Re: [cp-patches] [generics] RFC : implementation of java.lang.instrument package

[Nicolas Geoffray]

> Nicolas> Where this class should be : 
> Nicolas> gnu/java/lang/instrument/InstrumentationImpl.java?
>
> Yes, that is the correct package to use.  The name of the class is
> less strictly defined by Classpath, but that name sounds fine to me.
>  
>
>
> Yes, that name is OK. I don't know how much we can actually share
> between runtimes though. The interface is pretty minimal already, so a
> split between VMInstrumentationImpl and InstrumentationImpl that calls
> the VM class seems overkill.
The InstrumentationImpl.getObjectSize is clearly a native for vm.
And we also need a native redefineClass(Class, byte[]).

If we don't provide a VMInstrumentationImpl class, would
these method be native in InstrumentationImpl? (I thought
vm implementors only had to care about vm/reference/* files?)

> Do you think you can come up with something
> that can easily be shared by runtimes? Maybe you could hook into the
> default ClassLoader.defineClass() implementation to implement this?
>
>  
Because it's the vm that parses arguments, it knowns if the
application does instrumentation. Therefore, if it supports
redefining classes, it can set a flag or change definition of
VMClassLoader.defineClass. The vm then creates an
InstrumentationImpl object and calls premain.

The VMClassLoader.defineClass will be changed to call a
private InstrumentationImpl.defineClass method. This method will then call
each transformer.

Does that sound correct to you? This way I don't modify 
ClassLoader.defineClass.


Cheers,

Nicolas