Minutes from the meeting of 4/16 Attending: Serge,Roy,Vasilis,Jason,Yue, Michael,Janet From Roy: Here's a quick summary of the OEM view application discussion. * Last week, we talked about using views to impose some regular structure over an OEM database. We essentially came up with two different models. One would use Lore as is, essentially feeding restructured data into a relational/oo system (or front end). Alternatively, in a model where Lore is tightly integrated into an OO or SQL3 system, we wouldn't have to perform extra work to take advantage of the structure. * Today, we discussed the possibility of adding "with" to the query language, to provide much of the functionality of views without necessarily requiring a new workspace or maintenance. In addition to the with clause as defined in the view language, Michael proposed virtual attributes (such as computing age based on birthday). The semantics of a query involving with is that objects not specified by the "with" clause are not reachable when traversing the objects returned by the query. Serge pointed out that adding presentation specifications to the query language is yet another possible aspect to study. * Views could be useful in Tsimmis, to "pull" some work out of the wrappers and/or construct a more intelligible view over a database integrated from many heterogeneous sources. * We could cook up a spiffy demo based on the current UI, where different users could visually specify their own view over a large Web-site based Lore database (such as ESPN). Each user could specify the subject areas he/she is interested in, and we could show how updates are propagated/transformed/filtered into the different views. I think this is very applicable in the current state of the web-- too much work is required to find the specific information someone is interested in. * This begins to look a lot like Whips, though Whips clearly concentrates on a) heterogeneous sources and b) the relational model. Perhaps the two can be combined. * Finally, Serge argued that a demo need not necessarily have maintenance to be interesting, pointing out that restructuring static data is still be very useful. We didn't get around to how to demonstrate this within an application. ----------- More problems to be solved: 1. Implementation: focusing on implementing materialized view with basic maintenance for now. Major difference from LORE: * WITH clause * multiple workspace 2. Come up with a scenario for a demo (ESPN?) 3. Extending the query language * adding virtual attributes? * classification * adding and hiding attributes 4. maintenance 5. query subsumption 6. query rewrite [later]