The run-tests script puts the test output in the current working directory but the --print-failure and --clean-failure arguments look for test output in the tests/ directory. As a result the two only work together if the working directory for running the tests is effectively the tests/ directory. I found this out when running tests by calling run-tests by full path from wherever the cwd is: it would never print failures.

It would be better if the script would stick to one thing: either read test output from the cwd as well, or else produce it in the tests/ directory (which would be my favorite: not a fan of relying on cwd except for commandline usage). The latter could be extended by allowing to specify the output directory as an argument. Ideas of which way to go?