No working tree repository

I'd like to keep a local copy of the kernel repository mainly for searching the history. The working tree uses up nearly 400mb of disk space that I do not normally need. How can I remove the working tree, and keep only the packed repository so I can review the logs, and only check out a working copy if I actually want to edit or compile the sources?

Phillip Susi <psusi@cfl.rr.com> writes:

This is called a "bare" repository. Now, you have the keyword to RTFM ;-).

You can do something like

mv linux-repo linux-tree cd linux-tree mv .git ../linux-repo cd ../linux-repo git config core.bare true

(and optionnaly remove linux-tree/)

or you could have used "git clone --bare" when you cloned the repository.

On 6/15/2010 2:10 PM, Matthieu Moy wrote:

Ahh, that's the magic word I was groping for.

It seems that --bare on clone will prevent the checkout of the local working tree. If I decide I do want the sources today I can just check them out, but what is the proper way to do the reverse? I was thinking something like somehow empty the index file then do a git-reset or git-checkout-index to clean up the working tree to match the empty index, but I can't figure out how to empty the index.

On Tuesday, June 15, 2010 12:33:32 Phillip Susi wrote:

I often want to have a regular tree, but only sometimes check it out, and I want it to be in-place (e.g. not a bare repository and a separate work- tree).

What I do is I make a local "empty" branch with no files or ancestory. For this I have a "git-emptybranch" script (attached), but you can easily do this by hand.

Then when I want files, I do:

$ git checkout master  ## or whatever branch

When I'm done, I do

$ git checkout empty

and all the files go away.

Even with empty checked out, I can fetch, push, run gitk --all, etc.

This works well for me.

Phillip Susi <psusi@cfl.rr.com> writes:

rm -fr * # use at your own risk rm -f .git/index

Then, you can get back a tree with "git checkout" or "git checkout HEAD -- .". The advantage of this solution over the empty-branch solution is that although you destroyed your index, HEAD is still there so "git log" and friends will show you where you are.

Am 6/15/2010 20:33, schrieb Phillip Susi:

git rm -r .

-- Hannes

Phillip Susi:

Then I would recommend creating another clone. I usually do a

   git clone --reference path/to/bare server:path/to/origin newcheckout

to create clones with working trees, without having to pay the penalty of having (yet) another copy of the objects under .git (and with .git in the order of 400 megabytes, and several working copies, it does pay off).

Maintaining this is a lot easier than constantly adding and removing a working tree from the repository.

If you only need a single file, "git show commit:path" should work, even on a bare repository.

Matthieu Moy <Matthieu.Moy@grenoble-inp.fr> writes:

Or shorter:

$ git clone --bare linux-repo

(which creates linux-repo.git and uses hardlinks, so it won't waste space)

Andreas.