Subject: RE: Review of thoughts on Ad Hoc Queries
Message text written by "Nieman, Scott" > Should we have a straw vote on the OASIS 12/20 query approach vs. XPATH vs. OQL? If so, it needs to be discussed, detailed with mappings to the RIM, then voted upon based on analysis. these mappings could make the RS document more complicated. e.g., Syntax Mapping to RIM OQL Direct XPATH Requires creating virtual document views of RIM OASIS ???? It took us one full day to do this for XPATH, will it take another f2f for the OASIS approach???? Regards, Scott< >>>>>>>>>>>>>>>>>>> Scott, Thanks for the assessment here. I think we've so far been missing something vital here - what Farrukh calls a 'breakthrough' or two. It's tough getting everything down on email - or even articulating this stuff clearly on calls or in meetings sometimes. Sometimes it takes a while for everyone to understand all the nuances and facets. Foreinstance I see the OASIS/NIST approach as basically a container in XML. Nothing more. It serves to deliver a simple set of locator terms. It does not mandate any particular query syntax. In following this theme in the proposal I posted, I allowed for four different locator terms. a) Simple XML tag + operator + value b) XPath locator (path/node expression only) + operator + value c) Specific focused queries on GUID, UID, URN, etc. d) ANY = HTML style - I don't care about context - show me all hits. e) ability to specify returned content mode. The difference is that this avoids the complex XPath extensions in favour of just the nodes and paths piece. Why would we do all this? What I see is that d) is the defacto standard for the Web today. By doing a) and b) we are giving people a better pair of tweazers to pick the right piece with. However we are avoiding the complexity of %like% (can I say I do not see anything to like about %like% for our business use?) Then c) is the obvious focused queries that we can clearly see are needed. What c) overcomes is the need to write complex joins and other cascading queries - because we can go straight to the right pieces. As Farrukh noted this works because registry is not a collection of random content - but content with classification with GUID and UID associations - and exploiting this directly allows us to significantly lower the bar in terms of complex generalized searching algorithms. Simply put - we do not need OQL at this time to get us a functional registry. Most especially when - by Farrukh's own criteria - the need is to be able to implement registry with even simple XML flat files a XML parser and sub-directories at the bottom most level. Then e) is a vital piece. OQL was never designed to work with returning pieces of XML content. We saw this shortcoming from the PoC in Tokyo - returning URL's to the actual content instead of the content itself. And then there are slices into the content within nodes - one of the key points of tree heirarchies and XML itself. Providing this is a mandatory need - and I believe part of what CC is asking for in their analysis to be able to group XML content returned. OK - that said - do we need to take two days here and some of us work on a short few pager draft that spells this out in more detail? Including a set of access models based on the RIM we feel we need (classification -> content locations, and content <-> content associations, et al). I'm still not clear on that part especially. And this seems to be the heart of the matter. What is the level of complexity we need and why? Maybe we could break this down. We could do that bit first, and then look at the three approaches, OQL, XPath, and containered focused querying (CFQ). Notice that CFQ extends the simple OASIS model to fit the ebXML RIM and requirements. My sense is we are much closer than we were five days ago, and that a few more days are worth spending at this point. Also five days ago we were struggling for clear examples in XML of content. I'm not sure we have that solved yet. Have we taken say the CPP for example, added some classification structure for them, added GUID and UID references to some sample company registrations XML content, and then run some business case queries - \industry = 'plumbing' AND \city='Cleveland' thru to sanity check it all? Let's clearly understand the alternatives here - I'm not sure reading all that's been presented thus far that I yet do. Thanks, DW.
Powered by eList eXpress LLC