ChatGPT解决这个技术问题 Extra ChatGPT

I ran into a merge conflict. How do I abort the merge?

I used git pull and had a merge conflict:

unmerged:   some_file.txt

You are in the middle of a conflicted merge.

How do I abandon my changes to the file and keep only the pulled changes?

I realise this is a super-old question, but do you want to abort the whole merge, and leave the branch you were merging unmerged, or just ignore this one file as part of a larger merge, letting all the other files merge in as normal? To me, your title implies the former, your question body wants the latter. The answers do both, without making things clear.
I got similar case on commit saying that automatic merge failed; fix conflicts and then commit the result: [rejected] gh-pages -> gh-pages (non-fast-forward)
Gwyn, it could be useful to select an accepted answer here. The top voted one is a bit less safe than some of the more up to date solutions, so I think it would help to highlight others over it :)

R
Robin Daugherty

Since your pull was unsuccessful then HEAD (not HEAD^) is the last "valid" commit on your branch:

git reset --hard HEAD

The other piece you want is to let their changes over-ride your changes.

Older versions of git allowed you to use the "theirs" merge strategy:

git pull --strategy=theirs remote_branch

But this has since been removed, as explained in this message by Junio Hamano (the Git maintainer). As noted in the link, instead you would do this:

git fetch origin
git reset --hard origin

Instead of doing a hard reset, you could bring it to a more granular level by doing: git fetch origin --> git reset origin (soft reset, your changes are still present) --> git checkout file_to_use_their_version_of another_file (steamroll your own changes back to match the origin) I never use git pull any more. Since in a fight between my latest code and the origin, the origin should always win, I always git fetch and git rebase origin. This actually makes my merges and conflicts few and far between.
I agree. I also like to fetch first, and then examine the upstream changes (git log ..@{upstream} or git diff ..@{upstream}). After that, like you, I'll rebase my work.
As noted in a more recent answer, as of version 1.6.1, it is possible to use 'git reset --merge'
I used git merge -X theirs remote_branch instead of git pull --strategy=theirs remote_branch as theirs looks like an option of recursive
git merge --abort is far preferable.
t
the Tin Man

If your git version is >= 1.6.1, you can use git reset --merge.

Also, as @Michael Johnson mentions, if your git version is >= 1.7.4, you can also use git merge --abort.

As always, make sure you have no uncommitted changes before you start a merge.

From the git merge man page

git merge --abort is equivalent to git reset --merge when MERGE_HEAD is present.

MERGE_HEAD is present when a merge is in progress.

Also, regarding uncommitted changes when starting a merge:

If you have changes you don't want to commit before starting a merge, just git stash them before the merge and git stash pop after finishing the merge or aborting it.


Interesting - but the manual scares me. When exactly is it appropriate to use? When would you have to specify the optional <commit>? #GitMoment :-o
You'd typically use this when you want to redo the merge from the start. I have never had to specify the optional commit myself, so the default (no optional ) is just fine.
I wish this answer had more votes! At this point, it seems like the most relevant solution in many cases.
Even with uncommited changes git was able to restore the state before the merge. Nice!
Is git merge --abort just a synonym for git reset --merge? The name certainly makes more sense, but does it have the same functionality?
H
Honest Abe
git merge --abort

Abort the current conflict resolution process, and try to reconstruct the pre-merge state. If there were uncommitted worktree changes present when the merge started, git merge --abort will in some cases be unable to reconstruct these changes. It is therefore recommended to always commit or stash your changes before running git merge. git merge --abort is equivalent to git reset --merge when MERGE_HEAD is present.

http://www.git-scm.com/docs/git-merge


This has been available since git v1.7.4. It's an alias for git reset --merge.
P
Peter Mortensen

I think it's git reset you need.

Beware that git revert means something very different to, say, svn revert - in Subversion the revert will discard your (uncommitted) changes, returning the file to the current version from the repository, whereas git revert "undoes" a commit.

git reset should do the equivalent of svn revert, that is, discard your unwanted changes.


D
Dielson Sales

For git >= 1.6.1:

git merge --abort

For older versions of git, this will do the job:

git reset --merge

or

git reset --hard

C
CB Bailey

In this particular use case, you don't really want to abort the merge, just resolve the conflict in a particular way.

There is no particular need to reset and perform a merge with a different strategy, either. The conflicts have been correctly highlighted by git and the requirement to accept the other sides changes is only for this one file.

For an unmerged file in a conflict git makes available the common base, local and remote versions of the file in the index. (This is where they are read from for use in a 3-way diff tool by git mergetool.) You can use git show to view them.

# common base:
git show :1:_widget.html.erb

# 'ours'
git show :2:_widget.html.erb

# 'theirs'
git show :3:_widget.html.erb

The simplest way to resolve the conflict to use the remote version verbatim is:

git show :3:_widget.html.erb >_widget.html.erb
git add _widget.html.erb

Or, with git >= 1.6.1:

git checkout --theirs _widget.html.erb

thanks for the hint. doesn't this smack of a poor git user interface, though?
@Peter: I'm not convinced. The desired result is achievable with a few basic commands with simple options. What improvements would you suggest?
I think the git 1.6.1 command makes a lot of sense, and is good. That's exactly what I would have wanted. I think the pre-1.6.1 solution is inelegant and requires knowledge about other parts of git that should be separated from the merge resolution process. But the new version is great!
Well, without commenting on whether it's "poor" or not, let me say that I would never have guessed that a syntax involving 'show' and :1:, :2:, and :3: was the way to recover the base and two "tip" files, but I am immensely glad to know of this technique, so: thank you very much!
D
DARK_C0D3R

You can either abort the merge step:

git merge --abort

else you can keep your changes (on which branch you are)

git checkout --ours file1 file2 ...

otherwise you can keep other branch changes

git checkout --theirs file1 file2 ...

M
Martin G

Comments suggest that git reset --merge is an alias for git merge --abort. It is worth noticing that git merge --abort is only equivalent to git reset --merge given that a MERGE_HEAD is present. This can be read in the git help for merge command.

git merge --abort is equivalent to git reset --merge when MERGE_HEAD is present.

After a failed merge, when there is no MERGE_HEAD, the failed merge can be undone with git reset --merge, but not necessarily with git merge --abort. They are not only old and new syntax for the same thing.

Personally, I find git reset --merge much more powerful for scenarios similar to the described one, and failed merges in general.


what is here actually meant by "failed merge"? Merge with conflicts or something else? Or to rephrase it: when is MERGE_HEAD not present? My follow-up question is there to understand better use of "git reset --merge".
@Ewoks git stash apply caused a merge conflict for me but git merge --abort did not help while git reset --merge did.
Had to use this to resolve a failed merge, I had no MERGE_HEAD for some reason so git merge --abort didn't work.
Heads up to anyone else considering this: MERGE_HEAD is not present when your merge happened because of stash pop; and reset --merge will delete your untracked files. See stackoverflow.com/a/67099267/1623757
What I needed after a failed merge from a tag to master...
G
Geo

If you end up with merge conflict and doesn't have anything to commit, but still a merge error is being displayed. After applying all the below mentioned commands,

git reset --hard HEAD
git pull --strategy=theirs remote_branch
git fetch origin
git reset --hard origin

Please remove

.git\index.lock

File [cut paste to some other location in case of recovery] and then enter any of below command depending on which version you want.

git reset --hard HEAD
git reset --hard origin

Hope that helps!!!


A
Alain O'Dea

An alternative, which preserves the state of the working copy is:

git stash
git merge --abort
git stash pop

I generally advise against this, because it is effectively like merging in Subversion as it throws away the branch relationships in the following commit.


I found this approach useful when I accidentally merged to a git-svn branch, which doesn't handle that nicely. Squash merges or cherry picks are better when working with git-svn tracking branches. In effect my solution turns a merge into a squash merge after the fact.
Best answer to the question
How would this be different to a soft reset? A soft reset also resets the repository to head but doesn't touch the working copy.
Yes, but does "git reset --soft someref" spring to mind when your goal is "How can I abort the merge?". How do you know what to use for someref? Hence "git merge --abort" which does the right thing and is obviously named which is refreshing for Git.
P
Peter Mortensen

Since Git 1.6.1.3 git checkout has been able to checkout from either side of a merge:

git checkout --theirs _widget.html.erb

D
Daniel

Might not be what the OP wanted, but for me I tried to merge a stable branch to a feature branch and there were too many conflicts. I didn't manage to reset the changes since the HEAD was changed by many commits, So the easy solution was to force checkout to a stable branch. you can then checkout to the other branch and it will be as it was before the merge.

git checkout -f master

git checkout side-branch


m
marc_s

To avoid getting into this sort of trouble one can expand on the git merge --abort approach and create a separate test branch before merging.

Case: You have a topic branch, it wasn't merged because you got distracted/something came up/you know but it is (or was) ready.

Now is it possible to merge this into master?

Work in a test branch to estimate / find a solution, then abandon the test branch and apply the solution in the topic branch.

# Checkout the topic branch
git checkout topic-branch-1

# Create a _test_ branch on top of this
git checkout -b test

# Attempt to merge master
git merge master

# If it fails you can abandon the merge
git merge --abort
git checkout -
git branch -D test  # we don't care about this branch really...

Work on resolving the conflict.

# Checkout the topic branch
git checkout topic-branch-1

# Create a _test_ branch on top of this
git checkout -b test

# Attempt to merge master
git merge master

# resolve conflicts, run it through tests, etc
# then
git commit <conflict-resolving>

# You *could* now even create a separate test branch on top of master
# and see if you are able to merge
git checkout master
git checkout -b master-test
git merge test

Finally checkout the topic branch again, apply the fix from the test branch and continue with the PR. Lastly delete the test and master-test.

Involved? Yes, but it won't mess with my topic or master branch until I'm good and ready.


M
Malcolm Boekhoff

I found the following worked for me (revert a single file to pre-merge state):

git reset *currentBranchIntoWhichYouMerged* -- *fileToBeReset*

N
Nimantha

Sourcetree

Because you not commit your merge, then just double click on another branch (which mean checkout it) and when sourcetree ask you about discarding all changes then agree.

Note

This answer is addressed to those who use SourceTree as git client.


The OP is obviously using the command line. There is zero value to your post.
I agree that this is obvious that OP use command line - but people who not use command line but sourcetree and have similar problem find this question at the top in google (like I) - so for such persons this answer has value - this is why I left it here :)
No really because a merge conflict could be for different reasons. Nevertheless behind what you do on GUI in SourceTree, are running command line command. On top I would say be careful with GUI because any unexpected such low internet connection will cause your GUI to go out of sync with git index and you may face a corrupted flow with a git.lock that is hard to "fix" except to remove it, and abort the merge conflict