uxn/doc/tal.md

# TAL Assembler

See also [the UXN virtual machine](./uxn.md)

TAL is a Forth like two-pass assembler language translating directly to UXN memory images.

## Words
Words are up to 63 consecutive non-whitespace characters.
For instance `0x75786E00` (ascii UXN\0) would be one TAL "word" although its value is many bytes.
`foo`, `bar-baz` and `quix/qux` would all be examples of words.

Words starting with `_` are defined to be relative references.
Words starting with `,` are 

## Comments
Comments in TAL are written `( ... )` and support nesting. Eg. `( () )` is a valid comment. `( ( )` is not.
TAL does not have a way to "close all start comments" like Java and some other languages do.

## Includes
TAL files can include other files by writing `~<filename>`.
For instance the `uxnasm.tal` file writes `~projects/library/string.tal` to include implementations of string functions.
As with other preprocessor and assembler languages, TAL does not support namespacing, renaming or selective importing.

- All included code is assembled at the point where it is included.
- TAL does not support multiple definition or idempotent includes, and will error on repeated or recursive inclusion.

## Macros
Macros are sequences of instructions which may be repeated.
Macros are defined by writing `%macro-name { ... }`.
The canonical UXNASM does not allow macros to exceed 64 words in size.

When macros are invoked by using the macro-name as a bare word, the contents of the macro will be inserted.
Sub-macro references are supported and will be expanded with no recursion guards or limit.

## Padding

`|<number>` "pad-absolute" pads the resulting UXN rom to a given absolute address.
For instance `|0x0000` would explicitly align the assembler's point to `0x0000`.

`$<number>` "pad-relative" pads the UXN rom by the specified number of words (bytes).
For instance `$2` would move the assembler's point forwards two words.

## Labels

`@<word>` defines a top-level label.
For instance `@foo` would make the word `foo` a valid symbol for use elsewhere.
Defining a top-level word establishes a scope within which sub-labels may be defined.

`&bar` following `@foo` would create the label `foo/bar`.
This can be used to create semantic tables.

## References

Labels may be referenced in one of seven ways:
- Literal byte zero-page - `.label`
- Raw byte relative - `_label`
- Literal byte relative - `,label`
- Raw byte absolute - `-label`
- Raw short absolute - `:label` or `=label`
- Literal short absolute - `;label`

Literal labels are inserted with a `LIT` or `LIT2` as appropriate.
Raw labels are inserted directly into bytecode.

Absolute labels are double quantities.
Relative labels are single signed byte quantities with a ±127 range.

The zero page (`#00XX`) is used for system devices, along other things.
It's common to see labels such as `.Mouse/state`, being a reference to the address `#0096`.
However as UXN has a special `LDZ` operation for loading from the zero page, this address can be specified as simply `#96` to save a byte.
As the last device is mapped to `#CX`, it is common to see `#DX`, `#EX` and `#FX` used for program-global variables for ease of access.

Literal byte relative references ala `,foo` are used for control flow.
Using only a single byte, these references have a range of ±127 instructions.
A typical opcode sequences would be `,loop JMP`, eg. emit a relative address value to the loop label and perform a computed relative jump.
For bytecode compactness, UXN programs tend to use computed rather than absolute jumps.

The difference between single and double word references is critical, because the `LDR` instruction is a computed relative load, whereas `LDA` is an absolute short address load.

## Brackets
`[` and `]` are treated as whitespace, and may be used for visual grouping.
While they have semantics in traditional Forth, they have no semantics in TAL.

## Literals

Hex constants are written `#[0-9a-f]{1,4}`.
For instance `#00` or `#ffff` would be valid hex constants, the first assembling to one word, the second to two.
One and two byte literal quantities may also be provided without the `#` prefix.

Words may be captured as ASCII formatted strings.
Such strings are written `"<word>`.
For instance `"foo` would cause the bytes `#66 #6f #6f #00` to be literally inserted into the memory image.

As `"` notation cannot capture whitespace, the `#20` (space), `#0a` (newline) and `#09` (tab) character constants are common.