tree-sitter grammar: missing interface/type alias/as cast + scanner bug killing ternary parsing

[Opaius]

Summary

The tree-sitter grammar for ripple (used by the Zed tsrx extension for syntax highlighting) was missing several TypeScript features and had a bug in the external scanner that broke ternary expression parsing. These issues caused syntax highlighting to fail on many .tsrx files using standard TypeScript patterns.

Bugs fixed

1. Missing interface_declaration (highest impact)

The interface keyword was completely absent from the declaration rule. Any file with export interface TransitionOptions { ... } would produce a parse ERROR at line 1, cascading into a catastrophic failure that killed ALL syntax highlighting for the rest of the file.

Fix: Added interface_declaration, interface_body, method_signature, and index_signature grammar rules. interface_body uses optional(choice($._semicolon, ',')) to separate members.

2. Missing type_alias_declaration

type Foo = ... declarations were not recognized.

3. Missing as_expression

TypeScript as type casts like (el as any).__ripple_exit failed.

4. object_type only allowed comma separators

Object type literals like { x: number; y: string } use semicolons, but object_type only accepted commas.

5. Scanner bug: AUTOMATIC_SEMICOLON handler prevented TERNARY_QMARK (ROOT CAUSE of ternary failure)

In scanner.c, the tree_sitter_ripple_external_scanner_scan function had the checks ordered with AUTOMATIC_SEMICOLON first:

if (valid_symbols[AUTOMATIC_SEMICOLON]) {
    bool ret = scan_automatic_semicolon(lexer);
    if (!ret && !valid_symbols[TERNARY_QMARK] && lexer->lookahead == '?') {
        return false;
    }
    return ret;  // returns before TERNARY_QMARK is ever checked
}

When the scanner saw ? after an expression, the AUTOMATIC_SEMICOLON handler would run first and return before TERNARY_QMARK ever had a chance. This meant the scanner never produced TERNARY_QMARK, making every a ? b : c ternary fail.

6. Missing generic_type in primary_expression

new Map<string | number, DOMRect>() failed because generic_type (Identifier) was only in the type rule (for type annotations), not in primary_expression. Added it with necessary conflict declarations.

Files modified (in grammars/tree-sitter/)

• grammar.js — grammar rules (+interface_declaration, +type_alias_declaration, +as_expression, +generic_type in primary_expression, object_type separator fix, added sepBy helper, added precedences array)
• src/scanner.c — scanner control flow fix (TERNARY_QMARK now properly checked after AUTOMATIC_SEMICOLON falls through)

Verification

All tests pass with 0 parse errors:

• transitions.tsrx (440 lines, interfaces, ternaries, generics, as casts): 0 errors (was ~10+ before fixes)
• +Page.tsrx (JSX-heavy): 0 errors
• new Map<string | number, DOMRect>(): 0 errors
• const x = a ? b : c;: 0 errors
• interface Foo { bar: string; baz(): void; [key: string]: any }: 0 errors
• a ?? b, a?.b, a < b: all still work

Remaining pre-existing issues

• a?.[b] (optional element access) — pre-existing, rare
• Type parameters in call_expression (e.g. foo<Type>()) — not supported yet