Result: done
Files: crates/nikaia-std/src/tools/parser_grammar.nika crates/nikaia-std/src/tools/lowered/parser_grammar.rs crates/nikaia-std/src/lib.rs tests/compiler/tests/grammar_bound_peek.nika tests/compiler/tests/grammar_bound_peek.stdout tests/compiler/tests/parser_grammar_numbers.nika tests/compiler/tests/parser_grammar_numbers.stdout tests/compiler/tests/parser_grammar_literals.nika tests/compiler/tests/parser_grammar_trivia.nika tests/compiler/tests/parser_grammar_words.nika
Commit: Parser grammar in Nikaia: the number rules FLOAT, FLOAT_END, number_lit, negative_number_lit
Body: Translates by hand the Rust rules FLOAT, float_end, number_lit and negative_number_lit (with number_prefix and number_at) into tools/parser_grammar.nika; the actions call parse_numbers.nika as the Rust ones do. NUMBER_PREFIX is Rust’s number_prefix (- in front, E+/E-, ASCII letters, digits, _, .); NUMBER_READ binds it with a bound peek (#1201, winnow-grammar 7235ce46) and hands it to number_at; NUMBER_AT refuses Number::Not with a guard named digits and consumes consumed bytes with any{n}, so a refusal’s caret stands on the offending character, as in Rust, and the action throws parse_numbers’ words. float_end is FLOAT_END here: a lowercase rule skips trivia before its first element, so 1.5 G came out refused as a float with a scale; in capitals it reads only what touches the digits, as the Rust extern rule does. One difference stays: an action’s failure carries winnow-grammar’s normal priority (Diagnostics::external), the Rust rules gave a refused number PRIO_STRUCTURAL; it only matters where another alternative fails at the very same offset, and fail(..) takes no computed message. The lowered file adds two clippy lints to the tools allow list in crates/nikaia-std/src/lib.rs (let_unit_value for a unit action that throws, identity_op for the bound any{0 + n}). parser_grammar_numbers runs the rules over ADR-285 D16’s and ADR-322’s literals and prints each refusal beside what nikaia check (the Rust grammar) says of the same literal: words and caret equal (the check’s column is 12 to the right, after let x = ). grammar_bound_peek is #1201’s probe. The three older parser_grammar_* tests now copy parse_numbers.nika and ledger.nika’s slice beside the grammar file, which calls them; their output is unchanged.
Fixes: #1201
Part: #557
Pushed: local 8a23fc6dc2a87593a5ee40efdfe7f7b8538f7440
Tests: grammar_bound_peek and parser_grammar_numbers new (the bound peek is the dependency bump e265d24f, so not run against the old winnow-grammar; the number rules are new); it filtered parse/grammar/span/range_spelling/shipped_ledger: 163 passed; lower-std fixpoint holds (only parser_grammar.rs changed); tests/compiler –both-settings: 591 passed + the 3 older parser_grammar tests, which now also copy parse_numbers.nika and slice, re-run singly ok with unchanged stdout (594 total); tests/errors (it errors::) 38 passed; clippy clean; disk: project-tests cache 1.8G, 4.2G free, new cache entry rlib/af187e66 267 MB
Gate: green
Deviation: none
Found: none
Needs: none
Changelog: none
Owner: none
Disk freed: none
Rs2nika: none
Attribution:
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
Claude-Session: https://claude.ai/code/session_01GFeMqxRbtUMiGVyDMxsvEq
