On Sun, Aug 15, 2021 at 05:07:34PM -0600, jim.cromie@xxxxxxxxx wrote: > 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@xxxxxxxxxxxxxxxxxxxxx/ 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
