将一些git commit移到新的仓库中

Question

I am using this rest-more repository, which contains many implementations of the web-api (such as Facebook or Linkedin). I forked that repository, and in my branch I added my own implementation of another web api.

However now I find that it is pretty hard to keep updated with the original repository. I rebase my branch every time the original repository changes, which is pretty tedious. So I want to split out my own implementation to a standalone repository.

All my commits are in my branch grouped together. Only one commit is touching a file from the original repository.

Is there an easy way to move my commits into a new repository? I want to do this so I can preserve my commit histories.

Answer 1

Easiest: just stop accepting updates from the original repository. Remove the remote repository from your .git/config file.

Otherwise, from my notes:

Say you want to move dir1 from repository A to repository B :

Make a clone of repository A . To be safe, we break the link between the clone and A , to make sure none of our changes accidentally get sent back.

git clone git://git.foo.com/ProjectA.git NewProject
cd NewProject
git remote rm origin

Strip out everything except dir1 .

git filter-branch --subdirectory-filter dir1

This will have made dir1 the new root, which you might want to undo:

mkdir dir1
mv * dir1
git commit -a

Now merge this repository into repository B

git clone git://git.foo.com/ProjectB.git ProjectB
cd ProjectB
git remote add repo-A-branch ~/NewProject
git pull repo-A-branch master
git remote rm repo-A-branch

Answer 2

I would recommend keeping a single repository , and creating a second branch for your changes . You can then cherry pick your commits to this new branch, to isolate your changes, and merge/rebase upstream changes and your changes into a third branch as appropriate to create releases of your code. Keeping it in one repository makes it far easier to stay in sync with the upstream development, and having your changes on their own branch allows you to keep them isolated. The third branch can serve as a "your changes applied to the upstream code" branch.

Answer 3

Assuming that you have strictly linear history of your commits (no merges), you can simply export all your commits as follows:

git format-patch <start>..<stop>

Where start and stop is range of commits you need ( start is commit you started from, and stop is probably current HEAD ).

This will create lots of files named NNNNN-commit-description.patch , NNNNN is number starting from 00001.

Then, create new empty repository and import these patches:

git init newrepo
cd newrepo
find <oldrepo>/*.patch | xargs git am

Also, instead of starting off empty repo, you may want to make first commit with some original files that you depend upon, such that all your patches can apply cleanly.

Answer 4

I first open a new repository:

git init new-repo

I then pushed my branch to that repository:

cd rest-more
git push ../new-repo/ my_branch

cd ../new-repo
git co my_branch

I created a new empty initial commit:

git checkout --orphan newroot
git commit --allow-empty -m 'root commit'

Then I rebase my branch to this starting point, use interactive flag to pick only my commits:

git rebase --onto newroot --root my_branch -i

Need to resolve conflicts on the way, and finally I end up with a repository with only my commits.

(There are probably improvements to be made. Feel free to suggest anything.)