I assumed that the “message” parameter could be used to match any substring of a warning message, but it turns out it calls regex.match(), which is anchored at the beginning of the search string (but is not anchored at the end). I propose this patch to clarify the documentation, and add some tests.