Re: git format-patch -s enhancement

From: Jeff King <peff@peff.net>
Date: Aug 16, 2021, 16:36 UTC
Message-ID: <YRqUK3cRFJmANzDd@coredump.intra.peff.net>
In-Reply-To: <CAJfuBxxT_7weC8_O=KYScSbDcSeBdb3v5d_gtn-NXzW_fKLrsA@mail.gmail.com>

On Sun, Aug 15, 2021 at 05:07:34PM -0600, jim.cromie@gmail.com wrote:

Show 24 quoted lines

> git format-patch -s is sub-optimal :
> it appends the SoB,
> which falls after the snips
> ---
> changelog ...
> that the commit message may contain
> 
> 
> So it misfires on any maintainer scripts
> expecting the SoB above the 1st snip.
> 
> The workaround is manual SoBs above any snips.
> 
> I note this in -s doc,
> 
>            Add a Signed-off-by trailer to the commit message, using
> the committer identity of yourself.
>            See the signoff option in git-commit(1) for more information.
> 
> "trailer" is really "document current working behavior"
> (normative docu-speak, so to speak;)
> 
> Ideal behavior is to find 1st in-body  --- snip
> and insert there

The big disconnect here is that "---" snip lines are not meant to be meaningful within commit messages themselves. They are part of the process of sticking a commit message into an email. So format-patch and git-am know about them, but "git commit" for example doesn't.

So "git commit --signoff" probably shouldn't take them into account when deciding the end of a commit message. The user might or might not have meant "---" to be syntactically meaningful, depending on whether they plan to send the message with format-patch (and changing the behavior now is questionable).

Doing so with "git format-patch --signoff" is a slightly different question. The current behavior is working as intended, in the sense that it signs off just as "commit -s" would, and then separately sticks the result into the email. The fact that "---" in the commit message is indistinguishable from the ones added by format-patch is mostly an accident.

That said, it's kind of a useful accident for some workflows, exactly because you can carry these non-commit-message notes inside the commit message. And since we know how any in-commit-message "---" will be treated by git-am on the other side, it might be reasonable for format-patch to start considering them to be syntactically significant.

So I guess I would disagree that it's a bug exactly, in that the workflow you're advocating was never meant to be supported. But I don't see any reason we couldn't be a little friendlier to it, if somebody wanted to teach format-patch to do so.

An alternative workflow would be to use git-notes to attach the changelog data to the commit. Those are shown after the "---" by format-patch already. Unfortunately, keeping them up to date is kind of annoying. Ages ago, I had a patch to let you modify them while editing the commit message, which makes it pretty seamless:

  https://lore.kernel.org/git/20110225133056.GA1026@sigill.intra.peff.net/

I carried the patch in my local build for a while, but never really ended up using it. So I never polished it further. But I think it's still fundamentally a reasonable idea, if somebody is interested in carrying it forward. If so, here's the version I've been rebasing forward over the years:

  https://github.com/peff/git jk/commit-notes-wip

but it doesn't seem to actually pass its own tests anymore (so it may or may not be a helpful starting point. ;) ).

-Peff