From: Jonathan Nieder <jrnieder@gmail.com>
Date: Fri, 28 Jan 2011 17:13:59 GMT
Subject: Re: [PATCH v4] fast-import: treat filemodify with empty tree as delete
Message-ID: <20110128171359.GA26930@burratino>
In-Reply-To: <20110127204649.GB19378@m62s10.vlinux.de>

Peter Baumann wrote:
> On Thu, Jan 27, 2011 at 01:48:45PM -0600, Jonathan Nieder wrote:
>>> On Thu, Jan 27, 2011 at 12:07:49AM -0600, Jonathan Nieder wrote:

>>>> +	 empty_tree=$(git mktree </dev/null) &&
[...]
>>                               unless we hardcode the object name
>> (which I prefer not to do).
>
> Wny not? It *is* already hardcoded in the GIT source code (see
> grep -a1 EMPTY cache.h output).

I think it is okay for the git implementation to rely on an
implementation detail. ;-)  Likewise, t0000 checks that the empty tree
has id 4b825dc6.  Meanwhile I would like to see people's scripts
and other tests using the $(git mktree </dev/null) form, since it is
more self-explanatory and avoids hardcoding an implementation detail.

Of course this is not an absolute thing.

Hope that helps,
Jonathan

Further reading: t0000-basic.h --help:

	Note that this test *deliberately* hard-codes many expected object
	IDs.  When object ID computation changes, like in the previous case of
	swapping compression and hashing order, the person who is making the
	modification *should* take notice and update the test vectors here.

"Tips for Writing Tests" in t/README:

	However, other tests that simply rely on basic parts of the core
	GIT working properly should not have that level of intimate
	knowledge of the core GIT internals.  If all the test scripts
	hardcoded the object IDs like t0000-basic.sh does, that defeats
	the purpose of t0000-basic.sh, which is to isolate that level of
	validation in one place.  Your test also ends up needing
	updating when such a change to the internal happens, so do _not_
	do it and leave the low level of validation to t0000-basic.sh.