[Xesam] Sorting API to be more flexible

[Mikkel Kamstrup Erlandsen]

2008/6/25 Evgeny Egorochkin <phreedom.stdin at gmail.com>:
> В сообщении от Wednesday 25 June 2008 10:36:41 Mikkel Kamstrup Erlandsen
> написал(а):
>> 2008/6/24 Evgeny Egorochkin <phreedom.stdin at gmail.com>:
>> > В сообщении от Tuesday 24 June 2008 16:05:54 Mikkel Kamstrup Erlandsen
>> >
>> > написал(а):
>> >> 2008/6/24 Mikkel Kamstrup Erlandsen <mikkel.kamstrup at gmail.com>:
>> >> > 2008/6/13 Mikkel Kamstrup Erlandsen <mikkel.kamstrup at gmail.com>:
>> >> >> 2008/6/13 Jos van den Oever <jvdoever at gmail.com>:
>> >> >>> 2008/6/13 Urho Konttori <urho.konttori at nokia.com>:
>> >> >>> > We have been evaluating the Xesam for multiple uses at Nokia and
>> >> >>> > one of them is media player use. In many cases, where all the
>> >> >>> > songs need to be listed, they need to be listed with three sorting
>> >> >>> > criteria: Artist, Album, Track#. Xesam provides only two sorting
>> >> >>> > criteria. Now, you might argue that the application should do the
>> >> >>> > tertiary sorting, but then if we say that, then why two sorting
>> >> >>> > criteria, or one?
>> >> >>> >
>> >> >>> > I'm quite sure this is not the only use case where tertiary
>> >> >>> > sorting (or even more) would be beneficial.
>> >> >>> >
>> >> >>> > Also, the API looks a bit glued when you have primary and
>> >> >>> > secondary sorting as the session properties. Why not instead
>> >> >>> > change it to an array of sort criteria? So, as to call it just
>> >> >>> > sort.fields?
>> >> >>> >
>> >> >>> > It would be very good if this sort of change could still be done
>> >> >>> > to the xesam 1.0 spec, either next to the current primary and
>> >> >>> > secondary sorting, or better yet, as the only way to set the sort
>> >> >>> > order. It would be much cleaner way to do it.
>> >> >>> >
>> >> >>> > Anyway, I'm looking forward to comments on the subject.
>> >> >>>
>> >> >>> I agree with your suggestion.
>> >> >>> We would also need a way for the server to specify what type of
>> >> >>> sorting it supports.
>> >> >>> For example a readonly property: sortableFields and a property that
>> >> >>> says how many sorting fields can be used.
>> >> >>> Some servers may not support sorting at all (grep) or allow only
>> >> >>> sorting on one field.
>> >> >>
>> >> >> I am also for this, although I think it is quite a big feature
>> >> >> compared to our current freeze level.
>> >> >>
>> >> >> It is a good point you raise Jos. I do however think that it might be
>> >> >> acceptable to require the servers to be able to sort the hit data.
>> >> >> Otherwise one might have to pull massive amounts of hits over the
>> >> >> wire only to get those with the highest atime at the top.
>> >> >
>> >> > Consider this thread bumped. It is our list of blockers I recently
>> >> > posted about. We need a clear decision.
>> >> >
>> >> > I already made my point above.
>> >>
>> >> There has been some discussion on this on IRC which have led me to
>> >> believe that the following solution would be optimal for all:
>> >>
>> >>  * Scrap sort.primary and sort.secondary.
>> >>
>> >>  * Introduce three new session properties:
>> >>
>> >>   - sort.fields a read/write list of fields to sort after, in that
>> >> order. Default value ['xesam:relevancyRating']. The server should
>> >> truncate this list to to the size specified in vendor.maxSortFields if
>> >> too many sort fields are requested
>> >>
>> >>   - vendor.sort.maxfields a readonly uint. Specifies how many fields
>> >> in sort.fields will be taken into account, as mentioned, sort.fields
>> >> should be truncated at this length. This property must be 1 at
>> >> minimum. If this value is 0 any number of fields can be used. Default
>> >> value is undefined.
>> >>
>> >>  - vendor.sort.fields a readonly list of strings naming all fields the
>> >> server can sort after. If this list is empty all known fields can be
>> >> used for sorting. Default value is [].
>> >
>> > This parameter is superfluous. Suppose a client app determines the server
>> > doesn't support sorting by required field. Either app has to do sorting
>> > and doesn't really care about server sorting capabilities. If app doesn't
>> > implement sorting, then the server can ignore fields it can't sort on and
>> > that's all that can be done.
>> >
>> > Maybe there's another use case where app needs to know if server did sort
>> > or not... I can't think of one atm.
>>
>> Consider this: A paged client showing 20 hits/page. If xesam:title is
>> in vendor.sort.fields we make the list headers clickable so that the
>> user can sort hits alphabetically by title. Combining this case with a
>> result set of 50.000 hits should clear out the use case I think, I
>> would rather not pull in 50.000 hits and sort them if I can avoid it
>> (consider a grep-like or web-based server - although they may have
>> problems in any case). Or maybe there is a clever way around this that
>> I can not see..?
>
> Clickable column argument does sound sane.
>
> More likely than not we're going to see backends that don't sort at all like
> grep or db-driven backends that can sort all you want. The only possible gray
> area here are web services wrappers written by community... The services are
> very much likely running on top of a plain filesystem with no sorting or a
> db, but will they expose sufficient sorting capabilities?

Well. Lucene is also in a semi-gray area since it requires the fields
to be indexed to sort on them... I don't know about Xapian...

Cheers,
Mikkel