ShenandoahStrDedupStress.java test

[Aleksey Shipilev]

On 07/16/2018 06:40 PM, Zhengyu Gu wrote:
> On 07/16/2018 11:55 AM, Aleksey Shipilev wrote:
>> On 07/16/2018 04:04 PM, Zhengyu Gu wrote:
>>> What you observed is that, it takes longer than used to? or it simply takes too long?
>>
>> It simply takes too long, and that must tell us something about Shenandoah itself. Can you profile
>> the test with perf and see if we verify something excessively? Or there is some other peculiarity
>> that makes the test too long? Optimizing tests does pay off with development productivity.
>>
> The major issue is that, it contains *16* tests with different flag combinations, while each test
> takes a little over 1 minute.
> 
> I will break it up, and move most of experimental heuristics to tier3, and keep adaptive and
> traversal in tier2.
> 
> Make sense?

Breaking up is not very appealing, but I agree putting most important tests in tier2 is good. Can
you maybe introduce the "work size" parameter in the test, and make non-essential configuration run
with less work? See for example, LotsOfCycles test.

-Aleksey