While I cannot find the exact page anymore, I recently read a discussion of why running the tests twice is considered a good idea. The key issues cited were around the effects of the Cobertura byte code alteration on the accuracy of your tests. In certain cases the timing of your code execution might be important, the byte code alteration can cause tests that fail in JUnit to pass when run only in Cobertura and vice versa. For this reason, it was recommended that the tests be allowed to execute twice. Most of the examples cited were around multi-threaded behaviors, but I imagine that there could be other cases were the byte code alteration can cause issues in your tests. Having the tests execute both ways provides you with baseline results and also reduces the chances of sending you on a wild goose chase if Cobertura is in fact altering test success.
与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…