Arx Lexical Syntax

The authoritative token spellings, aliases, and editor-tooling inventory live in packages/arx/src/arx/lexer/syntax.json. This guide covers writing source and important restrictions, not a duplicate token catalog. Parser and semantic support is summarized in the capability matrix.

Source files and whitespace

• recognized extensions: .x and .arx
• indentation is significant
• the canonical indentation unit is 2 spaces
• tabs outside quoted literal contents are rejected; use spaces
• newlines delimit logical lines
• blank lines do not create indentation transitions

Files must be valid UTF-8 and no larger than 4 MiB of UTF-8 bytes. Token start locations use one-based lines and Unicode code-point columns. The lexer also limits delimiter nesting to 256, numeric tokens to 4,096 characters, and non-EOF tokens to 250,000. These are compiler-version resource limits, not a frozen language compatibility promise.

fn absolute(value: i32) -> i32:
  if value < 0:
    return 0 - value
  else:
    return value

Comments and docstrings

# starts a line comment. Block comments are not supported.

Triple backticks delimit Douki YAML docstrings and produce a dedicated docstring token:

```
title: Module documentation
summary: Optional description.
```

Docstring placement and schema validation are parser concerns documented in Docstrings.

Identifiers

Identifiers start with a Unicode letter or _ and continue with Unicode alphanumeric characters or _.

Reference pattern for Unicode-aware tooling:

(?:[_\p{L}])(?:[_\p{L}\p{N}])*

The exact Unicode categories currently follow the Python runtime. Matching is case-sensitive.

Reserved forms

Lexical reservation does not promise parsing or lowering. In particular, general const, then, and operator declarations remain unavailable even though their spellings are recognized.

Numeric literals

Supported forms:

• decimal integers: 0, 42
• decimal floats with one dot: 3.14, .5, 5.

Not supported:

• hexadecimal, binary, or octal prefixes
• exponent notation such as 1e9
• separators such as 1_000
• literal suffixes such as 42u32

Multiple dots are invalid, and . by itself is punctuation rather than a number.

Strings and characters

Double quotes create string literals. Single quotes create character literals. The lexer recognizes these escapes:

Raw strings, triple-quoted strings, and interpolation are not supported. Triple backticks are reserved for Douki docstrings, not ordinary string values.

Structural forms

Declaration modifiers

@[public, static, constant]
version: i32 = 1

Recognized modifiers are public, private, protected, static, constant, mutable, abstract, and extern.

Templates

@<T: i32 | f64>
fn identity(value: T) -> T:
  return value

Explicit template calls use angle brackets: identity<f64>(1.5).

Imports

Grouped imports use parentheses, require from, and allow a trailing comma:

import (
  sin,
  cos,
  tan as tangent,
) from math

Collection annotations

See type annotations. Literal ellipsis has different support boundaries for runtime-shaped tensors and runtime-schema DataFrames; it is not uniformly parameter-only.

Consistency rule

Changes to lexical syntax update the manifest first, then tokenization, parser, tests, and derived editor consumers. Update this short guide only for changed workflows or restrictions.

File, open, and dunder method names remain identifiers, not new keywords. Class lifecycle hooks do not add Arx with statements. Named calls use resolved signatures, not lexical privilege based on the callee's spelling.