Thanks Alan!
On 17 sep 2012, at 11:40, Alan Bateman <[email protected]> wrote:
> On 17/09/2012 10:16, Staffan Larsen wrote:
>> The test sometimes fails with a ratio somewhere in the low 100s. If the
>> timing was indeed linear in the number of listeners, the ratio would be
>> 20000. Increasing the allowed ration to 500 (from 100) should therefore
>> still catch the real regression, but make the test more robust.
>>
>> Thanks,
>> /Staffan
>>
>>
>> diff --git
>> a/test/javax/management/remote/mandatory/notif/ListenerScaleTest.java
>> b/test/javax/management/remote/mandatory/notif/ListenerScaleTest.java
>> --- a/test/javax/management/remote/mandatory/notif/ListenerScaleTest.java
>> +++ b/test/javax/management/remote/mandatory/notif/ListenerScaleTest.java
>> @@ -45,7 +45,7 @@
>> *
>> * As usual with timing-sensitive tests, we could potentially get
>> * sporadic failures. We fail if the ratio of the time with many
>> - * MBeans to the time with just one MBean is more than 100. With the
>> + * MBeans to the time with just one MBean is more than 500. With the
>> * fix in place, it is usually less than 1, presumably because some
>> * code was being interpreted during the first measurement but had
>> * been compiled by the second.
>> @@ -176,7 +176,7 @@
>> long manyMBeansTime = timeNotif(mbs);
>> System.out.println("Time with many MBeans: " + manyMBeansTime +
>> "ns");
>> double ratio = (double) manyMBeansTime / singleMBeanTime;
>> - if (ratio> 100.0)
>> + if (ratio> 500.0)
>> throw new Exception("Failed: ratio=" + ratio);
>> System.out.println("Test passed: ratio=" + ratio);
>> }
> Looks okay to me, and we can see over the next few weeks if this is good
> enough.
>
> -Alan
