RFR(T/S): 8236975 "compiler/graalunit tests fails with --illegal-access=deny"

Alan Bateman Alan.Bateman at oracle.com
Thu Mar 26 17:55:38 UTC 2020



On 26/03/2020 16:23, Ekaterina Pavlova wrote:
> On 3/26/20 1:11 AM, Alan Bateman wrote:
>> On 25/03/2020 22:04, Ekaterina Pavlova wrote:
>>> Hello,
>>>
>>> Please review the changes which allow to specify additional VM flags 
>>> to be used to run particular Graal unit tests.
>>> The fix makes Graal unit tests which previously failed with 
>>> --illegal-access=deny to pass now.
>>>
>>>
>>>       JBS: https://bugs.openjdk.java.net/browse/JDK-8206113
>>>    webrev: 
>>> http://cr.openjdk.java.net/~epavlova//8236975/webrev.00/index.html
>>>   testing: run graalunit tests with and without 
>>> "--illegal-access=deny" flag in mach5
>> The jtreg way to do this is `@modules java.base/java.lang:+open` and 
>> it would run the test with --add-open 
>> java.base/java.lang=ALL-UNNAMED. Is the reason for the approach in 
>> the webrev because the options aren't passed through to the VM run 
>> using the Graal compiler? 
>
> Yes, this is the case.
> Graal unit tests are launched by GraalUnitTestLauncher using 
> main/othervm mode.
> So specifying `@modules java.base/java.lang:+open` will not be passed 
> into the VM launched by GraalUnitTestLauncher.
>
>
>> They look okay of course and if you want to avoid using :SPACE then 
>> you can use '=' as an alternative (JEP 293 has more on this).
>
> I looked at java help and did not see it there, but yes I see it is 
> mentioned in the JEP 293
> "Tools that show command-line help for a long-form option that takes 
> an argument should normally show the option using a space to separate 
> the argument, and provide a note at the end that = may also be used as 
> a separator."
>
> Thanks for pointing into this, I have removed using :SPACE: and 
> regenerated the webrew.
>  http://cr.openjdk.java.net/~epavlova//8236975/webrev.00/index.html
This looks okay to me.

-Alan


More information about the hotspot-compiler-dev mailing list