easy repository initialization that turned out not to be that easy

I stumbled about some tasks that I expected to be easy but turned out not to be that easy ...

(with git version 1.5.3.rc2.29.gc4640f)

1) How can I switch a private repository to shared as if it
was created by 'git-init --shared' in the first place?

Just executing 'git-init --shared' in the repo doesn't adjust permissions as needed.

Cloning it to shared doesn't work either because the meaning of 'git-clone --bare --shared' is different from 'git-init --shared'.

So I created a fresh repository

    mkdir shared.git
    cd shared.git
    git --bare init --shared

and tried to naively fetch all by executing

    git fetch ../private.git

Hmm... it does a lot of things, but I ended without branches. fetch apparently has nothing like '--all'.

... but push has, so I finally did

    cd ../private.git
    git push --all ../shared.git
    git push --tags ../shared.git

Note, 'git push --all --tags' refused to work, so I needed to push branches and tags separately.

Is it really that hard, or did I miss something?

2) How can I set up an empty bare repository that shares
objects with an existing repository? I'd like to do that
to save space and bandwidth.

I started to search for something like 'git init --reference' or 'git init -l', similar to what 'git clone' provides.

I don't remember all approaches I tried, but finally I used

    git clone --bare -l existing.git new.git
    cd new.git
    mv refs/head/master .
    rm -rf refs/heads/* refs/tags/*
    mv master refs/heads

I remember that I needed to leave the master in place because otherwise a 'git push' didn't recognize that objects are already present and transferred everything, which is quite annoying over a slower network connection. With master in place the first push to new.git worked as expected. Only the additional objects were sent.

Did I miss something?

	Steffen