Spec: Longest-prefix-match vs. Exact-match

[nhiroki]

For a document "/foo", how do we prioritize scopes "/fo_", "/foo" and "/foo_"?

1) "fo*" < "foo" or "foo" < "fo*" ?
2) "foo*" < "foo" or "foo" < "foo*" ?

[jakearchibald]

Good point. I think exact-match wins always. Rough spec:

1. Let pathname be the request url's pathname
2. For each scope in _ScopeToServiceWorkerRegistrationMap for this origin
   1. If scope is an exact match for pathname
      1. Let serviceWorkerRegistration be _GetRegistration(matchingScope)
3. If serviceWorkerRegistration is null
   1. For each scope in _ScopeToServiceWorkerRegistrationMap for this origin
      1. If scope ends "" is a prefix match for pathname (minus the "")
         1. Let serviceWorkerRegistration be _GetRegistration(matchingScope)

Having said that, @slightlyoff, I'm still not convinced exact-path matches are useful. I'd rather get rid of "*" and assume it's a prefix match. Removes a lot of this complication.

[michael-nordman]

Good question. Is their practical use case for an exact-path only registration? If not, +1 removing complexity.

[jakearchibald]

Agreed in hangout to remove "*" from scope urls to make the matching longest-prefix-match.

Need to update spec.

[slightlyoff]

Discussed this today with @annevk, @jakearchibald, and @tobie. The current thinking is that we'll drop the globbing vs. exact semantics and always make matches longest-prefix matches. That means we can remove the "*"

[KenjiBaheux]

Done in Blink (http://crbug.com/398892)