Chapters

Hide chapters

Advanced Git

First Edition · Git 2.28 · Console

Section I: Advanced Git

Section 1: 7 chapters
Show chapters Hide chapters

2. Merge Conflicts
Written by Chris Belanger

The reality of development is that it’s a messy business; on the surface, it’s simply a linear progression of logic, a smattering of frameworks, a bit of testing — and you’re done. If you’re a solo developer, then this may very well be your reality. But for the rest of us who work on code that’s been touched by several, if not hundreds, if not thousands of other hands, it’s inevitable that you’ll eventually want to change the same bit of code that someone else has recently changed.

Imagine that your team’s project contains the following bit of HTML:

<p>Head over to the following link to learn how to get started with Git:</p>
<a href="http://guides.github.com/activities/hello-world/">link</a>

You’ve been tasked with updating all of the text of the links to something more descriptive, while your teammate has been tasked with changing HTTP URLs in this particular project to HTTPS.

At 9:00 a.m., your teammate pushes the following change to the piece of code to the project repository, to update http to https:

<p>Head over to the following link to learn how to get started with Git:</p>
<a href="https://guides.github.com/activities/hello-world/">link</a>

At 9:01 a.m. (because you were a little farther back in the coffee lineup that morning), you attempt to push the following change to the repository:

<p>Head over to the following link to learn how to get started with Git:</p>
<a href="http://guides.github.com/activities/hello-world/">GitHub’s Hello World project</a>

But, instead of Git committing your changes to the repository, you receive the following message instead:

 ! [rejected]        master -> master (fetch first)
error: failed to push some refs to 'https://github.com/supersites/git-er-done.git'
hint: Updates were rejected because the remote contains work that you do
hint: not have locally. This is usually caused by another repository pushing
hint: to the same ref. You may want to first integrate the remote changes
hint: (e.g., 'git pull ...') before pushing again.
hint: See the 'Note about fast-forwards' in 'git push --help' for details.

That’s something you’ve probably seen before. The remote has your teammate’s changes that you just haven’t yet pulled down to your local system. “Easy fix,” you think to yourself, so you execute git pull as suggested, and…

From https://github.com/supersites/git-er-done
   7588a5f..328aa94  master     -> origin/master
Auto-merging index.html
CONFLICT (content): Merge conflict in index.html
Automatic merge failed; fix conflicts and then commit the result.

Well, that didn’t go as planned. You were expecting Git to be smart to merge the contents of the remote, which contains the commits from your teammate, with your local changes. But, in this case, you and your teammate have changed the same line. And since Git, by design, doesn’t know anything about the language you’re working with, it doesn’t know that your changes won’t impact your teammate’s changes — and vice-versa. So Git plays it safe and bails, and asks you to do the work to merge the two files manually.

Welcome to the wonderful world of merge conflicts.

What is a merge conflict?

As a human, it’s fairly easy to see how two people modifying the same line of code in two separate branches could result in a conflict, and you could even argue that a halfway intelligent developer could easily work around that situation with a minimum of fuss. But Git can’t reason about these things in a rational manner as you or I would. Instead, Git uses an algorithm to determine what bits of a file have changed and if any of those bits could represent a conflict.

For simple text files, Git uses an approach known as the longest common subsequence algorithm to perform merges and to detect merge conflicts. In its simplest form, Git find the longest set of lines in common between your changed file and the common ancestor. It then finds the longest set of lines in common between your teammate’s changed file and the common ancestor.

Git aligns each pair of files along its longest common subsequence and then asks, for each pair of files, “What has changed between the common ancestor and this new file?” Git then takes those differences, looks again, and asks, “Now, of those changes in each pair of files, are there any sets of lines that have changed differently between each pair?” And if the answer is “Yes,” then you have a merge conflict.

To see this in action, you’ll start working through the sample project for this section of the book, and you’ll merge in some of your team’s branches in order to see that resolving merge conflicts isn’t quite as scary or frustrating as it looks on the surface.

Handling your first merge conflict

To get started, you’ll need to clone the magicSquareApp repository that’s used in this section of the book.

You can do this by way of the git clone command:

git clone https://github.com/raywenderlich/magicSquareJS.git

Once that’s done, navigate into the directory into which you cloned it.

Now, here’s the situation: Zach has been working on the front-end HTML of the magic square application to make it work with the back-end JavaScript. Zach isn’t a designer, so Yasmin has offered to lend her design skills to the project UI and style the front end so that it looks presentable.

As the project lead, you’re responsible for merging the various bits together and testing out the project. So, at this point, you’d like to verify that Zach’s HTML works properly with Yasmin’s UI. To do this, you’ll have to merge Zach’s work with Yasmin’s work, and then test the project locally.

Merging from another branch

Zach has been doing his work in the zIntegration branch, while Yasmin has been working in the yUI branch. Your job is to merge Yasmin’s branch with Zach’s branch and resolve any conflicts.

Pull down Yasmin’s branch so you have a local, tracked copy of the branch:

git checkout yUI

First, switch to Zach’s branch:

git checkout zIntegration

Open up index.html in a browser, to see what things look like in their current, pre-Yasminified state:

Well, it’s clear that Zach is no designer. Good thing we have Yasmin.

Now you need to merge in Yasmin’s UI branch:

git merge yUI

It appears that Zach and Yasmin’s work wasn’t completely decoupled, though, since Git indicates you have a merge conflict:

Auto-merging index.html
CONFLICT (content): Merge conflict in index.html
Automatic merge failed; fix conflicts and then commit the result.

Helpfully, Git tells you above what file or files contain the merge conflicts. Open up index.html and find the following section:

  <body>
    <h1>magicSquareJS</h1>
<<<<<<< HEAD
    <section>
      <input type="text" placeholder="Size"
        id="magic-square-size" />
      <a href="#"
        id="magic-square-generate-button">
          Generate Magic Square</a>
      <pre id="magic-square-display">
=======
    <section class="box">
      <input type="text" class="flex-item" placeholder="Size"/>
      <a href="#"
        class="flex-item btn" >Generate Magic Square</a>
      <pre class="flex-item" >
>>>>>>> yUI
      </pre>

OK, you admit that your HTML is a little rusty, but you’re pretty sure that <<<<<<< HEAD stuff isn’t valid HTML. What on earth did Git do to your file?

Understanding Git conflict markers

What you’re seeing here is Git’s representation of the conflict in your working copy. Git compared Yasmin’s file to the common ancestor, then compared Zach’s file to the common ancestor and found this block of code that had changed differently in each case.

In this case, Git is telling you that the HEAD revision (i.e., the latest commit on Zach’s branch) looks like the block between the <<< HEAD marker, and the === marker. The latest revision on Yasmin’s branch is the block contained between the === line and the >>> yUI marker.

Git puts both revisions into the file in your working copy, since it expects you to do the work yourself to resolve this conflict. If you were intimately familiar with the code in question, you might know exactly how to combine Zach’s and Yasmin’s code to get the desired result. But you skipped a few too many project design meetings, didn’t you?

No matter; you can ask Git to give you a few more clues as to what’s happened here. Remember that a merge in Git is a three-way merge, but by default Git only shows you the two child revisions in a merge conflict; in this case, Yasmin and Zach’s changes. It would be quite instructional to see the common ancestor for both of these child revisions, to figure out the intent behind each change.

Resolving merge conflicts

First, you need to return to the previous state of your working environment. Right now, you’re mid-merge, and you only have two choices at this point: Either go forward and resolve the merge, or roll back and start over. Since you want to look at this merge conflict from a different angle, you’ll roll back this merge and start over.

Reset your working environment with the following command:

git reset --hard HEAD

This reverts your working environment back to match HEAD, which, in this case, is the latest commit of your current branch, zIntegration.

A better way to view merge conflicts

Now, you can configure Git to show you the three-way merge data with the following command:

git config merge.conflictstyle diff3

Note: If you ever wanted to change back to the default merge conflict tagging, simply execute git config merge.conflictstyle merge to get rid of the common ancestor tagging.

To see the difference in the merge conflict output, run the merge again:

git merge yUI

Git explains patiently that yes, there’s still a conflict. In fact, this is a good time to see what Git’s view of your working tree looks like, before you go in and fix everything up. Execute the git status command, and Git shows you its understanding of the current state of the merge:

On branch zIntegration
Your branch is up to date with 'origin/zIntegration'.

You have unmerged paths.
  (fix conflicts and run "git commit")
  (use "git merge --abort" to abort the merge)

Changes to be committed:

  new file:   css/main.css

Unmerged paths:
  (use "git add <file>..." to mark resolution)

  both modified:   index.html

Most of that output makes sense, but the last bit is rather odd: both modified: index.html. But there’s only one index.html, isn’t there? Why does Git think there is more than one?

A consolidated git status

Remember that Git doesn’t always think about files, per se. In this case, Git is talking about both branches that are modified. To see this in a bit more detail, you can add the -s (--short) and -b (--branch) options to git status to get a consolidated view of the situation:

git status -sb

Git responds with the following:

## zIntegration...origin/zIntegration
A  css/main.css
UU index.html

The first two columns (showing A and UU) represent the “ours” versus “theirs” view of the code. The left column is your local branch, which currently is the mid-merge state of the original zIntegration branch mixed with the changes from the yUI branch. The right column is the remote branch. So this abbreviated git status command shows the following:

  • You have one file added (A) on your local branch; this is css/main.css that Yasmin must have added in her work. But it’s not in conflict with your work.
  • On the other hand, you have not one, but two revisions of a file that are unmerged (U) in your branch. This is the original index.html from the zIntegration branch, and the index.html from your yUI branch.

These files are considered unmerged because Git has halted partway through a merge, and put the onus on you to fix things up. Once you’ve fixed them up, committing those changes will continue the merge.

Editing conflicts

Open up index.html and have a look at the conflicted block of code now, with the new diff3 conflict style:

  <body>
    <h1>magicSquareJS</h1>
<<<<<<< HEAD
    <section>
      <input type="text" placeholder="Size" id="magic-square-size" />
      <a href="#" id="magic-square-generate-button">Generate Magic Square</a>
      <pre id="magic-square-display">
||||||| 69670e7
    <section>
      <input type="text" placeholder="Size"/>
      <a href="#">Generate Magic Square</a>
      <pre>
=======
    <section class="box">
      <input type="text" class="flex-item" placeholder="Size"/>
      <a href="#" class="flex-item btn" >Generate Magic Square</a>
      <pre class="flex-item" >
>>>>>>>yUI
      </pre>

There’s a new section in there: ||||||| 69670e7. This shows you the hash of the common ancestor of both Yasmin and Zach’s changes; that is, what the code looked like before each created their own branch. A visual comparison of HEAD (which is Zach’s branch) against the common ancestry shows the following changes:

  • Added id="magic-square-size" to the input tag
  • Added id="magic-square-generate-button" to the a (anchor) tag
  • Added id="magic-square-display" to the pre tag

A quick visual comparison of Yasmin’s changes against the common ancestor shows the following changes:

  • Added class="box" to the section tag
  • Added class="flex-item" to the input tag
  • Added class="flex-item btn" to the a tag
  • Added class="flex-item" to the pre tag

It looks like it will be less work to migrate Zach’s changes into Yasmin’s code. So edit index.html by hand, moving Zach’s new id attributes, from the first block of code in the conflicted section, into the third block in the conflicted section, which is Yasmin’s code.

When you’ve moved those three id attributes into Yasmin’s code, you can now delete the entire first two blocks from the conflicted section, from <<< HEAD all the way to ===. Then, delete the >>> yUI line as well. When you’re done, this section of code should look like the following:

  <body>
    <h1>magicSquareJS</h1>
    <section class="box">
      <input type="text" id="magic-square-size" class="flex-item" placeholder="Size"/>
      <a href="#" id="magic-square-generate-button" class="flex-item btn" >Generate Magic Square</a>
        <pre id="magic-square-display" class="flex-item" >
        </pre>

Save your work and return to the command line.

Completing the merge operation

You’ve finished resolving the conflict, so you can stage your changes with the following:

git add index.html

Execute the condensed version of git status -sb to see what Git thinks about your merge attempt:

## zIntegration...origin/zIntegration
A  css/main.css
M  index.html

There you are; one new file and one modified file. Git’s noticed that you’ve resolved the outstanding conflicts, so all that’s left to do to complete the merge is to commit your staged changes.

Commit those changes now, this time letting Git provide the merge message via Vim:

git commit

Type :wq inside of Vim to accept the preconfigured merge commit message, and Git dumps you back to the command line with a brief status, showing you that the merge succeeded:

[zIntegration af33aaa] Merge branch 'yUI' into zIntegration

Now, open index.html in a browser to see the changes:

That looks quite good. It’s not fully functional at the moment, but you can see that Yasmin’s styling changes are working. You’re free to delete her branch and merge this work into master.

First, delete the yUI branch:

git branch -d yUI

Switch to the master branch:

git checkout master

Now, attempt a merge of the zIntegration branch:

git merge zIntegration

Git takes you straight into Vim, which means the merge had no conflicts. Type :wq to save this commit message and complete the merge. Git responds with the results of the merge:

Merge made by the 'recursive' strategy.
 css/main.css | 268 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 index.html   |  16 +++++++----
 js/main.js   |  85 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 363 insertions(+), 6 deletions(-)
 create mode 100644 css/main.css

You’re now able to delete the zIntegration branch, so do that now:

git branch -D zIntegration

Note: The -D switch forces the local deletion of the branch regardless of its status. If you used the normal -d switch here, you’d see a warning about the branches changes not having been pushed to the remote.

Challenge

Challenge: Resolve another merge conflict

The challenge for this chapter is straightforward: resolve another merge conflict.

Xanthe has an old branch with some updates to the documentation; this work is in the xReadmeUpdates branch. You want to merge that work to master.

The steps are as follows:

  • Check out the xReadmeUpdates branch and look at the README.md file to see Xanthe’s version.
  • Check out master, since this is the destination for your merge.
  • Resolve any merge conflicts by hand.
  • Stage your changes.
  • Commit your changes.
  • Delete the xReadmeUpdates branch.

If you get stuck, or want to check your solution, you can always find the answer to this challenge under the challenge folder for this chapter.

Key points

  • Merge conflicts occur when you attempt to merge one set of changes with another, conflicting set of changes.
  • Git uses a simple three-way algorithm to detect merge conflicts.
  • If Git detects a conflict when merging, it halts the merge and asks for manual intervention to resolve the conflict.
  • git config merge.conflictstyle diff3 provides a three-way view of the conflict, with the common ancestor, “their” change, and “our” change.
  • git status -sb gives a concise view of the state of your working tree.
  • To complete a merge that’s been paused due to a conflict, you need to manually fix the conflict, add your changes, and then commit those changes to your branch.

Where to go from here?

In practice, merge conflicts can get pretty messy. And it might seem that, with a bit of intelligence, Git could detect that adding HTML attributes to a tag is not really a conflict. And there are, in fact, lots of tools, such as IDEs and their plugins, that are language-aware and can resolve conflicts like this easily, without making you make all the edits by hand. But no tool can ever replace the insight that you have as a developer, nor can it replace your intimate understanding of your code and its intent. So even though you may come across tools that seem to do most of the work of resolving merge conflicts for you, at some point you’ll find that there is no other way to resolve a merge conflict except by manual code surgery, so learning this skill now will serve you well in the future.

Up to now, your workflow has been constrained to the “happy path”: you can create commits, switch between branches, and generally get along quite well without being interrupted. But real life isn’t like that; you’ll more often than not be partway through working on a feature or a fix, when you want to switch your local branch to take a look at something else. But because Git works at the atomic level of the commit, it doesn’t like leaving things in an uncommitted state. So you need to stash the current state of your work somewhere, before you switch branches. And git stash, covered in the next chapter, does just that for you.

Have a technical question? Want to report a bug? You can ask questions and report bugs to the book authors in our official book forum here.
© 2024 Kodeco Inc.