Table of Contents

Resources

• My Github specific cheatsheet
• https://devhints.io/git-tricks
• Good one-liner cheatsheet (orga.cat)
• Another good cheatsheet (shortcutfoo)
• git ready - Simple, easy to understand, tips at a time
• gitignore templates
  • You can pull one of these in via the CLI, with gitignore NPM package: npx gitignore python
• .gitconfig - handy git bash aliases
• agis/git-style-guide
• Zak Laughton - Three ways to time travel in Git to undo destructive mistakes
• Git History (pomber/git-history)
  • Timeline / slideshow view of a single file's Git history
    • Just replace github.com with github.githistory.xyz
  • There is also a VSCode extension
  • And a CLI!
    • npx git-file-history path/to/file.ext
• Git Command Explorer

Often used commands cheat table

Cloning | git clone --depth {DEPTH} {REMOTE} | How many commits to retrieve when cloning.

Useful for if you need to quickly explore a repo or fetch just for testing - you can use depth of 1 for just the last commit. Remotes | git push {REMOTE} {LOCAL_BRANCH_NAME}:{REMOTE_BRANCH_NAME}
Or, if branch names are the same
git push {REMOTE} {BRANCH_NAME} | Push a branch to a specific remote

Most of the above commands are covered more in-depth in subsections below.

Note: git gui and gitk commands depend on the corresponding programs being installed (git-gui and gitk, respectively), which normally are bundled with the installation of git, but might not be on your system.

Get commit ___ # of commits ago

HEAD~1 = second to last commit

You can change 1 to how many commits back you need to look

For example, to diff between the last commit and the one before that:

git diff HEAD HEAD~1

Get the last commit SHA hash, or priors

# Full SHA
git rev-parse HEAD
# Short SHA
git rev-parse --short HEAD

# 2nd to last SHA
git rev-parse HEAD~1

Relevant S/O: https://stackoverflow.com/q/949314

Meta cheatsheet - actually setting up Git

📘 Docs: "Getting Started - First-Time Git Setup"

• Start from scratch (create empty repo in current folder): git init
• Set committer name and email
  • git config --global user.name "Alan Turing"
  • git config --global user.email "aturing@example.com"
• Checking installed version
  • git --version
• Check installed path
  • git --exec-path
• If you are getting auth issues ("logon failed", etc.), there could be a few reasons (especially if you are using 2FA):
  • your version of git is probably out of date
    • For example, on Windows, a standard combo is Git Credential Manager for Windows + Git for Windows. There is a known issue with the credential manager and older versions of GFW.
  • Windows credentials got "unset" as the cred provider, or something glitched with it
    • Try git config --global credential.helper manager AND git config --system credential.helper manager to reset Windows Credentials as the provider, then retry your operation, and if necessary, relogin
      • Note: This is no longer wincred!
      • See below about how to verify this setting took hold
    • You can use git config --list --{global || system} to check settings (see this)
• Checking your full git config
  • git config --list will show all, from all levels (system, global, and local)
  • To view just one level, include it as a flag:
    • git config --list --global
  • You can use git config --get to view specific key-value pairs
  • Check contents of ~/.gitconfig
• Checking the user config object
  • You can use the RegEx option to view the user config key-value pairs: git config --get-regexp 'user\..+'
• Bulk ignoring commits for git blame
  • Commit and use a .git-blame-ignore-revs file (will be supported on GitHub's web UI soon)

Dealing with Linkage

Adding a named remote

git remote add {name} [REMOTE_ORIGIN_URL]

Hint: You probably want to fetch after adding a remote. E.g. git remote add steve-contrib [URL] && git fetch steve-contrib

Hooking an existing local repo to a remote origin

(example - created git repo from cmd instead of github.com gui, now want to link up to existing Github repo)

1. Add already existing (but empty) github URL as remote repo and set as remote
   • git remote add origin [REMOTE_ORIGIN_URL]
2. Verify that it is linked
   • git remote -v
3. Now push up
   • git push origin main

Setting upstream (for forks)

This is really the same steps as adding a new remote origin:

git remote add upstream [REMOTE_UPSTREAM_URL]

Making sure your fork is up to date before making a PR

1. Make sure you have fresh data on upstream
   • git fetch upstream
2. Merge the fresh upstream main, into:
   • Your feature branch
     • git merge upstream/main
   • Your main branch
     • git checkout main && git merge upstream/main

Here is an advanced version that will fetch upstream main, merge into local main, and then merge into feature you are on, without switching branches. (as long as it can fast-forward) (only works if you are not on main currently):

git fetch upstream main:main && git merge main

Or (works regardless if you are on main or not currently):

git pull upstream main:main && git merge main

details

Change the link to remote

(for example, if you change the repo name on Github, or fork a repo and want to set origin to your new fork.)

1. First check list of remotes
   • git remote -v
2. Then remove the one you want to
   • git remote rm [NAME_OF_REMOTE||Example:Origin]
3. Double check that it was removed
   • git remote -v
4. Add new link
   • git remote add [NAME_OF_REMOTE||Example:Origin] [REMOTE_URL]

Branch Linkage

• Show how local branches are linked up to origin (e.g. showing tracking links)
  • git branch -vv
• List only remote branches:
  • git branch -r
    • or:
  • git ls-remote --heads origin
• Fetch (and switch to) remote branch that does not exist locally (yet)
  • git checkout --track origin/[REMOTE_BRANCH]
• Link a local branch to a remote TRACKING branch, that DOES NOT EXIST YET - set upstream (very common annoyance)
  • git push --set-upstream origin [LOCAL_BRANCH_NAME]:[NEW_REMOTE_BRANCH_NAME]
    • Or, even shorter
  • git push -u origin [LOCAL_BRANCH_NAME]:[NEW_REMOTE_BRANCH_NAME]
    • Or, even shorter (assuming same names)
  • git push -u origin git c[NEW_REMOTE_BRANCH_NAME]
• Link a local branch to a remote TRACKING branch that ALREADY exist
  • Different branch (already exists):
    • git branch -u origin/[TRACKING_BRANCH] [LOCAL_BRANCH]
  • Different branch (does not already exist)
    • git branch --track [LOCAL_BRANCH] origin/[TRACKING_BRANCH]
      • Or
    • git checkout [TRACKING_BRANCH] (notice did not use origin/ before it)
  • Current branch:
    • git branch --set-upstream-to origin/[TRACKING_BRANCH]
      • Or
    • git branch -u origin/[TRACKING_BRANCH]
• Push all local branches up to origin, regardless if they exist on origin or not yet
  • git push origin --all
    • Or
  • git push origin --all -u
    • Use the -u flag to set-upstream, which makes pulling from branches later easy
• UNLINK a branch that is tracking remote
  • git branch --unset-upstream

Good S/O answer about upstream. And this

In git push, when you only specify one branch name, instead of both remote and local, git assumes the branch names are the same!!! See notes under linking a local branch to a remote that does not yet exist.

Branches

Miscellaneous Branch Stuff

• Push to origin
  • Current branch
    • git push
  • Another branch
    • git push origin {branch}
  • All branches
    • git push origin --all
    • See above notes under "branch linkage"
• Merge branches without switching to them! (Only for fast-forward merges, use with caution) (Details)
  • Merge local into local
    • git fetch . {localBranchA}:{localBranchToMergeAInto}
  • Merge remote branch into local branch
    • git fetch origin {remoteBranch}:{localBranchToMergeInto}
      • Or:
    • git fetch . origin/{remoteBranch}:{localBranchToMergeInto}
  • Practical example: merge origin:main into local:main, and then merge that into yours, all without checking out!
    • git fetch origin main:main && git merge main
  • Extended example: merge origin main into local main, merge main into current branch feature, then merge current branch feature into local main, then push local main back up to origin.
    • Steps:
      • git fetch origin main:main && git merge main
      • git fetch . feature:main
      • git push origin main
    • This is basically the full update cycle for an org where main is source of truth
    • As one line:
      • git fetch origin main:main && git merge main && git fetch . feature:main && git push origin main

Get commits between branches

It's important to note that you can pass HEAD instead of {branchA}, if you just want to compare against the current branch you are on.

• Full
  • git log {branchA} ^{branchB}
• Short
  • git log --oneline {branchA} ^{branchB}
• Just the count (number of commits different between branches)
  • git rev-list --count {branchA} ^origin/main

Get the current branch (name)

Couple of options:

git symbolic-ref --short HEAD

Or:

git rev-parse --abbrev-ref HEAD

Note: Neither of these options actually captures the name as a variable or does anything with it. If you wanted to use the branch name with another command, you would need to use piping / redirection / variable capture.

My notes on Bash variable capturing and redirection are here.

View all untracked files, not just dir names

git status -u

Viewing Changes / Diffing

Get summary of lines changed by file

• git diff [version A] [OPT Version B] --stat

Get just summary of total lines changed, etc.

• git diff [version A] [OPT Version B] --stat | grep '^\s*.*files\schanged.*$'

Get a list of changed files (filenames)

• During pre-commit (staged files)
  • git diff --cached --name-only --diff-filter=ACMRTUXB
    • You can modify the filters to change which files show up
• During post-commit / find the files changed in the very last commit
  • With truly only names (credit):
    • git diff --name-only HEAD HEAD~1
    • You can combine with diff-filter: git diff --name-only --diff-filter=ACMRTUXB HEAD HEAD~1
  • To see operations for each file (modified, deleted, etc.)
    • git diff --name-status HEAD HEAD~1
  • Without using diff (for example, if there is only one commit in the repo):
    • git show HEAD --name-only --format=%b
    • Note that there will be spacing around the filenames

To pipe filenames from Git that might contains spaces, you need to take some extra steps - use the -z with diff to get null terminators, and then use -0 or --null with xargs, to tell it that input items are demarcated by null termination characters.

Exclude a file from a diff

• git diff [OTHER] . ":(exclude)[FILEPATH_TO_EXCLUDE]"
• Sample:
  • git diff 7cc297d5baf2e305b709fcf93e3fe93284fb18e1 --stat -- . ":(exclude)package-lock.json"

Better Diffs for Text-Based Files

The bare-bones git diff command works well for code, but often is almost unusable when trying to view the differences between revisions of a text-based / prose file, like in Markdown documents.

Aside from using a different diff'ing tool (such as diffr), their are some options for git diff that can vastly improve the text diff viewing experience:

💡 If you are looking for the shortest command with the biggest impact, it might be git diff --color-words="{REGEX}"

Example: git diff --word-diff-regex="\w+"

• These commands are functionally equivalent
  • git diff --color-words
  • git diff --word-diff=color
• You can also customize the pattern for the word-based diff views:
  • git diff --word-diff-regex="\w+"
  • You might want to combine this with the color mode, so either:
    • git diff --color-words="\w+"
      • Or...
    • git diff --word-diff-regex="\w+" --color-words

Amend the last commit - [!!! - Danger - REWRITING HISTORY - !!!]

• For if you just want to change the message
  • git commit --amend -m "my new commit message to replace old"
• For if you forgot to add files / stage (stage first with add before running)
  • git commit --amend
• Same as above, but without interactive confirmation prompt to change message
  • git commit --amend --no-edit

Removing a file that was added in the last commit?

• git rm {file} and then git commit --amend like normal

Cherry-Picking / Selective Git Merging

cherry pick a commit from another branch to add to current

• git cherry-pick [COMMIT_HASH]
  • Use the --no-commit option if you want to manually merge changes, and/or combine multiple cherry-picks into one commit

Merge up to a specific commit / Cherry-pick multiple with shared history

• git merge [COMMIT_FROM_OTHER_BRANCH_HASH]
  • This is the automatic version of cherry-picking commits manually from where a branch diverged, up to the point you want
  • Very cool feature; just pass in the hash of the last commit on the other branch!
  • Unlike cherry-pick, this is going to require shared history, just like regular merging
  • Relevant S/O

Cherry-pick, but for specific files

This is rather complicated, especially if you are trying to preserve history... in many cases, you are better off using the "grab files" approach (e.g. checkout, see further down below).

In the case that you really want to preserve history, you basically only have two options:

• Manually find all the commits that affected the file and cherry-pick with them, removing the changes for other files as you go
  • You can use git cherry-pick -n to stage but not commit
• Or, create a diff patch based on the target commit(s), and then git apply them
  • This S/O answer shows how to do it
    1. git show {SHA} -- {filename(s)} | git apply -
    2. git add {filename(s)}
    3. git commit -c {SHA}

Grab files from another branch (or commit) and merge into yours, without branch merging

• Specific files:
  • git checkout {branchOrSHA} -- {filename(s)}
  • Interactively:
    • git checkout -p {branchOrSHA} -- {filename(s)}
    • git checkout -p {branchOrSHA}
• Grab everything (all files on branch or commit):
  • git checkout {branch} -- .

Selective Git Staging / Interactive

A neat tip is that you don't always have to stage an entire file - you can add individual lines! This is great to remember when maybe you need to comment out something that breaks your local build, but needs to stay in the code base for someone else at the moment.

The easiest way to do this is with your IDE. In VSCode, all you have to do is select the lines you want to stage, then open the command palette (CTRL + P) and select "Git: Stage Selected Ranges".

From the CLI, you can do this by running git add --patch {filename}. Details. You could alternatively run git add -i and then select patch (details).

Revise the last commit (different from revert - this is essentially like undo)

• if you want to undo the last commit, but keep changes locally so you can edit and then re-commit
  • git reset HEAD~1
  • If you want to re-commit with original message
    • Interactive commit message editor
      • git commit -c ORIG_HEAD
    • No editor
      • git commit -C ORIG_HEAD
  • Else if you want to just nuke it
    • git reset --hard HEAD

Nuke it - reset to remote main

1. git fetch origin
2. git reset --hard origin/main

You can use this to reset to head of any branch really. Just make sure you have fetched. For example,

git reset --hard or git reset --hard HEAD.

Useful for when you accidentally diverge and want to revert to the origin as source of truth.

This won't actually remove/delete untracked files. To do that, see below section.

Removing and Unstaging Files

Remove untracked files

• Interactively
  • git clean -i -fd
• Remove all untracked files and directories
  • git clean -fd
• Also smart to use this flag: --dry-run [OR] -n

Good combo: git clean -fd -n to preview, and then git clean -fd to finalize

Remove a file from git tracking after adding it to the gitignore

• Use --cached flag with rm
  • git rm --cached <file>

"unstage" a file (do the reverse of git add)

Unstage single file

• git reset -- [FILEPATH_TO_UNSTAGE]
  • Note that the "--" is because git reset can also be used with branches instead of files, so "--" is to specify this is only for files

unstage all added files

• git reset .

Submodules

• git-scm page
• Good submodule cheatsheet: vogella.com/tutorials/GitSubmodules

Add a submodule

• With SSH access
  • git submodule add git@github.com:joshuatz/j-prism-toolbar.git [LOCAL_FOLDER_PATH]/[NEW FOLDER THAT DOES NOT EXIST]
• With standard Github credentials
  • $ git submodule add https://github.com/joshuatz/j-prism-toolbar.git [LOCAL_FOLDER_PATH]/[NEW FOLDER THAT DOES NOT EXIST]
• If something goes wrong, don't be afraid to manually edit .gitmodules and .git/CONFIG
  • StackOverflow

Update a submodule / init after clone

• To clone a repo, and include submodules from the get-go
  • git clone --recurse-submodules {originUrl}
• If you forgot to clone with recurse on, you can initialize submodules after the fact by using:
  • git submodule update --init or git submodule update --init --recursive

Updating (fetch)

• Easiest to remember is to just treat git subdir as real git repo
  • cd to the directory, then run git fetch and git merge origin/main
  • Then cd back up, and commit the update to the parent repo
• Alternative is to use shorthand command
  • git submodule update --remote

Remove a submodule

• Most complete option - https://stackoverflow.com/a/36593218
  1. Remove the submodule entry from .git/config
     • git submodule deinit -f path/to/submodule
  2. Remove the submodule directory from the superproject's .git/modules directory
     • rm -rf .git/modules/path/to/submodule
  3. Remove the entry in .gitmodules and remove the submodule directory located at path/to/submodule
     • git rm -f path/to/submodule

Jump to a specific moment in time, and commit

DANGER!

• git reset --hard [COMMIT_HASH]

TEMPORARILY jump to a specific moment in time

• Very cool, you can just checkout the commit!
  • git checkout 778de63b25d66b576beba53b2ca0506ced9dded7
• If you want to jump back to tip after, just checkout the branch name again
  • git checkout main

Merging a branch the default Github PR way

git merge [BRANCH_TO_MERGE] --no-ff

The reason why you can see commits grouped together with a specific PR / branch merge on Github is because when you click the "merge" button, instead of just doing git merge [BRANCH_TO_MERGE] it uses git merge [BRANCH_TO_MERGE] --no-ff

• --no-ff means "no fast forward":
• Default merge uses fast forward, which basically says that if the branch you are merging into shares a common history with the branch you are merging, it will "fast forward" the base branch until it points to the last commit on the branch you are merging.
• "no fast forward" means all the commits that make up the feature branch you are merging are kind of lumped together (or treated as children) as a new EXPLICIT merge commit. This is why when you merge a PR on github, it forces you to create a new specific merge commit.

Side Benefit: Merging this way means that you can point to a specific commit that brought in a set of feature changes (or an entire feature). This provides a bunch of different benefits:

• To undo the merge of a feature, you just need to revert one commit, instead of having to do some crazy stuff with finding the commit before the merge was done, or cherry picking commits, etc.
• You can easily visualize branch history, and see where a feature was specifically worked on separately
• Git GUI's (like Github) will treat it like a "true merge" if you do it this way

Moving Files

• Why use git mv?
  • most of the time, Git can guess renames/moves vs new files based on contents and filename, but not 100% of the time. Git mv is a little more foolproof, since you are explicitly telling Git where your files are moving to / getting renamed
    • git mv also automatically takes care of the "git rm" for the old file, and "git add" for the renamed/moved file
• Methods:
  • One by one
    • git mv oldfiledir/oldfile.h newfilepath/oldfile.h
  • Bulk
    • Generally, you can just move the files yourself, and when you "git add" or "git add -A", it should detect rename vs new files
    • Make sure you do git add after moving the files, but BEFORE changing contents. Since the git rename detection works by content hash.

Show which files are being ignored:

On git v1.7.7 and up (SO),

git status --ignored

Stashing / Temporary Staging

Docs:

• git-scm
• Atlassian tutorial

Stashing - Reminders:

• stash will grab both staged and unstaged files, by default
  • If you only want certain files (e.g. staged), you need to use patch and/or push. See "partial stashes" notes below.
• Stashes are NOT synced with origin / remote

Stashing - The basics

• Stash:
  • git stash
  • Include untracked or ignored files:
    • git stash -u
    • git stash -a
• Pop:
  • git stash pop
• See all stashes:
  • git stash list

Partial stashes

SO answers: 1, 2, 3

• Interactive (hunk picker)
  • git stash -p (or git stash --patch)
• Single file:
  • (Still interactive, but just hit a to stash all hunks in current displayed file)
    • git stash -p index.html
    • You can even include a message!
      • git stash -p index.html -m "trying out different CDN"
  • Newer syntax (> v2.13) - non interactive
    • git stash push {flags} -- {pathspec}
    • Example:
      • git stash push -m "trying out different CDN" -- index.html

    • Order matters! -- {pathspec} must come last.

git stash -p is an internal alias for git stash push -p

Reminder: There is not "cost" to just using a temporary branch to stash changes, and is usually a better alternative to stashing. Or, explore using patches or actual temp files with arbitrary extension that is then gitignored (e.g. ".tempdump")

Dumping changes (git diff) as a patch file

If you want to quickly dump all the differences between two branches (lets say between feature and main), or just the changes in your current branch, as a backup, you could export the diff as a patch file.

git diff main > {filename}.patch

Or, to create a patch based on difference between current and staged, just leave off the branch:

git diff > {filename}.path

And to restore from that patch, you could use:

git checkout main && patch -p1 < {filename}.patch

You can use --dry-run with patch for a preview of changes

Use patch -p0 if --no-prefix was used with git diff

Alternative option: git diff > {filename}.patch and then git apply {filename}.patch

🚨 WARNING: This won't capture untracked changes. As a workaround, some might prefer to stage all their changes and then use git diff --cached > {filename}.patch.

🚨 WARNING: If you want to capture binary change (e.g. JPG files, EXEs, etc.), you need to use --binary when creating the patch; `git diff

Credit: S/O answer

Metadata (commits, files, etc.)

Show BOTH the authorDate and commitDate

• git log --format=fuller

Get 'last modified' timestamps:

• git show -s --format=%at
  • -s = --no-patch = suppress diff info
  • --format=%at = --pretty[=<format>] = pretty-print
    • %at = 'author date, UNIX timestamp'
      • Alternative: %ct = 'committer date, UNIX timestamp'
  • For a specific thing (commit, tree), just put it last:
    • git show -s --format=%at {thing}
• git log --pretty=format:%at | head -n 1
  • Gets the log, formats as UNIX stamps, sorts, then limits to one line
  • This can be for a file, etc:
    • git log --pretty=format:%at -- myfile.js | head -n 1
  • Alternative: git log --pretty=format:%at | sort | tail -n 1
• git log -1 --pretty=format:%at
  • Same as above, but uses the -{numLimit} git option to pre-limit the result to one line
    • Now we don't have to use head or tail to limit

committer date vs author date - there should only ever be one author date, and it corresponds to when the code was actually first committed/authored with git commit. However, there can be multiple commit dates, and they correspond to when the commit was modified in the process of applying it through merge/rebase. See "Issues with dates and rebasing" on this page for details.

Get date 'created' timestamps:

• git log --pretty=format:%at --follow -- {thing} | tail -n 1
  • --follow will make sure it captures renames in history
  • SO

Warning: You cannot combine --reverse with --follow; this is a known bug

Git Attributes (gitattributes)

Check Attributes

• git check-attr {flag} {path}

Example (shows all attributes)

• git check-attr -a cheatsheets/vue.md

Git Hooks

Reminder: Git hooks are not committed into your repo by default. The recommended way to share hooks in a repo is to create a checked-folder, like /.githooks (but this can be named anything), and automate copying (or even better, symlinking) the scripts to the real githooks location (.git/hooks/*). This could be done with Bash scripts, BAT, MAKEFILE, etc. See this and this. Reminder: Add "shebang" to file and make sure executable (chmod +x).

Git Hooks - Resources

• Atlassian Guide
• githooks.com

Git Hooks - Tips

• Pre-commit
  • You can add files to the commit, without the user needing to interact or approve, by just calling git add within the git hook
• Post-Commit
  • Be very careful using further git commands within a post-commit git hook. Very easy to accidentally write an endless loop
    • For example, if you touch a file and then amend it to the last commit, that will actually trigger the hook itself and you will end up looping

Git Tags

As usual, Atlassian has one of the best guides: here

Tip: Something that a lot of tutorials gloss over or don't even mention, but what I feel like should be bullet point number #1, is that when you create a tag and don't explicitly attach it to a specific commit, by default it gets attached to the current commit that the HEAD is pointing at.

Creating a tag does not actually check new code into VC or save the "state" of your environment. It is more like a pointer to commit.

In brief:

git describe --tags can return a tag in a format that looks like v2.4-28-a402cbee4 - that is: `{tag_name}-{number_of_commits_since_tag}-{commit_hash}

Finding commit that a tag points to

One command you can use is git rev-list -n 1 {tagName}, which will get you the hash of the commit the tag points to. You can then use something like git show to get the full commit details (author, changes). Or, as a one liner:

git rev-list -n 1 {tagName} | xargs git show

What do I use for a tag name?

Up to you! A lot of people use Semantic Versioning with tags to correspond to releases. In fact, Github will automatically create new releases under the corresponding tab of your repo if you add tags.

A sample tag might be something like v1.4.23 or v1.4.23-beta.1+autobuild24.

Annotated vs lightweight

Essentially, lightweights can only be the name of the tag, and nothing more. Whereas with annotated, you can add a message, sign with PGP key, and more.

Rebasing

Some explanations / resources:

• https://www.youtube.com/watch?v=f1wnYdLEpgI
• https://www.atlassian.com/git/tutorials/rewriting-history/git-rebase
• https://www.atlassian.com/git/tutorials/merging-vs-rebasing

💡 If you use rebase a lot, you might want to consider configuring rebase to always use autoStash, so you don't have to manually stash changes for an unclean working directory before rebasing.

git config rebase.autoStash true

or

git config --global rebase.autoStash true

What Does Rebasing Do?

If (like me), you have trouble remembering and conceptualizing what rebase actually does, just think about the name... "re-base" - you are re-setting the base of the of a bunch of commits to a new one.

Here are some guides:

• Rebasing vs merging (Atlassian)
• git rebase (Atlassian)
• git rebase (git-scm)

Rebase Commands

** Where {baseRef} is one of standard ref (id, branchName, tag, HEAD, etc.)

• Rebase non-interactively
  • git rebase {baseRef}
• Rebase interactively
  • git rebase -i {baseRef}

Example: Rebase branch alpha onto main

• git checkout alpha && git rebase main
  • Or, if you want to rebase while on main:
• git rebase main alpha

Rebasing: Quick Fixup Commits

If you want to revise a specific commit instead of going through the normal full interactive rebase process, there is a slightly faster way to revise it using a "fixup" commit.

1. Stage your changes, and then run git commit -a --fixup {HASH_OF_COMMIT_YOU_WANT_TO_FIX}
2. Have git handle moving and squashing for you: git rebase --autosquash --interactive {HASH_OF_COMMIT_BEFORE_ONE_YOU_WANT_TO_FIX}
3. Save and quit when Git opens the editor

Rebase vs ff merge

A rebase that is used just to reset the base (and not make other edits) is very similar, in results, to a "fast forward merge" (through git merge (if it can ff and defaults have not been changed) or git merge --ff-only). Both methods result in a linear history, that makes it look like the feature branch commits were applied directly to main (without merge commits).

Example rebasing of feature onto main

git checkout feature
git rebase main # Or, git rebase -i main

# One-liner
git rebase main feature

The main difference seems to be that rebase is really replaying or copying commits onto the new base, versus merge ff, which is more like a pointer move. The result is that, although the output looks the same, rebase can end up with different commit hashes, since copying a commit results in a slightly different hash.

Note: The one-liner syntax for rebase still performs a checkout first, changing your working directory; it just does it for you, behind the scenes. If you want to rebase without modifying working directory, you will need to use something like worktree

Atlassian has an in-depth explainer on the differences: "Merging vs Rebasing"

Issues with dates and rebasing

Since you are technically re-commiting when you rebase (by re-writing history), the default thing that happens is that the Author and AuthorDate stay the same as before, but the Commit (author) and CommitDate reflect yourself and the current time of the rebase.

This can make it look like old commits have just been made, and since Github goes by CommitDate for ordering and display within a repo (1, 2, 3), this can really mess with the commit order in a PR!

Rebase - Fixing date order

We can use the same rewriting properties of rebase to fix the very problem it created, and edit the past commit dates.

If we have already used rebase, and now our dates are messed up, we can run rebase again, but this time using the --committer-date-is-author-date flag, like so:

git rebase --committer-date-is-author-date {commitHash}

For picking the commitHash to rebase from, pick the first commit that has a correct date, where the author and commit dates match, and use it.

Credit goes to this S/O.

Warning 🚨: This flag is easy to misinterpret; it essentially sets the committer date to whatever the author date (usually this means pretending it happened earlier than it did).

If you want the opposite, to force the author date to match the committer date, you will want to use -i, --ignore-date, or --reset-author-date (v > = 2.29). Similar to the above trick, you will also need to pick a commit hash that comes before the one you want to fix, if the commit you want to fix is the current HEAD.

Squashing on Github - Issues

Github has two advanced options for merging PRs - "squash and merge" and "rebase and merge" can both lead to complicated issues.

For example, squashing and rebasing often changes the metadata about who committed the code, and in the case of squashing, it can attribute large quantities of code by dozens of authors, to a new single author.

In late 2019, Github improved how attribution works with squash merges, by automatically adding everyone who touched the code in the PR as a co-author. You can see this automated text in the bottom of the squashed commit text.

Tools like Gitlens often do not play nice with complex code history, where there are multiple authors (co-authors), a different commiter vs author, etc.

Github's "squash and merge" option basically leaves two commit trails. The branch that you merged from will have the full commit history. The branch you are merging into will have just a single commit that bundles up all the commits that made up the PR.

Rebase - Opinions

Something I find funny is rebase goes against a lot of best practices around source control (source of truth, preserving history, etc.), and yet so many devs advocate for using it. I understand its value, but it also seems like it is often more trouble than it can be worth. Not to mention that rewriting history and permanently screwing up a shared history is a great way to make devs you are working with irate.

Even on Github, you'll find pages that discourage using git rebase.

When in doubt, make temp local backup branches before attempting anything complex.

General Rule: Here is the general rule about rebase that seems to be the consensus: Rebasing on your own stuff (your branch of shared repo, feature, solo repo) that is not yet part of shared code is fine. Rebasing on shared branches or commits that are already pushed, or other people's branches, is not.

Git styles, standards, best practices

Good reference: agis/git-style-guide.

Branch naming

Here is a good S/O thread on the topic. And here is a Dev discussion.

Easy, good rules to remember:

• No caps
• Use hyphens instead of spaces
• Try to keep branch names short
  • Use "grouping" tokens and slashes to help
• Use issue / ticket IDs when applicable

Examples:

• feature/OMT-4215/adding-gps-locator
• feature/adding-gps-locator/joshua
• joshua/OMT-4215/adding-gps-locator

You should try to avoid using master as your main branch, and use something like main, production, or whatever else makes logical sense. "Master", at least in the origins of Git (and its intention to replace Bitkeeper), has roots in a "master / slave" relation context, and in addition, often isn't a great descriptor of what the branch is for anyways! And, it often doesn't take that much effort to change.

Writing Commit Messages

• To encourage writing good commit messages, you might want to avoid using commit -m "my message", and switch to using commit, which will launch your default text editor and encourage you to write a full commit message.
  • To set VSCode as the default commit editor, it should be as easy as: git config --global core.editor "code --wait"
• There is no single "standard", but a very common format is one outlined by Tim Pope in a blog post. This is further summarized and outlined in many other posts:
  • McKenzie: How to write a Git commit message properly
  • Painsi: Painsi/commit-message-guideline.md (GH Gist)
  • Ayodeji: How to Write Good Commit Messages: A Practical Git Guide
  • Beams: How to Write a Git Commit Message
• Here is my own summary of the Tim Pope standard:
  1. Heading: Short (<= 50 chars), capitalized, but no period
  2. Empty Line
  3. Body
     • This can be a single paragraph, multiple paragraphs, bulleted lists, or just a few sentences. BUT, they should adhere to the following format rules:
       • Lines should wrap at 72 characters (including bullet items)
       • Bullets should use * or -
       • Empty lines should be placed between multiple paragraphs, but are optional for between bullet list items
     • Use imperative grammar: Add test suite, not Added test suite
• My own personal preferences, and notes:
  • If you are using a ticketing system, make sure to include the ticket ID either in the head or body
    • Most systems, like Jira, will auto-link based on this
    • Github actually does this as well
      • if you write something like addresses #4 in your commit message - where there is an issue with id #4 - pushing the commit up will auto-associate it with the issue (or PR)
      • You can even close issues via commit messages - once the commit is merged into your default branch (details).
  • I prefer short bullet lists over long paragraphs
    • Bullet lists are also much easier to combine in rebase / merges, over paragraphs or sentences
    • Easy way to avoid 72 char wraps
  • If you did something "weird" in your commit (disabled a test, reverted critical code, etc.) - SAY SO
  • If you are going to insist on using Emoji in your headings, try to be consistent; take a look at emoji-log

Finding a force push on Github

1. Generate a token to use with the Github API if you don't already have one handy
2. Make a request to https://api.github.com/repos/{OWNER}/{REPO}/events
3. Scan / filter to events with pushEvent as the type
4. FURTHER filter to those where the new payload.head does not match any of the SHAs in the payload.commits array
   • Sample code:

   const filtered = data.filter((obj)=> {
   	let wasForcePush = false;
   	if (obj.type === 'PushEvent') {
   		const payload = obj.payload;
   		if (Array.isArray(payload.commits) && payload.commits.length) {
   			wasForcePush = payload.commits[payload.commits.length - 1].sha !== payload.head;
   		}
   	}
   	return wasForcePush;
   });

5. Once you have found the commit that forced a new diverged tip, you can create a new branch based off the commit right before that forced push (on Github.com via API)
6. Finally, you can check out the newly created branch locally, and if you want to, merge back.

Finding a Force Push - Further reading:

• https://objectpartners.com/2014/02/11/recovering-a-commit-from-githubs-reflog/
• https://stackoverflow.com/a/28958418
• https://stackoverflow.com/a/48020000

Wiping Git History - Squashing to a Single Commit

In certain cases, you might want a branch or repo to be devoid of history. In this case, what you are usually looking to do is squash the entire codebase into a single commit.

There are several ways to do this, enumerated here and here. The "orphan branch" solution, described in this response seems the most straight-forward and least error-prone of all the solutions listed.