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.
Had this idea, the ClassBus and ClassIdentity (or whatever) is a bad idea. What I need is a single class in the logical area with a sub interface. The class can be called WhichClass and provides the means for potentially looking for classes that are dependencies of compilation. Since the call table for other classes must be determined before any native code is generated for them. This is why the classes will have a checksum based dependency system. No, forget that. The dependencies of methods can be determined at link time so the class will just have an import and export table, super class stuff would also be determined then also (if a class calls a non-static method in its far down superclass it will pretty much be a link back to self if it is not final or if it does not have such method it will correctly point to the superclass one). When first compiling a class, due to field volatile potential it would be unknown if a getfield or putfield should synchronize memory. So perhaps using some nop hacks I can instead have it so there is a potential for it to be volatile and if it really is not volatile then to just wipe out the machine code that synchronizes memory. However, one thing though is that it might only have to be done on volatile putfields. Because if other CPUs have a changed value in their cache for that field then a memory sync after a put would be enough before other CPUs read the value. That should statisfy the order whatever. That is, happens before something else. Although I would have to read up more on the CPU to determine this as it could vary. So I will need a possible sync before a field read or one after a field set and it could vary on the CPU. Though I could add code destroyers if the field is not volatile at all so when the native class is loaded it gets NOPed away. That would be the simplest solution as I would no longer need to load all of these classes just to figure out how to access a field, so that saves time so I no longer need to potentially load the entire class library at once.
Ok, so I need a good way to initialize SSA operations as I construct the code from the byte code.
All methods will have an intro and an endtro, the intro does basic stuff if needed (such as synchronized methods) and the endtro also frees the lock on exit. The endtro would also clean up any final references to things. I will want to have a new (for objects) that is capable of spawning objects in the heap or on the stack (depending on visibility and such) and then in the final cleanup for a method it can lower the reference count if it is in the heap or destroy the object completely if it resides on the stack. Introductions to areas that are capable of being exception handled will have to have their own endtros also to lower the count of any references that were lost when the exception was thrown.
Perhaps on the leave of a method there can be a method object reference table that is ran through to clear up any old references.
Actually, I can have two stacks, the normal method stack and the referenced object stack. When the method ends it goes through the object stack and lowers the reference count and if it is zero it clears it. The only objects that get added there are ones that are on the heap and not the stack. Also for optimize of new, each method will need a hint as to what it does with the argument (if it is an object). That hint could propagate to callers of the method when they perform the actual new. The hints would be simple such as Stack or Heap. A heap can never get downgraded to a stack but a stack could be upgraded to a heap. At linking time it is determined how an object should act based on its exposure status. It would completely depend on stuff like the super class so when I design all of my classes I must avoid this references being set to fields and such. So a method could allocate an object which it never sets to a field and never externally exposes. Now it all depends on the object, whether it exposes itself (or potentially can do it). Exposure is pretty much getting assigned to a field or an array. However the array can vary because if the array itself is not exposed then all of its members do not get upgraded, however if the array is exposed then all the members that are added to it must get their exposure set. The byte code compiler would perform basic work while the SSA optimization system would determine object exposures and such.
Currently going to setup Jenkins for automated building and testing so that way I can test a bunch of things.