James M Snell wrote:
> 
> Eric Scheid wrote:
> 
>> pub:control as a wrapper allows for order of control instructions to be
>> preserved. Most often that won't be needed, but consider an entry whose
>> content is an image POSTed from a mobile phone... order might be
>> important,
>> why rule it out?
>>  
>>
> I'm not sure I follow your mobile phone example but you're right, order
> *might* be important.  But can we honestly claim that it *might* be
> significant in 80% of the cases or are we talking about a few nebulous
> edge cases that fall somewhere in the 20% range?  Sure, let's not rule
> it out, but the Atom format does allow extensions to declare that order
> is significant within those extensions, so we would not be ruling it by
> not providing a pub:control container in the core.
> 
> Is order significance of control elements the only argument we have left
> in favor of pub:control?

We went over metadata v directives a while back and agreed (I thought)
that there was some overlap but there's value in preserving a
distinction. These are qualitatively distinct data we're talking about
and I don't think that fact that we can put them the same place
structurally in the entry obscure that. Again as a datapoint, I'll
mention out that XML itself makes the distinction between markup and
processing directives.

>From a 'modelling' perspective I guess it makes no difference where
control data lives. It's all markup.

>From a programming perspective it does. It's cleaner to be able to pull
out the directives in one shot, distinct from the other metadata because
fom a programming perspective you're likely have different code for
dealing with entry metadata which-are-metadata and entry-medata
which-are-PIs. It would suck if we made pipelined XML processing and
node iteration awkward. Hence I find this XPath compelling:

   atom:entry/pub:control

as it lowers code entropy and general spaghettiness!


Incidentally there's also the question of layering between data and
control in the protocol. If we weren't tied into a mindset of "single
XML document", I'd be up for sending control data alongside the entry
instead of littering the entry with protocol stuff. I guess this would
be a MIMEish position.

cheers
Bill

Reply via email to