Re: ContainerHierarchy again

On 4 Mar 2014, at 20:05, Sandro Hawke <sandro@w3.org> wrote:

> On 03/04/2014 01:16 PM, henry.story@bblfish.net wrote:
>> At the last meeting there was a resolution to move back to the previous positon 
>> on container hierarchies, the position called "the minimal consensual position"
>> in 
>> 
>>    https://www.w3.org/2012/ldp/wiki/ContainerHierarchy
>> 
>> But that wiki page shows how the minimal consensual position we have currently
>> + the decisions we have previously come to lead to hierarchy
>> 
>> IndirectContainer
>>    DirectContainer
>>       BasicContainer
>> 
>> In RDF subclass relations DO NOT mean dependence of one subclass
>> on another. You can easily deprecate classes without subclasses or
>> superclasses without this leading to legacy issues.
>> 
>> All the subclass relations mean is that you cannot have objects that are in one
>> class and not the superclass as shown by the picture
>> 
>>  https://www.w3.org/2012/ldp/wiki/ContainerHierarchy#LDP-BC_.3C:_LDP-DC_.3C:_LDP-IC_.3C:_LDPC
>> 
>> 
>> So I think it is misleading in fact to now show the classes as having no relation to each
>> other when we know they do.
>> 
>> Henry
> 
> Can we separate logic, pedagogy, and protocol?

Interesting question. :-)

> 
> What you're saying appears to be logically true, yes.   That was why we adopted the hierarchy for a while.
> 
> I'm ambivalent from a pedagogical perspective.  If we're going to include this hierarchy in the spec, I think we'd really need to include a Venn diagram with the actual features as points in the space, so that it's clear in what sense it's a class hierarchy.

yes, that makes sense. A Venn Diagram would be less likely to make people think in a OO manner.
That would be the second Venn diagram I drew up in the wiki page linked to above.

> From a protocol perspective, I think it's important that the hierarchy not matter.  In particular, clients who are looking for a DirectContainer must be told link rel=type DirectContainer, NOT BasicContainer.   Logically every BasicContainer is a DirectContainer but we've agreed not to require clients to do that kind of inference.  

That's a pragmatic issue. By specifying DirectContainer for a BasicContainer, a server would potentially loose some clients
that would be unwilling to look further. 

Protocols are in the space of pragmatics I believe. The above error would be a bit akin to someone saying 
"there is something outside", which while true may not have the same effect as saying "there is a friend outside". 
But that need not lead us to conclude that friends are not people or that people are not also things. 

>   So as long as we're not going back on that, the hierarchy has no impact on the protocol.

Agree, the 'hierarchy' does not have an impact on the protocol, but it can be a bit confusing if people look at a UML diagram
where the 3 classes are shown unrelated where they are in fact subsets of each other. This may lead someone to 
consider all the possibilities shown in the first Venn Diagram on the Wiki, where in fact only three of those possibilities exist. 

So +1 for adding a Venn Diagram from me :-)

> 
>        -- Sandro
> 

Social Web Architect
http://bblfish.net/