Re: port #?

[Mark Nottingham <[e-mail]>]

Fine by me, as long as people don't (yet) read this as precluding using another port if we have external, non-URI information (e.g, a DNS record).

Cheers,


On 08/06/2013, at 5:30 AM, Martin Thomson <[e-mail]> wrote:

> Everything that Ted says, plus I think that the suggested text isn't
> quite the right place.  We talk about using the same "http:" and
> "https:" schemes in Section 2.  It would be relatively easy to add
> "...and ports" to the following statement:
> 
> OLD:
>   HTTP/2.0 uses the same "http:" and "https:" URI schemes used by HTTP/1.1.
> ADD:
>   HTTP/2.0 also shares the same default port numbers: 80 for "http:"
> URIs and 443 for "https:" URIs.
> 
> That would address option 5, remove any ambiguity, etc...
> 
> On 7 June 2013 13:17, Ted Hardie <[e-mail]> wrote:
>> Hi Eliot,
>> 
>> Some comments in-line.
>> 
>> 
>> On Fri, Jun 7, 2013 at 1:02 AM, Eliot Lear <[e-mail]> wrote:
>>> 
>>> 
>>> Hi everyone,
>>> 
>>> I note that we still haven't cleaned up the connection model
>>> sufficiently.  When someone implements a specification they need to know
>>> at least the port number to connect to. This is the document that has to
>>> specify at least at a bare minimum how that happens.  This can be
>>> handled in at least one of four ways:
>>> 
>>> 1.  We refer to RFC-2616 normatively.  This implies that we will not
>>> obsolete 2616 at this time.  If we do so later we would need to pull the
>>> HTTP URI definition out and update the IANA definition.
>> 
>> 
>> Other httpbis documents obsolete 2616, so we should refer to those, rather
>> than 2616.
>> 
>>> 
>>> 2.  We pull the HTTP URI definition out and produce a small document for
>>> it separately and refer to that, updating RFC-2616.
>>> 
>>> 3.  We include the URI definition in the HTTP2 draft.
>> 
>> 
>> If it needs to be re-iterated, I think having the reiteration within the
>> HTTP2 draft is fine.  But simply referring to whatever RFC
>> draft-ietf-httpbis-p1-messaging-13 becomes seems simpler.  That reinforces
>> the idea that HTTP2 and HTTP share the same URI synatx.
>> 
>> 
>>> 
>>> 4.  We abstract the connection model entirely from the document.
>>> 5.  We specify that unless specified within a URI, the default protocol
>>> is TCP and the default port is 80.
>>> 
>>> This all came to light because of interest to do some work with HTTP2
>>> using something other than TCP.  Thus, one might thing that [4] is the
>>> appropriate thing to do, but my experience with BEEP is that it lends
>>> itself to an ugly set of documents and violates the KISS principle.  To
>>> that end, I recommend the text in [5] be added for now, and that as
>>> HTTP2 matures we consider [2] later.
>>> 
>> So, I think saying that new transports may mint new URI schemes
>> (http.newfangled) is safe enough; they may.  But I'm not sure whether that
>> adds much value.  What's the harm in simply referring to
>> draft-ietf-httpbis-p1-messaging for the URI syntax and leaving it at that
>> for the moment?
>> 
>> regards,
>> 
>> Ted
>> 
>> 
>>> 
>>> Specifically, OLD:
>>> 
>>>   The HTTP/2.0 session runs atop TCP ([RFC0793]).  The client is the
>>>   TCP connection initiator.
>>> 
>>> NEW:
>>> 
>>>   Unless otherwise specified within a URI, an HTTP/2.0 session runs
>>>   atop TCP ([RFC0793]) and a client initiates a server on port 80.
>>> 
>>> Eliot
>>> 
>> 
> 

--
Mark Nottingham   http://www.mnot.net/