`Version` can be exhaustively matched

I feel like having Version be a load of consts rather than an enum was probably so that it could be extended without breakage. Unfortunately it seems rustc is too smart and sees through the ruse?

fn foo(version: Version) {
    match version {
        Version::HTTP_09 => http09(),
        Version::HTTP_10 => http10(),
        Version::HTTP_11 => http11(),
        Version::HTTP_2 => http2(),
        v => panic!("unsupported http version {:?}", v) // unreachable pattern lint triggered
    }
}

[seanmonstar]

Well shoot, that is unexpected. It was definitely hoped that by being an opaque struct, new versions could be added without making a breaking change. Has rustc always been able to do this? Or is it recent?

[carllerche]

Yeah, this sucks... but at least there won't be a new HTTP version added in the near term!

[choubacha]

https://github.com/hyperium/http/blob/master/src/version.rs#L43

Pretty sure it can tell based on the limited number of versions in the enum (thus there is a limited number of possible Version structs). I've seen crates that use an __Reserved enum before that is private:

/// Represents a version of the HTTP spec.
#[derive(PartialEq, PartialOrd, Copy, Clone, Eq, Ord, Hash)]
pub struct Version(Http);

impl Version {
    /// `HTTP/0.9`
    pub const HTTP_09: Version = Version(Http::Http09);

    /// `HTTP/1.0`
    pub const HTTP_10: Version = Version(Http::Http10);

    /// `HTTP/1.1`
    pub const HTTP_11: Version = Version(Http::Http11);

    /// `HTTP/2.0`
    pub const HTTP_2: Version = Version(Http::H2);
}

#[derive(PartialEq, PartialOrd, Copy, Clone, Eq, Ord, Hash)]
enum Http {
    Http09,
    Http10,
    Http11,
    H2,
    // Left unpublished to allow for future versions of http to occur
    __Reserved,
}

This should force you to implement a _ or v since not all outputs are exposed. Of course, this still kinda requires that the versions exist in the crate since in order to actually match on them you'd need to upgrade the crate.

[seanmonstar]

Yes, the fix here is to add a variant. It would technically break anyone that has an exhaustive match on the version, so I've tossed it into the 0.2 milestone as a breaking change (shucks, the first one!).

[seanmonstar]

Thanks to #159, this is fixed in the unreleased 0.2.x branch!