Re: git diff-files weirdness (bug?)

From: Linus Torvalds <torvalds@linux-foundation.org>
Date: May 23, 2008, 16:50 UTC
Message-ID: <alpine.LFD.1.10.0805230942480.3081@woody.linux-foundation.org>
In-Reply-To: <alpine.LFD.1.10.0805230906110.3081@woody.linux-foundation.org>

On Fri, 23 May 2008, Linus Torvalds wrote:

> 
> And I think "setup_diff_no_index()" is broken exactly because it sees that 
> '--' and then totally ignores it and says "if I see two paths, we'll say 
> it's the non-index case".

Actually, I think it's broken not so much because it ignores '--', but because it's just way too eager to do the "non-git" case.

Basically, it is trying to act like a traditional "diff", but in doing so it disables the normal "git diff" semantics. Which is incorrect, when you actually _want_ the normal "git diff" behaviour.

So I would suggest that we make it *much* harder to trigger the "make it act like a traditional 'diff'" thing.

I would suggest that we *not* invoce the traditional 'diff' behaviour when:

 - we're called as "git-diff-files". That's clearly a git thing. Don't try 
   to make it act like an external non-git diff. Only do the special case 
   for plain "git diff" itself.

 - even for plain "git diff", make it much harder to trigger non-git 
   behaviour. Don't do it if the files don't exist. Don't do it if there 
   is '--' there.

In fact, maybe we should remove that thing entirely, or *require* a flag to enable it (at least if we're in a git directory). It's a cute hack, but when the cute hack actually makes it impossible to do certain real git operations, it's a cute hack that is detrimental.

		Linus