Automate Merge Conflicts in git

Question

In our project we have parallel branches. Let’s suppose we have Branch_A and Branch_B. If a developer merges a change to file abc.js in Branch_A and another developer merges their changes to the same abc.js file in Branch_B.

Currently we gather all the merge conflicts weekly and developers are required to merge their changes manually from Branch_A to Branch_B. This manual merging is tedious during busy weeks and causes delay in meeting deadlines for Branch_B.

Is there any way to automate code merge conflict resolution.

We couldn’t come up with a automated code conflict resolution framework.

Answer 1

Merge conflicts cannot be resolved automatically, because Git already performs merging automatically, and merge conflicts are exactly those merges which failed to be merged automatically.

So, the statement that merge conflicts cannot be resolved automatically is tautologically true because the definition of the term merge conflict is that it is a merge which couldn't automatically be performed by Git.

There are, however, ways in which you could improve Git's abilities to automatically.

rerere

The simplest option is to activate Git's rerere (reuse recorded resolution) feature.

Rerere can not automatically resolve merge conflicts, but it can make sure that the same conflict only needs to be manually resolved once. The way rerere works is that it records how a conflict was resolved manually and then reuses this manual resolution every time the same merge conflict appears again.

This will not get rid of merge conflicts, but it will decrease the number of repeatedly having to resolve the same conflict. Any git command that performs merges will automatically run git rerere for you if it is enabled.

merge drivers

A merge driver tells Git how to merge whole files. There are a few builtin low-level drivers, but you can easily supply your own.

If you have a better way of merging files than simply doing it textually, you can write your own driver and use that for merging these files.

For example, if you add a column to a CSV file, this will change every line in the file. If you have another change which deletes one row, Git's standard line-based merge driver will not be able to merge those two changes automatically. But a merge driver that specifically knows about the semantics of CSV files could easily do that.

Another example would be XML, HTML, JSON, YAML, or other structured and semi-structured data serialization formats. Having a merge driver that operates on the object graph instead of the text files could be a major improvement.

And of course, for code written in a programming language, a custom merge driver could detect Refactorings such as Pull Up Method or Change Function Declaration and perform merging by composing those refactorings instead of composing line-based text changes.

merge strategies

Similar to your own merge driver, you can also supply your own merge strategy. A merge strategy foo is nothing more than an executable utility git-merge-foo placed in the current $PATH.

If you have additional context information about your branches that Git does not have, supplying your own merge strategy can allow you to merge changes that Git on its own does not know how to merge.

diff algorithms

You can choose different diff algorithms. While it is not 100% clear to me how diff algorithms influence the behavior of merges, the documentation indicates they they do.