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.
12:11
New month now. I wonder if my diff counter is meaningful. I stopped using it long ago. Actually the script no longer exists so I must have removed it. Actually had it reversed, it still exists. There were lots of changes so overall it will appear a bit funny. Subtractions: 96031, Additions: 102994. sloccount gives me:
SLOC Directory SLOC-by-Language (Sorted)
129469 java-library-4 java=129469
76540 java-library-1 java=76540
41351 java-jsr231 java=41351
25629 java-library-3 java=25629
13574 java-library-2 java=13574
12770 java-jsr926 java=12770
7984 java-jsr926-sun-compat java=7984
5910 java-jsr148 java=5910
3762 narf-decompiler-java java=3762
3443 java-jsr353 java=3443
3179 script-forth java=3179
3167 kernel-binary-base java=2725,pascal=394,ansic=48
2787 c-preprocessor java=2748,ansic=39
2706 squirrel-quarrel java=2706
2639 hairball java=2639
2239 ssa-compiler java=2239
1813 kernel-binary-asm java=1813
1759 doclet-java java=1759
1709 java-jsr080 java=1709
1560 kernel-binary-asm-java java=1560
1399 kernel-binary-type java=1399
1350 hairball-os-construct java=1350
1100 kernel-binary-load-java java=1100
616 c-compatibility-posix ansic=412,sh=201,java=3
519 sun-doclet-api java=519
425 image-decoder-xpm java=425
415 kernel-binary-asm-ppc java=415
293 kernel-k8 java=273,ansic=20
172 c-compiler ansic=123,java=30,sh=19
93 std-compliance java=93
70 freetanx java=70
58 x11-on-swing java=58
52 kernel-platform-base java=52
50 k8-tests java=50
27 kernel-platform-ieee1275-ppc asm=15,java=12
19 hairball-os-construct-gui java=19
6 script-ecmascript java=6
1 kernel-interface java=1
1 kernel-platform-ieee1275 java=1
0 debug-protocol (none)
0 debug-protocol-gdb (none)
0 debug-protocol-jdwp (none)
0 developer-blog (none)
0 k8-gui-resources (none)
0 kernel-binary (none)
0 kernel-binary-asm-mips (none)
0 kernel-binary-asm-sparc (none)
0 kernel-platform-ieee1275-sparc (none)
0 top_dir (none)
21:41
The always zero register in MIPS is a good idea. Even though there is one less register, it is a common and the same way for all operations to be a bit bucket and a common source of zero without XOR with self.
23:01
Ok, I have a general idea how garbage collection and exception handling will work. There will need to be a register which is a dedicated sequence point which determines which exception handling to jump into and to copy which variables to where. The start of an exception zone will just be a try/catch/finally as normal. That will be settled/initialized by a lookup table which knows the sequence number and the type to handle. Now to detect exceptions happening in the middle of a sequence set, even numbers would be valid sequences while odd would be in the middle of. So say if one is thrown while the values are being restored into their correct locations for the exception handler, the sequence number will identify it. Another thing is that the stack will need to store all registers which are not explosive. When a method is called, an exception is thrown (via throw), or the process manager thing throws an exception for a thread; the registers will be saved to a specific location on the stack. The temporary registers will not be saved. On RISC based platforms where there are lots of registers where only work can be done in the registers themselves, some registers will be set to temporaries which will just not be saved/restored when exceptions are caught. Now for this to work, all the registers must be saved before a call to a method since an exception can be thrown in a called method which propogates to the caller and so on. So in all of these cases, the registers needed for restore are saved as required. When it comes to sequence points, some complex methods will require a ton of them for entire exception blocks. So larger methods are worse because they complicate things. At least with sequence points, I should be able to shuffle register assignments to variables as needed. At the change of a variable assignment spot the sequence point will be incremented, then the values will be stored and restored as needed, then the sequence point will be incremented again.
23:12
Rather tired though currently, but that is how exception handling will work in the code. Now as for garbage collection, that is a bit complicated. I suppose it would be best to do some kind of simple reference usage flagging and mark and sweep for more detailed stuff. When an object is allocation it just gets an automatic implied, was used in a method mark. Now when that object is set to a field or inside an array of an object, it is... on another note. If a field is set to an object then that object gets marked with either a static or (forgot the word, but something with an i) flag. The word would be instance. The main thing though is actual classes, they will get all the marks. For arrays however, that will require another flag too since those are independent. There will also be another method flag, one associated with invoked as an argument. Say for example that the code generator knows that an object is only referenced locally and never set to any fields, it can just free those objects at the end if the only thing they exist as is being in the method and nothing was ever called with them on. That can only be done if the object flags remain at the method level however. The invoke flag works because if an object is passed in from a calling method but not set anywhere else it will just have the invoke flag.
23:26
At the end of methods, such inner handlers will have to be used to check the flags for a quick free.