Stack notation: "<stack before> -- <stack after>". Rightmost is top of stack
(TOS). For example, in "a b -- c d", b is TOS before, d is TOS after. "R:" means
that the Return Stack is modified. "I:" prefix means "IMMEDIATE", that is, that
this stack transformation is made at compile time.

DOES>: Used inside a colon definition that itself uses CREATE, DOES> transforms
that newly created word into a "does cell", that is, a regular cell ( when
called, puts the cell's addr on PS), but right after that, it executes words
that appear after the DOES>.

"does cells" always allocate 4 bytes (2 for the cell, 2 for the DOES> link) and
there is no need for ALLOT in colon definition.

At compile time, colon definition stops processing words when reaching the
DOES>.

Example: ": CONSTANT CREATE HERE @ ! DOES> @ ;"

Word references (wordref): When we say we have a "word reference", it's a
pointer to a words *code link*. For example, the label "PLUS:" in this unit is a
word reference. Why not refer to the beginning of the word struct? Because we
actually seldom refer to the name and prev link, except during compilation, so
defining "word reference" this way makes the code easier to understand.

Atom: A word of the type compiledWord contains, in its PF, a list of what we
call "atoms". Those atoms are most of the time word references, but they can
also be references to NUMBER and LIT.

Words between "()" are "support words" that aren't really meant to be used
directly, but as part of another word.

"*I*" in description indicates an IMMEDIATE word.

*** Defining words ***
: x ...     --              Define a new word
;           R:I --          Exit a colon definition
,           n --            Write n in HERE and advance it.
' x         -- a            Push addr of word x to a. If not found, aborts
['] x       --              *I* Like "'", but spits the addr as a number
                            literal. If not found, aborts.
(           --              *I* Comment. Ignore rest of line until ")" is read.
ALLOT       n --            Move HERE by n bytes
C,          b --            Write byte b in HERE and advance it.
CREATE x    --              Create cell named x. Doesn't allocate a PF.
[COMPILE] x --              Compile word x and write it to HERE. IMMEDIATE
                            words are *not* executed.
COMPILE x   --              Meta compiles. Kind of blows the mind. See below.
CONSTANT x  n --            Creates cell x that when called pushes its value
DOES>       --              See description at top of file
FIND        a -- a f        Read at a and find it in dict. If found, f=1 and
                            a = wordref. If not found, f=0 and a = string addr.
IMMED?      a -- f          Checks whether wordref at a is immediate.
IMMEDIATE   --              Flag the latest defined word as immediate.
LITN        n --            Write number n as a literal.
VARIABLE c  --              Creates cell x with 2 bytes allocation.

Compilation vs meta-compilation. When you compile a word with "[COMPILE] foo",
its straightforward: It writes down to HERE wither the address of the word or
a number literal.

When you *meta* compile, it's a bit more mind blowing. It fetches the address
of the word specified by the caller, then writes that number as a literal,
followed by a reference to ",".

Example: ": foo [COMPILE] bar;" is the equivalent of ": foo bar ;" if bar is
not an immediate. However, ": foo COMPILE bar ;" is the equivalent of
": foo ['] bar , ;". Got it?

Meta-compile only works with real words, not number literals.

*** Flow ***
Note about flow words: flow words can only be used in definitions. In the
INTERPRET loop, they don't have the desired effect because each word from the
input stream is executed immediately. In this context, branching doesn't work.

(br)        --              Branches by the number specified in the 2 following
                            bytes. Can be negative.
(?br)       f --            Branch if f is false.
[           --              Begin interetative mode. In a definition, words
                            between here and "]" will be executed instead of
                            compiled.
]           --              End interpretative mode.
ABORT       --              Resets PS and RS and returns to interpreter
ABORT" x"   --              *I* Compiles a ." followed by a ABORT.
AGAIN       I:a --          *I* Jump backwards to preceeding BEGIN.
BEGIN       -- I:a          *I* Marker for backward branching with AGAIN.
ELSE        I:a --          *I* Compiles a (fbr) and set branching cell at a.
EXECUTE     a --            Execute wordref at addr a
IF          -- I:a          *I* Compiles a (fbr?) and pushes its cell's addr
INTERPRET   --              Get a line from stdin, compile it in tmp memory,
                            then execute the compiled contents.
QUIT        R:drop --       Return to interpreter prompt immediately
RECURSE     R:I -- R:I-2    Run the current word again.
THEN        I:a --          *I* Set branching cell at a.
UNTIL       f --            *I* Jump backwards to BEGIN if f is *false*.

*** Parameter Stack ***
DROP        a --
DUP         a -- a a
OVER        a b -- a b a
ROT         a b c -- b c a
SWAP        a b -- b a
2DROP       a a --
2DUP        a b -- a b a b
2OVER       a b c d -- a b c d a b
2SWAP       a b c d -- c d a b

*** Return Stack ***
>R          n -- R:n        Pops PS and push to RS
R>          R:n -- n        Pops RS and push to PS
I           -- n            Copy RS TOS to PS
I'          -- n            Copy RS second item to PS
J           -- n            Copy RS third item to PS

*** Memory ***
@           a -- n          Set n to value at address a
!           n a --          Store n in address a
?           a --            Print value of addr a
+!          n a --          Increase value of addr a by n
C@          a -- c          Set c to byte at address a
C!          c a --          Store byte c in address a
CURRENT     -- a            Set a to wordref of last added entry.
HERE        -- a            Push HERE's address
H@          -- a            HERE @

*** Arithmetic / Bits ***

+           a b -- c        a + b -> c
-           a b -- c        a - b -> c
-^          a b -- c        b - a -> c
*           a b -- c        a * b -> c
/           a b -- c        a / b -> c
MOD         a b -- c        a % b -> c
/MOD        a b -- r q      r:remainder q:quotient
AND         a b -- c        a & b -> c
OR          a b -- c        a | b -> c
XOR         a b -- c        a ^ b -> c

*** Logic ***
=           n1 n2 -- f      Push true if n1 == n2
<           n1 n2 -- f      Push true if n1 < n2
>           n1 n2 -- f      Push true if n1 > n2
CMP         n1 n2 -- n      Compare n1 and n2 and set n to -1, 0, or 1.
                            n=0: a1=a2. n=1: a1>a2. n=-1: a1<a2.
NOT         f -- f          Push the logical opposite of f

*** Strings ***
LIT         --              Write a LIT entry. You're expected to write actual
                            string to HERE right afterwards.
LIT< x      --              Read following word and write to HERE as a string
                            literal.
LITS        a --            Write word at addr a as a atring literal.
SCMP        a1 a2 -- n      Compare strings a1 and a2. See CMP
SCPY        a --            Copy string at addr a into HERE.
SLEN        a -- n          Push length of str at a.

*** I/O ***

A little word about inputs. There are two kind of inputs: direct and buffered.
As a general rule, we read line in a buffer, then feed words in it to the
interpreter. That's what "WORD" does. If it's at the End Of Line, it blocks and
wait until another line is entered.

KEY input, however, is direct. Regardless of the input buffer's state, KEY will
return the next typed key.

PARSING AND BOOTSTRAP: Parsing number literal is a very "core" activity of
Forth, and therefore generally seen as having to be implemented in native code.
However, Collapse OS' Forth supports many kinds of literals: decimal, hex, char,
binary. This incurs a significant complexity penalty.

What if we could implement those parsing routines in Forth? "But it's a core
routine!" you say. Yes, but here's the deal: at its native core, only decimal
parsing is supported. It lives in the "(parsed)" word. The interpreter's main
loop is initially set to simply call that word.

However, in core.fs, "(parsex)", "(parsec)" and "(parseb)" are implemented, in
Forth, then "(parse)", which goes through them all is defined. Then, "(parsef)",
which is the variable in which the interpreter's word pointer is set, is
updated to that new "(parse)" word.

This way, we have a full-featured (and extensible) parsing with a tiny native
core.

(parse)     a -- n          Parses string at a as a number and push the result
                            in n as well as whether parsing was a success in f
                            (false = failure, true = success)
(parse.)    a -- n f        Sub-parsing words. They all have the same signature.
                            Parses string at a as a number and push the result
                            in n as well as whether parsing was a success in f
                            (0 = failure, 1 = success)
(parse*)    -- a            Variable holding the current pointer for system
                            number parsing. By default, (parse).
(print)     a --            Print string at addr a.
.           n --            Print n in its decimal form
.X          n --            Print n in its hexadecimal form. In hex, numbers
." xxx"     --              *I* Compiles string literal xxx followed by a call
                            to (print)
                            are never considered negative. "-1 .X -> ffff"
C<          -- c            Read one char from buffered input.
EMIT        c --            Spit char c to output stream
IN>         -- a            Address of variable containing current pos in input
                            buffer.
KEY         -- c            Get char c from direct input
PC!         c a --          Spit c to port a
PC@         a -- c          Fetch c from port a
WORD        -- a            Read one word from buffered input and push its addr

There are also ascii const emitters:
BS
CR
LF
SPC