RFR(T) : 8195060 : vm/mlvm/anonloader/stress/byteMutation intermittently times out

[Igor Ignatyev]

actually, it's the opposite, the test wants to generate invalid classes, as its goal to check that loading of invalided classes as anonymous classes won't lead to a crash and/or hang. generating only invalid classes requires using of classfile fuzzy framework and will make this test even more complicated.

I've updated the RFE[1] w/ the request to harden this test, so it will always generate invalid classfiles. 

-- Igor

[1] https://bugs.openjdk.java.net/browse/JDK-8079699

> On Feb 13, 2019, at 9:57 AM, Vladimir Kozlov <vladimir.kozlov at oracle.com> wrote:
> 
> So the test filters out invalid class file before executing? Good then.
> 
> Thanks,
> Vladimir
> 
> On 2/13/19 9:35 AM, Igor Ignatyev wrote:
>> in its current state, the test changes 4 random bytes to random values, and w/ <clinit> being[1] I don't see it possible to get valid classfile w/ an infinite recursion in <clinit>.
>> Thanks,
>> -- Igor
>> [1]
>>>   static {};
>>>     descriptor: ()V
>>>     flags: (0x0008) ACC_STATIC
>>>     Code:
>>>       stack=6, locals=0, args_size=0
>>>          0: ldc           #8                  // String %65500c%X
>>>          2: iconst_2
>>>          3: anewarray     #9                  // class java/lang/Object
>>>          6: dup
>>>          7: iconst_0
>>>          8: bipush        99
>>>         10: invokestatic  #10                 // Method java/lang/Character.valueOf:(C)Ljava/lang/Character;
>>>         13: aastore
>>>         14: dup
>>>         15: iconst_1
>>>         16: invokestatic  #11                 // Method vm/mlvm/share/Env.getRNG:()Ljava/util/Random;
>>>         19: invokevirtual #12                 // Method java/util/Random.nextLong:()J
>>>         22: invokestatic  #13                 // Method java/lang/Long.valueOf:(J)Ljava/lang/Long;
>>>         25: aastore
>>>         26: invokestatic  #14                 // Method java/lang/String.format:(Ljava/lang/String;[Ljava/lang/Object;)Ljava/lang/String;
>>>         29: putstatic     #2                  // Field theDrumIsTheDrumIsTheDrumIsTheDrumIsTheDrumIsTheDrumIsTheDrumIsTheDrumIsTheDrumIsTheDrumIsTheDrumIsTheDrumIsTheDrumIsTheDrumIsTheDrumIsTheDrumIsTheDrumIsTheDrumIsTheDrumIsTheDrumIsTheDrumIsTheDrumIsTheDrumIsTheDrumIsTheDrum:Ljava/lang/String;
>>>         32: return
>>>       LineNumberTable:
>>>         line 31: 0
>>>         line 32: 10
>>>         line 31: 32
>>> On Feb 13, 2019, at 9:28 AM, Vladimir Kozlov <vladimir.kozlov at oracle.com> wrote:
>>> 
>>> Can it randomly generate infinite recursions in <clinit> method too so your fix would be not enough? Or only <init>?
>>> 
>>> Thanks,
>>> Vladimir
>>> 
>>> On 2/12/19 7:20 PM, Igor Ignatyev wrote:
>>>> http://cr.openjdk.java.net/~iignatyev/8195060/webrev.00/
>>>>> 2 lines changed: 0 ins; 0 del; 2 mod;
>>>> Hi all,
>>>> could you please review this one-line patch for vmTestbase/vm/mlvm/anonloader/stress/ tests? the goal of these tests is to check that loading of anonymous classes won't lead to a crash and/or a hang, byteMutation test uses a randomly changed class file, which in some cases might lead to infinite recursions in <init> method. as the tests don't need to create an instance of loaded anonymous class to achieve their goal, the patch replaces Class::newInstance call w/ Unsafe::ensureClassInitialized to guarantee that we try to load, link and initialize classes.
>>>> webrev: http://cr.openjdk.java.net/~iignatyev/8195060/webrev.00/
>>>> JBS: https://bugs.openjdk.java.net/browse/JDK-8195060
>>>> testing:
>>>> - vmTestbase/vm/mlvm/anonloader/stress/ tests
>>>> - manually verified that the tests still can fail by running w/ Xverify:none
>>>> Thanks,
>>>> -- Igor