[Nepomuk] heavy duty query visualization

[Sebastian Trüg]

Hi Marco,

On 10/14/2011 10:13 PM, Marco Martin wrote:
> in Plasma Active one things like the activity screen that shows all the 
> nepomuk resources, the ui to search and add a new one, the image viewer (that 
> lists all objects of type image) use the plasma metadata dataengine.
> this one is basically just a wrapper around Nepomuk QueryClient.
> due to the architecture of plasma dataengines, this is not so much efficient, 
> since a quite too much of data copy around is involved.
> 
> in the next releases we want to do more fancy stuff, like a central kick ass 
> document browser that lets browse stuff by basically anything, type, date, 
> rating,tags,activities, whatever and combining all of this.

nice.

> what i'm trying to do now, is a QAbstractItemModel (working prototype in the 
> plasma-mobile repo, mart/nepomukmodel branch) that is done expressively to 
> show resources from a query, to be instantiated from QML, so some questions 
> surface, related both to performance issues and nepomuk usage.
> 
> * lazy loading: if i do a query with really a lot of restults, would be 
> possible to limit it like a relational database, to know how many results 
> there are, asking for the first 10, then from 11 to 20 and so on? (if a query 
> has something like 2000 results that would be hady;)

There are Query::setLimit and Query::setOffset which can be used here.

> * sorting: is it possible to directly sort the results based on one of the 
> properties? in this case i wouldn't have to load the whole model in memory 
> from the beginning

Sorting is quite complex since you can sort by pretty much anything. In
the query API this is done by setting sort weights on the ComparisonTerm
instances.

> * property watching: if i understood correctly to dynamically update property 
> values i have to use Nepomuk::ResourceWatcher that stays 
> nepomukdatamanagement, still semi private in kde-runtime, right?

Right. I will get to that one after all the bug fixing.

> * query construction: since this will have to be used from QML/JavaScript, 
> unfortunately the c++ query api can't be used directly, and full javascript 
> bindings to it doesn't sound like fun ;) as isn't brutally exposing sparql to 
> qml. What i'm thinking to do here is:
>  * pass queries by string, so with the more limited desktopquery language.

Then this would have to be extended to include all kinds of things like
sorting and limits and so on. That seems rather pointless.
Don't we have some automatic binding creation like we have for python?

>  * expose to js some functions to limit the results, like limitByType(), 
> limitByActivity() limitByTag() etc. this will with the c++ api add the needed 
> parameters to a query precedently created from a string

Impressive how ugly things need to become just so this js stuff works. :P

> * what queries are possible? how? I think many of the queries we will need 
> will be to build things tagcloud-ish, so besides a proper tagcloud that has 
> been possible to do in nepomuk since a while, other needed things could be 
> series of result like:
>   * resource type, number of occurrencies
>   * rating, number of occurrencies
>   * date interval (like month) number
> and so on. is this easily feasible? (this will probably go in the dataengine, 
> not in the model)

Do you mean feasible in C++? Then the answer if of course "yes". If you
say feasible in the desktop query language, then "no".

Cheers,
Sebastian

> * for the write part, like assigning a score, a tag, connect to activity: this 
> will be done exactly like now with a plasma service, in this case created not 
> by the dataengine but by the model
> 
> 
> comments? ideas? opinions?
> 
> Cheers,
> Marco Martin
> _______________________________________________
> Nepomuk mailing list
> Nepomuk at kde.org
> https://mail.kde.org/mailman/listinfo/nepomuk
>