RFR(XS): 8183233: [MVT] "unexpected value type klass" assert in ValueTypeNode::is_loaded()

Tobias Hartmann tobias.hartmann at oracle.com
Wed Jul 5 14:23:36 UTC 2017


Hi Karen,

On 05.07.2017 15:46, Karen Kinnear wrote:
> It seems wrong to me to enable MVT if CompileTheWorld is true.
> It makes more sense to me to explicitly disable compiling java.lang.__Value if !EnableMVT && !EnableValhalla.

Yes, that's correct but I was concerned that there are other points of friction where we may encounter __Value although MVT/Valhalla is not enabled. For example, in Class::isValueClass() we reference __Value [1]. Looking at it again, it shouldn't be a problem (it actually is the same with -Xcomp).

That said, I agree that we should probably just disable compilation of __Value if CompileTheWorld is true. I'll prepare a new fix.

Thanks,
Tobias

[1] http://cr.openjdk.java.net/~mchung/valhalla/webrevs/8181425/revised-fix/jdk/src/java.base/share/classes/java/lang/Class.java.sdiff.html

>> On Jul 5, 2017, at 9:18 AM, Tobias Hartmann <tobias.hartmann at oracle.com> wrote:
>>
>> Thanks Roland!
>>
>> (I'll look at your changes as soon as possible, just very busy with other stuff atm).
>>
>> Best regards,
>> Tobias
>>
>> On 05.07.2017 15:09, Roland Westrelin wrote:
>>>
>>>> http://cr.openjdk.java.net/~thartmann/valhalla/vt_prototype/webrev.27/
>>>
>>> Looks ok to me.
>>>
>>> Roland.
>>>
> 


More information about the valhalla-dev mailing list