# sanctuary (working title)

sanctuary is a 64-bit subroutine threaded forth for amd64 linux systems.

## stack effect notation
labels outside of the ones listed here are specific to a certain word's
documentation and will be obvious or documented in the description.

- `a`: memory address
- `c`: one byte value
- `e`: error code
- `n`: signed integer
- `u`: unsigned integer
- `z`: null-terminated string
- `?`: boolean flag
- `xt`: execution token
- `ht`: header token
- `""`: string in input buffer
- `|`: 'or'
- `,`: separates multiple stack effects when multiple are needed

## Glossary

the following is a list of words available in this forth.
certain constants for linux system interaction are not documented;
they are identical to their C versions.

### `!   ( u a -- )`
store the 64 bit value u into the memory address a.

### `#   ( u -- u' )`
add a numeric digit to the numeric output buffer
by dividing u by `base` and appending the remainder.

### `#>   ( u -- a u )`
drop the remaining digit processing debris from the stack
and return the numeric output buffer contents as a string.

### `#pad   ( -- u )`
length of the pad buffer.

### `#tib   ( -- u )`
variable containing the amount of characters in the input buffer.

### `'   ( "word" -- xt )`
read a word from the input buffer,
push to the stack its execution token.

### `'h   ( "word" -- ht )`
read a word from the input buffer,
push to the stack its header token.

### `(   ( -- ) IMMEDIATE`
start a comment which lasts until the next closed bracket.
if the unclosed bracket in the description above bothers you,
have a closing bracket: ).

### `(0handler)   ( -- )`
the very early error handler, which simply quits the program.

### `(abort")   ( a u -- )`
perform the runtime actions of `abort"`:
print string to error and abort.

### `(create)   ( -- )`
the default behaviour of a word made by `create`,
which simply pushes the address following the definition to the stack.
this messes with the return stack and is not meant to be called
outside of its specific context.

### `(defer)   ( -- )`
activate a deferred word from the address following this word's call.
this messes with the return stack and is not meant to be called
outside of its specific context.

### `(does>)   ( -- )`
run non-default behaviour of a `create`d word.
pushes the data location onto the stack and calls the word
immediately following the `(does>)` call.
this messes with the return stack and is not meant to be called
outside of its specific context.

### `(header)   ( a u -- ht )`
create a dictionary header for a word named the provided string.
this word does not set the code field.
this word does not update latest.

### `(hide)   ( ht -- )`
set the smudge bit on the header ht.

### `(unhide)   ( ht -- )`
unset the smudge bit on the header ht.

### `*   ( u1 u2 -- u )`
multiply u1 and u2.

### `*/mod   ( n1 n2 n3 -- n4 n5 )`
multiply n1 and n2, divide the result by n3.
remainder is in n4, result is in n5.

### `+   ( u1 u2 -- u )`
add u2 to u1.

### `+!   ( a -- )`
add one to the value at memory address a.

### `+to   ( comp: "name" -- , intr: u "name" -- ) IMMEDIATE`
compile or execute (depending on `state`) code to add u to the contents
of a `value`. (in compile mode u is whatever was on the stack already.)

### `,   ( u -- )`
write a 64 bit value to user memory and increment the user memory pointer.

### `-   ( u1 u2 -- u )`
subtract u2 from u1.

### `-!   ( a -- )`
subtract one from the value at memory address a.

### `-to   ( comp: "name" -- , intr: u "name" -- ) IMMEDIATE`
compile or execute (depending on `state`) code to subtract u from the contents
of a `value`. (in compile mode u is whatever was on the stack already.)

### `-rot   ( u1 u2 u3 -- u3 u1 u2 )`
rotate the three topmost values on the stack so that the topmost value
is moved to the third highest.

### `."   ( -- ) IMMEDIATE COMPILE-ONLY`
compile into the current definition the following string (terminated by `"`)
being written to output.

### `/buffer   ( -- u )`
the size of an input buffer.

### `/linebuf   ( -- u )`
the size of a line buffer.

### `/mod   ( u1 u2 -- u3 u4 )`
divide u1 by u2. result is in u4, remainder is in u3.

### `[   ( -- ) IMMEDIATE`
set the system to interpret mode.

### `[']   ( "word" -- ) IMMEDIATE COMPILE-ONLY`
read a word from the input buffer,
compile into the current definition a stack push of the xt of the word.

### `[compile]   ( "word" -- ) IMMEDIATE COMPILE-ONLY`
compile into the current definition a call to a normally immediate word.

### `]   ( -- ) IMMEDIATE`
set the system to compiling mode.

### `:   ( "name" -- )`
start compilation of the word 'name'.

### `:noname   ( -- xt )`
start compiling a nameless, headerless word and yield its xt.

### `;   ( -- ) IMMEDIATE`
end compilation of the currently compiling word.

### `\   ( -- ) IMMEDIATE`
start a comment that lasts until the end of the current line.

### `@   ( a -- u )`
fetch the 64 bit value at memory address a.

### `=   ( n1 n2 -- ? )`
return true if n1 and n2 are equal.

### `<   ( n1 n2 -- ? )`
return true if n1 is less than n2.

### `<#   ( -- )`
initialise the system for numeric output.

### `<=   ( n1 n2 -- ? )`
return true if n1 is less than or equal to n2.

### `<>   ( n1 n2 -- ? )`
return true if n1 and n2 are not equal.

### `<mark   ( -- a )`
mark the destination of a backward branch.

### `<resolve   ( a -- )`
mark the source of a backward branch.

### `>   ( n1 n2 -- ? )`
return true if n1 is greater than n2.

### `>=   ( n1 n2 -- ? )`
return true if n1 is greater than or equal to n2.

### `>body   ( ht -- xt )`
yield the code field of header token.

### `>defer   ( xt -- a )`
get the xt storage address of the deferred execution token.

### `>errno   ( u -- val err )`
transform the result of a system call into a value/error pair.
if no error occured, err is zero and val is the result,
if an error has occurred, val is zero and
err is a negative integer
(the exact value depends on the error)

### `>in   ( -- a )`
variable containing the index of the first unparsed character
in the input buffer.

### `>mark   ( -- a )`
mark the source of a forward branch.

### `>r   ( u -- ) ( R: -- u )`
move a value from the working stack to the return stack.

### `>resolve   ( a -- )`
mark the destination of a forward branch.

### `?allocate   ( u -- a e )`
allocate a dynamic block of memory, producing an error on failure.

### `?branch   ( -- )`
compile into user memory an incomplete conditional branch.
if the value on the stack is zero the branch is taken.
a 32 bit branch offset must be written immediately after.

### `?dup   ( n -- 0 | n n )`
if n is not zero, perform `dup`.

### `?find   ( a u -- ht )`
look in the dictionary for the word a (of u characters).
if a word was found,
its link field address is returned along with the true flag.
if no word was found or the string is of length zero, abort.

### `?componly?   ( -- )`
produce a compile-only error.

### `?notfound?   ( -- )`
produce a word not found error.

### `?overflow?   ( -- )`
produce a stack overflow error.

### `?underflow?   ( -- )`
produce a stack underflow error.

### `0=   ( n -- ? )`
return true if n is equal to zero.

### `0<   ( n -- ? )`
return true if n is less than zero.

### `0<=   ( n -- ? )`
return true if n is less than or equal to zero.

### `0<>   ( n -- ? )`
return true if n is not equal to zero.

### `0>   ( n -- ? )`
return true if n is greater than zero.

### `0>=   ( n -- ? )`
return true if n is greater than or equal to zero.

### `1+   ( u -- u')`
add one to u.

### `1-   ( u -- u')`
subtract one from u.

### `2drop  ( u1 u2 -- )`
remove the two topmost values from the stack.

### `2dup   ( u1 u2 -- u1 u2 u1 u2 )`
duplicate the two topmost values on the stack.

### `abort   ( -- )`
call the error handler
(the address of which is in the variable `handler`)

### `abort"   ( -- ) IMMEDIATE COMPILE-ONLY`
write the message that follows (terminated by `"`)
to standard out then call the error handler.
(the address of which is in the variable `handler`)

### `again   ( -- ) IMMEDIATE COMPILE-ONLY`
complete an infinite loop began by the word `begin`.

### `allocate   ( u -- a e )`
allocate a dynamic block of memory.

### `allot   ( u -- )`
reserve u bytes of user memory.

### `and   ( u1 u2 -- u )`
perform bitwise AND on u1 and u2.

### `base   ( -- a )`
a variable containing the current numeric input/output base.
by default this is 10.

### `base-buffer   ( -- a )`
the address of the base input buffer which reads from the terminal.

### `base-linebuffer   ( -- a )`
the address of the base line input buffer.

### `begin   ( -- ) IMMEDIATE COMPILE-ONLY`
mark the beginning of a begin-again, begin-until,
or begin-while-repeat loop.

### `binary   ( -- )`
set current base to binary.

### `branch   ( -- )`
compile into user memory an incomplete branch.
a 32 bit branch offset must be written immediately after.

### `brk@   ( -- a )`
yields current program break.

### `bye   ( -- )`
exits the forth system.

### `c,   ( c -- )`
write an 8 bit value to user memory and increment the user memory pointer.

### `c!   ( u a -- )`
store the 8 bit value u into the memory address a.

### `c@   ( a -- c )`
fetch the 8 bit value at memory address a.

### `cell+   ( u -- u' )`
increment u by the size of one cell.

### `cell-   ( u -- u' )`
decrement u by the size of one cell.

### `cells   ( u -- u' )`
transform an amount of cells into an amount of bytes.

### `char   ( "c" -- c )`
yield the value of the first character of the next word in the input stream.

### `close-file   ( fd -- e )`
close the file at fd.

### `cmove   ( a1 a2 u -- )`
copy u bytes of memory from a1 to a2.
bytes are copied in low memory to high memory order.

### `cmove,   ( a u -- )`
copy u bytes of memory from a1 to `here`, then increment `here` appropriately.
bytes are copied in low memory to high memory order.

### `cmove>   ( a1 a2 u -- )`
copy u bytes of memory from a1 to a2.
bytes are copied in high memory to low memory order.

### `compile,   ( xt -- )`
compile a call to xt into user memory.

### `compile-only   ( -- )`
mark the most recently defined word as compile-only.

### `compile-only?   ( ht -- ? )`
true if ht is marked compile-only, false otherwise.

### `constant   ( u "name" -- )`
create a word that pushes a cell value u to the stack.

### `create   ( "name" -- )`
create a word in the dictionary that, by default,
pushes the address directly following the header to the stack.
this behaviour can be modified with `does>`.

### `d,   ( n -- )`
write a 32 bit value to user memory and increment the user memory pointer.

### `d!   ( u a -- )`
store the 32 bit value u into the memory address a.

### `decimal ( -- )`
set current base to decimal.

### `defer   ( "name" -- )`
create a new word, the behaviour of which can be controlled with
`defer!`, `defer@`, `is` and `action-of`.
initially it is set to yield an error.

### `defer!   ( xt1 xt2 -- )`
set the deferred word xt2's behaviour to xt1.

### `defer@   ( xt -- xt' )`
retrieve the xt' which the deferred word xt is set to execute.

### `does>   ( -- )`
modify the behaviour of the most recent `create`d word.
(non-`create`d words will be corrupted.)

### `dp   ( -- a )`
a variable that contains the lowest free byte of memory in user memory.

### `dp0   ( -- )`
a variable that contains the first byte of user memory.

### `dp$   ( -- )`
a variable that contains the last available byte of user memory.

### `drop   ( u -- )`
remove the value at the top of the stack.

### `dup   ( u -- u u )`
duplicate the value at the top of the stack.

### `e."   ( -- ) IMMEDIATE COMPILE-ONLY`
compile into the current definition the following string (terminated by `"`)
being written to error output.

### `else   ( -- ) IMMEDIATE COMPILE-ONLY`
update the current if statement to branch here
when the flag is false,
and skip to `then` if the corresponding `if` was true.

### `emit   ( c -- )`
print the single character c to output.

### `executable   ( a u -- )`
marks the u bytes starting at address a as executable.
this is used primarily to mark the program break,
which is used as the user memory space.

### `execute   ( xt -- )`
call the word xt.

### `exit   ( -- ) IMMEDIATE COMPILE-ONLY`
compile into the current definition a return instruction.

### `false   ( -- u )`
a cell with no bits set.

### `find   ( a u -- a u 0 | a -1 )`
look in the dictionary for the word a (of u characters).
a zero is returned along with the original given string
if no word was found. if a word was found,
its link field address is returned along with the true flag.

### `free   ( a u -- e )`
free the given block of memory created by `allocate`.

### `grow   ( u -- )`
grows, and marks as executable, the user memory space by u bytes.

### `handler   ( -- a )`
variable containing the address of the current error handler.

### `here   ( -- a )`
yields the address of the first available byte in user memory.

### `hex   ( -- )`
set current base to hexadecimal.

### `hide   ( "word" -- )`
set the smudge bit on the given word.

### `hijacks   ( xt "word" -- )`
'hijack' an existing definition to perform the action of xt.
this word *will* corrupt the dictionary if used outside
its very specific context (replacing core assembly words
with better versions in forth), so it should be avoided
in favour of `defer` and friends.

### `hld   ( -- a )`
the address of the beginning of the used section of the pad buffer.

### `hold   ( c -- )`
add the given charater into the numeric output buffer.

### `if   ( ? -- ) IMMEDIATE COMPILE-ONLY`
if the flag is true, execute the following if statement,
terminated by `else` or `then`.

### `init-source   ( -- n )`
yield the value of source-id when processing the initialisation script.

### `immediate   ( -- )`
mark the most recently defined word as immediate.

### `immediate?   ( ht -- ? )`
true if ht is marked immediate, false otherwise.

### `interpret   ( -- )`
interprets the contents of the terminal input buffer
until it runs out.

### `invert   ( u -- u' )`
invert all bytes in u.

### `is   ( xt "name" -- ) IMMEDIATE`
set the deferred word name to execute xt.

### `latest   ( -- a )`
a variable containing the execution token of
the most recently created word.

### `literal   ( n -- ) IMMEDIATE COMPILE-ONLY`
compile a push of the literal value n into the currently compiling word.

### `mmap   ( offset fd flags prot u a -- u ) `
perform a mmap(2) system call.

### `munmap   ( u a -- u ) `
perform a munmap(2) system call.

### `nip   ( u1 u2 -- u2 )`
drop the second-highest value from the stack.

### `nonaming   ( -- ? )`
a `value`: true if the currently compiling word is a `:noname` word.

### `number   ( a u -- n -1 | 0 )`
convert given string into a number along with a flag.
if parsing a number fails then 0 (false) is returned
and no number is provided.

### `octal   ( -- )`
set current base to octal.

### `open-file   ( z mode -- fd e )`
open the given file with the given mode (probably r/w, r/o or w/o)
and yield the resulting file descriptor
note that this word uses the pad.

### `or  ( u1 u2 -- u )`
perform bitwise OR on u1 and u2.

### `over   ( u1 u2 -- u1 u2 u1 )`
copy the second-highest value on the stack and move it to the top of the stack.

### `pad   ( -- a )`
the address of the start of the pictured numeric output buffer.

### `pad$   ( -- a )`
the address of the end of the pictured numeric output buffer.
useful because numeric output uses the buffer high-memory first.

### `parse   ( "name<c>" c -- a u )`
parse one word from the input buffer,
separated by a newline or the character c,
and return as a string.

### `parse-name   ( "<ws>name<ws>" -- a u )`
parse one whitespace-separated word from the input buffer,
and return as a string.
tabs (ascii 0x09), newlines (ascii 0x10), and spaces (ascii 0x20)
are considered whitespace.

### `postpone   ( "name" -- ) IMMEDIATE COMPILE-ONLY`
compile the execution behaviour of a word into the current definition.
if the word is immediate, that will execute the word at runtime
(like `[compile]`).
if the word is not immediate, this will compile code that compiles that word.

### `private{   ( -- )`
mark the start of a private section closed by `}private`
and activated with `privatise`.

### `}private   ( -- )`
mark the end of a private section opened by `private{`
and activated with `privatise`.

### `privatise   ( -- )`
activate a private section.

### `r/o   ( -- 0 )`
a constant, meaning 'read only', used for file I/O.

### `r/w   ( -- 2 )`
a constant, meaning 'read and write', used for file I/O.

### `r>   ( -- u ) ( R: u -- )`
move a value from the return stack to the working stack.

### `rdrop   ( R: u -- )`
remove the value at the top of the return stack.

### `read-file   ( a u fd -- u' e )`
read u bytes from fd into memory location a.
u' is the number of bytes read.

### `repeat   ( -- ) IMMEDIATE COMPILE-ONLY`
in a begin-while-repeat loop, loop back to the condition.

### `rot   ( u1 u2 u3 -- u2 u3 u1 )`
rotate the top three values on the stack so that the third highest value is moved to the top.

### `rp   ( -- a )`
yield the address of the return pointer.
note that the address points to the return stack *before*
this word was called.

### `rp0   ( -- a )`
a variable containing the value of the return stack at the beginning of the program.

### `s"   ( "string" -- , COMPILES: -- a u ) IMMEDIATE COMPILE-ONLY`
compile into the definition code to push the given string,
terminated by a double quote.
the string data and length are stored inline in the definition.

### `s>z,   ( a u -- a )`
compile into user memory a copy of the given regular string
converted to a null-terminated string.

### `sign   ( n -- )`
add a minus sign to the numeric output buffer if n is less than zero.

### `smudge   ( -- )`
toggles the smudge bit on the xt in latest.

### `source-id   ( -- n )`
yield either the file descriptor of the current input file,
-1 if the current input is from a string,
or -2 if the current input is the initialisation script
built into the binary.

### `sp   ( -- a )`
yield the address of the stack pointer.
note that the address points to the stack *before*
this value is pushed.

### `sp-reset   ( -- )`
reset the working stack pointer to its starting value.

### `state   ( -- a )`
a variable containing a boolean value.
if 0 (false), the system is in interpreting mode,
if -1 (true), the system is in compiling mode.

### `stderr   ( -- 2 )`
push the file descriptor of stderr to the stack.

### `stdin   ( -- 0 )`
push the file descriptor of stdin to the stack.

### `stdout   ( -- 1 )`
push the file descriptor of stdout to the stack.

### `string-source   ( -- n )`
yield the value of source-id when processing a string.

### `swap   ( u1 u2 -- u2 u1 )`
swap the two topmost values on the stack.

### `sys-close   ( fd -- n )`
perform a `close(2)` system call on the given file descriptor.

### `sys-open   ( mode flag z -- n )`
perform an `open(2)` system call on the file named
in the null-terminated string z, using the given file mode and flags.

### `sys-read   ( u a fd -- n )`
perform a `read(2)` system call, reading into the buffer `u a`
from file descriptor `fd`.
n is the resulting value of the register `rax`.

### `sys-write   ( u a fd -- n )`
perform a `write(2)` system call, writing the string `u a`
to file descriptor `fd`.
n is the resulting value of the register `rax`.

### `syscall0   ( rax -- u )`
perform the syscall with the id in `rax`,
and push the value of the `rax` register to the stack.

### `syscall1   ( rdi rax -- u )`
perform the syscall with the id in `rax`,
taking one parameter placed in `rdi`,
and push the value of the `rax` register to the stack.

### `syscall2   ( rsi rdi rax -- u )`
perform the syscall with the id in `rax`,
taking two parameters placed in `rdi` and `rsi`,
and push the value of the `rax` register to the stack.

### `syscall3   ( rdx rsi rdi rax -- u )`
perform the syscall with the id in `rax`,
taking three parameters placed in `rdi`, `rsi` and `rdx`,
and push the value of the `rax` register to the stack.

### `syscall4   ( r10 rdx rsi rdi rax -- u )`
perform the syscall with the id in `rax`,
taking four parameters placed in `rdi`, `rsi`, `rdx` and `r10`,
and push the value of the `rax` register to the stack.

### `syscall5   ( r8 r10 rdx rsi rdi rax -- u )`
perform the syscall with the id in `rax`,
taking five parameters placed in `rdi`, `rsi`, `rdx`, `r10` and `r8`,
and push the value of the `rax` register to the stack.

### `syscall6   ( r9 r8 r10 rdx rsi rdi rax -- u )`
perform the syscall with the id in `rax`,
taking six parameters placed in `rdi`, `rsi`, `rdx`, `r10`, `r8` and `r9`,
and push the value of the `rax` register to the stack.

### `then   ( -- ) IMMEDIATE COMPILE-ONLY`
conclude an if statement.

### `tib   ( -- a )`
a variable containing the address of the current input buffer.

### `to   ( comp: "name" -- , intr: u "name" -- ) IMMEDIATE`
compile or execute (depending on `state`) code to modify the contents
of a `value`.

### `true   ( -- u )`
a cell with all bits set.

### `tuck   ( u1 u2 -- u2 u1 u2 )`
place a copy of the highest value on the stack
below the second highest value on the stack.

### `type   ( a u -- )`
write u characters at a to output.

### `u<   ( u1 u2 -- ? )`
return true if u1 is less than u2.

### `u<=   ( u1 u2 -- ? )`
return true if u1 is less than or equal to u2.

### `u<>   ( u1 u2 -- ? )`
return true if u1 and u2 are not equal.

### `u>   ( u1 u2 -- ? )`
return true if u1 is greater than u2.

### `u>=   ( u1 u2 -- ? )`
return true if u1 is greater than or equal to u2.

### `until   ( ? -- ) IMMEDIATE COMPILE-ONLY`
if the given flag is true, loop back to `begin`.

### `value   ( u "name" -- )`
create a value called name, the initial value of which is u.

### `variable   ( "name" -- )`
create a variable word, which yields an address that can be written and read.

### `w/o   ( -- 1 )`
a constant, meaning 'write only', used for file I/O.

### `warn   ( a u -- )`
write u characters at a to error output.

### `while   ( ? -- ) IMMEDIAT COMPILE-ONLYE`
if given flag is true, continue the current begin-while-repeat loop,
otherwise branch to after.

### `write-file   ( a u fd -- u' e )`
write u bytes from a into fd.
u' is the number of bytes written.

### `xor   ( u1 u2 -- u )`
perform bitwise XOR on u1 and u2.

### `z"   ( "string" -- , COMPILES: -- a ) IMMEDIATE COMPILE-ONLY`
compile into the definition code to push the given string,
terminated by a double quote.
the string is null terminated and does not store a length;
this is meant for interfacing with the linux system.

### `zstrlen   ( a -- u )`
the length of a null terminated string in bytes.
the ending null byte is not counted.

## dictionary format

note that the string length of one byte limits a word's name to 255 characters.

| field | size |
| :---- | :--- |
| link to previous word | 8 bytes |
| flag field | 1 byte |
| string length | 1 byte |
| string | <256 bytes |
| code | variable length |

## reserved registers

the register `r15` is reserved for the parameter stack pointer.

## differences from standard forth

for the most part this forth intends to be in line with standards
but it diverges in a few notable places:

- the most visally obvious one by far,
    this forth uses lower case word names for core words.
- `find` takes `a u` instead of a counted string,
    and does not return 1 for immediate words.
- PNO words (`<# # #>` etc.) work with single cell numbers.
    this is because this forth has no double number support.
    (128 bit integer arithmetic does not seem all that useful to me)
- the dynamic allocation `free` word requires a length.
    this is because munmap requires a length.
- `abort"` does not take a flag and always executes.

## license

public domain.
although: a modified version of john hayes' test suite is used
which is under 'distribute but you have to include the copyright notice'
it's included in the source