RFR(S): 8199532: AbstractMethodErrorTest.java test failed with -Xcomp

[David Holmes]

For the record I think what was done here to get this to run okay is 
truly awful. I think this is the wrong approach for dealing with tests 
that need to explicitly test interpreter behaviour or JIT behaviour. If 
the incoming jtreg flags can disrupt the test then we need to run the 
test in a way that is isolated from those flags e.g by directly exec'ing 
a VM with the right flags -Xint, -Xcomp etc.

David

On 15/03/2018 6:30 AM, Lindenmaier, Goetz wrote:
> Hi Vladimir,
> 
> Thanks for pushing!
> And sorry for causing the trouble.
> 
> Best regards,
>    Goetz.
> 
>> -----Original Message-----
>> From: Vladimir Kozlov [mailto:vladimir.kozlov at oracle.com]
>> Sent: Wednesday, March 14, 2018 7:04 PM
>> To: Lindenmaier, Goetz <goetz.lindenmaier at sap.com>; David Holmes
>> <david.holmes at oracle.com>; hotspot-runtime-dev at openjdk.java.net
>> Subject: Re: RFR(S): 8199532: AbstractMethodErrorTest.java test failed with -
>> Xcomp
>>
>> Good. I added webrev to bug report.
>> I will let you know when testing finished.
>>
>> Vladimir
>>
>> On 3/14/18 10:58 AM, Lindenmaier, Goetz wrote:
>>> Hi Vladimir,
>>>
>>> I added the changes you proposed to a new webrev:
>>> http://cr.openjdk.java.net/~goetz/wr18/8199532-
>> fixAmeExMsg/webrev.03/
>>>
>>> I also added all the flags to the IncompatibleClassChangeErrorTest.java.
>>> It needs as least compMode==Xmixed, it fails with Xint. It passes with -
>> Xcomp etc.
>>> If you think it's not necessary I'll just remove it again.
>>>
>>> Best regards,
>>>    Goetz.
>>>
>>>
>>>> -----Original Message-----
>>>> From: Vladimir Kozlov [mailto:vladimir.kozlov at oracle.com]
>>>> Sent: Wednesday, March 14, 2018 5:51 PM
>>>> To: Lindenmaier, Goetz <goetz.lindenmaier at sap.com>; David Holmes
>>>> <david.holmes at oracle.com>; hotspot-runtime-dev at openjdk.java.net
>>>> Subject: Re: RFR(S): 8199532: AbstractMethodErrorTest.java test failed
>> with -
>>>> Xcomp
>>>>
>>>> Thank you, Goetz
>>>>
>>>> You need additional check TieredStopAtLevel == null in @require
>>>> otherwise it does not execute even without any external flags:
>>>>
>>>> - * @requires !(os.arch=="arm")
>>>> + * @requires !(os.arch=="arm") & vm.flavor == "server" &
>>>> !vm.emulatedClient & vm.compMode=="Xmixed" & (!vm.graal.enabled |
>>>> vm.opt.TieredCompilation == true) & (vm.opt.TieredStopAtLevel == null |
>>>> vm.opt.TieredStopAtLevel==4)
>>>>
>>>> I also tested with Graal JIT and it passed with above changes but it
>>>> require Tiered switched ON otherwise it will timeout (since Graal run in
>>>> Interpreter until it compiles itself). That is why there is additional
>>>> Graal's conditions in @require.
>>>>
>>>> But it is not enough when Tiered is off. For that you need to increase
>>>> iterations (10000 is at the border of triggering C2 compilation) or
>>>> better add -XX:CompileThreshold=1000 flag to @run command - it would
>>>> require less time to run test:
>>>>
>>>> - *                   -XX:-BackgroundCompilation -XX:-Inline
>>>> + *                   -XX:CompileThreshold=1000
>>>> -XX:-BackgroundCompilation -XX:-Inline
>>>>
>>>> I started testing with this changes
>>>>
>>>> Thanks,
>>>> Vladimir
>>>>
>>>> On 3/14/18 7:53 AM, Lindenmaier, Goetz wrote:
>>>>> Hi,
>>>>>
>>>>> this webrev adds @requires statements so that the test is skipped in the
>>>>> test runs below. With C1 the explicit compilation won't work, anyways.
>>>>>
>>>>> Please review:
>>>>> http://cr.openjdk.java.net/~goetz/wr18/8199532-
>>>> fixAmeExMsg/webrev.02/
>>>>>
>>>>> Best regards,
>>>>>      Goetz.
>>>>>
>>>>>
>>>>>> -----Original Message-----
>>>>>> From: hotspot-runtime-dev [mailto:hotspot-runtime-dev-
>>>>>> bounces at openjdk.java.net] On Behalf Of David Holmes
>>>>>> Sent: Mittwoch, 14. März 2018 01:10
>>>>>> To: Vladimir Kozlov <vladimir.kozlov at oracle.com>; hotspot-runtime-
>>>>>> dev at openjdk.java.net
>>>>>> Subject: Re: RFR(S): 8199532: AbstractMethodErrorTest.java test failed
>>>> with -
>>>>>> Xcomp
>>>>>>
>>>>>> On 14/03/2018 7:41 AM, Vladimir Kozlov wrote:
>>>>>>> Thank you, Goetz
>>>>>>>
>>>>>>> -Xcomp passed now.
>>>>>>>
>>>>>>> I tried other flags we use for testing. And it failed when Tiered is
>>>>>>> off: -XX:-TieredCompilation or -XX:+TieredCompilation
>>>>>>> -XX:TieredStopAtLevel=1.
>>>>>>>
>>>>>>> Stupid question. Why we have different messages for compiler and
>>>>>>> interpreter? The problem is the same.
>>>>>>
>>>>>> The difference is whether information on the receiver class is available.
>>>>>>
>>>>>> David
>>>>>>
>>>>>>> -Xcomp flag also set -Xbatch flag so you don't need it in comment:
>>>>>>>
>>>>>>> +    // Seen if test started with -Xcomp -Xbatch:
>>>>>>>
>>>>>>> Thanks,
>>>>>>> Vladimir
>>>>>>>
>>>>>>> On 3/13/18 2:14 PM, Lindenmaier, Goetz wrote:
>>>>>>>> Hi,
>>>>>>>>
>>>>>>>> AbstractMethodErrorTest fails with -Xcomp because
>>>>>>>> other exception messages are printed if the error happens
>>>>>>>> in comiled code.
>>>>>>>> I hardened the test.
>>>>>>>> Please review:
>>>>>>>> http://cr.openjdk.java.net/~goetz/wr18/8199532-
>>>>>> fixAmeExMsg/webrev.01/
>>>>>>>>
>>>>>>>> Best regards,
>>>>>>>>       Goetz.
>>>>>>>>