RFR(M) 8208999 Some use of Klass* should be replaced by InstanceKlass*

[Ioi Lam]

On 8/7/18 5:30 AM, coleen.phillimore at oracle.com wrote:
>
>
> On 8/7/18 12:31 AM, Ioi Lam wrote:
>> Hi Coleen,
>>
>> Thanks for the review. Updated webrev is here:
>>
>> http://cr.openjdk.java.net/~iklam/jdk12/8208999_klass_should_be_instanceklass.v02/ 
>>
>>
>> More comments below
>>
>>
>> On 8/6/18 3:36 PM, coleen.phillimore at oracle.com wrote:
>>>
>>> http://cr.openjdk.java.net/~iklam/jdk12/8208999_klass_should_be_instanceklass.v01/src/hotspot/share/classfile/systemDictionaryShared.cpp.udiff.html 
>>>
>>>
>>> 1007 return entry != NULL ? entry->instance_klass() : 
>>> (InstanceKlass*)NULL;
>>>
>>> do you need this cast here?
>>>
>> Removed.
>>> http://cr.openjdk.java.net/~iklam/jdk12/8208999_klass_should_be_instanceklass.v01/src/hotspot/share/memory/metadataFactory.hpp.udiff.html 
>>>
>>>
>>> + static void free_array(ClassLoaderData* loader_data, const 
>>> Array<T>* data) {
>>>
>>>
>>> But we are modifying the contents of the data (?).  This is for 
>>> deleting secondary_supers?   I think maybe adding the const for them 
>>> might be overkill.
>>>
>>> http://cr.openjdk.java.net/~iklam/jdk12/8208999_klass_should_be_instanceklass.v01/src/hotspot/share/oops/klass.hpp.udiff.html 
>>>
>>>
>>> - Array<Klass*>* _secondary_supers;
>>> + const Array<Klass*>* _secondary_supers;
>>>
>>> I think the improvement outweighs this messy part of this. This 
>>> looks like a difficult thing to change.
>>>
>> I agree. I reverted the "const" for Klass::secondary_supers. I also 
>> removed Array::as_const_array_of() since that will probably be abused 
>> by people who don't read the comments. Instead, I added this in 
>> InstanceKlass::compute_secondary_supers:
>>
>> set_secondary_supers((Array<Klass*>*)(address)interfaces);
>>
>> it's a little ugly but the effects are localized.
>
> I agree, and you can have a short comment above just that part why the 
> cast is needed, and the if statement where you cast them both to address?
>


I added these comments. What do you think?

     // The secondary super list is exactly the same as the transitive 
interfaces, so
     // let's use it instead of making a copy.
     // Redefine classes has to be careful not to delete this!
     // We need the cast because Array<Klass*> is NOT a supertype of 
Array<InstanceKlass*>,
     // (but it's safe to do here because we won't write into 
_secondary_supers from this point on).
     set_secondary_supers((Array<Klass*>*)(address)interfaces);

   if (secondary_supers() != NULL &&
       secondary_supers() != Universe::the_empty_klass_array() &&
       // see comments in compute_secondary_supers about the following cast
       (address)(secondary_supers()) != 
(address)(transitive_interfaces()) &&


Thanks
- Ioi


> The as_const_array_of() will just give someone to something to puzzle 
> over in the future without really increasing safety in my opinion.
>>
>>> http://cr.openjdk.java.net/~iklam/jdk12/8208999_klass_should_be_instanceklass.v01/src/hotspot/share/oops/arrayKlass.cpp.udiff.html 
>>>
>>>
>>> + set_super(Universe::is_bootstrapping() ? (InstanceKlass*)NULL : 
>>> SystemDictionary::Object_klass());
>>>
>>>
>>> Do you need this cast?  Object_klass() returns InstanceKlass - all 
>>> of these do.
>>>
>> Removed.
>
> Thanks!  Thank you for doing this change!
> Coleen
>>
>> Thanks
>> - Ioi
>>
>>>
>>> This looks really good. Thank you for doing this!
>>>
>>> Coleen
>>>
>>>
>>>
>>> On 8/6/18 5:26 PM, Ioi Lam wrote:
>>>> Hi, please review this clean up
>>>>
>>>> https://bugs.openjdk.java.net/browse/JDK-8208999
>>>> http://cr.openjdk.java.net/~iklam/jdk12/8208999_klass_should_be_instanceklass.v01/ 
>>>>
>>>>
>>>> I changed the following APIs so that we can avoid many unnecessary 
>>>> type casts
>>>> from Klass* -> InstanceKlass*.
>>>>
>>>>     Klass::java_super()                    -> now returns 
>>>> InstanceKlass*
>>>>     InstanceKlass::local_interfaces()      -> Array<InstanceKlass*>*
>>>>     InstanceKlass::transitive_interfaces() -> Array<InstanceKlass*>*
>>>>
>>>> I also modified APIs in SystemSictionary and SystemSictionary to 
>>>> return InstanceKlass*
>>>> where possible.
>>>>
>>>> Most changes are mundane, but here are some notable changes:
>>>>
>>>> 1. To save space, Klass::_secondary_supers (an Array<Klass*>*) may 
>>>> be pointing
>>>>    to the same array as InstanceKlass::_transitive_interfaces (an
>>>>    Array<InstanceKlass*>*).
>>>>
>>>>    To make this happen, I added Array::as_const_array_of().
>>>>    See InstanceKlass::compute_secondary_supers() and array.h.
>>>>
>>>>    I've added enough 'const' and asserts to catch mis-uses of this 
>>>> type
>>>>    of casting. If someone has a better suggestion, please let me know.
>>>>
>>>> 2. I changed Klass::_secondary_supers to 'const Array<Klass*>*' 
>>>> because
>>>>    of #1.
>>>>
>>>>    There are many pointers returned by Klass/InstanceKlass that 
>>>> should be
>>>>    'const', as most callers should read them only without 
>>>> modifying, but that's
>>>>    outside the scope of this RFE.
>>>>
>>>> 3. SystemDictionary::resolve_super_or_fail() currently accepts the 
>>>> format
>>>>    of "Ljava/lang/Object;" for the "super_name" parameter. I am 
>>>> keeping the
>>>>    same behavior in my patch. However, I wonder if this actually 
>>>> correct
>>>>    per JVMLS. I filed JDK-8209030.
>>>>
>>>>    See the call to resolve_instance_class_or_null_helper from 
>>>> resolve_super_or_fail.
>>>>
>>>>
>>>> Testing: hs tiers 1/2/3/4
>>>>
>>>> Thanks
>>>> - Ioi
>>>
>>
>