I stumbled upon this while working on DFDL-777. "testBinaryLongRunningForProfiling" was failing due to the expected result not matching the actual, somehow this was causing a roughly 700% increase in performance in "testLongRunningForProfiling", which is the test that immediately follows the failed binary test.

As soon as I fixed testBinaryLongRunningForProfiling so that the expected result was correct, testLongRunningForProfiling went back to its usual speed.

It seems very interesting to me that the result of a test would effect the performance of a different test, especially to such a degree. Also, why is the test so much faster when the previous test fails? Is TDMLRunner holding something back when the test gets run correctly?

This should be easily reproducible, all you need to do is change the expected result for the "PCAP1" parserTestCase in daffodil-perf/src/test/resources/edu/illinois/ncsa/daffodil/profiling/Profiling.tdml