[processing] Pass a copy of the feature to processFeature, not

the original layer's feature

Otherwise python processFeature implementations may return a
modified shallow copy of this feature, which means that the
test for modifications fails, since we're comparing against
the same feature which was modified.

Fixes some python algorithms do not modify features when used
in-place mode.

python/plugins/processing/gui/AlgorithmExecutor.py

@@ -227,7 +227,10 @@ def execute_in_place_run(alg, active_layer, parameters, context=None, feedback=N
             field_idxs = range(len(active_layer.fields()))
             feature_iterator = active_layer.getFeatures(QgsFeatureRequest(active_layer.selectedFeatureIds())) if parameters['INPUT'].selectedFeaturesOnly else active_layer.getFeatures()
             for f in feature_iterator:
-                new_features = alg.processFeature(f, context, feedback)
+                # need a deep copy, because python processFeature implementations may return
+                # a shallow copy from processFeature
+                input_feature = QgsFeature(f)
+                new_features = alg.processFeature(input_feature, context, feedback)
                 new_features = make_features_compatible(new_features, active_layer)
                 if len(new_features) == 0:
                     active_layer.deleteFeature(f.id())