> It is often easier to write expected-to-fail tests this way (so that
> they can all look the same), rather than have to have, for example, an
> extra driver that converts expected errors into success codes for the
> automake test harness.

What do you mean precisely by “an extra driver”?