Creating job with InputDir set to non-existant directory locks up jobtracker

This, in the very least, affects anything using the default listFiles() from InputFormatBase. If no files are enumerated, an exception is thrown... but the JobTracker keeps attempting to run listFiles() for this job. Trying to stop the job with hadoop job -kill job_name just results in timeouts, and further started jobs also don't progress. This happens every single time with, say, "wordcount", and a non-existent input path.