Our current test coverage for vertical text is somewhat lacking, as evident from e.g. issue 6387. That regression could easily have been avoided if the `taro` test-case would have been an `eq` test, as well as an `text` test.