DISCLAIMER: These notes are from the defunct k8 project which precedes SquirrelJME. The notes for SquirrelJME start on 2016/02/26! The k8 project was effectively a Java SE 8 operating system and as such all of the notes are in the context of that scope. That project is no longer my goal as SquirrelJME is the spiritual successor to it.
I was going to have two math operations one for register and the other for a constant, but I figured that is pointless as they can just be combined into one and checked accordingly.
I do wonder about GETSTATIC and GETFIELD, right now my plan is to pass an input register with a null value. But that would not work too well. I can just have the op have two constructors so that one is static and the other is not.
One bit of code has more than 512 exception handlers, ouch. Actually that is due to this buffer clearing code not doing what as is intended. Well, the clear is correct so there must really be tons of exception handlers.
Now the next instruction I must decode is INVOKEDYNAMIC. Requires decoding of the CONSTANT_InvokeDynamic_info structure in the constant pool. Which references a structure in the class file, which is very external to the method. So I suppose a way to do such a thing would be to defer handling of fields and methods and handle class attributes first. That can easily be done by flipping some buffers and such around. Then the class attributes can be given to the field and method handlers so they can use it as required. It might be best to make the bootstrap information be a part of a method rather than a class so that such things can be recompiled at will without requiring the FormClass to be specified.
I could always just get the obtiained information and then just figure out how invokedynamic truly works at the time of class recompilation, although that is the next step it would allow me to finish off writing the byte code decoding code as needed.
The constant used by invokedynamic contains the descriptor of the method that is called, so the arguments which are used can be known as needed. Each invokedynamic instruction gets its own bootstrap method.
The invoke dynamic stuff can be simplified to just being contained within a specific set of classes one initializes. That being the DynamicLink.Dynamic so that must contain the correct stuff.
Looking at the decoded form of BootstrapMethods in javap and it all is starting to come together now.
Writing this bit of the handle code gave me a dejavu feeling for an odd reason.
Well, implementing INVOKEDYNAMIC decoding took 7 hours. Well at least it is done now.
Thinking about it, I might go straight to PowerPC generation rather than C since I have been waiting a year. Writing an unneeded at this time C output would slow things down. Running k8 on existing operating systems is not the intended result. The sooner this is self hosting, the better. Although I would need to write a Java compiler to be able to be self hosting however.
Probably will need to implement StackMapTable now so that some exception states will work. Also my local reduction is incorrect, I assume that the order of locals is linear so I treat it like a stack. I must check ALL of the local types and only the later may be null as needed.
Based on some reading elsewhere, the StackMapTable only lists the state of things for jump targets. Which makes sense because normal running through the program handles such cases in between where operations are not part of jumps. My current stopage at running is that the target jump to an exception handler has a single difference of a local. Hopefully the StackMapTable for that contains a rule where the locals can be trimmed to fit as needed.
May have to end up rewriting all of the JavaCodeBank stuff as it is rather messy, however I can do that later (if ever) due to the modularity of the dynamic recompilation system. It would probably also be best to write this one first so I know how all of the class reading stuff works. Then I can use such knowledge to write a newer better version since I know stuff now that I did not know before.