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 believe for my multi-root system, for compatibility with POSIX that I will have a PWR (process working root, like PWD) and use // for selecting other roots. So say if one were to change the working directory to one in //, it would select that volume as the new root. However, that may be a problem for scripts that jump to a directory then jump back because it is assumed that there is a single root. So the PWR would have to be just one of those roots and then you would have to do a chroot to change the default root that you have selected. Although I could still use my existing setup that I plan since I do like that better. The POSIX compatibility library will just extrapolate roots and such much like the compatibility layers on Windows.
Now I am back to recompiling of the code, with this revamped code and the new order of things I can more quickly build and test specific bits. One thing I will have to do in the future is split off vmjrt and put it in vmjar with the 4 separated profile bits. That is, I can compile profile1 first, then continue on as needed. Then with that in the future I can explicitely state which profile is to be included when the system is built. So if one desires a lighter runtime say for an embedded system things do not have to get insane with building a giant library for no reason, although it could still be built. Internally, the kernel itself should only depend on anything from profile1 and not everything else. So this can be handy to prevent the kernel from say using something from AWT accidently. It would also prevent compilation of the entire class library when it is not needed for the kernel.
I had previously made some timing stuff in the past. Current on a single core PowerBook G4 the steps run in 50 seconds. On my i3 laptop they take only 10 seconds. And despite doing a little bit more, the total run-time for incomplete work is much less as less stuff is duplicated and fails before hand rather than after the fact. Using the Java bootstrap code makes it much more efficient as it is no longer being interpreted in a fork heavy shell script. And my desktop takes only 40s with its Dual PowerPC CPUs. Before it took 3 minutes and 17 seconds. So about 20% original time take or in other words, five times faster. So these huge savings will considerably help in the speed of testing things.
My Kindle takes 2 minutes 14 seconds.
My Atom 1.8GHz takes 38 seconds. Compared to my desktop, the atom has the advantage in memory speed but my desktop has an SSD in it. I should also note that the code is compiling in Oracle's VM on Wine. Anything non x86 however is using JamVM with its cached intrepreter so it is not as bad as a pure interpreter. So to measure it correctly on my i3, I should switch to JamVM. And that takes 12 seconds on first run, and 12 on the next. So the JIT is only slightly faster since it could just be running everything in an interpreter of sorts to profile the code. Although going back to it, it takes 7 seconds on average. So JamVM is only about 5 seconds slower or 1.7x slower.
OK, so I am back to where I was before. I was rewriting the allocator to be better. My plans then were to take saved and unsaved states. The saved spots are in the same registers and stack locations, wheras unsaved locals can be put anywhere and cycled around because they are dropped when exceptions are called. So I do not currently see anything wrong with this. I do however will need to have volatile register sets. I have something called the green zone, which are registers to be saved when a method is called for example.
One handy thing my OS will really need, is JAR injection. Like DLL and SO injection on other systems, except that existing methods may be replaced when the virtual machine creates a process. This could be used for general hacking, patching, and some other things.