dunglas / go-urlpattern / 24663216038

92%

main: 92%

Pull #9

github

spec: conform to latest WHATWG URL Pattern WPT test data

Refresh testdata/urlpatterntestdata.json from the current WPT snapshot
(365 cases, up from 313) and align the implementation with the new
expectations:

- createComponentMatchResult now bounds iteration by the group name
  list length so user-provided regex groups (e.g. :foo((?<x>a))) no
  longer index past the name list and panic. Groups without a name
  are simply not exposed on result.Groups, matching WPT test 364.
- canonicalizeHostname: drop the broad reject-on-'/ ? # \\' guard.
  The URL parser now canonicalizes those boundary chars correctly;
  keep only the ':' check since without it "bad:hostname" would be
  silently split into host + port.
- hostnameParser: drop WithFailOnValidationError so tab/LF/CR in a
  hostname are stripped per the WHATWG URL spec instead of errored.
- canonicalizePort: rewritten to
    1. skip leading ASCII tab / LF / CR and require the first
       significant byte to be a digit (to reject "invalid80"),
    2. always parse against a non-special scheme when no protocol
       is supplied so the library doesn't collapse default ports
       like 80/443 to empty,
    3. remove the former "portValue != canonicalized" rejection
       which also rejected valid canonicalizations like "8\t0" -> "80"
       and "80x" -> "80" (state-override truncation).

All 365 WPT cases pass (2 skipped for advanced unicode features Go's
regexp engine does not support).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

28 of 29 new or added lines in 2 files covered. (96.55%)

1491 of 1627 relevant lines covered (91.64%)

1124.09 hits per line