Change for 8248135: Build microbenchmarks with --enable-preview makes other non-preview JMH benchmarks to fail

[Peter Levart]

A quick work-around for anyone wanting to run the microbenchmarks is to 
pass --enable-preview explicitly. For example:


make test TEST="micro:java.util.stream.ops" 
MICRO="JAVA_OPTIONS=--enable-preview"


Regards, Peter


On 7/7/20 10:23 AM, Peter Levart wrote:
>
> On 7/7/20 10:13 AM, David Holmes wrote:
>> Hi Peter,
>>
>> cc Claes
>>
>> On 7/07/2020 5:59 pm, Peter Levart wrote:
>>> Hi,
>>>
>>>
>>> Recently I proposed and pushed a change for [1] which adds 
>>> --enable-preview option to javac compilation of JMH microbenchmarks 
>>> in general to enable running a benchmark that uses preview feature 
>>> (Records). This makes the class files produced marked with version 
>>> 60.65535. The benchmark that uses preview feature executes without 
>>> problems because it explicitly specifies the following in its code:
>>>
>>>
>>> @Fork(value = 1, warmups = 0, jvmArgsAppend = "--enable-preview")
>>>
>>>
>>> Recently I wanted to run JMH benchmarks for Stream ops with:
>>>
>>>
>>> make test TEST="micro:java.util.stream.ops"
>>>
>>>
>>> ...but all of them fail to run with the following exception:
>>>
>>>
>>> java.lang.UnsupportedClassVersionError: Preview features are not 
>>> enabled for 
>>> org/openjdk/bench/java/util/stream/ops/value/generated/NoneMatchShort_seq_start_jmhTest 
>>> (class file version 60.65535). Try running with '--enable-preview'
>>>
>>>
>>> What shall we do? Add similar annotation to all of them? Is there a 
>>> way to specify that all micro benchmarks should be run with 
>>> --enable-preview option passed to java?
>>
>> So this breaks running all non-preview using benchmarks? If so I say 
>> we need to backout the change for 8248135 while a proper solution is 
>> found.
>
>
> I guess it does break (at least the way I tried to run them). The 
> problem is that this little change:
>
>
> --- a/make/test/BuildMicrobenchmark.gmk    Wed Jun 24 01:02:19 2020 +0200
> +++ b/make/test/BuildMicrobenchmark.gmk    Wed Jun 24 11:05:09 2020 +0200
> @@ -90,10 +90,11 @@
>      TARGET_RELEASE := $(TARGET_RELEASE_NEWJDK_UPGRADED), \
>      SMALL_JAVA := false, \
>      CLASSPATH := $(MICROBENCHMARK_CLASSPATH), \
> -    DISABLED_WARNINGS := processing rawtypes cast serial, \
> +    DISABLED_WARNINGS := processing rawtypes cast serial preview, \
>      SRC := $(MICROBENCHMARK_SRC), \
>      BIN := $(MICROBENCHMARK_CLASSES), \
>      JAVA_FLAGS := --add-modules jdk.unsupported --limit-modules 
> java.management, \
> +    JAVAC_FLAGS := --enable-preview, \
>  ))
>
>
> ...was pushed as part of larger fix for 8247532 which has already been 
> forward and backported. So I think backing out the whole patch (which 
> is perfectly OK by itself) would cause more problems then fixing this 
> particular problem in a followup, given that we can find a fix 
> quickly. Its has been 14 days since the above was pushed and nobody 
> noticed until now, so I guess this is not a big problem?
>
>
> Regards, Peter
>
>
>
>>
>> David
>> -----
>>
>>>
>>> [1] https://bugs.openjdk.java.net/browse/JDK-8248135
>>>
>>>
>>> Regards, Peter
>>>
>>>