RFM: v2 lexer carries no layout, so a newline before '-' merges two statements silently

[gunbai-bot]

Files one gunbc.recurring_failure_mode row, and nothing else.

The class: a statement followed by a line that begins with - is one binary expression in v2 (x then -1 becomes x - 1). The seed refuses that source as ambiguous (v1.compiler.parse is_ambiguous_prefix_infix_newline_boundary). So the parsers disagree, and v2's reading is silently different code.

Why it's a lexer-model gap, not a production fix: v2.std.compilers.lexing Token carries no layout, and the dag lex rules emit no newline token. No grammar production can tell x - 1 from x newline -1, and a parser special case would be a second, positional authority for a fact the token stream doesn't carry.

Scope: only operators that can also start an expression (today -, the seed's operator_continuation_dual_role_excluded_symbols). Infix-only operators can't begin a statement, so they aren't this class.

Rung: silent wrongness. The ceiling is structurally guaranteed, and the trigger is that the token stream carries layout and the grammar refuses a newline-preceded dual-role operator at an operand boundary, with a typed ParseRefusalReason.

This row gates the block-headed left-operand repair (quiet-gull-780's lane): that change would add block statements to the population, so it waits on this row's trigger.

🤖 Generated with Claude Code