Allow non-RFC 3986-compliant URLs

[c960657]

HTTP::Cookie uses uri which parses URLs strictly according to RFC 3986.

Many HTTP clients allow requesting URLs which are not compliant with RFC 3986. In practice, this usually works. E.g. cURL supports a wider range of URLs described as “RFC 3986 plus”. Apart from 8-bit characters, the requested URL is more or less passed on verbatim to the server, even if it contains characters not permitted by RFC 3986. The same applies to web browsers.

Trying to “fix” non-compliant URLs is tricky. See e.g. httprb/http#758.

In the spirit of Postel's law (“be conservative in what you do, be liberal in what you accept from others”), I suggest HTTP::Cookie accepts a wider range of URLs.

With this PR, HTTP::Cookie does not parse URLs from strings if they are represented by an URI-like object, e.g. HTTP::URI from http.rb or Addressable::URI. This also reduces the work connected with converting a URI-like object to a string and parsing it back into another URI object.

Fixes #24.

[knu]

Thanks for the PR.

How about normalizing non-RFC-compliant URIs to solve the problem? Here's the approach I used in the past that worked pretty well without having to fix everywhere:

https://github.com/huginn/huginn/blob/2d5fcafc507da3e8c115c3479e9116a0758c5375/lib/utils.rb#L27

I am concerned that introducing polymorphism here may over-complicate the internals.

[c960657]

How about normalizing non-RFC-compliant URIs to solve the problem?

AFAICT RFC 6265 does not mention normalising URLs, so I assume they should be compared byte for byte, i.e. so that /a and /%61 are treated as two different paths. I have made a new PR that takes this approach: #45

Addressable::URI is very eager when normalising and sometimes modifies URLs in a way that is not treated as equivalent by the server (see sporkmonger/addressable#472). I assume this is not a big problem in practice for this library, though, so perhaps it is a pragmatic approach to just parse everything with Addressable::URI, even though this is not mandated by the RFC.

What do you think?

[knu]

Superseded by #45.