W3C home > Mailing lists > Public > public-media-fragment@w3.org > July 2010

Re: Media Fragments URI parsing: pseudo algorithm code

From: Yves Lafon <ylafon@w3.org>
Date: Thu, 1 Jul 2010 05:10:29 -0400 (EDT)
To: Philip Jägenstedt <philipj@opera.com>
cc: Jack Jansen <Jack.Jansen@cwi.nl>, Silvia Pfeiffer <silviapfeiffer1@gmail.com>, Raphaël Troncy <raphael.troncy@eurecom.fr>, Media Fragment <public-media-fragment@w3.org>
Message-ID: <alpine.DEB.1.10.1007010506410.30546@wnl.j3.bet>
On Wed, 30 Jun 2010, Philip Jägenstedt wrote:

> On Wed, 30 Jun 2010 16:22:15 +0200, Yves Lafon <ylafon@w3.org> wrote:
>
>> On Wed, 30 Jun 2010, Philip Jägenstedt wrote:
>> 
>>> You cannot write a robust MF parser based on this grammar, because 
>>> t=1&foo=bar is not a valid production, meaning that any future extension 
>>> foo of MF will cause that parser to fail completely. Either the grammar 
>>> itself must be relaxed, or the parsing must be defined normatively and 
>>> handle some things which are not valid productions of the grammar.
>> 
>> What do you mean by "robust" ?
>
> I mean that it doesn't stop working completely for future additions to the 
> syntax, that it should degrade gracefully. If browsers shipped with a parser

Graceful degrdation should not be mistaken with "betraying intent", while 
graceful degradation is wonderful in many cases, you always have to be 
careful.
ex: http://www.example.com/football.movie?xywh=10,20,30,40&action=track 
may mean "highlight this part (a ball), and track it", a MF aware client 
will just crop the identified part. That's not graceful degradation, that 
is betraying intent (regardless of the fact that the extra action=track 
might be a bad design).
In CSS, properties with unknown values are ignored, to allow both graceful 
degradation (it doesn't impact _other_ properties) and forbid betraying 
intent.

> based on the ABNF of MF 1.0, then #t=1 will work and authors will use it. MF 
> 2.0 then adds a foo dimension, but authors won't be able to use #t=1&foo=bar 
> because the t=1 part would also be ignored until all browsers have upgraded 
> to a MF 2.0 parser. That's the opposite of graceful degradation.
>
>>>> With the current grammar, it is allowed only in track and id productions.
>>>> So it is perfectly compatible with the processing defined in rfc3986 and 
>>>> perfectly allows #track=A%20%26%20B&t=10
>>> 
>>> No disagreement that we need to define it, thankfully. The disagreement is 
>>> only where to decode percent-encoding.
>> 
>> RFC3986 gives the answer, after the URI components are parsed (and we 
>> define here how to split out in components).
>
> The disagreement here is only for which components to decode 
> percent-encoding, RFC3986 will not help us.
>
>>> <issue>
>>> 
>>> MF parsing must be defined normatively in the MF spec itself, meeting 
>>> these conditions:
>>> 
>>> 1. should handle all valid productions of the ABNF syntax correctly and, 
>>> where necessary, input which is not valid per the syntax.
>>> 
>>> 2. must be forward-compatible, so that future extensions to MF do not 
>>> break existing MF parsers. (Compare to how new HTML elements and 
>>> attributes or CSS properties degrade in implementations that don't 
>>> understand them.)
>> 
>> I completely disagree with this, as it may preclude other uses than 
>> mediafragment to use an "a=b" syntax.
>
> If you expect not only future revisions of MF but also completely unrelated 
> uses to co-exist in the same fragment component, then that's all the more 
> reason for parsers to not fail when encountering unknown name-value pairs. 
> That aside, since it is only the MIME registrations that have the authority 
> to say that MF applies, one can safely assume they won't do something crazy 
> like allowing conflicting syntaxes in the same component. Can you please give 
> a concrete example of a problem that might arise?
>
>>> 3. should match as closely as possible how query components on the form 
>>> a=1&b=2 are parsed by existing server-side software (e.g. ASP, PHP, JSP, 
>>> Perl CGI)
>>> 
>>> </issue>
>>> 
>>> An implementation that conforms exactly to the (current, non-normative) 
>>> ABNF fails condition 2 (e.g. t=1&foo=bar) and is not an option.
>> 
>> As I completely disagree with 2, strict ABNF makes perfect sense.
>> 
>
> Are there any existing experimental implementations that actually take this 
> approach? I'd love to experiment with them to show how fragile it is.
>
>

-- 
Baroula que barouleras, au tiéu toujou t'entourneras.

         ~~Yves
Received on Thursday, 1 July 2010 09:10:40 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Wednesday, 21 September 2011 12:13:39 GMT