Re: Structured Headers: URI type (#782)

On 02.05.2019 09:25, Matthew Kerwin wrote:
> On Thu, 2 May 2019 at 15:23, Mark Nottingham <mnot@mnot.net> wrote:
>>
>> (Editor hat on)
>>
>> <https://github.com/httpwg/http-extensions/issues/782>
>>
>> PHK and I have discussed this, and I think we agree that this issue should be closed without any change to the specification.
>>
>> Any further discussion? We'd like to get this spec shipped.
>>
>> Thanks,
>>
>
> For what it's worth, I'm still recovering from my ordeal with file://
> so the thought of touching URLs again is unappealing.  The simplest
> thing to do (aside from "nothing") would be to use whatever already
> exists.  So if we were to define a URI type I would recommend using
> whatever "Link:" already uses.  Looking at RFC 8288, that's
> "URI-Reference" from 3986.

I would be ok with that, but I would also agree to just allow anything,
and to delegate error handling up one layer.

> However, I think Link is a bad example to build off because it is a
> "parameterised" thingy, but we've already restricted the set of
> parameterisable thingies in SH to just sh-tokens.  So we couldn't
> recreate Link in SH even if we wanted to, without even more work.  So
> why bother adding a new type for it?  And we don't need a structured
> type for Location, because that's not structured per se.
>
> So I'm for closing with no action.
> ...

That's actually a good point, but then there's the recent
<https://github.com/httpwg/http-extensions/issues/797>:

> in Signed Exchanges, @jyasskin asks about having Parameterised Lists whose parameterised identifiers are things other than Tokens.

Best regards, Julian