I want an honest, technical comparison of two git merge workflows when an AI (LLM) is doing the conflict resolution. Be skeptical; point out where either claim is wrong.
WORKFLOW A - "Typical AI merge":
- Developer works on a feature branch (commits f1, f2) forked from main at commit B.
- Developer merges main into the feature branch (merge commit M1). Conflicts are resolved, and those decisions live inside M1.
- Developer continues on the feature branch (f3, f4).
- Repeat steps 2-3 several times (M2, M3...). Main may also receive squash-merges of other branches, some of which this feature branch had pulled in.
- Finally an AI resolves the last merge. It's given what a typical tool gives it: the conflict markers (ours/theirs hunks), maybe surrounding file context, maybe git log.
WORKFLOW B - "Reset-fork-point 3-way merge" (as in github.com/deanhiller/webpieces-ts):