Situation with a view without unique (and indexed) key is bad database model, or bad query. Using window function is unefficient, but i can't find another way to make unique column on query with duplicities in records. In query without duplicities is possible to use oid (but this needs tables with oids), or ctid (but ctid is slow).
Sorting in row_number can be defined in (ORDER BY ...). But is unefficient (as you say). But works well for display smaller parts of data (thousands of features). Materialized view can be indexed. But this is about use case and it isnt a "pretty solution". Je; On Tue, Sep 01, 2015 at 11:36:53AM -0400, James Keener wrote: > Please stop saying this. It's fine for certain situations, but it is not a > permanently unique identifier for a row. It may change when the underlying > table is altered. Sure, it's unique if you read the results and keep them in > memory and never talk about it again, but qgis does that on its own already. > > Moreover querying on it is extremely inefficient. Querying against it forces a > sequence scan on the table. Depending on the table and version, it could also > try to materialize the view fully in memory before doing the scan. > > Jim > > On September 1, 2015 11:31:32 AM EDT, "Leknín Řepánek" > <godzilalal...@gmail.com> wrote: > > Workaround about primary key in view, i had use sever times is using > window function "row_number() over()". It works in views and in query > from database manager. > > Je; > > On Tue, Sep 01, 2015 at 04:09:49PM +0200, Sandro Santilli wrote: > On Tue, Sep 01, 2015 at 02:37:22PM +0200, Jürgen E. Fischer wrote: > Hi Sandro, > > On Tue, 01. Sep 2015 at 13:48:33 +0200, Sandro Santilli wrote: > I agree with Luca this should have been better not > backported to 2.8.3. > Only proper bugs should be backported, and this was a > (debatable) > GUI enhancement, as far as I can > tell. > > We intend to only backport fixes and not bugs. ;) > > You were always supposed to select the key column - preselecting > the first > column was the bug (also debatable). And #11317 is a ticket > that demonstrates > there were unaware users. > > Reading #11317 I looks to me that the reported bug was about > "Add PostGIS Layer" not giving the user full detail of why a > layer could not be loaded: "is an invalid layer - not loaded". > In this I agree with Aren here: > http://hub.qgis.org/issues/11317#note-6 > > That the first column often happens to be the primary key and > and the combobox > is not lexically sorted is somewhat pure luck - and unless you > avoid having the > key verified (using "use estimated metadata"), > keeping a wrongly select > column will make the layer to insert invalid. > > I agree that the reported regression was based on the false > expectation > that QGIS would pick a primary key automatically, but in the > (unlikely?) case a user was aware of that and properly coded the view > to ensure primary key was first (or only) numeric the change was > indeed a degradation of the experience. > > But I agree that the tooltip that you get on disabled lines (not > only for the > key selection, but also geometry type and srid) might not be > visible enough > (but that IMHO would be just a GUI enhancement). > > There should maybe be another rule about LTS backports being: > debatable fixes/enhancement need to be debated more on list ? > > --strk; > > ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ > > Qgis-user mailing > list > Qgis-user@lists.osgeo.org > http://lists.osgeo.org/mailman/listinfo/qgis-user > > ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ > > Qgis-user mailing list > Qgis-user@lists.osgeo.org > http://lists.osgeo.org/mailman/listinfo/qgis-user > > > -- > Sent from my Android device with K-9 Mail. Please excuse my brevity. _______________________________________________ Qgis-user mailing list Qgis-user@lists.osgeo.org http://lists.osgeo.org/mailman/listinfo/qgis-user
