WHATWG URLPattern

Elide implements the WHATWG [URL Pattern Standard][spec] URLPattern global — a routing-oriented URL matcher with named captures, per-component regex groups, and the full {...} grouping / modifier grammar. The underlying regex engine is graal-js's TRegex, so user-provided (regex) groups get full ECMAScript RegExp semantics (including Unicode property escapes and variable-length lookbehind).

[spec]: https://urlpattern.spec.whatwg.org/

Construction

new URLPattern(input, baseURL?, options?)   // string form
new URLPattern(init, options?)               // init-dict form

URLPatternInit

interface URLPatternInit {
  protocol?: string;
  username?: string;
  password?: string;
  hostname?: string;
  port?: string;
  pathname?: string;
  search?: string;
  hash?: string;
  baseURL?: string;
}

Any component not specified defaults to the pattern "" (matches an empty component). The optional baseURL field lays down defaults for unspecified components.

Readonly attributes

Operations

pattern.test(input?, baseURL?)     // → boolean
pattern.exec(input?, baseURL?)     // → URLPatternResult | null

• input may be a URL string or a URLPatternInit dict. When it's a string and baseURL is provided, the string is resolved against the base.
• exec returns null on mismatch; otherwise a URLPatternResult:

interface URLPatternResult {
  inputs: Array<string | URLPatternInit>;
  protocol: { input: string; groups: Record<string, string | undefined> };
  username: { input: string; groups: Record<string, string | undefined> };
  // ... one entry per URL component
}

Pattern grammar

Canonicalization

Pattern and input strings are both canonicalised per the URL Standard before matching, so a pattern written against a canonical form matches a loosely-typed user URL:

• protocol — ASCII-lowercased.
• hostname — IDNA toASCII + lowercased. IPv6 literals pass through verbatim.
• port — leading zeros stripped, range-validated (0..65535).
• pathname / search / hash / userinfo — percent-encoded per the URL spec's per-component encode sets (fragment set for hash, query set for search, path set for pathname, userinfo set for username / password). Already-well-formed percent triplets pass through without double-encoding.

Pattern inputs with embedded metacharacters (: { } ( ) * ? + \) preserve those characters verbatim during canonicalization so they're seen by the pattern tokenizer.

Examples

// Named path segment.
const p = new URLPattern({ pathname: "/users/:id" });
p.test({ pathname: "/users/42" });                // true
p.exec({ pathname: "/users/42" }).pathname.groups; // { id: "42" }

// Full URL with wildcard host.
const api = new URLPattern("https:<<>>
api.test("https://v2.example.com/api/users");     // true
api.exec("https:<<>>

<<>>
const numeric = new URLPattern({ pathname: "/items/:id(\\d+)" });
numeric.test({ pathname: "/items/abc" });         <<>>
numeric.test({ pathname: "/items/123" });         <<>>

<<>>
const opt = new URLPattern({ pathname: "/users/:id?" });
opt.test({ pathname: "/users" });                 <<>>
opt.test({ pathname: "/users/42" });              <<>>

<<>>
const multi = new URLPattern({ pathname: "/files/:path+" });
multi.exec({ pathname: "/files/a/b/c" }).pathname.groups.path;  <<>>

<<>>
const rel = new URLPattern("/items/:id", "https://example.com/");
rel.test("https://example.com/items/7");          // true

Notes on the implementation

• Plain-call URLPattern(...) is routed to TypeError via a manual construct?-ternary patch in the generated JSURLPatternClassBase.java (same pattern as the other Phase 2 classes). elide-dev/bindgen emits this branch automatically on the next regenerate.
• Compiled regexes are produced via RegexCompilerInterface.compile(...) — graal-js's TRegex compiler — so custom (regex) groups get full ECMAScript semantics, not Java's near-compatible subset.
• Capture names are tracked in a parallel list rather than baked into the regex as (?...). This is necessary because the spec's + / * modifier forms require an outer repeating group that ECMA doesn't allow to re-bind a named group on each repetition; the two-list design keeps group extraction a single array lookup.
• The class is installed via installAsGlobal (not putBuiltinClass) for the same reason as AbortSignal / AbortController: graal-js's per-realm shape-slot array is fixed at JSContext creation and our non-ordinary URLPatternObject layout collides with graal-js's own JSOrdinaryObject shapes at high indices. The custom installer uses JSObjectFactory.createBound to keep the shape on the factory itself.