Re: [core] Adam Roach's Discuss on draft-ietf-core-links-json-07: (with DISCUSS and COMMENT)

Carsten and Adam,

> On 22 Apr 2017, at 09:32, Carsten Bormann <cabo@tzi.org> wrote:
> 
> Interesting.
> Julian Reschke reminds me that this is repaired in draft-nottingham-rfc5988bis-05.txt:
> 
> 3.  Link Serialisation in HTTP Headers
> 
>   The Link header field provides a means for serialising one or more
>   links into HTTP headers.
> 
>   The ABNF for the field value is:
> 
>     Link       = #link-value
>     link-value = "<" URI-Reference ">" *( OWS ";" OWS link-param )
>     link-param = token BWS "=" BWS ( token / quoted-string )
> 
>   Note that any "link-param" can be generated with values using either
>   the "token" or the "quoted-string" syntax, and therefore recipients
>   MUST be able to parse both forms.  Individual "link-param"s specify
>   their syntax in terms of the value after any necessary unquoting (as
>   per [RFC7230], Section 3.2.6).

I think this document should recognize this change and just reference draft-nottingham-rfc5988bis normatively. (I will be AD sponsoring it shortly anyway.) Adam, will this address this concern

> Unfortunately, this doesn’t retroactively fix RFC 6690, which still has restrictive ABNF for rt, if, title, anchor.  Hmm.

Well, we just have to special case them. The list of special attributes is fixed.
> 
> Grüße, Carsten
> 
> 
>> On Apr 22, 2017, at 10:03, Carsten Bormann <cabo@tzi.org> wrote:
>> 
>> Hi Adam,
>> 
>> thank you very much for this deep review.
>> This will allow us to improve the document a lot.
>> 
>> Let me pick one issue here that I think goes beyond the details of links-json.
>> 
>>> Let's say some future document defined a link relation ("smellslike")
>>> that included a parameter with ABNF syntax like:
>>> 
>>> "smellsys" "=" quoted-string
>> 
>> My knee-jerk reaction is “then don’t do that”.
>> 
>> But the more fundamental problem is that RFC 5988 actually defines a data model and its serialization without ever saying so.  Any reasonably defined link parameter will either have a syntax that is limited to ptoken or allow both ptoken and quoted-string serialization.
>> The fact that we then use ABNF to define the link parameter syntax obscures that property and allows the above deviation.  We already made the mistake for [anchor, title, rt, if] (see MUSTBEQUOTED in the reference implementation).  There is no need to make it again,  but the use of ABNF for what should be a data model definition does make it too easy to shoot ourselves in the foot.
>> 
>> Note that the above issue hits any implementation that wants to generate RFC 5988 syntax, not just one that converts links-json into RFC 5988 syntax.
>> 
>> So my takeaway summary is that links-json only works for reasonably defined link parameters (after grandfathering in anchor, title, rt, if) and probably should say so much more explicitly.  But I don’t know how to fix RFC 5988 about this issue.
>> 
>> Grüße, Carsten
>> 
>> _______________________________________________
>> core mailing list
>> core@ietf.org
>> https://www.ietf.org/mailman/listinfo/core
>> 
>> 
> 