Summary

1. when a test (or any function called from the test) flushes or cleans the output buffer, this incorrectly changes the output making tests pass that fail with expectOutputString and/or would cause a PHP notice when the code is actually executed
2. if a test ends but starts a new output buffer, no error is reported, since only the ob level is compared instead of comparing the output
3. when a test (or any function called from the test) register a non-removable ob callback, this leads to complete failure/unexpected behavior
4. non-removable ob callbacks cannot currently be tested
5. if stopping output buffering fails, it will try again creating the same notices twice

I will provide a PR.