Skip to content

char-literal-escapes.md

Latest commit

 

History

History
25 lines (15 loc) · 2.09 KB

char-literal-escapes.md

File metadata and controls

25 lines (15 loc) · 2.09 KB

< All Decisions

Char Literal Escape Sequences — Design Rationale

Panel Deliberation

Five panelists voted on one question. Resolves gap: "char-literal escape set not defined in spec".

Q: What escape sequences should the lexer accept inside '...' char literals? (4-1 for Option A)

  • Systems: Option A. Common control codes (\n, \t, \0) must resolve at parse time to a clean uint32_t constant; forcing Char.from_code_point(10) for a newline turns a compile-time constant into a function call.
  • Web/Scripting: Option A. Every scripting language developers use (Python, Ruby, Rust, Go) accepts \n-style escapes in character literals; diverging buys nothing but surprise.
  • PLT: Option A. Consistency with string-literal escapes is Principle of Least Astonishment; swap \" for \', drop \{ \} (interpolation inapplicable to scalars), add \0 (NUL genuinely useful).
  • DevOps: Option A. '\t' and '\n' need to Just Work in config parsers, CSV tooling, CLI flag handling without importing Char methods; deferring unicode escapes alongside task 19v5gb keeps the escape story coherent across both literal kinds.
  • AI/ML: Option C (dissent). Tokenizer and vocab code leans on emoji and non-BMP codepoints; '\u{1F600}' reads better than Char.from_code_point(0x1F600). Concede deferring with hex-escape work is defensible.

Decision

Char literals accept: \n \r \t \\ \b \f \0 \' — same as string escapes swapping \" for \' and dropping \{ \}. Unicode/hex escapes deferred alongside string hex escapes (task 19v5gb); non-ASCII chars use Char.from_code_point(...) until then.

Spec Amendment

Add to sections/02_syntax.md after the string escape table:

Char literals ('x') accept: \n (newline), \r (carriage return), \t (tab), \\ (backslash), \b (backspace), \f (form feed), \0 (NUL), \' (single quote). Any other escape is a lexer error. \u{...} and \x.. are deferred (task 19v5gb). A char literal must contain exactly one Unicode scalar value; '' and 'ab' are errors. Surrogate codepoints (0xD800–0xDFFF) are invalid.