Uri::from_parts to return SlashInPathMissing

[lukaspiatkowski]

When running this code:

let mut parts = Parts::default();
parts.scheme = Some(Scheme::HTTP);
parts.authority = Some(Authority::from_str("www.example.com").unwrap());
parts.path_and_query = Some(PathAndQuery::from_str("my_path").unwrap());

let uri = Uri::from_parts(parts).unwrap();
println!("{}", uri);

the intention would most likely be to print "http://www.example.com/my_path", but in reality it is "http://www.example.commy_path" (notice the missing slash).

Rust playground here: https://play.rust-lang.org/?gist=903f0d760c9b94b358a21540301c2a02&version=stable&mode=debug&edition=2015

This can lead into annoying bugs and furthermore is against the RFC3986 sections 3 and 3.3 where it states that:

URI         = scheme ":" hier-part [ "?" query ] [ "#" fragment ]
hier-part   = "//" authority path-abempty
              / path-absolute
              / path-rootless
              / path-empty
path-abempty  = *( "/" segment )

[seanmonstar]

Oh my, yea, that's definitely wrong! Thanks for filing this. I wonder, should perhaps PathAndQuery::from_str("my_path") actually be the error? A path should always start with a slash, right?

[lukaspiatkowski]

According to the section 3 of the RFC mentioned above:

   The scheme and path components are required, though the path may be
   empty (no characters).  When authority is present, the path must
   either be empty or begin with a slash ("/") character.  When
   authority is not present, the path cannot begin with two slash
   characters ("//").  These restrictions result in five different ABNF
   rules for a path (Section 3.3), only one of which will match any
   given URI reference.
   The following are two example URIs and their component parts:

         foo://example.com:8042/over/there?name=ferret#nose
         \_/   \______________/\_________/ \_________/ \__/
          |           |            |            |        |
       scheme     authority       path        query   fragment
          |   _____________________|__
         / \ /                        \
         urn:example:animal:ferret:nose

the "urn:example:animal:ferret:nose" is a valid URI with schema "urn" and path "example:animal:ferret:nose" with no leading "/".

Taking into account that this is URI implementation for http crate one might argue that it is irrelevant to support generic URI, but then if we want to support relative references as defined in section 4.2. of the aforementioned RFC then

relative-part = "//" authority path-abempty
              / path-absolute   ; begins with "/" but not "//"
              / path-noscheme   ; begins with a non-colon segment
              / path-empty      ; zero characters

allows creation of such URIs that can be used in this kind of html:

<a href="sub_page">Relative link</a>

which kind of makes sense to support by a crate whose name is "http".

OTOH, neither hyper client nor server would have any use for a URI whose schema is not HTTP or HTTPS and if schema is provided then path can be either empty or has to have a leading "/".

[lukaspiatkowski]

tl;dr What I am saying is that it is up to your ambition whether you want to support generic URI or to be more hyper specific in which case a lot of code from https://github.com/hyperium/http/blob/master/src/uri/scheme.rs and this lines https://github.com/hyperium/http/blob/master/src/uri/mod.rs#L189-L193 can go away (unless I am missing something and there actually is a use case for relative references and generic URI in hyper)

[seanmonstar]

I'm not decided which makes more sense... @carllerche?

[carllerche]

I believe the goal of this Uri type is only to support cases from HTTP 1.x and 2.0. For 1.x, this would be the request line and for 2.0, it would be representing the related pseudo headers.

I guess, the question is what is valid for HTTP 1.x, because 2.0 is pretty strict.

[carllerche]

In this case, at the very least, Uri::from_path should validate that if there is an authority, path is empty or starts with /. That seems like a good next step.