Re: [LDP Paging] Comparison to other techniques of pagination

> - The 2NN status might wreak less havoc with caches than alternatives, 
but I'm not confident it will be a lot less

I think that's an interesting point.  To date we've focused on the focused 
on the effects on the "application client" rather than intermediaries. 
I'll simply note a few things:

1: Paging is not dependent on 2NN, which is one reason that having it At 
Risk is "ok" with the working group.  You could completely ignore 2NN, use 
303 (paying the extra round-trip latency penalty), and every existing 
intermediary that behaved correctly before and that implements HTTP 
correctly (conforms) will still behave correctly, would be the working 
group's assertion.  Counter-examples heartily welcomed.

2: Informally I think people have higher expectations for the correctness 
of cache implementations, and they expect that "anything that works for a 
client works for a cache [playing the role of HTTP client in an end to end 
flow]".  Getting out my RFC-monger's microscope, [1] appears to cover this 
case when it says "a recipient MUST NOT cache a response with an 
unrecognized status code."  FWIW, the equivalent statement is in the 
superseded RFC 2616 (section 6.1) as well.  Any caches violating something 
in the RFCs for 15+ years would not receive much sympathy from me, 

- I understand that RDF stores needn't provide predictable order for 
triples, but don't personally follow how a such a resource's triples could 
then be paged in any useful way, which seems to bring us back to content 

Funny you should "ask", I had the same problem until recently (within oh 
the last 6 months) until I convinced Sandro to provide an example if it's 
so easy - and he did.  I captured his example in [2].  The basic pattern 
is that *the server* has some algorithm for slicing the graph into pieces, 
and the breadcrumb for the boundary(ies) gets encoded in the server's page 
URLs ... the client does not, should not be, is not, aware of it.  It just 
needs to be repeatable and to follow the guarantee of 6.2.7 etc.  If you 
hark back to the assets/liabilities/advisors example in LDP, an equally 
viable strategy that a server could use is: page 1 = all the "net worth 
but not container" triples, 2 = assets, 3= liabilities, 4=advisors. 
Clearly that's artificial, but it's an illustration that once you see the 
pattern the server can probably find assumption it already relies on to 
make things work efficiently enough. 
Just as clearly, that does not suffice in cases where the client wishes to 
impose some ordering on the triples in the response - classic example, 
returning the list of a container's members sorted in some 
client-controlled manner; LDP/LDP Paging do not provide a standard way for 
the client to specify its order, so that has to be layered on top via 
extensions to LDP.  Once an ordering known to the client can be imposed on 
the things, other interesting things happen - some of your comments on LDP 
Paging Ben were poking at exactly that aspect I think.


Best Regards, John

Voice US 845-435-9470  BluePages
Cloud and Smarter Infrastructure OSLC Lead

Received on Wednesday, 17 September 2014 18:55:42 UTC