[15] RFR(T) : 8249697 : java/lang/invoke/RicochetTest.java should use @requires instead of @ignore

[Mandy Chung]

webrev.03 looks good.

Mandy

On 7/20/20 12:22 PM, Igor Ignatyev wrote:
> Hi Mandy,
>
> you are right, it's better to have just one @run, and as I don't think 
> that 7197210 changes '-XX:-VerifyDependencies' nor '/timeout=3600' are 
> needed anymore, I suggest to restore the test to its original version 
> w/  `@run junit/othervm -DRicochetTest.MAX_ARITY=255 
> test.java.lang.invoke.RicochetTest`, so the patch 
> (http://cr.openjdk.java.net/~iignatyev//8249697/webrev.03) would be just:
>
>> -/* @test
>> +/*
>> + * @test
>>   * @summary unit tests for recursive method handles
>> - * @run junit/othervm/timeout=3600 -XX:+IgnoreUnrecognizedVMOptions 
>> -XX:-VerifyDependencies -DRicochetTest.MAX_ARITY=10 
>> test.java.lang.invoke.RicochetTest
>> - */
>> -/*
>> - * @ignore The following test creates an unreasonable number of 
>> adapters in -Xcomp mode (7049122)
>>   * @run junit/othervm -DRicochetTest.MAX_ARITY=255 
>> test.java.lang.invoke.RicochetTest
>>   */
>
> and then the bug's summary would be smth like 'remove temporary fixes 
> from java/lang/invoke/RicochetTest.java' .
>
> sure there is no reason for it to be pushed into 15, I've retargeted 
> to 16.
>
> -- Igor
>
>> On Jul 20, 2020, at 11:57 AM, Mandy Chung <mandy.chung at oracle.com 
>> <mailto:mandy.chung at oracle.com>> wrote:
>>
>> Hi Igor,
>>
>> OK.  Should this revert the change by 7049122 then? i.e. simply 
>> change -DRicochetTest.MAX_ARITY=10 to 255
>>
>> Your proposed patch adds a new @run instead of modifying the existing 
>> @run command:
>>
>>   * @run junit/othervm/timeout=3600 -XX:+IgnoreUnrecognizedVMOptions 
>> -XX:-VerifyDependencies -DRicochetTest.MAX_ARITY=10 
>> test.java.lang.invoke.RicochetTest
>>
>> I looked at the history and this @run was modified by JDK-7197210 
>> that adds -XX:+IgnoreUnrecognizedVMOptions -XX:-VerifyDependencies 
>> options and reduce MAX_ARITY from 50 to 10.
>>
>> This issue is not critical to target for 15.  It may worth 
>> considering target this test fix for 16.  Just a suggestion.
>>
>> Mandy
>>
>> On 7/20/20 10:13 AM, Igor Ignatyev wrote:
>>> Hi Mandy,
>>>
>>> that's actually the opposite, the 2nd subtest is run only in modes 
>>> other than Xcomp, as w/ Xcomp the test creates lots of adapters and 
>>> used to lead to JVM failure as described in 7049122. I tried to 
>>> reproduce this failure, but in vain,..  after a bit more historical 
>>> digging, I realized that the underlying problem was 7009641, which 
>>> has been fixed in hs25/jdk8. so I've changed the fix for 8249697 to 
>>> simply return run w/ '-DRicochetTest.MAX_ARITY=255': 
>>> http://cr.openjdk.java.net/~iignatyev//8249697/webrev.02
>>>
>>> I've verified that the test passes w/ Xcomp and
>>>  - -XX:+TieredCompilation (c1 + c2);
>>>  - -XX:-TieredCompilation (c2-only);
>>>  - -XX:+NeverActAsServerClassMachine (emulated-client, c1-only)
>>>
>>> the test was run 100 times on {linux,windows,macos}-x64 w/ 0 failures.
>>> Thanks,
>>> -- Igor
>>>
>>>> On Jul 18, 2020, at 9:32 PM, Mandy Chung <mandy.chung at oracle.com 
>>>> <mailto:mandy.chung at oracle.com>> wrote:
>>>>
>>>>
>>>>
>>>> On 7/17/20 8:54 PM, Igor Ignatyev wrote:
>>>>> http://cr.openjdk.java.net/~iignatyev/8249697/webrev.00/
>>>>>
>>>>
>>>> I suggest to change this:
>>>>   32  * @comment The following test creates an unreasonable number 
>>>> of adapters in -Xcomp mode (7049122)
>>>>
>>>> To:
>>>>
>>>>    @bug 8249697
>>>>    @summary verify very high number of adapters in -Xcomp mode
>>>>
>>>> Otherwise, looks fine.
>>>>
>>>> Mandy
>>>>> Hi all,
>>>>>
>>>>> could you please review this small and trivial patch for java/lang/invoke/RicochetTest.java test?
>>>>> from JBS:
>>>>>> a run of java/lang/invoke/RicochetTest.java w/ MAX_ARITY=255 was removed from all configurations by JDK-7049122, yet the problem manifests itself only w/ Xcomp. as now we have @requires to filter out tests from certain configurations, the test can be updated to run MAX_ARITY=255 in all configs but Xcomp.
>>>>> the patch splits the test into two subtests, each one w/ one @run, and use @requires to exclude one w/ MAX_ARITY=255 from execution if Xcomp flag is used.
>>>>>
>>>>> JBS:https://bugs.openjdk.java.net/browse/JDK-8249697
>>>>> webrev:http://cr.openjdk.java.net/~iignatyev/8249697/webrev.00/
>>>>> testing: java/lang/invoke/RicochetTest.java on {linux,windows,macos}-x64 w/ and w/o -Xcomp; Xcomp runs, as expected, had only 1 test run
>>>>>
>>>>> Thanks,
>>>>> -- Igor
>>>>>
>>>>> JDK-7049122 :https://bugs.openjdk.java.net/browse/JDK-7049122
>>>>
>>>
>>
>