Local shallow git cloning with hard-links

Question

On my local filesystem, I want to be able to only clone the head of git repo (A) so no history comes along to a new git repo (B). But I want the benefits of hard-links for the files now in B to save on space. Is there a way to do this? Do the hard-links even help once repo A changes?

Thanks!

Answer 1

It appears that it is impossible to do local shallow cloning with hard links among object databases, at least as of git 1.7.12. git clone --depth 1 --single-branch explicitly warns that --depth is ignored in local clones, and to use file://. So you will need to choose between hard links and shallow cloning.

The hard links do work even when repository changes, at least for a time, because git adds new objects to new files, and never modifies existing files. However, it does occasionally repack the object database for efficiency, and I don't see how the hard links could be preserved then.

If you choose shallow cloning, you can create the clone with git clone --single-branch --depth 1 file://old_repo_dir options. I find it annoying that --depth 1 means 1 item of history, so you'll get not only the latest commit, but also its parent (or parents if it's a merge). The parent gets the commit message from the original repository, but the commit message lies because the commit in actuality contains the creation of the entire tree.

I prefer to start off with a single commit with a commit message of my choosing that creates the initial tree. This is obtained by first creating a new branch without history in the old repo, and then pulling that branch into the new empty repo. I tested this on a huge repo with a 664MB object database with 673k objects (the Emacs bzr repository converted to git). When the new repo received the pull, it had a 36MB object database with 3477 objects — so the excess content was apparently pruned. Here are the exact steps:

# at the old repo:
git checkout --orphan tmp-snapshot
git commit -m "Initial commit."

# at the new repo location:
git init
git pull OLD_REPO_DIR tmp-snapshot:master

# back at the old repo:
git branch -D tmp-snapshot   # no longer serves a purpose

Now the master branch of the new repo contains a single commit with a tree identical to the tree of the old repo, and without any history.