[Qgis-developer] Fwd: [Qgis-user] QGIS problem with GeoPackage database views

[Even Rouault]

Le lundi 27 juin 2016 02:47:09, Marcus Blake a écrit :
> @Richard: thanks for redirecting this question in to the correct people…and
> apologises for the size of the file…lack of specification on my part to a
> over enthusiastic team member ;-). I’ll reduce the size and put in a issue
> report.
> 
> I’m in total agreement with you on the tremendous potential of GeoPackage
> to support the sharing of all types of data (statistical, geospatial,
> metadata, imagery…). Certainly the statistical community (the UNGGIM), is
> grappling with the issue of integrating statistical and geospatial data
> models and I think GeoPackage can provide one important part of the
> solution.
> 
> As well as publishing 2016 Census data...One of my aspiration is to use
> GeoPackage to publish the full data model that the ABS uses to internally
> store our geographic classification. This would provide a point of truth
> for all users…and they can tell us where we have gone wrong ;-)
> 
> But to do both of these things well, views are very valuable and would
> prevent a lot of data duplication.
> 
> @Even Thanks for the quick reply…and all your important support of GDAL.
> 
> One of my team (more knowledgable than I) thought that it was an ID
> issue…so it good to have that confirmed. But it doesn’t look like there is
> a easy solution.
> 
> It interesting that the despite the lack of PKID (and therefore a correct
> FID) both the classification and symbolisation for a choropleth maps and
> the Identify Features Tool work as expected….thoughts?

I didn't check in the code, but likely rendering "just" iterates over features 
and doesn't care about their FID. And for identification, it is a spatial 
query, which ends up being an iteration over all features too (since, with 
views, spatial indexes are not - at least currently - used. Could probably be 
improved too since the Spatialite driver can use the underlying table spatial 
index for Spatialite views. But Spatialite has an explicit declaration of 
which pkid to use and the geometry column of the base table in its 
views_geometry_columns table).
On the contrary the attribute table works with a feature cache based on FID.

Even

-- 
Spatialys - Geospatial professional services
http://www.spatialys.com