This document identifies the stable editor-facing boundary and the current implementation limits. It does not standardize one editor extension.
Editors, formatters, syntax highlighters, and language servers must use the
same UTF-8, token, package, import, and span rules as the compiler. Diagnostic
positions use the versioned diagnostic schema in
DIAGNOSTIC_SCHEMA_V2.md; tools must not silently
substitute UTF-16, byte, or display-cell columns where that schema requires a
different coordinate.
The current language-server implementation provides diagnostics and selected navigation/completion behavior by invoking compiler-owned parsing and semantic analysis. Its feature set is informative and may be incomplete. A successful editor operation does not establish program conformance, and a missing editor feature does not make valid source invalid.
Any future stable LSP surface must document, per request, whether it is syntax- only or semantic, project-root discovery, overlay/unsaved-buffer behavior, cancellation, partial results, position encoding, error handling, and behavior for unsupported compiler states.
Secondary grammars are editor aids, not language authorities. The audited tree-sitter grammar accepts several historical or permissive forms that do not match the current P-374/S-205 grammar, including broader parameter, interpolation, loop, and arena shapes. Consumers must not use it to decide whether a program is valid. It should be generated from or tested against the normative lexical and syntactic grammar before it can be described as conformant.
At minimum, editor-grammar tests must include every lexer token, contextual keyword, nested block comments, comment attachment, interpolation boundaries, all declaration families, precedence tiers, recovery at malformed delimiters, and explicit negative cases for removed syntax.