Do not decode path before matching

[trowski]

This reverts #4. I believe https://datatracker.ietf.org/doc/html/rfc3986.html#section-2.4 was mis-interpreted. We should not be decoding the path until after matching, otherwise "/" is interpreted as separating path segments.

Matched route parameters set in attributes are decoded. This is similar to our handling of query parameters, which are also decoded before returning from Request::getQueryParameter(), etc.

See https://chat.stackoverflow.com/transcript/message/57474623#57474623.

This is a BC break and will require a new major, since routes will need to be defined using the URL encoded version. I don't anticipate this would affect many users.

[Nek-]

Hello, I'm indeed hitting this issue too!

Here is my case:

• Path is /.well-known/mercure/subscriptions/https%3A%2F%2Fexample.com%2Fmy-other-topic
• The current code is transforming it to /.well-known/mercure/subscriptions/https://example.com/my-other-topic

It prevents fastroute to route correctly with the route /.well-known/mercure/subscriptions/{topic}. Maybe adding a test case with my case would be nice?

Furthermore, I'm not sure this is a BC break. To me it's only a (serious) fix.

[kelunik]

Closing this as discussed in chat: We need to decode before matching (skipping %2f), otherwise things like encoded "a" won't match as well.

[kelunik]

@Nek- This limitation is explicitly documented and there's a workaround: https://github.com/amphp/http-server-router?tab=readme-ov-file#limitations