Re: [PATCH] git-diff: Add --staged as a synonym for --cached.

From: Johannes Schindelin <johannes.schindelin@gmx.de>
Date: Nov 12, 2008, 11:10 UTC
Message-ID: <alpine.DEB.1.00.0811121205100.30769@pacific.mpi-cbg.de>
In-Reply-To: <20081112083353.GB3817@coredump.intra.peff.net>

Hi,

On Wed, 12 Nov 2008, Jeff King wrote:

Show 18 quoted lines

> On Mon, Nov 10, 2008 at 11:04:42PM -0500, Avery Pennarun wrote:
> 
> > Speaking just for myself, I would find this all a lot less confusing 
> > if "staged" were a refspec of some sort, not an option at all.
> > 
> >    git diff HEAD..STAGED
> >    git diff STAGED..WORKTREE
> >    git grep pattern STAGED HEAD sillybranch WORKTREE ^ignorebranch --
> > path/to/files
> 
> I agree that such a thing is reasonably intuitive. I have thought about 
> "magic" refspecs before; my local git has an "EMPTY" refspec which 
> points to the empty tree for diffing. However, that was trivial to 
> implement (since it turns into a sha1), and yours is very hard (since 
> you will have to pass these "pretend" objects around).
> 
> So I think it is a neat idea, but I am not volunteering to work on it.
> :)

Just in case anybody thought about creating tree objects on the fly and use their SHA-1s: that won't fly, as you can have unmerged entries in the index. So STAGED.. is a _fundamentally_ different thing from HEAD^..

Maybe we could play tricks with a special staged_commit (pretending to be a commit with SHA-1 000000... so that git log STAGED.. would do the same as plain git log, the rationale being that STAGED is no commit, so ^STAGED should be a nop).

"git diff" would then have some special handling for the case that there are exactly two revs, exactly one of them negative, and exactly one of them being the staged_commit, passing off to the respective diff backends.

Ciao, Dscho