Re: [PATCH 3/6] git-gnu-progs-Makefile: git Makefile update

From: Junio C Hamano <junkio@cox.net>
Date: Jul 11, 2005, 20:14 UTC
Message-ID: <7virzhrtfy.fsf@assigned-by-dhcp.cox.net>
In-Reply-To: <42D2CBA2.8060705@yahoo.com>

Bryan Larsen <bryanlarsen@yahoo.com> writes:

> Last night, I couldn't think of alternatives to these, but I obviously
> didn't try very hard.  xargs -r can probably happen via a temporary
> file and cp -u can probably be simulated using rsync.

The only user of "xargs -r" in the Linus GIT is git-prune-script which tries not to run "rm -f" with an empty argument list, like this:

    git-fsck-cache --cache --unreachable "$@" |
    sed -ne '/unreachable /{
        s/unreachable [^ ][^ ]* //
        s|\(..\)|\1/|p
    }' | {
            cd "$GIT_OBJECT_DIRECTORY" || exit
            xargs -r $dryrun rm -f
    }

Not tested on a BSD, and it is probably as ugly as it can get, but we could:

    {
        echo 'unreachable nosuch/file';
        git-fsck-cache --cache --unreachable "$@" 
    } |
    sed -ne '/unreachable /{
        s/unreachable [^ ][^ ]* //
        s|\(..\)|\1/|p
    }' | {
            cd "$GIT_OBJECT_DIRECTORY" || exit
            xargs $dryrun rm -f
    }

The only user of "cp -l" in the Linus GIT is git-clone-script local optimization. I could revert it to the version that I originally sent to the list, which uses cpio -pld, if your cpio groks that flag.

I do not speak for Pasky, but to me "cp -u" sounds just like an optimization, so maybe defining CP_U='cp -u' and detect missing support at config time and falling back on the simple "cp" would be an option?

GNU "cp -a" states that is the same as "-dpR" (never follow symlinks, preserve link, mode, ownership, and timestamps), so that can be rewritten as a shell function in cg-Xlib that is called say cg_copy_tree, whose implementation runs two tar processes piped together when "cp -a" is not available. Using a tarpipe unconditionally is also fine.