Lexical Rules

Norm source files use UTF-8. The compiler decodes a file into Unicode text before establishing source positions.

Whitespace and newlines

Spaces, tabs, and newlines separate tokens. A newline usually does not end an expression; grammar structure and operators determine whether it is complete. See ref<T> reference syntax for a statement boundary involving line-initial dereferences. An ending semicolon may be omitted. The first specification discourages several statements on one line.

Block call chains require their connecting head to remain on the same line; a newline ends eligibility for a connection without a dot.

Identifiers

An identifier starts with a Unicode letter or underscore. Later characters may be Unicode letters, decimal digits, or underscores. Keywords cannot be identifiers. Names use NFC normalization; two names equal after normalization are duplicate declarations.

String displayName
Integer retry_count

Public APIs should prefer ASCII identifiers for tool and ecosystem compatibility. The compiler should warn about easily confusable cross-script characters.

Number and string tokens

A numeric token includes integers, decimal points, exponents, and digit-separating underscores. The literal and expected type determine its concrete type. String uses double quotes and backslash escapes. ${expression} contains an ordinary Norm expression, while \$ means a literal $. The current lexer does not recognize // or /* */ as comments; see Status for delivery state.

Source positions

SourceSpan stores a file and half-open range of UTF-16 code units. Line and column positions use the same unit and align with LSP. A protocol needing original-file byte positions must separately derive a byte range from the UTF-8 input, not conflate it with SourceSpan.