There was no check on context > stdout, so requesting more context than
was actually printed by the test could result in a negative value.
Python "helpfully" interpreted this as a negative index, resulting in
somewhat random context lengths.
This, combined with my tendency to just default to a large number like
--context=100, led to me thinking a test was printing much less than it
actually was...
Don't get me wrong, I love Python, and I think Python's negative indices
are a clever way to add flexibility to slice notation, but the
value-dependent semantics are a pretty unfortunate footgun...
Christopher Haster
c3dc7cca10