Three-Way Merge

A three-way merge combines two divergent versions by comparing each against their common ancestor — the merge base. Knowing the ancestor is what lets the algorithm distinguish "this side changed it" from "that side changed it", and to combine non-overlapping changes automatically.

The merge base is why comparisons must be taken against the ancestor rather than against the current tip of the other branch. Diffing a branch against the current main attributes every change made on main since the branch was cut to the branch itself. The three-dot form of a diff selects the merge base and is almost always what is wanted.

Two things merges cannot do.

They cannot detect semantic incompatibility. Non-overlapping changes are combined without complaint even when they contradict each other; see Semantic Conflict. Conflict absence is not agreement.

They cannot resolve genuine overlap, which is a Merge Conflict and requires a human decision.

There is also a duplication hazard: two branches that each added the same content in the same place can merge with no conflict markers and produce it twice, which a compiler catches and a reader might not.