My approach with RS was slightly wrong: RS' TOP was always containing current
IP. It worked, but it was problematic when came the time to introduce
RS-modifying words: it's impossible to modify RS in a word without immediately
messing your flow.
Therefore, what used to be RS' TOS has to be a variable that isn't changed
midway by RS-modifying words. I guess that's why RS is called *return* stack...
Big one.
This allows us to write higher order words directly in Forth, which is much
more convenient than writing post-immediate (see "NOT" structure in diff if
you want to see what I mean) structures in ASM.
These structures can then be written to ROM (rather than loaded in RAM for
definitions loaded at run-time).
That's quite a bit of tooling that was added, 2 compilations stages, but I
think it's well worth it.
Things are better now, but immediates inside colons are broken. However,
IF/THEN/ELSE are now immediates and it's much cleaner this way. Still, this
commit has too much stuff in it, I need to commit, I don't want to lose this
step.
The goal was to be able to implement "(" in forth, but I realised that my
INTERPRET approach was wrong. Compiling the line beforehand is, after all,
not good. I'll have to change it again.
Previous approach was broken with regards to defined word using CREATE.
Also, reduce name length by one to make space for a new flags field for
"immediate" (which isn't used yet).
This scheme of "when we handle line-by-line, compile one word at a time then
execute" so that we could allow words like "CREATE" to call "readword" before
continuing was a bad scheme. It made things like branching outside of a colon
definition impossible.
This commit implement a new "litWord". When an undefined word is encountered at
compile time, it is included as-is in a string literal word. It is at run time
that we decide what to do with it.
