Show Feature Count shows wrong numbers with rule-based classification when rules were copied/pasted via the contextmenu

[pathmapper]

Describe the bug

Show Feature Count shows wrong numbers with rule-based classification when rules were copied/pasted via the contextmenu.

How to Reproduce

Data used in the test cases of this ticket is from
WFS: https://geoservices.krzn.de/security-proxy/services/wfs_verb_grenzen
Layer: gis:gemeindegrenzen

1. Add the WFS layer mentioned above to the legend.
2. Add rule-based classification with copied/pasted rules like shown in Show Feature Count shows wrong numbers with rule-based classification when rules were copied/pasted via the contextmenu #40206 (comment) . The result will be the same as in this sample_layer_definition_file.zip.
3. Open contextmenu of the layer and click on Show Feature Count.
4. See error: Feature Count for Viersen, Krefeld and Wesel is wrong, e.g. click on "Test" in the "Edit Rule" window, compare with attribute table or visually count the features on the canvas.

QGIS and OS versions

QGIS version	3.16.0-Hannover	QGIS code revision	43b64b1
Compiled against Qt	5.11.3	Running against Qt	5.11.3
Compiled against GDAL/OGR	2.4.0	Running against GDAL/OGR	2.4.0
Compiled against GEOS	3.7.1-CAPI-1.11.1	Running against GEOS	3.7.1-CAPI-1.11.1 27a5e771
Compiled against SQLite	3.27.2	Running against SQLite	3.27.2
PostgreSQL Client Version	11.9 (Debian 11.9-0+deb10u1)	SpatiaLite Version	4.3.0a
QWT Version	6.1.4	QScintilla2 Version	2.10.4
Compiled against PROJ	5.2.0	Running against PROJ	Rel. 5.2.0, September 15th, 2018
OS Version	Debian GNU/Linux 10 (buster)
Active python plugins	ClipMultipleLayers; quick_map_services; scriptrunner3; LAStools; canvas_clipper; savedialog; loadthemall; clipper; kmltools; qgis-maptiler-plugin; MapsPrinter; Discovery; mmqgis; pluginbuilder3; vector_tiles_reader; layerGroupFilter; plugin_reloader; realcentroid; mask; MetaSearch; db_manager; processing

Additional context

The similiar issue #39318 was closed because it was not reproducible.

There's something wrong in general with Show feature count for rule-based classification: #27893, #27892

[gioman]

@pathmapper same as #37799 ?

[roya0045]

@gioman I'd say yes.

[pathmapper]

same as #37799 ?

@gioman I'd say no, because if I understood it correct #37799 is not about showing wrong numbers, instead it's about showing the total number of features even if a filter is applied. I'd say that #37799 is a duplicate of this feature request: #29458

The issue reported in this ticket here is about showing wrong numbers. The count number 23 shown is totally wrong and has nothing to do with the total number of features. 23 is the total number of features minus all features which are not "KREIS" = 'Kleve'.

Correct would be:

Kreis	feature count
Viersen	9
Krefeld	1
Wesel	13
Kleve	16
total	39

[roya0045]

Did you toggle the count when you had only two classes? This would indicate so.

[pathmapper]

Did you toggle the count when you had only two classes?

@roya0045 no, just reproduced the issue:

[roya0045]

What if you use the category symbology first to get the classes and then switch to rule-based? Or used the refine selected rules button instead of doing it all one by one?

[pathmapper]

What if you use the category symbology first to get the classes and then switch to rule-based?

In this case the numbers are correct:

Or used the refine selected rules button instead of doing it all one by one?

If you mean something like the following, it works also:

Here is another one using the Add Rule button:

---

So it looks like the bug here which leads to wrong feature count numbers is regarding the use of copy/paste of rules via the contextmenu:

[roya0045]

Ok I think this is enough to isolate the issue.

When counting features the rulekeys for feature is used

QGIS/src/core/qgsvectorlayerfeaturecounter.cpp

Line 75 in 213df8d

const QSet<QString> featureKeyList = mRenderer->legendKeysForFeature( f, renderContext );

In the rule based render the keys are generated at the start

QGIS/src/core/symbology/qgsrulebasedrenderer.cpp

Line 52 in 842e765

mRuleKey = QUuid::createUuid().toString();

But to maintain some usability those keys are not changed when cloning.

QGIS/src/core/symbology/qgsrulebasedrenderer.cpp

Line 1039 in 842e765

// normally with clone() the individual rules get new keys (UUID), but here we want to keep

@nyalldawson would it be a good idea to use the filter expression as a key? I know that this can be more complex with nested symbols as there can be duplicates in different branches. But my older PR has code to get a full comprehensive expression string for each 'node', this would solve that issue.

[beerockxs]

Using 3.22-1, I have this issue without having copy-pasted rules. I created a rule based symbology and refined it for two additional levels using the "Refine Selected Rules" button and adding further categories:

[pathmapper]

But to maintain some usability those keys are not changed when cloning.

Yep, if the style is exported as .qml, the first three rules are having the same key.

If you manually edit the .qml so that the keys are unique, the style works as expected.

[lp-dj]

I would like to add that I sometimes have the problem with single symbol styling as well: