How do I split a Git repository cleanly?

This question took me part of the way, but I'm still missing something.

I have a single-user repository (so no problems with rewriting history vs push/pull) where I have developed product code and support code and documents together.

After a year-or-so's work, I have finally seen the light, and want to split my repo in two. This is what I tried:

1. Create two new directories under the project root; let's call them Project/Code/ and Project/Support/
2. git mv all the Code stuff into Code/
3. git mv all the Support stuff into Support/
4. copy and modify .gitignore appropriately
5. commit everything

Now, when I run what I thought the right command:

git filter-branch -d /cygdrive/z/temp/ --subdirectory-filter Code/ -- --all

I just get a detached branch with a single commit, my original branch is still there (as original/refs/head/brent), and all the other branches are as they were.

What I really need is something to rewrite history as if the Code/ directory was "always there", and the Support/ stuff wasn't, and vice versa for the other side.

I'm a relative Git noob, so I'm sure I've missed some important concept -- please fill me in.

---

OK, I've done that, and it appears to do what I need very well, for which much thanks. I have a couple of supplementaries:

1. I added a -r to allow removal of complete directories. Is this correct? The actual command I used was:

   git filter-branch --index-filter \

   'git rm -r --cached --ignore-unmatch ' \

   -- --all

2. The repository now has about twice as many commits as I expected. These appear in gitk as being chained back from thing with names like "original/refs/heads/TestSys".

I assume that these are the original commit tree; how do I get rid of them? They don't show up in fsck, and neither prune nor gc --aggressive gets rid of them.

开发者_高级运维

---

Uh-oh, just read the fine manual; going to try git clone to prune the cruft.

---

OK, now I've done the clone, and the repository has shrunk back to its original size. The clone only seemed to copy branch information for the current (source repo) HEAD, and I had to set up the other branches manually with this:

for n in `git branch -r|grep -v 'HEAD'` ; do git branch ${n/origin?/} $n ; done 

Is there a simple way to do this at clone time?

Plus this: I don't wish to seem ungrateful, but there are now "empty" commits which used only to touch files that are now gone.

Is there a simple way to eliminate these?

Really grateful for your help and patience.

---

I suppose moving the files into the subdirectory is the problem – git doesn't follow them before the move. I would suggest cloning two new repositories from the old one and running following in each of them:

git filter-branch --index-filter 'git rm --cached --ignore-unmatch <list of files that shouldnt be here>' -- --all