Re: [PATCH] Add -r flag to show-diff for diff-cache/diff-tree like output.

LT> Why not just make this the default? Who really cares about LT> show-diff?

Me, to care enough about it to send patches in ;-).

LT> So why not just make "rational" the standard format, and then make LT> "diff-tree-helper" warn about unmerged ("U") files?

LT> As far as I can tell, that is really what _everybody_ wants.

I do not think of a good reason to forbid people from getting a patch out of core GIT without going through the wrapper layer, and the code is already there. That said, I agree that scripts usually do not use patch generating form. I checked Cogito before doing the patch to make sure it is not affected by this, and my JIT core tools do not use it either.

I think making the diff-tree/cache compatible output default would be fine. The current diff-producing behaviour can be made into an option (yes, I want to keep it).

LT> And calling "-r" "rational", when it means "recursive" for LT> diff-tree and diff-cache doesn't sound rational to me.

Well, the truth is that -r does not stand for anything. It just is there to match useless -r to diff-cache (the change you made to it to accept the same command line parameters -r (and optionally -z) people would always give diff-tree. You can call it recursive if you want.

How about me doing the following and resubmitting?

 - Make this diff-tree/cache compatible output the default.

 - Take but ignore -r flag like diff-cache.

 - Add U warning to diff-tree-helper.

 - Add -p flag (patch) and have it cause the patch generating
   behaviour. 

Later I'll add -p flag to diff-cache and diff-tree, so the usage of these three commands match. That is:

    show-diff  -r (useless) -z           | diff-tree-helper -z
    diff-cache -r (useless) -z tree      | diff-tree-helper -z
    diff-tree  -r           -z tree tree | diff-tree-helper -z

    show-diff  -p [-r (useless)]
    diff-cache -p [-r (useless)]
    diff-tree  -p [-r]

With the GIT_EXTERNAL_DIFF envioronment variable, I suspect that wrapper scripts do not have to use the diff-tree-helper but directly use the -p form; but that will come later. Thoughts?