yarlpattern 0.2.0

WHATWG URLPattern for Python. 100% specification strict, pure Python, optimized and yarl-compatible.

yarlpattern

WHATWG URLPattern for Python — 100% conformance to the upstream WPT corpus: 469 / 469 cases passing across all five test suites, the same files Chromium, Safari, and Firefox validate against.

Pure Python on top of yarl — immutable pattern objects, component properties named after their URL counterparts, zero non-Python dependencies. The pattern is the API: compile once, then ask .test(url) or .exec(url) from anywhere a yarl.URL lives.

from yarlpattern import URLPattern

# Multi-tenant API: the subdomain identifies the tenant, the path
# captures the API version and the resource tail — all extracted in
# one match call.
pat = URLPattern({
    "hostname": ":tenant.myapp.com",
    "pathname": "/api/v:version/*",
})

result = pat.exec("https://acme.myapp.com/api/v2/users/42")
result.hostname["groups"]["tenant"]    # 'acme'
result.pathname["groups"]["version"]   # '2'
result.pathname["groups"]["0"]         # 'users/42'

pat.test("https://foo.example.com/api/v2/users")  # False — wrong host
pat.test("https://acme.myapp.com/api/users")      # False — no version

That's the differentiator. Flask-style :id routers match the path component in isolation; URLPattern matches across protocol, hostname, port, path, and search at once, returning structured named groups per component.

Conformance

469 / 469 upstream Web Platform Tests pass (100%) across every WHATWG URLPattern test suite — the same files Chromium, Safari, Firefox, Ada, and rust-urlpattern validate against. The WPT corpus is SHA-pinned by scripts/fetch_references.sh to commit dd54691 (2026-05-11) so the pass count is reproducible at any future date.

Suite	Source	Status
urlpattern.any.js	WPT  ·  urlpatterntestdata.json	✅   366 / 366
urlpattern-constructor.any.js	WPT (inline)	✅   4 / 4
urlpattern-hasregexpgroups.any.js	WPT  ·  urlpattern-hasregexpgroups-tests.js	✅   55 / 55
urlpattern-compare.tentative.any.js	WPT  ·  urlpattern-compare-test-data.json	✅   25 / 25
urlpattern-generate.tentative.any.js	WPT  ·  urlpattern-generate-test-data.json	✅   19 / 19
yarlpattern unit tests	this repo  ·  tokenizer / parser / parts / regex / engine / pattern	✅   130 / 130
Total		✅   599 / 599

→ Full per-case conformance report (regenerate via just compliance-report)  ·  Documented deviations and stricter-than-yarl rules

What we get right that's easy to miss

The 100% number is the headline. Equally load-bearing — and easy to skip past — are the per-component canonicalisation rules the WHATWG URLPattern spec quietly requires. yarlpattern enforces all of them; a stdlib-only port that goes through urllib.parse cannot:

• WHATWG URL parsing end-to-end via yarl, not urllib.parse (which is not WHATWG-conformant).
• IDNA2008 / UTS46 hostname canonicalization via the third-party idna package, not Python's stdlib idna codec (which is IDNA2003 and not spec-compliant for modern IDN labels).
• Strict port parsing — "8080xyz" is rejected as the WHATWG URL parser's port-state requires; webhook-validation patterns that constrain on exact ports stay robust against junk suffixes.
• Case-preserving %XX passthrough in pattern literals — caf%c3%a9 round-trips as itself, where yarl would normalise to uppercase (WPT cases 146 / 148 pin this).
• U+FFFD substitution for unpaired surrogates before UTF-8 percent-encoding, where yarl silently drops them (WPT case 157).
• Hostname-pattern truncation at ? / # / / / \, matching browser engine behaviour for hostnames that were pasted from full URLs.

Stdlib-only mode. Under stdlib re without the [regex] extra, conformance on urlpattern.any.js is 364 / 366 (99.5%). The two outlier patterns — [a&&b] (intersection) and [a--b] (difference) from the JS v-flag — require Matthew Barnett's regex package; they're marked xfail with an install hint when it's absent. pip install yarlpattern[regex] activates them.

API surface

Every stable and tentative method in the WHATWG URLPattern IDL is implemented: URLPattern(input | string, baseURL?, options?), test, exec, all eight component properties, has_regexp_groups, URLPattern.compare_component, and the tentative generate(component, groups). The IDL camelCase spellings (hasRegExpGroups, compareComponent) are kept as aliases so code ported verbatim from the spec or browser JS reads identically. See SPEC_DEVIATIONS.md for the intentional Python-flavour choices.

How this differs from aiohttp.web.UrlDispatcher

aiohttp.web.UrlDispatcher is a mature path-router shaped around web-request dispatch. yarlpattern is a predicate: it matches across all eight URL components (not just the path), works standalone (no server context required), and uses the same WHATWG pattern syntax browsers, Deno, Bun, and Cloudflare Workers all implement.

Use UrlDispatcher if you're building an aiohttp service. Use yarlpattern if you're matching URLs outside a server context, need to constrain on hostname / port / scheme alongside path, or want patterns that match what browsers do.

→ Full comparison

How this differs from yarl

yarl is a URL parser / builder; yarlpattern is a URLPattern matcher. They're complementary — yarlpattern depends on yarl for URL parsing and IDNA hostname encoding, accepts yarl.URL directly in .test(...) and .exec(...) calls (no str() round-trip), and uses WHATWG component names (protocol / hostname / pathname / search / hash) rather than yarl's (scheme / host / path / query / fragment).

Where the WHATWG URLPattern spec is stricter than yarl, yarlpattern enforces the spec — see the Conformance section above and SPEC_DEVIATIONS.md.

Component-name mapping for muscle-memory porting:

yarl	yarlpattern	WHATWG / browser JS
scheme	protocol	protocol
user	username	username
host	hostname	hostname
path	pathname	pathname
query (MultiDict)	search (str)	search
fragment	hash	hash

→ Full comparison, including the WPT cases that pin down each strictness rule, the with_* ergonomics, and the encoding philosophy yarlpattern shares with the rest of aio-libs.

Install

pip install yarlpattern            # stdlib re backend
pip install 'yarlpattern[regex]'   # full 100% conformance — see Conformance § above

Bring your own regex engine

The matcher's regex backend is pluggable behind a @runtime_checkable Protocol. Two adapters ship in-tree — stdlib re (always available; default fallback) and regex (auto-detected when yarlpattern[regex] is installed; closes the [a&&b] / [a--b] gap).

Selection priority: explicit engine= argument › URLPATTERN_REGEX_ENGINE env var › auto-probe (prefers regex when importable, falls back to re). See src/yarlpattern/_regex_engine/protocols.py for the Protocol definitions; a future PyO3-backed engine slots in as one new adapter module.

Quick start

uv sync --all-groups
uv run pytest                  # full test suite
just check                     # lint + types + tests (requires `just`)

from yarlpattern import URLPattern

# Dict form, fully wildcarded except path
api = URLPattern({"pathname": "/api/v:version/users/:id(\\d+)"})
api.test({"pathname": "/api/v2/users/42"})              # True
api.exec({"pathname": "/api/v2/users/42"}).pathname     # {'input': '...', 'groups': {'version': '2', 'id': '42'}}

# String form with base URL
route = URLPattern("/posts/:slug", "https://blog.example.com")
route.test("https://blog.example.com/posts/hello")      # True

# Match a full URL against the constructed pattern
pat = URLPattern("https://*.shop.example/products/:sku")
pat.test("https://eu.shop.example/products/SKU-991")    # True