Hi Philippe,
> Add a __git_trailer_tokens function to list the configured trailers
> tokens, and use it in _git_commit to suggest the configured tokens,
> suffixing the completion words with ':' so that the user only has to add
> the trailer value.
Makes sense.
I've never dabbled in the completion scripts, so take the following with
some salt.
> +__git_trailer_tokens ()
> +{
> + git config --name-only --get-regexp trailer.\*.key | awk -F. '{print $2}'
> +}
The rest of this script uses `__git config` rather than `git config`.
The purpose of `__git` seems to be to respect options given on the
command line, so I think we would want to use it here.
These "." in "trailer." and ".key" will match any character. We also
don't anchor this at beginning and end. Maybe tighten this a bit and use
'^trailer\..*\.key$' to behave better in the face of config such as
this:
[strailer]
skeying = "s"
[trailerx]
keyx = "x"
Another thing. Consider such a config:
[trailer "q.p"]
key = "Q-p-by"
The "trailer.q.p.key" config above ends up completing as just "q"
because of how you use `print $2`. I see that `git commit --trailer=`
itself is fairly relaxed here, so `--trailer=q` effectively ends up
picking up "q.p" in the end. Tightening that is obviously out of scope
here and I have no opinion if the current behavior there is intended.
But maybe we should be a bit less relaxed here and complete to "q.p"? At
any rate, it gets weird when you also have "trailer.q.x.key" in your
config but we still just suggest the one "q".
I see your patch is in next, but maybe some of this tightening might be
worthwhile doing on top of it?
Martin
