Notes and Tips for Git

$HOME/.gitconfig my global .gitconfig

• Run git config WITHOUT --global, or
• Edit path/to/git_repo/.git/config instead of $HOME/.gitconfig.
• You may want to set the global ones to the most conservative values (or leave them empty), for example:

  # global/default
  git config --global user.name "lgfang"
  git config --global user.email "no-spam@lungang.cn"
  # project specific/private
  git config user.name "myhandle"
  git config user.email "mymail@mycompany.com"
  

In $HOME/.gitcofig, add lines in the "alias" section. For example:

[alias]
    topo = log --graph --pretty=format:"%d" --simplify-by-decoration --all

For example:

git remote set-url origin https://username@github.com/...
git config --global credential.helper "cache --timeout=3600"

Put file name globs (one per line) into path/to/git_repo/.gitignore, for instance:

*~
*.pyc

my global .gitignore

To push/fetch a project after a proxy server.

# Examples of setting proxy
git config http.proxy http://proxyuser:proxypwd@proxy.server.com:8080
git config http.proxy localhost:8888
# To clear
git config http.proxy ""

I defined two aliases ("hp" and "nohp") for the commands, refer to my configure file.

• "pull" is basically a combination of "fetch" and "merge". It will download all, but only merge current branch.
• "pull" is NOT recommended, use "fetch" + "rebase" (or the equivalent command pull --rebase) instead.

All branches, tags are NOT pushed by default.

• For tags

  # push tag "mytag"
  git push origin mytag
  # push all tags
  git push --tags
  

• For branches
  1. "fetch" always retrieve all stuff (both branches and tags) from remote.
  2. All "tracking branches" are pushed by default.
  3. "fetched" branches are all tracking branches.
  4. Locally created branches become tracked ones (and are pushed) by the following commands.

     git push -u origin branch_name
     # Or
     git push -u origin --all
     

http://git-scm.com/book/en/Git-on-the-Server-Getting-Git-on-a-Server

If a handful of people are working on the same project, chances are that you want a git repository on a server which everyone can pull and push.

The major characteristic of this kind of respositorise is they contain no working directory.

To set up this, just setup a "normal" repository and then clone it with 'bare'.

cd ~/tmp/autoit && git init && cd ~/.repo git clone --bare ~/tmp/autoit auto.git

# later on, others can on the same machine
git clone file:///lgfang/.repo/auto.git
# or, on other machines with ssh to the_host
git clone login@the_host:///lgfang/.repo/auto.git

It complains "bash: git-upload-pack: command not found" because There is less "env" when running commands remotely using ssh non-interactive session.

Solution given in above link as well.

Sometimes, we don't care the past of a project. Since each commit creates new copies of corresponding files, we want to clone only commits we want so that we not only get a cleaner history but also save a little bit disk space. This is achieved by --depath N.

git clone --depth 1 <remote repository>

git clone -b branch_name --single-branch <remote repo> <local dir>

git archive -o archive.zip HEAD

This one makes a zip file content of which does not contain ".git" directory. Another way is to clone a single branch without history. I.e.

git clone -b branch-name --single-branch --depth 1 <remote repo> <local dir>

Example:

git branch -d foo
git push origin --delete foo

Example:

git branch -m foo bar

To rename current branch, simply:

git branch -m bar

You may want to create an alias for it.

git log --graph --pretty=format:"%d:%h" --simplify-by-decoration --all

In short, use git stash.

1. git stash
2. git co another_branch
3. do some work there
4. git co master
5. git stash pop

Or, as a more complex scenario, if both branches use stash, specify stash explicitly.

git stash list
git stash pop stash@{1}

Note that, stashing increases size of ".git" as well until a garbage collection run (either automatically or manually).

• "add" and "commit"

  git add file1 file2
  git add file3
  git commit
  

• Changes made after git add are NOT committed

  Hence you usually "add" right before "commit".

• Commit without "add"

  In fact, For files already being tracked, you can do "add" and "commit" in one step. Just run the following command.

  git commit file1 file2 ...
  # Or, use the brave "commit -a"
  

Simply checkout the file again, i.e.

git checkout -- file_name

If you already run "git add", then you have to "reset" first.

git reset file_name
# Or just "git reset"

I.e. delete/undo the latest commits.

git reset --hard HEAD~1

I.e. undo a previous commit but keep commits newer than it.

git revert HEAD~1
git revert e0c0698

With two-stage commit (git add), you can stage your changes without having to commit them.

1. You are working on a complicated problem.
2. You make some progress, then you git add to stage your code.
3. You continue and make more changes.
4. You find the changes are inappropriate.
5. You git checkout to discard the inappropriate changes and restart from where staged with another try.
6. After several rounds, all done.
7. You commit your code.

When a conflict happened during pulling, there are some "tricky" things to manually fix the conflict and check-in.

1. You cannot git commit a_file_with_conflicts, you have to git add all_files_with_conflicts and then git commit. The rational is that, in this case, you should commit all or none in this case.
2. Alternatively, git commit -a should combine those two steps (add and commit) for you.

git add -p file_name

Refer to tip 6 of http://www.sitepoint.com/10-tips-git-next-level/

You can specify multiple commits at a time.

git show b64dd26
git show HEAD~3
git show HEAD~1 HEAD~2

This works for either another version or another branch.

git show hash#:path/to/that/file/from/root/of/git/repo
git show branch:path/to/that/file/from/root/of/git/repo

or between certain commits and current status

git diff HEAD~2
git diff HEAD~1..HEAD~3
git diff d151d9d..9a72743
git diff branch1..branch2 [path/to/afile]

Like cvs annotate

git blame file_name