---

agent: builder model: opus max_turns: 110 budget_usd: 22 effort: high —

Brief: #557 step 3, the grammar language gets same_line

Blocks: the parser grammar port (#557 step 4: parser/mod.rs, 2,728 Rust lines, the largest file left in Rust)

Score: decided (ADR-339 D2, Part II 10.8 table row same_line); steps 1-2 are on develop (2ca9841, 9545797). One built-in, checker and lowering.

Facts: Where things live, Traps already met, Commands

Decided

• ADR-339 D2 (copied): “same_line matches nothing and fails where the whitespace skipped just before it held a line break. The generated parser records, by position, what its last WS skip held, so the answer is a fact about the input and survives backtracking.” ADR-339 D3: no grammar declares a state or an extern rule; the lowering may write both into the generated grammar!.
• Only a grammar that writes same_line (and declares no rule of that name) changes; every other generated grammar lowers byte for byte as before.
• A line break anywhere in the skipped run counts, inside a block comment too. The Rust Trivia counts only breaks outside comments, so a /*\n*/ [0] is an index there and not here. Name this difference in Body: (the port compares on every tests/compiler program).

Facts

• The Rust model (crates/nikaia/src/parser/mod.rs, grep, lines drift): fn same_line (i.current_token_start(), then Trivia::began_a_line(at)), state Trivia;, extern rule same_line -> ();, the recording rule WSE = w:multispace1 @ at -> { _state.user().ran(at, w.contains('\n')) }, the entries building ParseInput::<Trivia> with ParseContext::<Trivia>. The record is PrTrivia::ran / began_a_line in tools/parser_leaves.nika. The winnow-grammar source (state, extern rule, _state.user()) is under ~/.cargo/git/checkouts/winnow-grammar-*/.
• Where the generated parser is written: emit_grammar.nika (emit_grammar_over, emit_grammar_body); step 2’s template rules __nikaia_span / __nikaia_here and emit_writes_a_built_in are the pattern. The checker side: check_grammar.nika beside span / here.
• The lowering writes, in a grammar that uses same_line: (a) a state of one small record type in nikaia_std::grammar (fields end, newline; ran, began_a_line as PrTrivia’s; one type, no second copy of the logic); (b) a WS that records the run it skipped (the grammar’s own WS renamed under a recording WS, or the default whitespace written out the same way); (c) a template rule __nikaia_same_line (an extern rule over a hand-written function after the block); (d) entries building the parse input with that state. Over Bytes and text alike (needs no text::Pos, so no NK1187).

To build

1. same_line in the checker (arity 0, binds nothing; a grammar’s own rule same_line wins) and in the lowering as above. Tests, each a .nika with its .stdout in tests/compiler/tests/: grammar_same_line.nika (postfix = name (same_line "[" n "]" | same_line "(" n ")")* over text and over Bytes: a\n[0], f\n(x) are not an index / call, a [0], f (x) are; a backtracking case: an alternative that skips the break and fails, then one asking same_line at the same place gives the same answer) and grammar_same_line_own_ws.nika (own rule WS = (multispace1 | "//" until(line_ending) | "/*" until("*/") "*/")*: a // c\n[0] not an index, a /* c */ [0] an index, a /*\n*/ [0] not an index). One commit with the lowered files, Part of #557.

Done when

tests/compiler with both settings (emit_* changed), it tests filtered on grammar, the builder gate, clippy, lower-std fixpoint.

Stop and report if

winnow-grammar cannot carry a state in a generated grammar over input bytes, or its entries cannot be handed one (say what it offers); an existing test’s .stdout/.stderr changes; a change in grammar_run.rs, build_time.rs or parser/mod.rs is needed.

This site is open source. Improve this page.