In meas_algorithms tests/testPsfDetermination.py has a test testRejectBlends which does not operate as expected. When it calls pcaPsfDeterminer it results in no usable psf candidates BEFORE blends are rejected. Formerly this resulted in a numpy array named "sizes" containing one uninitialized value, which might raise an unexpected exception or raise the desired exception, depending on whether that value was negative or positive.
DM-3117 I pushed a fix for the bug that caused the invalid "sizes" array, but the unit test is now reliably broken because no viable psf candidates raises the wrong exception and does not test blend rejection in any case. So on this same ticket I have commented out the bad test for now.