Hi,
we have had this ticket open for some time about how numberMatched is not exactly working as
it should in WFS 2.0:
http://jira.codehaus.org/browse/GEOS-6187
Looking at it I’ve reviewed the standards and the implementation to find a good overall solution.
The current code will return -1 for numberMatched under some conditions, see my other thread
about the need to actually return “unkonwn” if the count is not known.
I’ve prepared a pull request that should fix that issue for most cases in which we return -1 now,
but of course, more work will need to be done to fix the case where the store actually does not know:
https://github.com/geoserver/geoserver/pull/502
Basically the patch uses the approach Justin suggested in a comment, to do the total count later
only if necessary, but rolling a list of CountExecutor(s) that are collected as we fetch the feature
collections, and defer the decision on whether we really need to do a total count at the end.
If “count” (featuresReturned) is less than maxFeatures, then we don’t have to compute anything,
and regardless we try to keep around the result of the count for the first feature type,
to avoid that extra call.
If instead we realize we’re returning less features than matched, or we have a base offset,
then we will have to do the total count running the count executors.
This should give us good performance in the common case, and correctness in the general one.
Implementing it I’ve also noticed that we try to avoid doing the count itself for WFS 1.0, assuming
there is a single feature type requested, and looking at the WFS 1.1 schemas, I’ve noticed that
we could extend the optimization to WFS 1.1 too, since numberOfFeatures is optional:
<xsd:attribute name=“numberOfFeatures”
type=“xsd:nonNegativeInteger”
use=“optional”/>
Shall we do that? One less query 
The same cannot apply to WFS 2.0, where both numberMathced and numberReturned
and mandatory, and numberReturned is a “nonNegativeInteger”.
The ticket also talks about the hits behavior for wfs 1.0 and 1.1, and suggests to do a total
count regardless (basically, behaving as if the “ignoreMaxFeaturesHits” flag is always up).
Reading the specs I don’t agree and believe we should maintain our current behavior,
that allows the admin to configure how things should behave.
Here are the salient bits out of the wfs 1.1 spec, when it comes to describing how
WFS 1.
–
== Our support, Your Success! Visit http://opensdi.geo-solutions.it for more information ==
Ing. Andrea Aime
@geowolf
Technical Lead
GeoSolutions S.A.S.
Via Poggio alle Viti 1187
55054 Massarosa (LU)
Italy
phone: +39 0584 962313
fax: +39 0584 1660272
mob: +39 339 8844549
http://www.geo-solutions.it
http://twitter.com/geosolutions_it