Question

While helping a friend with a git problem today, I had to introduce a
branch that needed to be totally separate from the master branch.
The contents of this branch really had a different origin from what
had been developed on the master branch, but they were going to be
merged into the master branch at a later time.

I remembered from reading John Wiegley’s Git from the bottom
up how
branches are essentially a label to a commit that follows a certain
convention and how a commit is tied to a tree of files and, optionally
to parent commits. We went to create a parentless commit to the
existing repository using git’s plumbing:

So we got rid of all files in the index …

$ git rm -rf .

… extracted directories and files from a tarball, added those to the
index …

$ git add .

… and created a tree object …

$ git write-tree

(git-write-tree told us the sha1sum of the created tree object.)

Then, We committed the tree, without specifying parent commits…

$ echo "Imported project foo" | git commit-tree $TREE

(git-commit-tree told us the sha1sum of the created commit object.)

… and created a new branch that points to our newly created
commit.

$ git update-ref refs/heads/other-branch $COMMIT

Finally, we returned to the master branch to continue work there.

$ git checkout -f master

This seems to have worked as planned. But this is clearly not the kind
of procedure I would recommend to someone who is just getting started
using git, to put it mildly. Is there an easier way of creating a
new branch that is entirely unrelated to everything that has happened
in the repository so far?

Answer 1

There is a new feature (since V1.7.2) which makes this task a little more high-level than what’s in any of the other answers.

git checkout now supports the --orphan option. From the man page:

git checkout [-q] [-f] [-m] --orphan <new_branch> [<start_point>]

Create a new orphan branch, named
<new_branch>, started from
<start_point> and switch to it. The
first commit made on this new branch
will have no parents and it will be
the root of a new history totally
disconnected from all the other
branches and commits.

This doesn’t do exactly what the asker wanted, because it populates the index and the working tree from <start_point> (since this is, after all, a checkout command). The only other action necessary is to remove any unwanted items from the working tree and index. Unfortunately, git reset --hard doesn’t work, but git rm -rf . can be used instead (I believe this is equivalent to rm .git/index; git clean -fdx given in other answers).

---

In summary:

git checkout --orphan newbranch
git rm -rf .
<do work>
git add your files
git commit -m 'Initial commit'

I left <start_point> unspecified because it defaults to HEAD, and we don’t really care anyway. This sequence does essentially the same thing as the command sequence in Artem’s answer, just without resorting to scary plumbing commands.