Dubious format-patch options

Git-ites,

Hmmm... There are a ton of dubious options being accepted by the "git format-patch" command due to it being a skinny front to "git log". Like, the -S and pick-axe crew:

Yeah, this is context-less, but my point is that we get totally different behavior from the -S"string" here:

    $ git log -SC alt~1..HEAD
    commit bca87c386c8868664e64d397108b8e6deda1e574
    Author: Jon Loeliger <jdl@jdl.com>
    Date:   Thu Jun 12 17:57:57 2008 -0500

	C
versus

    $ git format-patch -SC alt~1..HEAD
    0001-C.patch
    0002-D.patch
    0003-Z.patch
    0004-F.patch

Oh yeah.

And --color is a cool thing too. I mean, I was pleasantly pleased with:

    $ git format-patch -1 --stdout --color master

But I seriously doubt that we _really_ want to send a patch that has the embedded escape sequences in it:

    From: Jon Loeliger <jdl@jdl.com>
    Date: Thu, 12 Jun 2008 18:13:10 -0500
    Subject: [PATCH] F

    ---
     fileESC[m |    1 ESC[32m+ESC[m
     1 files changed, 1 insertions(+), 0 deletions(-)ESC[m

    ESC[1mdiff --git a/file b/fileESC[m
    ESC[1mindex 6ac15b9..65f21da 100644ESC[m
    ESC[1m--- a/fileESC[m
    ESC[1m+++ b/fileESC[m
    ESC[36m@@ -5,3 +5,4 @@ D
    ESC[m X
    ESC[m Y
    ESC[m Z
    ESC[mESC[32m+ESC[mESC[32mFESC[m
    -- 
    1.5.6.rc0.46.gd2b3

Oh yeah.  Another Vodka Tonic please while git-send-email that...

jdl

Jon Loeliger <jdl@jdl.com> writes:

Amusing ;-).

On Thu, 12 Jun 2008, Jon Loeliger wrote:

Hey, you asked for it, you got it. I hope you're not complaining.

		Linus

Linus Torvalds <torvalds@linux-foundation.org> writes:

Well, having said and heard all that, coming back to our more serious selves...

I think being able to do

    $ git -p format-patch --stdout --color -2 HEAD

makes some sense.  Being able to produce an unusable patch by saying

    $ git format-patch --stat old..

at the first glance is of quite dubious value, but even that would make sense as a good input source for "commit log automailer".

I think I know where Jon is coming from and where he wants to go. While I am somewhat sympathetic to the cause of adding some warning or safety valve to prevent nonsense option combinations from being given, I am not sure we can draw a line to classify options into black and white.

On Thu, Jun 12, 2008 at 08:01:45PM -0700, Junio C Hamano wrote:

I am against a safety valve here. For example, I have the following alias:

  format-patch --pretty=format:%s%n%n%b

which obviously does not generate a valid email, but which I use for pulling "how about this?" patches directly into existing emails (otherwise, I have to delete the header cruft myself). Yes, I _could_ do this with git-log, but format-patch has slightly different command line semantics. And it most clearly expresses what I want: the patch submission format, but with the header bits pretty-printed differently.

My point being that this flexibility _is_ useful, and I am not sure it is worth removing it in favor of people who get confused about why

  git format-patch --pretty=format:%s%n%n%b origin | git send-email

doesn't work.

-Peff