Re: [PATCH 0/2] Make gc a builtin.

From: Linus Torvalds <torvalds@linux-foundation.org>
Date: Mar 12, 2007, 19:14 UTC
Message-ID: <Pine.LNX.4.64.0703121202560.9690@woody.linux-foundation.org>
In-Reply-To: <20070312133612.GD4372@thunk.org>

On Mon, 12 Mar 2007, Theodore Tso wrote:

> 
> I'm not sure I understand the portability argument?  All of the
> platforms that git currently supports will handle shell scripts,
> right?

Git "supports" MinGW, or at least wants to. And yes, you can put bash in there, but we'd be *so* much better off if we had no shell scripting at all.

Another thing I find annoying (even as a UNIX user) is that whenever I do any tracing for performance data, shell is absolutely horrid. It's *so* much nicer to do 'strace' on built-in programs that it's not even funny.

It's also sad how many performance issues we've had with shell, just because even something really simple (like a few hundred refs) is just too slow for shell scripting.

> Heck, git-commit is still a shell script, and that's a rather, ah,
> fundamental command, isn't it?

Yeah, and that's probably my pet peeve. I'd love to see a built-in "git commit" and "git fetch". The "fetch--tool" thing in next gets rid of some of the latter (and apparently the worst performance problems), but it's sad how we have a really nice builtin "push", but our "fetch" is still mostly really hairy shell-code (not just "git-fetch.sh" itself, but "git-parse-remote.sh".

A gold star for whoever gets rid of any of of commit/clone/fetch or ls-remote

(ls-remote isn't that big or hairy, but I mention it because it's a user of "parse-remote", so making even just ls-remote built-in is probably going to help with fetch/clone eventually).

		Linus