RFR: 8170595: Optimize Class.isAnonymousClass

[Joseph D. Darcy]

Hello,

I think work like this merits some additional test development to make 
sure the new, faster implementation doesn't introduce any regressions in 
correctness.

 From a quick look in the java/lang/Class test directory, I didn't see 
any existing tests probing the correctness of isAnonymous; although 
there are some uses of that method.

Thanks,

-Joe

On 12/1/2016 9:34 AM, Claes Redestad wrote:
> Hi,
>
> sadly, avoiding getSimpleBinaryName is where most of the gain comes 
> from, avoiding
> getSimpleName only improves marginally on the baseline:
>
> Benchmark                       Mode  Cnt    Score    Error  Units
> Clazz.isAnonymousClass_Anon     avgt   10  192.522 ± 33.214  ns/op
> Clazz.isAnonymousClass_Regular  avgt   10  115.770 ± 19.953  ns/op
>
> Wrapping the checkPermission call inside getEnclosingMethod so that
> Reflection.getCallerClass() doesn't happening improves things a bit, 
> but still
> not close:
>
> Clazz.isAnonymousClass_Anon     avgt   10  187.074 ± 38.128  ns/op
> Clazz.isAnonymousClass_Regular  avgt   10   96.196 ± 15.709  ns/op
>
> /Claes
>
> On 2016-12-01 17:57, Mandy Chung wrote:
>> Hi Claes,
>>
>> What is the performance difference if this method calls 
>> getSimpleBinaryName?  Your patch exposes the implementation details 
>> and sensitive to the change there, if any.
>>
>> Mandy
>>
>>> On Dec 1, 2016, at 5:38 AM, Claes Redestad 
>>> <claes.redestad at oracle.com> wrote:
>>>
>>> Hi,
>>>
>>> due to recent interest to optimize Class.isAnonymousClass[1] I took
>>> a look at the implementation and found that we can further improve
>>> performance of this method, especially when asking non-anonymous
>>> classes[2].
>>>
>>> As such calls are a common occurrence during startup and bootstrap
>>> of lambdas this actually appears rather worthwhile:
>>>
>>> Webrev: http://cr.openjdk.java.net/~redestad/8170595/webrev.01/
>>> Bug: https://bugs.openjdk.java.net/browse/JDK-8170595
>>>
>>> Thanks!
>>>
>>> /Claes
>>>
>>> [1] 
>>> http://mail.openjdk.java.net/pipermail/core-libs-dev/2016-December/045116.html
>>> [2]
>>> Benchmark                       Mode  Cnt    Score    Error Units
>>> Clazz.isAnonymousClass_Anon     avgt   50  200.900 ± 15.503 ns/op
>>> Clazz.isAnonymousClass_Regular  avgt   50  136.896 ±  9.605 ns/op
>>>
>>> Clazz.isAnonymousClass_Anon     avgt   50  186.564 ± 12.219 ns/op
>>> Clazz.isAnonymousClass_Regular  avgt   50   33.878 ±  1.524 ns/op
>>>
>>> See bug for benchmark source
>