Comment by skydhash

You’re looking at it from the perspective of a human reasoning. But a computer is a simple machine (what it can do, not how it does it). What seems obvious to you could be a complicated algorithm.

Git store all its information as a directed acyclic graph (a tree) of commits. The leaves of that tree have names, and are what we called branches. Each commit points to a tree (also a tree data structure) where the nodes are blobs (files) and sub trees. But that tree only stores the files that has been changed since the last commit. Git does not store diffs. Diffs are computed as needed.

This why the common ancestor commit is important. From there, a version of the working directory is computed for each branch (main-with-squashed-A and PR B). Files that have not been changed since PR A are ok, but everything else will be different, especially if you’ve modified the same lines.

Squashed A is a brand new commit with a new tree that PR B does not know about. You need to recompute PR B on top of Squashed A, (which will create new commits for PR B).