[vmkit-commits] [PATCH] registerNatives

[Will Dietz]

Starting a new thread for both our sanity :).

Also, note the change of email address--feel free to contact me on
either, but being double-subscribed to the ML was silly.

On Mon, Oct 24, 2011 at 1:10 AM, Nicolas Geoffray
<nicolas.geoffray at gmail.com> wrote:
>
> Should we think about concurrent updates and lookups on registeredNatives?
> Unfortunately, I think we should :(
>

Hmm, I think you're right.  Certainly ensures the correct behavior
regardless.  I'll use the classloader's lock as you mention.

>> +  word_t getRegisteredNative(const JavaMethod * meth) const {
>> +    std::map<const JavaMethod*,word_t>::const_iterator I =
>> +      registeredNatives.find(meth);
>> +    if (I == registeredNatives.end()) return 0;
>> +    return I->second;
>
> Could that be replaced with registeredNatives[meth]?
>

No, can't call operator[] on a const map, although the const
correctness isn't really very important, since we never have a const
JCL--so maybe clarity/simplicity of code should win here.

Will change.

>>
>> +  }
>> +
>>   friend class Class;
>>   friend class CommonClass;
>>   friend class StringList;
>> diff --git a/lib/J3/VMCore/Precompiled.cpp b/lib/J3/VMCore/Precompiled.cpp
>> index e469eaf..9a85e99 100644
>> --- a/lib/J3/VMCore/Precompiled.cpp
>> +++ b/lib/J3/VMCore/Precompiled.cpp
>> @@ -102,6 +102,7 @@ extern "C" word_t vmjcNativeLoader(JavaMethod* meth) {
>>   char* buf = (char*)threadAllocator.Allocate(
>>       3 + JNI_NAME_PRE_LEN + 1 + ((mnlen + clen + mtlen) << 3));
>>   word_t res = meth->classDef->classLoader->nativeLookup(meth, j3, buf);
>
> Could you please add a comment on what's happening here? It might get tricky
> to understand.
>

Sure thing.  I'll include that in next revision.

Thanks!

~Will