I mean... does every single patch need a regression test? If I did free(p); p[i] = 1; and then I fixed it by doing p[i] = 1; free(p); do I really need a regression test to trigger the dumb use-after-free I'd introduced?
Adding this kind of non regression test is costly, but it protects against source code management mistakes.
You mean the answer could be "yes, every single patch must have a regression test"?