I apologize, but I cannot recreate the issue I saw last night. It's very strange given that the 'bad' build and 'good' build were installed with different --prefix flags. So they are 'make install'ed in completely separate directories.
Though my building directory is the same.
Here are the sequence of events:
1. I update my build to
e5e42ce diff-hunk-kill independent of point inside headers
2. Start a fresh emacsclient session, and see that issue.
3. I keep that session open, checkout the branch that last worked fine for me:
8c0fcaf Avoid inefficient regex in diff-refine-hunk
4. I take a screenshot of the bad build and kill that session.
5. Start another fresh emacsclient session using the build of the 'good' branch.
6. And everything looks good again.
Sorry guys, as the issue magically went away, I fear it will arrive again. Will open a bug report next time with more info as possible.