Git Tips

May 25, 2019 811 words 4 minutes

Contents

Basics

Git - more like a mini filesystem - thinks about its data more like a stream of snapshots, unlike other delta-based version control systems (Subversion.)
Unlike CVCS, nearly every operation in Git is local.
Git has integrity. Everything is checksummed. It uses SHA-1 to store everything in it’s DB.
Git generally only adds data.
Git has three stages: modified, staged, and committed. And it has three main sections:
- The working tree: a single checkout of one version of the project. Files are pulled out of the compressed DB in the Git directory and placed on the disk for modifications.
- The staging area: - aka index - a file that stores info about what will go into the next commit.
- The Git directory: - .git - stores the metadata and object DB.

git commit --amend:

1
2
3


git commit -m "Initial commit"
git add forgotten_file
git commit --amend

Then we’ll have one single commit.

When we want to ignore certain files or folders, create a .gitignore file and add the file/folder name inside:
1 2 3

file.txt # a single file folder/ # a folder named `folder` *.txt # all txt files
If a file is already tracked, use git rm --cached foo.txt to remove it first from the git repo.

When we want to get a simpler version of the log, use git log --pretty=oneline.

Compare the diff between two branches:

1
2
3


git diff branch_1..branch_2
# Or if already in branch_1:
git diff ..branch_2

Compare the commits between two branches:

1
2
3


git log master..branch-X
# Or if already in branch-X:
git log master..

Use git checkout . to revert to the last commit before adding the new changes.
Use git checkout plus the first 6 characters of the reference ID to check out the old commits. This enters the detached HEAD state. And it’s best not to make changes when checking out old commits. Use git checkout master to go back the master branch.
Use git reset --hard plus the first 6 character of the reference ID to reset the project to the old commit.
git reset HEAD README.md to unstage the file.
git checkout -- README.md to discard the changes for the file.
git checkout ffffff -- foo.txt bar.txt reverts the two files back to commit ffffff.

git remote -v to see all the remotes verbosely with URLs. origin is the default name.
git pull = git fetch + git merge.
git push <remote> <branch>: git push origin master.
git remote show <remote> to see details of a remote.
git remote rename <old_remote> <new_remote> and git remote remove <remote>.

git branch <branch_name> creates a new branch. git checkout <branch_name> moves HEAD to the branch. The shorter version is git checkout -b <branch_name> to do both at the same time.
git log --oneline --decorate shows the branch pointers.
git log --oneline --decorate --graph --all shows the divergence.
Branches are cheap since they are essentially a file that has the 40-character checksum of the commit pointed to.
git checkout master + git merge hotfix will fast-forward (or use the recursive strategy) the master branch to the match the hotfix branch. Then git branch -d hotfix deletes it.
In case of merge conflict, we need to choose one side or merge the contents ourselves. Then add and commit the file again.
git branch -v shows the last commit of each branch. git branch --merged and git branch --no-merged shows the branched that are already merged to the current branch or not yet respectively. Or git branch --no-merged master.
git checkout experiment + git rebase master + git checkout master + git merge experiment gives a fast-forward merge.
git stash is a convenience tool. But since branching is cheap, we can always branch and then delete it.
git branch -a to see all the branches.
git fetch and git checkout -b test remotes/origin/test creates a local test branch and connects it with the remote branch.