RHS literal branch has no yang guard: adjacent literals parse without whitespace between them

[thedavidmeister]

Unit

src/lib/parse/LibParse.sol parseRHS branch chain: the literal branch (char & CMASK_LITERAL_HEAD > 0, ~line 398) calls pushLiteral without checking FSM_YANG_MASK, then sets yang itself.

Intent oracle

The yang FSM exists so that adjacent tokens require a separator. The RHS word branch enforces it explicitly (~line 231):

if (char & CMASK_RHS_WORD_HEAD > 0) {
    // If yang we can't start a new word.
    if (state.fsm & FSM_YANG_MASK > 0) {
        revert UnexpectedRHSChar(state.parseErrorOffset(cursor));
    }

and the LHS item branch does the same (~line 155). Maintainer ruling (2026-08-24): literals should have a gap too.

Violated property

After a literal completes, the state is yang, but the next iteration's literal branch admits any CMASK_LITERAL_HEAD char (0-9, -, ", [) without the yang check. Consequently adjacent literals parse with no whitespace:

• _ _: 1-2; → literals 1 and -2 (driven by - being in CMASK_NUMERIC_LITERAL_HEAD)
• 1"a" → literal 1 then string literal "a"
• 1[x y] → literal 1 then sub-parseable literal

Proposed fix

Mirror the word branch: at the top of the literal branch, if (state.fsm & FSM_YANG_MASK > 0) revert UnexpectedRHSChar(state.parseErrorOffset(cursor)); — plus tests pinning that 1 2 parses and 1-2/1"a"/1[x y] revert.

Triage note

This tightens the accepted grammar: any existing rainlang text relying on literal adjacency (e.g. 1-2 meaning two literals) would start reverting. No known corpus depends on it, but the change is observable to every downstream parser consumer.

Found by the 2026-08-24 rain.string CMask consumer-oracle audit (the mask value is correct; the missing guard is at this call site).

🤖 Generated with Claude Code