Re: [RFC] series headers

From: Theodore Tso <tytso@mit.edu>
Date: Jul 10, 2007, 13:24 UTC
Message-ID: <20070710132401.GJ2343@thunk.org>
In-Reply-To: <7vejjgsq3y.fsf@assigned-by-dhcp.cox.net>

On Mon, Jul 09, 2007 at 11:57:21PM -0700, Junio C Hamano wrote:

Show 5 quoted lines

> I suspect that temptation to touch rev_info.commit_format arises
> purely because you are thinking about making 0/N a (perhaps
> fake) commit.  I do not see a point in that.
> 
> What is the workflow?

In general a patch series goes through multiple cycles of improvements, where people send it out for review/comment, and then it gets fixed up, etc. etc. So you don't want to just do this:

Show 10 quoted lines

>  $ work work work, commit commit commit, reorder and perfect
>  $ git format-patch --with-cover origin..HEAD
>  ... which notices --with-cover, and perhaps does
>  ... $ git-shortlog origin..HEAD
>  ... $ git diff --stat --summary origin..HEAD
>  ... $ echo "*** BLURB HERE ***"
>  ... to create 0/N which it did not do so far in
>  ... 0000-cover-letter.txt
>  $ $EDITOR 0000-cover-letter.txt
>  $ git-send-email 0*.txt

Because you'll be sending out the 0000-cover-letter.txt multiple times, refining it (and the patches) as you go along.

Some people will use quilt for this process; others will use guilt or stg in a seprate branch. The way I would probably solve it is by making the Patch 0/N cover letter be a empty commit with no changes, and simply storing the comment in the commit log. Yes, that means you have to go back and edit the 0/N message after you've finished doing your patch series, but the general work flow will require you to go back and clean up patch 3/27 given some comment that Cristoph Hellwig made, so you'll be going back and forth using some tool like guilt or stg anyway. So going back to edit the 0/N cover-letter is not a big deal.

The one advantage of putting the 0/N cover-letter at the end is that it makes it easier to drop it if the patch series ends up being pushed via git, but again, given that the patch series is going to be reworked multiple times during its life cycle, reworking it one last time to drop the cover letter doesn't seem like a major issue --- and if it is ends up getting applied by the maintainer via e-mail and git-am, the cover letter will get dropped by the maintainer simply not including it in his mailbox of patches to apply.

	  	     	    	       	       - Ted