[v0.3] Specify what is meant by "not syntactically valid"

[wingo]

A number of the interfaces that turn strings into typed values, e.g. fields.from-list, fields.set, request.set-method, and so on can return a failure result if the value being set is not "syntactically valid". However, the specification does not define what this means. Suggest adding wording such as, "Syntactically valid methods are defined in RFC 9110§9.1".

Also, is it an error if a WASI HTTP implementation does not impose syntactic constraints on header/method/etc values?

[pchickey]

Yes we should amend the spec to clarify in that way, and yes the implementation should not be spec compliant if it fails to impose those constraints (so word it as a MUST).

[lukewagner]

Agreed this should be added. IIUC, the relevant grammatical rules are field-name (RFC 9110§5.1) and field-value (RFC 9110§5.5) and, thankfully, these grammars are actually real world Web compatible.

[wingo]

For path-and-query, what is the syntax for the path part? Is it path-abempty as mentioned in RFC9110 §4.2.1? That would require that the path start with /, which doesn't seem to be what Wasmtime currently implements.

[wingo]

Should we check that percent-encoding is valid? I would guess not, but we could require that the user do so.

[wingo]

For path-and-query, #780 is a related issue.

[wingo]

For path-and-query, do we forbid fragments? Wasmtime currently accepts # in a path-and-query.

[pchickey]

Not sure if we should forbid fragments or if we permit them to be normalized away. What does whatwg fetch do with a fragment?

[lann]

What does whatwg fetch do with a fragment?

Fragments are permitted and removed for requests.