Re: [PATCH 1/1] commit: add support to provide --coauthor

[Jeff King <peff@peff.net>]

On Thu, Oct 10, 2019 at 10:49:23AM +0200, Toon Claes wrote:

> > Yeah, I'd agree that we should start first with a generic trailer line.
> 
> IIUC you are suggesting something like this?
> 
>  git commit --trailer="Co-authored-by: <coauthor>"
> 
> I really want to consider this, but I do not understand how that improves
> the user experience compared to adding that trailer manually when typing the
> commit message in your $EDITOR?

I agree that it's a lot worse to type than "--coauthor". And I don't
really have a problem with us ending up with "--coauthor". My reasoning
in starting with a generic form was mostly:

  - by having _any_ way to do this on the command-line, it makes it
    possible to use in aliases, etc.

  - having a generic form, even if we later add syntactic sugar on
    top, lets people easily experiment with their own trailers

> > There might be some advantage to building trailer-specific intelligence
> > on top of that (for instance, it would be nice for coauthor trailers to
> > expand names the way --author does). But that can come after, and might
> > not even be in the form of specific command-line options. E.g., if the
> > coauthor trailer could be marked in config as "this is an ident", then
> > we we would know to expand it. And the same could apply to acked,
> > reported, etc.
> 
> Wouldn't making it a generic --trailer option make this more complex? I can
> image users might even want to use the --trailer argument to indicate which
> issue the commit closes:
> 
>  git commit --trailer="Closes: $BUGNUMBER"
> 
> So, how can we make the config understand it has to expand Co-authored-by
> and not Closes?

We already have config blocks for specific trailers to describe how they
should be parsed or added. I was thinking that you'd set an option like
"trailer.co-authored-by.ident" to "true". And possibly that could be
used in other places, too (e.g., interpret-trailers code could make sure
it's syntactically valid, but I didn't really think through the
implications there).

And of course we could bake in the defaults for particular trailers if
we wanted to (I think we already do for trailer.signoff.*).

> > I wonder how we are supposed to use this trailer in the Git project, in
> > particular in combination with Signed-off-by. Should all (co)authors
> > sign off as well?  Or will Co-authored-by imply Signed-off-by?
> 
> For this purpose I think it's useful git understands what "Co-authored-by"
> means, so when you run:
> 
>  git commit -s --coauthor=<coauthor>
> 
> The following trailer will be added:
> 
>  Co-authored-by: <coauthor>
>  Signed-off-by: <author>
>  Signed-off-by: <coauthor>
> 
> So I'm still pro of adding a --co-author option, but I do understand the
> concerns to avoid adding an option for all the possible trailers found in
> the link above.

Yes, I agree that ordering and de-duplication rules are useful, too.
Some of that can be expressed already in trailer.* config, but I don't
know if it would be capable enough to do everything you want (though
again, it would be really nice to _make_ it capable enough so that other
types besides co-authored-by can make use of them).

I don't have a hard belief that we have to do it that way (generic
before specific), and I can believe that when you get down to the
details that it might be hard to express some of this stuff in config
rather than C code. But I think we should at least take a look at
whether it's possible, because the benefits of having a generic solution
are nice.

-Peff