Re: What's cooking in git.git (Jun 2013, #07; Thu, 20)

From: Ramkumar Ramachandra <artagnon@gmail.com>
Date: Jun 22, 2013, 06:17 UTC
Message-ID: <CALkWK0=FaC9CqiTLE=UJA+Xg04=ZS45XddQ-RmPe_HSej2s5bA@mail.gmail.com>
In-Reply-To: <7vvc57kxxt.fsf@alter.siamese.dyndns.org>

Junio C Hamano wrote:

> Because the implementation is too ugly.

I know :) The only reason I sent it with code is because I didn't get any responses to an email without code. If you agree that it is a useful feature, we have to figure out a beautiful implementation.

Show 6 quoted lines

> I would however can imagine an alternative implementation which
> might be more palatable. It may go like this:
>
>         [alias "lgF"]
>                 command = "log --oneline --boundary --first-parent"
>                 help = "show the first parent chain, one line per commit"

I'm not sure what value this adds. If I ever forget what my alias is called, I `git rp --help` to get the expansion and then look up the manpage. It would be ideal if I could `man git lgF` though: I'm sick of having to type out `man git for-each-ref` everytime I need the manpage for fer (obviously an unscripted `man` won't work: `git help` will do the translation).

>                 completion = log

Again, unsure what value it adds. I already have plenty of aliases that complete fine. The completion only fails if I have an !-command; in that case, this solution is a hack: we should work towards not requiring an !-command in the first place.

> so that not just alias.c code can take notice of alias.lgF.command
> to expand it,

How? Fundamentally, alias_lookup_cb() is a fired off by the config-parsing infrastructure which calls a tolower() on everything: alias.c has control over nothing, unless we re-implement the entire config-parsing infrastructure specifically for aliases (Bad idea). I don't see how changing from alias.<name> to alias.<name>.command helps anything, especially when the other alias.<name>.* keys aren't even useful.

> but we can later extend it to help "git help lgF"

Yeah, this is a good idea.

> and
> bash/zsh completion (i.e. they would learn "lgF parameter would
> complete in a way similar to 'log'" from alias.lgF.completion).

Solved problem: see __git_aliased_command() in git-completion.bash.