First-class conflicts

Introduction

Conflicts can happen when two changes are applied to some state. This document is about conflicts between changes to files (not about conflicts between changes to bookmark targets, for example).

For example, if you merge two branches in a repo, there may be conflicting changes between the two branches. Most DVCSs require you to resolve those conflicts before you can finish the merge operation. Jujutsu instead records the conflicts in the commit and lets you resolve the conflict when you feel like it.

Data model

When a merge conflict happens, it is recorded as an ordered list of tree objects linked from the commit (instead of the usual single tree per commit). There will always be an odd number of trees linked from the commit. You can think of the first tree as a start tree, and the subsequent pairs of trees to apply the diff between onto the start. Examples:

• If the commit has trees A, B, C, D, and E it means that the contents should be calculated as A+(C-B)+(E-D).
• A three-way merge between A and C with B as base can be represented as a commit with trees A, B, and C, also known as A+(C-B).

The resulting tree contents is calculated on demand. Note that we often don't need to merge the entire tree. For example, when checking out a commit in the working copy, we only need to merge parts of the tree that differs from the tree that was previously checked out in the working copy. As another example, when listing paths with conflicts, we only need to traverse parts of the tree that cannot be trivially resolved; if only one side modified lib/, then we don't need to look for conflicts in that sub-tree.

When merging trees, if we can't resolve a sub-tree conflict trivially by looking at just the tree id, we recurse into the sub-tree. Similarly, if we can't resolve a file conflict trivially by looking at just the id, we recursive into the hunks within the file.

See here for how conflicts are stored when using the Git commit backend.

Conflict simplification

Remember that a 3-way merge can be written A+C-B. If one of those states is itself a conflict, then we simply insert the conflict expression there. Then we simplify by removing canceling terms. These two steps are implemented in Merge::flatten() and Merge::simplify() in merge.rs.

For example, let's say commit B is based on A and is rebased to C, where it results in conflicts (B+C-A), which the user leaves unresolved. If the commit is then rebased to D, the result will be (B+C-A)+(D-C) (D-C comes from changing the base from C to D). That expression can be simplified to B+D-A, which is a regular 3-way merge between B and D with A as base (no trace of C). This is what lets the user keep old commits rebased to head without resolving conflicts and still not get messy recursive conflicts.

As another example, let's go through what happens when you back out a conflicted commit. Let's say we have the usual B+C-A conflict on top of non-conflict state C. We then back out that change. Backing out ("reverting" in Git-speak) a change means applying its reverse diff, so the result is (B+C-A)+(A-(B+C-A)), which we can simplify to just A (i.e. no conflict).