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.
09:40
Still very cold.
10:08
Actually having a dedicated class for instructions for the ArchitectureData and one for registers can have it more easily cached. I can then just reference that instead. They can be a type of map however.
10:21
What I can do is modify the Disassembler, Assembler, and MutableOpCode to use a future moved DataAccessor (I would have to move it out of kernel-bin) and then use that for placing instructions and decoding them. So for example instead of having MutableOpCode return a byte array to later be placed in, it could just write it directly so to speak.
10:26
Then from the InstructionMap
class I can have disassembler ready information
such as images and mappings so that they can be decoded as required a bit
faster and having it cached so it can be collected when it is no longer
referenced. So I believe the first thing to do is move and cleanup (split off
the DataStream) DataAccessor
. Then I can engineer the Assembler/Disassembler
around that along with MutableOpCode. I also wonder if I can have a
disassembler view of an operating similar to MutableOpCode
but is derived
from the DataAccessor. When disassembling, if an operation is mutable might
not matter, there also is a full copy which is required when changing it. I
can also potentially handle something similar in the assembler. However
thinking about this, it is possible to have a single stream for both the
assembler and disassembler. So it would essentially combine both and allow it
to be modified (if it can be written to). It could be a moving window view so
that stuff can be shared.
11:26
Ok, so the first thing to do would be to move the DataAccessor
out and into
its own project. Then I can also have a GrowableByteArrayAccessor
which is
essentially a byte array backed accessor which can increase linearly in size
as instructions are output via the assembler.
13:12
Figured out a name, I can just have it be BinaryBlob
since that fits well
as a better name than DataAccessor
.
13:48
Right now BinaryBlob
like DataAccessor
is limited to 2GiB, I should perhaps
remove that limit so that there is more utility to the classes.
17:08
ArchitectureData can probably go best without guessInstruction and instead make that part of the disassembler so that it only contains data and does not have any actual decoding logic.
17:12
I wonder if there is a type of hash sum for very small potential values. I can do a linear search through the instruction table when disassembling, however that would be a bit costly. I can architecture specific lookups however there can be on the order of tons of instructions in how they are handled for each architecture. So what I can do is instead in the architecture manifest is to have a major op specifier of sorts which identifies the mask of the input bytes which define the instruction image in general.
17:21
Well, I can have a merged assembler and disassembler (which would be quite powerful) so that there would instead just be a single class which does both.
17:23
The class would essentially edit programs, so perhaps ProgramEditor
. I would
have to potentially support going back and forth between it or have unified
positions and such. Also the MutableOpCode.Value
could be wholly cached and
such so that once the instruction is determined it can create values as they
are needed rather than all the time. So the instruction could be guessed
without the creation of any value fields in the map. Then if the fields of
the instruction are cared about then they can be stored in a kind of reference
map which can share the field mapping values. So essentially the value for
field #3 in an instruction, if the instruction is changed then that field
association (which bits it maps to and how) will be different. So storing the
values as a kind of cache would be invalid. However, in short, the fields
would likely be sorted by shift order so that way only a handful of fields are
needed at once in memory. Essentially, the new Value
for a field would end
up being mapped to the field number. Any interaction with field values can
just do something similar to MutableOpCode
where they directly modify the
bits of the program. The assembler could also support a re-interpret which
forces the fields to be interpreted as another operation when it really is not
that operation. What this means is that the op field for aaload for
example when the instruction is reinterpreted to aastore will still use the
op value as it would be for aaload. However for some instructions such
as the wide variants, the fields are mapped to shift values so that the narrow
version will probably start at byte 1, the wide value field will then be mapped
to byte 2.
17:31
Then MutableOpCode could still be used, but it would essentially just be a
ProgramEditor
with a semi-fixed position. However thinking about it, some
architectures (such as Java byte code) may have specific alignment requirements
for instructtions depending on the position of the instruction in the program.
So these special instructions would appear wider and narrower depending on
their position. There are also variable sized instructions. Fields could be
generated based on this. The only use of MutableOpCode
outside of the
assembler and disassembler is to store a working operation before it is passed
to the assembler. The disassembler loads into it, but the editor could just
then have a nextInstruction(). I could have something similar to a queue where
having putInstruction
writes an instruction and then moves past it, while
having an offerInstruction
being similar to a put but without moving. The
same could be said for disassembling where there are poll
and peek
.
Going backwards in a program would be tricky however because previous bytes of
a program may have changed. But when implementing it I shall figure out the
true and better names to use than the ones named similar to queue.
21:13
Was attacked by a dog today, it was biting my right wrist. Luckily for me due to the cold weather I was wearing a thick coat so my skin was undamaged. I would suppose that this drives me more to never want any pets at all. Was a rather horrifying experience. If my wrist were to be badly damaged then I would be unable to type as fast as I do and would probably have to do it one handed which would severely cut my speed while programming. However I would probably adept by switching to an editor where I can use lots of macros.
21:16
However, I will calm myself by writing the ProgramEditor
and watching some
videos since I rarely watch videos.
22:02
Just tired and out of mind, cannot really think about programming.