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 wonder what the best way to represent the current allocation state is.


Need to make these things more memory efficient so k8 and the recompiler will complete faster and use less resources on more constrained systems. Though I need to actually measure this usage and such.


I should wait on that to see if it can be optimized in the future when I get to actually running it, but the Java decoder could probably use it.


The register assignments, the ones which are available currently get regenerated every time a method needs to be recompiled. Although dynamic in standard usage it will be a waste really. Also allowing compiler settings to be changed (anything) would lead to a bunch of exploits if one can just change a setting producing an incompatible binary. So I will need a structure that is capable of storing compilation settings, the CPU used and other things. The one thing that may be changed without issue are the optimization passes and not the settings that directly affect the CPU specific compiler. Also an assembler does not always need an allocator either so that has to be decoupled from that. The DynaRec however, requires an allocator to be used.


So yes, that Option map is going to have to go and be replaced by something else that is more effective and can be stored into the recompiled output as needed. So the DynaRec and such will use this new CompilerProfile which states common CPU setting information. DynaRec is also getting a ton of generic parameters slapped onto it also. So I need some kind of clean system where I can do dynamic recompilation for various systems without producing tons of duplicated state.


Actually what I can do is move stuff in NARF around so that I take advantage of the asmfmt factory. I can code it in a way so that the narf packages loses the dynarec stuff so things will be a bit more lean and such. So do away with the DynaRec sort of and just make it leaner and more generic. Then also add support in FormCodeBank for compiler settings used to "compile" the resulting method code (if any). I do wonder though if I could change FormClass to be an representitive interface for input and output code. Well the main thing is just methods. The Form stuff does work as is however. I just am currently disliking how some of the dynamic recompiler stuff is done. Not going to clear away much code, just will be a refactor. The BinaryFormatFactory is for loading FormClasses (so future stuff like ELF could be supported), although what I would like is some way to change things back or patch over it rather than it just being purely immutable. Would have to figure that out.


Thought about a mutable FormClass, but that would reduce speed of the running system as it would have to check for changes. Adding and removing methods should be illegal. So what I really need is a type of window into a binary stream for code implementations of methods. Essentially just a byte buffer over the native code and such.


I know one thing though, FormCodeBank is really ugly.


FormCodeBank could benefit from making more of it immutable and having property sets for exceptions and such defined in the bank area, and perhaps make it final and not have every class that supports it extend off of it. Then remove the FormExecutableCodeBank and just have FormCodeBank be alone. So the first thing to do then is to move DynaFactory into asmformat into a classes which provide directionality information as to what they do and how they do it rather than using a by name and other sort of ugly thing.


As I write this Transcompiler class, something hit me, I can also merge passes into it also.


Hmm a question, do I make FormCodeBank immutable?


After I get this MIPS code redone I am just going to drop it for now and instead work on PowerPC. The main reason being is that I can actually debug PowerPC and I know enough of the firmware interface in IEEE1275 to make it work, as I have done it before. Or I can just drop it now and work on PowerPC at this moment.


I wonder if it is a bad idea to have a map of FormCodeBanks assigned by language in the method. So perhaps it would be best to do it so that it requires iteration when requested. This way I can have many code banks with the same architecture but perhaps with different recompilation settings.


Bootstrap methods must also be part of forms as they are used for invokedynamic everywhere.