jcushman/git-merge-tokenize

git-merge-tokenize

git-merge-tokenize is an experimental git merge driver that performs merges by splitting files based on arbitrary tokenizers, rather than based on linebreaks.

This is experimental software. Don't use it for anything important.

Install

npm install -g jcushman/git-merge-tokenize

Configure

In a git repo:

Add new merge drivers to .git/config:

echo '[merge "words"]
    name = git-merge-tokenize merge -- word level
    driver = git-merge-tokenize -t words -o %A %O %A %B
    recursive = binary
[merge "chars"]
    name = git-merge-tokenize merge -- character level
    driver = git-merge-tokenize -t chars -o %A %O %A %B
    recursive = binary' >> .git/config

Specify which files should use the alternate merge drivers in .git/info/attributes. For example, this will cause .xml files to use character-by-character merging, and .txt files to use word-by-word merging:

echo '*.xml merge=chars' >> .git/info/attributes
echo '*.txt merge=words' >> .git/info/attributes

Use

Let's see how this works using a new git repo configured as above:

$ echo 'Hello world.' > test.txt
$ git add test.txt
$ git commit -am 'init'
[master (root-commit) 5cec723] init
 1 file changed, 1 insertion(+)
 create mode 100644 test.txt

We're now on branch master and test.txt has contents Hello world.

Now let's make a change in a branch:

$ git checkout -b exclaim
Switched to a new branch 'exclaim'
$ echo 'Hello world!' > test.txt
$ git commit -am 'add exclamation'
[exclaim 3a1fcd4] add exclamation
 1 file changed, 1 insertion(+), 1 deletion(-)

On master, text.txt is Hello world.
On exclaim, text.txt is Hello world!

Now let's make a change to the same line on master:

$ git checkout master
Switched to branch 'master'
$ echo 'Goodbye world.' > test.txt
$ git commit -am 'Hello to goodbye'
[master 605f70d] Hello to goodbye
 1 file changed, 1 insertion(+), 1 deletion(-)

On master, text.txt is Goodbye world.
On exclaim, text.txt is Hello world!

If we were using the normal git merge driver, this would be a conflict. But we can now merge exclaim back to develop and get both changes:

$ git merge exclaim -m 'merge exclaim branch into master'
Auto-merging test.txt
Merge made by the 'recursive' strategy.
test.txt | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
$ cat test.txt
Goodbye world!

The same merge driver also lets us revert earlier changes. Here we'll revert the earlier hello-to-goodbye change while preserving the exclamation point change:

$ git revert HEAD~1
[master 9772b01] Revert "Hello to goodbye"
 1 file changed, 1 insertion(+), 1 deletion(-)
$ cat test.txt
Hello world!

Conflicts

Conflicts are reported similarly to the default git merge driver. Let's create a conflict where both branches try to add a word in the same place:

$ git checkout -b awesome
Switched to a new branch 'awesome'
$ echo 'Hello awesome world!' > test.txt
$ git commit -am "awesome"
[awesome a375a92] awesome
 1 file changed, 1 insertion(+), 1 deletion(-)
$ git checkout master
Switched to branch 'master'
$ echo 'Hello cool world!' > test.txt
$ git commit -am 'cool'
[master 11f1a7c] cool
 1 file changed, 1 insertion(+), 1 deletion(-)
$ git merge awesome
Auto-merging test.txt
CONFLICT (content): Merge conflict in test.txt
Automatic merge failed; fix conflicts and then commit the result.
$ cat test.txt
Hello <<<<<<<
cool
|||||||

=======
awesome
>>>>>>>world!

git-merge-tokenize's Issues

Allow plugging in of grammar specifications

Awesome work on this, thanks! I was stuck in the middle of an XML merge that was making me tear my hair out. I think it'd be really cool if you could specify grammars or arbitrary parser executables that produce a serializable parse tree of whatever you're merging. The merge operation could be performed on the object tree, then mapped back to character diffs. This would allow clean merges for many languages.