, etc.  But
I don't want to bother if that is not the intent.

Hoping that is clearer,
Jonathan