RFR: 8235257: Split 'assert(loader != NULL && oopDesc::is_oop(loader), "loader must be oop")'

[coleen.phillimore at oracle.com]

On 12/4/19 4:25 AM, Stefan Karlsson wrote:
> On 2019-12-03 15:39, coleen.phillimore at oracle.com wrote:
>> Actually, I was wondering if it would help also, to add an 
>> assert(Universe::heap()->is_oop(loader), "as well");
>
> It's one part of oopDesc::is_oop:
>
> bool oopDesc::is_oop(oop obj, bool ignore_mark_word) {
>   if (!Universe::heap()->is_oop(obj)) {
>     return false;
>   }
>
> but do you mean that you want us to explicitly check that before 
> calling oopDesc::is_oop?
>
> I'm going to push 8235257, and if this is important we can fix it as a 
> separate RFE.

I would have said this yesterday but that's fine.  You can push this.  
The failures in oopDesc::is_oop() had a non null oop so I thought this 
would help figure out that Universe::heap()->is_oop() was the reason, 
and not some transient zero for the markWord.  I don't think it's the 
latter though.

thanks,
Coleen

>
> StefanK
>
>>
>> Coleen
>>
>> On 12/3/19 9:32 AM, Stefan Karlsson wrote:
>>> Hi all,
>>>
>>> Please review this trivial patch to split this assert into two asserts.
>>>
>>> https://cr.openjdk.java.net/~stefank/8235257/webrev.01/
>>> https://bugs.openjdk.java.net/browse/JDK-8235257
>>>
>>> We keep hitting this assert in testing and it's unclear if the 
>>> loader is NULL or if it is a bad oop. I propose that we split it in 
>>> two.
>>>
>>> Thanks,
>>> StefanK
>>
>