what version of fasm is the minimalist?

i am currious of the size of the compiler if it compiles only instructions needed to compile itself without bugs.

Why don't you just modify it and make a version that does what you want? Rather than making wild guesses you can know for real.

Okay so putting just a little bit of thought into making a truly small "assembler" then all that is needed is to assemble db.

So doing only that and also limiting to something like a DOS .com file would allow the "assembler" to be very small, less than 1kB.

Of course, writing source code would be rather difficult, but that would be a small price to pay to save that precious RAM and HDD space.

let it be just one byte, just it will be a special very big byte

not to save space, but to run many assemblers together to do a sort of AI with real time compilation of asm source.
the PC to become a virtual machine or stuff like that.

why don't you use direct opcode generation then?

it's not only to generate opcodes, but to link datas.
create pointers into ram using the equates and labels
and script some asm sometimes to make data processing.

it is in order to have every files and processes under control.
first step should be to create a pointer based program.
instead of successive calls, i have successive pointers.
but these pointers should be modifiable in real time on the demand of the user, modified, created, and deleted.
the only secure way i know is to use fasm to link in real time everything.
fasm is fast enough to be used as a shell and kernel device.

edfed: Mind if you keep us updated on your progress from time to time?

Ooh, I just saw this response now.

You should consider the executable code as shareable, only the data needs to change. Much like the DLL system in Windows, you only need one copy of the code. So making it small would not be such an important goal, more important would be to make it thread safe.

You mean something like a self-compiler based on different data? (whatever the AI gets with sensors or whatever)

I think you'll need to code your own algo that does it (there's already one for 'data', albeit not very efficient, called Neural Networking). You need your own algo to do what you want and generate the code YOU need -- or do you expect the AI to write the source code FIRST and then compile it? Why?

Ive seen code go as low as 12byte here http://board.flatassembler.net/topic.php?t=9147 so, I just put over 4 and below 8.

Those are compilers? o_O



BTW wouldn't that be "after 8 and before 16"?