156

This question pertains not only to how to accomplish this task, but to whether doing so is good or bad practice with Git.

Consider that locally I do most work on the master branch, but I have created a topical branch I will call "topical_xFeature". In the process of working on "topical_xFeature" and switching back and forth to do other work on the master branch, it turns out that I have made more than one commit on the "topical_xFeature" branch, but between each commit, I have done no push.

First, would you consider this bad practice? Would it not be wiser to stick with one commit per branch per push? In what cases would it be good to have multiple commits on a branch before a push is made?

Second, how shall I best accomplish bringing the multiple commits on the topical_xFeature branch into the master branch for a push? Is it a nuisance to not worry about it and just do the push where multiple commits get pushed, or is it less annoying to somehow merge the commits into one and then push? Again, how to do this?

Machavity
  • 29,816
  • 26
  • 86
  • 96
Todd Hopkinson
  • 6,553
  • 5
  • 31
  • 34

6 Answers6

156

For your first question, no, there's nothing wrong with pushing multiple commits at once. Many times, you may want to break your work down into a few small, logical commits, but only push them up once you feel like the whole series is ready. Or you might be making several commits locally while disconnected, and you push them all once you're connected again. There's no reason to limit yourself to one commit per push.

I generally find that it's a good idea to keep each commit a single, logical, coherent change, that includes everything it needs to work (so, it does not leave your code in a broken state). If you have a two commits, but they would cause the code to be broken if you only applied the first one, it might be a good idea to squash the second commit into the first. But if you have two commits where each one makes a reasonable change, pushing them as separate commits is fine.

If you do want to squash several commits together, you can use git rebase -i. If you're on branch topical_xFeature, you would run git rebase -i master. This will open an editor window, with a bunch of commits listed prefixed by pick. You can change all but the first to squash, which will tell Git to keep all of those changes, but squash them into the first commit. After you've done that, check out master and merge in your feature branch:

git checkout topical_xFeature
git rebase -i master
git checkout master
git merge topical_xFeature

Alternatively, if you just want to squash everything in topical_xFeature into master, you could just do the following: 

git checkout master
git merge --squash topical_xFeature
git commit

Which one you choose is up to you. Generally, I wouldn't worry about having multiple smaller commits, but sometimes you don't want to bother with extra minor commits, so you just squash them into one.

Brian Campbell
  • 306,970
  • 56
  • 356
  • 335
  • 2
    After I merge with --squash, I am not able to delete the topic branch with `git branch -d topic`. Why is git not able to identify that all the changes are merged? – balki Oct 22 '12 at 11:20
  • 11
    @balki Because Git detects whether patches are merged based on whether they appear in the history of the given branch. Squashing commits changes them; they become a new commit, and while that new commit happens to do the same thing as the other ones, Git can't tell that, it can only tell if commits are the same if they have the same commit ID (SHA-1). So once you squash it, you need to tell git to delete the old branch with `git branch -D topic` to forcibly delete it. – Brian Campbell Oct 22 '12 at 13:44
79

This is the way I generally follow to combine multiple Commits into a single commit before I push the code.

To achieve this, I suggest you use 'squash' concept provided by GIT.

Follow the below steps.

1) git rebase -i master (instead of master you can also use a specific commit)

open the rebase interactive editor, where it will show all your commits. Basically where you need to identify the commits which you want to merge into a single commit.

Imagine these are your commits and shown something like this in the editor.

pick f7f3f6d changed my name a bit    
pick 310154e updated README formatting and added blame   
pick a5f4a0d added cat-file

It's important to note that these commits are listed in the opposite order than you normally see them using the log command. Means, the older commit will be shown first.

2) Change 'pick' to 'squash' for last committed changes. something like shown below. Doing that so, your last 2 commits will be merged with the first one.

pick f7f3f6d changed my name a bit         
squash 310154e updated README formatting and added blame   
squash a5f4a0d added cat-file

You can also use short form if you have lot of commits to combine:

p f7f3f6d changed my name a bit         
s 310154e updated README formatting and added blame   
s a5f4a0d added cat-file

for editing use 'i', it will enable the editor for insertion. Keep in mind top most(oldest) commit cannot be squashed as there is no previous commit to combine with. So it has to be picked or 'p'. Use 'Esc' to exit insert mode.

3) Now, save the editor with the following command. :wq

When you save that, you have a single commit that introduces the changes of all three previous commits.

Hope this will help you.

harshainfo
  • 441
  • 7
  • 9
Kondal Kolipaka
  • 3,281
  • 20
  • 26
  • 8
    Perhaps this is obvious to others but, when you say "git rebase -i", you also need to specify which commit you start at. This is something I didn't realize when I tried following this example. So, in this example, it'd be "git rebase -i xxxxx" where xxxxx is the commit right before f7f3f6d chronologically. Once I figured that out, everything worked out exactly as described above. – nukeguy Apr 22 '16 at 19:26
  • That's interesting @nukeguy, I didn't have any issue not specifying a specific commit. It just defaulted to what was there. – JCrooks Jan 25 '17 at 20:23
  • 1
    Maybe like @nukeguy, `git rebase -i HEAD~2` was a helpful place for me to start. Then this answer was helpful. Then, my `git status` showed "Your branch and 'origin/feature/xyz' have diverged, and have 1 and 1 different commits each, respectively." So I needed to `git push origin feature/xyz --force-with-lease` See https://stackoverflow.com/a/59309553/470749 and https://www.freecodecamp.org/forum/t/how-to-squash-multiple-commits-into-one-with-git-squash/13231 – Ryan Jun 19 '20 at 17:15
  • What if there are a huge amount of `pick` to replace with`squash`? – Eduardo Pignatelli Aug 17 '21 at 10:24
12

Switch to the master branch and make sure you are up to date.

git checkout master

git fetch this may be necessary (depending on your git config) to receive updates on origin/master

git pull

Merge the feature branch into the master branch.

git merge feature_branch

Reset the master branch to origin's state.

git reset origin/master

Git now considers all changes as unstaged changes. We can add these changes as one commit. Adding . will also add untracked files.

git add --all

git commit

Ref: https://makandracards.com/makandra/527-squash-several-git-commits-into-a-single-commit

shiva kumar
  • 11,024
  • 4
  • 22
  • 26
12

First: nothing tells you to only have one commit per branch per push: a push is a publication mechanism allowing you to publish a local history (i.e. a collection of commits) on a remote repo.

Second: a git merge --no-ff topical_xFeature would record on master as a single commit your topic work, before pushing master.
(That way, you keep topical_xFeature around for further evolutions, that you can record on master as a single new commit on the next merge --no-ff.
If getting rid of topical_xFeature is the goal, then git merge --squash is the right option, as detailed in Brian Campbell's answer.)

Community
  • 1
  • 1
VonC
  • 1,129,465
  • 480
  • 4,036
  • 4,755
  • I think that `--squash`, not `--no-ff` is what you want. `--no-ff` would create a merge commit, but also leave all of the commits from `topical_xFeature`. – Brian Campbell Apr 19 '11 at 19:59
  • @Brian: I agree and upvoted your answer, but I first thought of the --no-ff option because I wanted to keep `topical_feature` branch around, and just record a single commit on `master` branch. – VonC Apr 19 '11 at 20:15
11

  1. First choose which commit you want everything to come after.

    git reflog
5976f2b HEAD@{0}: commit: Fix conflicts
80e85a1 HEAD@{1}: commit: Add feature
b860ddb HEAD@{2}: commit: Add something

  2. Reset to your selected head (I have chosen HEAD@{2})

    git reset b860ddb --soft

  3. git status (just to be sure)

  4. Add your new commit

    git commit -m "Add new commit"

Note: HEAD@{0} & HEAD@{1} Are now merged into 1 commit, this can be done for multiple commits also.

git reflog again should display:

git reflog
5976f2b HEAD@{0}: commit: Add new commit
b860ddb HEAD@{1}: commit: Add something
Machavity
  • 29,816
  • 26
  • 86
  • 96
Eddy Ekofo
  • 461
  • 5
  • 11
1

A Tool To Automate Multiple Commits Into One

as Kondal Kolipaka says. Using "git rebase -i"

The logic of "git rebase"

When using "git rebase -i", git generates the git-rebase-todo file in the current .git/rebase-merge directory, and then invokes the git editor to let users edit the git-rebase-todo file for processing. So the tool needs to meet:

  1. Modify the git editor to the tool which we provided;
  2. The tool processes the git-rebase-todo file.

Modify the default git editor

git config core.editor #show current default git editor
git config --local --replace-all  core.editor NEW_EDITOR # set the local branch using NEW_EDITOR as git editor

So, the tool need change the git editor and process the git-rebase-todo file. The tool using python below:

#!/usr/bin/env python3
#encoding: UTF-8

import os
import sys

def change_editor(current_file):
    os.system("git config --local --replace-all  core.editor " + current_file) # Set current_file as git editor
    os.system("git rebase -i") # execute the "git rebase -i" and will invoke the python file later with git-rebase-todo file as argument
    os.system("git config --local --replace-all core.editor vim") # after work reset the git editor to default

def rebase_commits(todo_file):
    with open(todo_file, "r+") as f:
        contents = f.read() # read git-rebase-todo's content
        contents = contents.split("\n")
        first_commit = True
        f.truncate()
        f.seek(0)
        for content in contents:
            if content.startswith("pick"):
                if first_commit:
                    first_commit = False
                else:
                    content = content.replace("pick", "squash") # replace the pick to squash except for the first pick
            f.write(content + "\n")

def main(args):
    if len(args) == 2:
        rebase_commits(args[1]) # process the git-rebase-todo
    else:
        change_editor(os.path.abspath(args[0])) # set git editor

if __name__ == "__main__":
    main(sys.argv)

Ref: https://liwugang.github.io/2019/12/30/git_commits_en.html

liwugang
  • 21
  • 3