How it's tested

The fixtures in spec/fixtures are the source of truth. The spec says what should happen, the fixtures pin it down, and every port has to pass all of them. Both packages do: 12,965 of 12,965, with empty skip lists, in CI on every push.

Handwritten: 1,148 cases

Each case cites the spec rules it exercises, and 62 of them spell out the complete output, field by field, so nothing drifts unnoticed.

Group	Cases	What's in there
Regressions	273	Every test from two real-world parsers this library replaces, and the known failures of get-artist-title and friends
Versions and brackets	90	Every version keyword, generic heads (Radio Edit → radio), multi-version groups, feat inside brackets
Artists	76	Joiners, guards (AC/DC, Mumford & Sons, Malcolm X), knownArtists, dedup
Title credits and producers	57	Title ft. X, (prod. by X), self-produced artists
Separators and suffixes	63	Dash variants, - Remastered 2015, unspaced and asymmetric dashes
Prefixes, junk, flags	111	Track numbers vs 2 Unlimited, cue times, every junk kind, years, explicit/clean
Modes	65	YouTube pipes, "Title" by Artist, uploader fallback, file names
Unicode and normalisation	71	NFC, zero-width characters, fullwidth brackets, Cyrillic, CJK, Turkish İ
format, parseArtists, unknown IDs	94	Round trips, canonical joiners, ID - ID
Tracklist mode	54	Real lines that single-line parse() gets wrong, and real lines with no separator at all
Tracklist lines and times	70	LF/CRLF/CR, trailing spaces, line kinds, every time form, brackets, gap separators, invalid times
Tracklist positions	13	A real 81-line list numbered without dots, and the 2 Unlimited guard
Tracklist warnings	36	Each code with its near-miss, and real out-of-order lists with their suggestions
Tracklist edits	75	Style keeping, shift, clamp, sort, insert/delete line endings, invalid edits, locality

Generated: 11,817 cases

A deterministic generator builds inputs from small "atoms": an artist, a joiner, a feat placement, a version group, a junk suffix. Each atom carries its own expected contribution, and the expected output is assembled from those atoms. It never comes from running a parser, so the generator can't quietly agree with the code it's testing. Combinations whose answer would depend on a rule the atoms don't model are rejected instead of guessed.

Scenario	Cases	Scenario	Cases
artists × joiners	1,811	kitchen sink (pairwise + random)	1,589
versions	1,650	junk × YouTube	806
dash suffixes	1,174	file names	780
dedup	1,008	unicode	674
feat placement	922	prefixes	577
tracklists	826

Properties

fast-check (JS) and Hypothesis (Python) throw thousands of random strings at the parser to check the invariants from the spec:

• it never throws, on any input, and a 10,000-character input stays linear;
• normalize is idempotent;
• whitespace, dash and zero-width noise don't change the result;
• every output validates against the JSON Schema;
• no word of the input goes missing from the output;
• parse(format(parse(s))) round-trips.

For tracklists, on random multi-line text and random edit lists:

• raw + eol over all entries gives back the input, and lead + time + gap + rest gives raw;
• applyEdits(t, []) returns t, and an edit never touches a line it doesn't name;
• shifting forward and back restores the text, unless something was clamped or upgraded;
• 10,000 lines parse in linear time (the JS suite allows 3 s; it takes about 0.6 s).

Port against port

A fixture only checks the fields it asserts. So compare-ports.mjs runs every fixture input through both packages and diffs the complete JSON output, tracklists and edits included. The difference count is zero.

Real tracklists

The tracklist rules were written against a private corpus of real podcast tracklists, a little over 5,000 lines. It can't be published, so it stays out of the repository. Every surprising line in it became a handwritten fixture instead, which is why the tracklist fixtures talk about "real lines".

Run it yourself

pnpm install
pnpm test          # fixtures, properties, generator tests
pnpm conformance   # pass rate per fixture file and per spec rule

cd python && uv run pytest   # the same fixtures and invariants in Python