Using Vector Layers

This section summarizes various actions that can be done with vector layers.

Iterating over Vector Layer

Iterating over the features in a vector layer is one of the most common tasks. Below is an example of the simple basic code to perform this task and showing some information about each feature. the layer variable is assumed to have a QgsVectorLayer object

iter = layer.getFeatures()
for feature in iter:
    # retrieve every feature with its geometry and attributes
    # fetch geometry
    geom = feature.geometry()
    print "Feature ID %d: " % feature.id()

    # show some information about the feature
    if geom.type() == QGis.Point:
      x = geom.asPoint()
      print "Point: " + str(x)
    elif geom.type() == QGis.Line:
      x = geom.asPolyline()
      print "Line: %d points" % len(x)
    elif geom.type() == QGis.Polygon:
      x = geom.asPolygon()
      numPts = 0
      for ring in x:
        numPts += len(ring)
      print "Polygon: %d rings with %d points" % (len(x), numPts)
    else:
      print "Unknown"

    # fetch attributes
    attrs = feature.attributes()

    # attrs is a list. It contains all the attribute values of this feature
    print attrs

Attributes can be referred by index.

idx = layer.fieldNameIndex('name')
print feature.attributes()[idx]

Iterating over selected features

Convenience methods.

For the above cases, and in case you need to consider selection in a vector layer in case it exist, you can use the features() method from the built-in Processing plugin, as follows:

import processing
features = processing.features(layer)
for feature in features:
    # do whatever you need with the feature

This will iterate over all the features in the layer, in case there is no selection, or over the selected features otherwise.

if you only need selected features, you can use the :func: selectedFeatures method from vector layer:

selection = layer.selectedFeatures()
print len(selection)
for feature in selection:
    # do whatever you need with the feature

Iterating over a subset of features

If you want to iterate over a given subset of features in a layer, such as those within a given area, you have to add a QgsFeatureRequest object to the getFeatures() call. Here’s an example

request=QgsFeatureRequest()
request.setFilterRect(areaOfInterest)
for f in layer.getFeatures(request):
    ...

The request can be used to define the data retrieved for each feature, so the iterator returns all features, but return partial data for each of them.

# Only return selected fields
request.setSubsetOfAttributes([0,2])
# More user friendly version
request.setSubsetOfAttributes(['name','id'],layer.pendingFields())
# Don't return geometry objects
request.setFlags(QgsFeatureRequest.NoGeometry)

Modifying Vector Layers

Most vector data providers support editing of layer data. Sometimes they support just a subset of possible editing actions. Use the capabilities() function to find out what set of functionality is supported

caps = layer.dataProvider().capabilities()

By using any of following methods for vector layer editing, the changes are directly committed to the underlying data store (a file, database etc). In case you would like to do only temporary changes, skip to the next section that explains how to do modifications with editing buffer.

Add Features

Create some QgsFeature instances and pass a list of them to provider’s addFeatures() method. It will return two values: result (true/false) and list of added features (their ID is set by the data store)

if caps & QgsVectorDataProvider.AddFeatures:
  feat = QgsFeature()
  feat.addAttribute(0, 'hello')
  feat.setGeometry(QgsGeometry.fromPoint(QgsPoint(123, 456)))
  (res, outFeats) = layer.dataProvider().addFeatures([feat])

Delete Features

To delete some features, just provide a list of their feature IDs

if caps & QgsVectorDataProvider.DeleteFeatures:
  res = layer.dataProvider().deleteFeatures([5, 10])

Modify Features

It is possible to either change feature’s geometry or to change some attributes. The following example first changes values of attributes with index 0 and 1, then it changes the feature’s geometry

fid = 100   # ID of the feature we will modify

if caps & QgsVectorDataProvider.ChangeAttributeValues:
  attrs = { 0 : "hello", 1 : 123 }
  layer.dataProvider().changeAttributeValues({ fid : attrs })

if caps & QgsVectorDataProvider.ChangeGeometries:
  geom = QgsGeometry.fromPoint(QgsPoint(111,222))
  layer.dataProvider().changeGeometryValues({ fid : geom })

Adding and Removing Fields

To add fields (attributes), you need to specify a list of field definitions. For deletion of fields just provide a list of field indexes.

if caps & QgsVectorDataProvider.AddAttributes:
  res = layer.dataProvider().addAttributes([QgsField("mytext", QVariant.String), QgsField("myint", QVariant.Int)])

if caps & QgsVectorDataProvider.DeleteAttributes:
  res = layer.dataProvider().deleteAttributes([0])

After adding or removing fields in the data provider the layer’s fields need to be updated because the changes are not automatically propagated.

layer.updateFields()

Modifying Vector Layers with an Editing Buffer

When editing vectors within QGIS application, you have to first start editing mode for a particular layer, then do some modifications and finally commit (or rollback) the changes. All the changes you do are not written until you commit them — they stay in layer’s in-memory editing buffer. It is possible to use this functionality also programmatically — it is just another method for vector layer editing that complements the direct usage of data providers. Use this option when providing some GUI tools for vector layer editing, since this will allow user to decide whether to commit/rollback and allows the usage of undo/redo. When committing changes, all changes from the editing buffer are saved to data provider.

To find out whether a layer is in editing mode, use isEditing() — the editing functions work only when the editing mode is turned on. Usage of editing functions

# add two features (QgsFeature instances)
layer.addFeatures([feat1,feat2])
# delete a feature with specified ID
layer.deleteFeature(fid)

# set new geometry (QgsGeometry instance) for a feature
layer.changeGeometry(fid, geometry)
# update an attribute with given field index (int) to given value (QVariant)
layer.changeAttributeValue(fid, fieldIndex, value)

# add new field
layer.addAttribute(QgsField("mytext", QVariant.String))
# remove a field
layer.deleteAttribute(fieldIndex)

In order to make undo/redo work properly, the above mentioned calls have to be wrapped into undo commands. (If you do not care about undo/redo and want to have the changes stored immediately, then you will have easier work by editing with data provider.) How to use the undo functionality

layer.beginEditCommand("Feature triangulation")

# ... call layer's editing methods ...

if problem_occurred:
  layer.destroyEditCommand()
  return

# ... more editing ...

layer.endEditCommand()

The beginEditCommand() will create an internal “active” command and will record subsequent changes in vector layer. With the call to endEditCommand() the command is pushed onto the undo stack and the user will be able to undo/redo it from GUI. In case something went wrong while doing the changes, the destroyEditCommand() method will remove the command and rollback all changes done while this command was active.

To start editing mode, there is startEditing() method, to stop editing there are commitChanges() and rollback() — however normally you should not need these methods and leave this functionality to be triggered by the user.

Using Spatial Index

Spatial indexes can dramatically improve the performance of your code if you need to do frequent queries to a vector layer. Imagine, for instance, that you are writing an interpolation algorithm, and that for a given location you need to know the 10 closest points from a points layer, in order to use those point for calculating the interpolated value. Without a spatial index, the only way for QGIS to find those 10 points is to compute the distance from each and every point to the specified location and then compare those distances. This can be a very time consuming task, especially if it needs to be repeated for several locations. If a spatial index exists for the layer, the operation is much more effective.

Think of a layer without a spatial index as a telephone book in which telephone numbers are not ordered or indexed. The only way to find the telephone number of a given person is to read from the beginning until you find it.

Spatial indexes are not created by default for a QGIS vector layer, but you can create them easily. This is what you have to do.

  1. create spatial index — the following code creates an empty index

    index = QgsSpatialIndex()
    
  2. add features to index — index takes QgsFeature object and adds it to the internal data structure. You can create the object manually or use one from previous call to provider’s nextFeature()

    index.insertFeature(feat)
    
  3. once spatial index is filled with some values, you can do some queries

    # returns array of feature IDs of five nearest features
    nearest = index.nearestNeighbor(QgsPoint(25.4, 12.7), 5)
    
    # returns array of IDs of features which intersect the rectangle
    intersect = index.intersects(QgsRectangle(22.5, 15.3, 23.1, 17.2))
    

Writing Vector Layers

You can write vector layer files using QgsVectorFileWriter class. It supports any other kind of vector file that OGR supports (shapefiles, GeoJSON, KML and others).

There are two possibilities how to export a vector layer:

  • from an instance of QgsVectorLayer

    error = QgsVectorFileWriter.writeAsVectorFormat(layer, "my_shapes.shp", "CP1250", None, "ESRI Shapefile")
    
    if error == QgsVectorFileWriter.NoError:
      print "success!"
    
    error = QgsVectorFileWriter.writeAsVectorFormat(layer, "my_json.json", "utf-8", None, "GeoJSON")
    if error == QgsVectorFileWriter.NoError:
      print "success again!"
    

    The third parameter specifies output text encoding. Only some drivers need this for correct operation - shapefiles are one of those — however in case you are not using international characters you do not have to care much about the encoding. The fourth parameter that we left as None may specify destination CRS — if a valid instance of QgsCoordinateReferenceSystem is passed, the layer is transformed to that CRS.

    For valid driver names please consult the supported formats by OGR — you should pass the value in the “Code” column as the driver name. Optionally you can set whether to export only selected features, pass further driver-specific options for creation or tell the writer not to create attributes — look into the documentation for full syntax.

  • directly from features

    # define fields for feature attributes. A list of QgsField objects is needed
    fields = [QgsField("first", QVariant.Int),
              QgsField("second", QVariant.String)]
    
    # create an instance of vector file writer, which will create the vector file.
    # Arguments:
    # 1. path to new file (will fail if exists already)
    # 2. encoding of the attributes
    # 3. field map
    # 4. geometry type - from WKBTYPE enum
    # 5. layer's spatial reference (instance of
    #    QgsCoordinateReferenceSystem) - optional
    # 6. driver name for the output file
    writer = QgsVectorFileWriter("my_shapes.shp", "CP1250", fields, QGis.WKBPoint, None, "ESRI Shapefile")
    
    if writer.hasError() != QgsVectorFileWriter.NoError:
      print "Error when creating shapefile: ", writer.hasError()
    
    # add a feature
    fet = QgsFeature()
    fet.setGeometry(QgsGeometry.fromPoint(QgsPoint(10,10)))
    fet.setAttributes([1, "text"])
    writer.addFeature(fet)
    
    # delete the writer to flush features to disk (optional)
    del writer
    

Memory Provider

Memory provider is intended to be used mainly by plugin or 3rd party app developers. It does not store data on disk, allowing developers to use it as a fast backend for some temporary layers.

The provider supports string, int and double fields.

The memory provider also supports spatial indexing, which is enabled by calling the provider’s createSpatialIndex() function. Once the spatial index is created you will be able to iterate over features within smaller regions faster (since it’s not necessary to traverse all the features, only those in specified rectangle).

A memory provider is created by passing "memory" as the provider string to the QgsVectorLayer constructor.

The constructor also takes a URI defining the geometry type of the layer, one of: "Point", "LineString", "Polygon", "MultiPoint", "MultiLineString", or "MultiPolygon".

The URI can also specify the coordinate reference system, fields, and indexing of the memory provider in the URI. The syntax is:

crs=definition
Specifies the coordinate reference system, where definition may be any of the forms accepted by QgsCoordinateReferenceSystem.createFromString()
index=yes
Specifies that the provider will use a spatial index
field=name:type(length,precision)
Specifies an attribute of the layer. The attribute has a name, and optionally a type (integer, double, or string), length, and precision. There may be multiple field definitions.

The following example of a URI incorporates all these options

"Point?crs=epsg:4326&field=id:integer&field=name:string(20)&index=yes"

The following example code illustrates creating and populating a memory provider

# create layer
vl = QgsVectorLayer("Point", "temporary_points", "memory")
pr = vl.dataProvider()

# add fields
pr.addAttributes([QgsField("name", QVariant.String),
                    QgsField("age",  QVariant.Int),
                    QgsField("size", QVariant.Double)])

# add a feature
fet = QgsFeature()
fet.setGeometry(QgsGeometry.fromPoint(QgsPoint(10,10)))
fet.setAttributes(["Johny", 2, 0.3])
pr.addFeatures([fet])

# update layer's extent when new features have been added
# because change of extent in provider is not propagated to the layer
vl.updateExtents()

Finally, let’s check whether everything went well

# show some stats
print "fields:", len(pr.fields())
print "features:", pr.featureCount()
e = layer.extent()
print "extent:", e.xMin(), e.yMin(), e.xMax(), e.yMax()

# iterate over features
f = QgsFeature()
features = vl.getFeatures()
for f in features:
  print "F:", f.id(), f.attributes(), f.geometry().asPoint()

Appearance (Symbology) of Vector Layers

When a vector layer is being rendered, the appearance of the data is given by renderer and symbols associated with the layer. Symbols are classes which take care of drawing of visual representation of features, while renderers determine what symbol will be used for a particular feature.

The renderer for a given layer can obtained as shown below:

renderer = layer.rendererV2()

And with that reference, let us explore it a bit

print "Type:", rendererV2.type()

There are several known renderer types available in QGIS core library:

Type Class Description
singleSymbol QgsSingleSymbolRendererV2 Renders all features with the same symbol
categorizedSymbol QgsCategorizedSymbolRendererV2 Renders features using a different symbol for each category
graduatedSymbol QgsGraduatedSymbolRendererV2 Renders features using a different symbol for each range of values

There might be also some custom renderer types, so never make an assumption there are just these types. You can query QgsRendererV2Registry singleton to find out currently available renderers.

It is possible to obtain a dump of a renderer contents in text form — can be useful for debugging

print rendererV2.dump()

Single Symbol Renderer

You can get the symbol used for rendering by calling symbol() method and change it with setSymbol() method (note for C++ devs: the renderer takes ownership of the symbol.)

Categorized Symbol Renderer

You can query and set attribute name which is used for classification: use classAttribute() and setClassAttribute() methods.

To get a list of categories

for cat in rendererV2.categories():
  print "%s: %s :: %s" % (cat.value().toString(), cat.label(), str(cat.symbol()))

Where value() is the value used for discrimination between categories, label() is a text used for category description and symbol() method returns assigned symbol.

The renderer usually stores also original symbol and color ramp which were used for the classification: sourceColorRamp() and sourceSymbol() methods.

Graduated Symbol Renderer

This renderer is very similar to the categorized symbol renderer described above, but instead of one attribute value per class it works with ranges of values and thus can be used only with numerical attributes.

To find out more about ranges used in the renderer

for ran in rendererV2.ranges():
  print "%f - %f: %s %s" % (
      ran.lowerValue(),
      ran.upperValue(),
      ran.label(),
      str(ran.symbol())
     )

you can again use classAttribute() to find out classification attribute name, sourceSymbol() and sourceColorRamp() methods. Additionally there is mode() method which determines how the ranges were created: using equal intervals, quantiles or some other method.

If you wish to create your own graduated symbol renderer you can do so as illustrated in the example snippet below (which creates a simple two class arrangement)

from qgis.core import *

myVectorLayer = QgsVectorLayer(myVectorPath, myName, 'ogr')
myTargetField = 'target_field'
myRangeList = []
myOpacity = 1
# Make our first symbol and range...
myMin = 0.0
myMax = 50.0
myLabel = 'Group 1'
myColour = QtGui.QColor('#ffee00')
mySymbol1 = QgsSymbolV2.defaultSymbol(myVectorLayer.geometryType())
mySymbol1.setColor(myColour)
mySymbol1.setAlpha(myOpacity)
myRange1 = QgsRendererRangeV2(myMin, myMax, mySymbol1, myLabel)
myRangeList.append(myRange1)
#now make another symbol and range...
myMin = 50.1
myMax = 100
myLabel = 'Group 2'
myColour = QtGui.QColor('#00eeff')
mySymbol2 = QgsSymbolV2.defaultSymbol(
     myVectorLayer.geometryType())
mySymbol2.setColor(myColour)
mySymbol2.setAlpha(myOpacity)
myRange2 = QgsRendererRangeV2(myMin, myMax, mySymbol2 myLabel)
myRangeList.append(myRange2)
myRenderer = QgsGraduatedSymbolRendererV2('', myRangeList)
myRenderer.setMode(QgsGraduatedSymbolRendererV2.EqualInterval)
myRenderer.setClassAttribute(myTargetField)

myVectorLayer.setRendererV2(myRenderer)
QgsMapLayerRegistry.instance().addMapLayer(myVectorLayer)

Working with Symbols

For representation of symbols, there is QgsSymbolV2 base class with three derived classes:

  • QgsMarkerSymbolV2 — for point features
  • QgsLineSymbolV2 — for line features
  • QgsFillSymbolV2 — for polygon features

Every symbol consists of one or more symbol layers (classes derived from QgsSymbolLayerV2). The symbol layers do the actual rendering, the symbol class itself serves only as a container for the symbol layers.

Having an instance of a symbol (e.g. from a renderer), it is possible to explore it: type() method says whether it is a marker, line or fill symbol. There is a dump() method which returns a brief description of the symbol. To get a list of symbol layers

for i in xrange(symbol.symbolLayerCount()):
  lyr = symbol.symbolLayer(i)
  print "%d: %s" % (i, lyr.layerType())

To find out symbol’s color use color() method and setColor() to change its color. With marker symbols additionally you can query for the symbol size and rotation with size() and angle() methods, for line symbols there is width() method returning line width.

Size and width are in millimeters by default, angles are in degrees.

Working with Symbol Layers

As said before, symbol layers (subclasses of QgsSymbolLayerV2) determine the appearance of the features. There are several basic symbol layer classes for general use. It is possible to implement new symbol layer types and thus arbitrarily customize how features will be rendered. The layerType() method uniquely identifies the symbol layer class — the basic and default ones are SimpleMarker, SimpleLine and SimpleFill symbol layers types.

You can get a complete list of the types of symbol layers you can create for a given symbol layer class like this

from qgis.core import QgsSymbolLayerV2Registry
myRegistry = QgsSymbolLayerV2Registry.instance()
myMetadata = myRegistry.symbolLayerMetadata("SimpleFill")
for item in myRegistry.symbolLayersForType(QgsSymbolV2.Marker):
  print item

Output

EllipseMarker
FontMarker
SimpleMarker
SvgMarker
VectorField

QgsSymbolLayerV2Registry class manages a database of all available symbol layer types.

To access symbol layer data, use its properties() method that returns a key-value dictionary of properties which determine the appearance. Each symbol layer type has a specific set of properties that it uses. Additionally, there are generic methods color(), size(), angle(), width() with their setter counterparts. Of course size and angle is available only for marker symbol layers and width for line symbol layers.

Creating Custom Symbol Layer Types

Imagine you would like to customize the way how the data gets rendered. You can create your own symbol layer class that will draw the features exactly as you wish. Here is an example of a marker that draws red circles with specified radius

class FooSymbolLayer(QgsMarkerSymbolLayerV2):

  def __init__(self, radius=4.0):
    QgsMarkerSymbolLayerV2.__init__(self)
    self.radius = radius
    self.color = QColor(255,0,0)

  def layerType(self):
    return "FooMarker"

  def properties(self):
    return { "radius" : str(self.radius) }

  def startRender(self, context):
    pass

  def stopRender(self, context):
    pass

  def renderPoint(self, point, context):
    # Rendering depends on whether the symbol is selected (QGIS >= 1.5)
    color = context.selectionColor() if context.selected() else self.color
    p = context.renderContext().painter()
    p.setPen(color)
    p.drawEllipse(point, self.radius, self.radius)

  def clone(self):
    return FooSymbolLayer(self.radius)

The layerType() method determines the name of the symbol layer, it has to be unique among all symbol layers. Properties are used for persistence of attributes. clone() method must return a copy of the symbol layer with all attributes being exactly the same. Finally there are rendering methods: startRender() is called before rendering first feature, stopRender() when rendering is done. And renderPoint() method which does the rendering. The coordinates of the point(s) are already transformed to the output coordinates.

For polylines and polygons the only difference would be in the rendering method: you would use renderPolyline() which receives a list of lines, resp. renderPolygon() which receives list of points on outer ring as a first parameter and a list of inner rings (or None) as a second parameter.

Usually it is convenient to add a GUI for setting attributes of the symbol layer type to allow users to customize the appearance: in case of our example above we can let user set circle radius. The following code implements such widget

class FooSymbolLayerWidget(QgsSymbolLayerV2Widget):
  def __init__(self, parent=None):
    QgsSymbolLayerV2Widget.__init__(self, parent)

    self.layer = None

    # setup a simple UI
    self.label = QLabel("Radius:")
    self.spinRadius = QDoubleSpinBox()
    self.hbox = QHBoxLayout()
    self.hbox.addWidget(self.label)
    self.hbox.addWidget(self.spinRadius)
    self.setLayout(self.hbox)
    self.connect(self.spinRadius, SIGNAL("valueChanged(double)"), \
      self.radiusChanged)

  def setSymbolLayer(self, layer):
    if layer.layerType() != "FooMarker":
      return
    self.layer = layer
    self.spinRadius.setValue(layer.radius)

  def symbolLayer(self):
    return self.layer

  def radiusChanged(self, value):
    self.layer.radius = value
    self.emit(SIGNAL("changed()"))

This widget can be embedded into the symbol properties dialog. When the symbol layer type is selected in symbol properties dialog, it creates an instance of the symbol layer and an instance of the symbol layer widget. Then it calls setSymbolLayer() method to assign the symbol layer to the widget. In that method the widget should update the UI to reflect the attributes of the symbol layer. symbolLayer() function is used to retrieve the symbol layer again by the properties dialog to use it for the symbol.

On every change of attributes, the widget should emit changed() signal to let the properties dialog update the symbol preview.

Now we are missing only the final glue: to make QGIS aware of these new classes. This is done by adding the symbol layer to registry. It is possible to use the symbol layer also without adding it to the registry, but some functionality will not work: e.g. loading of project files with the custom symbol layers or inability to edit the layer’s attributes in GUI.

We will have to create metadata for the symbol layer

class FooSymbolLayerMetadata(QgsSymbolLayerV2AbstractMetadata):

  def __init__(self):
    QgsSymbolLayerV2AbstractMetadata.__init__(self, "FooMarker", QgsSymbolV2.Marker)

  def createSymbolLayer(self, props):
    radius = float(props[QString("radius")]) if QString("radius") in props else 4.0
    return FooSymbolLayer(radius)

  def createSymbolLayerWidget(self):
    return FooSymbolLayerWidget()

QgsSymbolLayerV2Registry.instance().addSymbolLayerType(FooSymbolLayerMetadata())

You should pass layer type (the same as returned by the layer) and symbol type (marker/line/fill) to the constructor of parent class. createSymbolLayer() takes care of creating an instance of symbol layer with attributes specified in the props dictionary. (Beware, the keys are QString instances, not “str” objects). And there is createSymbolLayerWidget() method which returns settings widget for this symbol layer type.

The last step is to add this symbol layer to the registry — and we are done.

Creating Custom Renderers

It might be useful to create a new renderer implementation if you would like to customize the rules how to select symbols for rendering of features. Some use cases where you would want to do it: symbol is determined from a combination of fields, size of symbols changes depending on current scale etc.

The following code shows a simple custom renderer that creates two marker symbols and chooses randomly one of them for every feature

import random

class RandomRenderer(QgsFeatureRendererV2):
  def __init__(self, syms=None):
    QgsFeatureRendererV2.__init__(self, "RandomRenderer")
    self.syms = syms if syms else [QgsSymbolV2.defaultSymbol(QGis.Point), QgsSymbolV2.defaultSymbol(QGis.Point)]

  def symbolForFeature(self, feature):
    return random.choice(self.syms)

  def startRender(self, context, vlayer):
    for s in self.syms:
      s.startRender(context)

  def stopRender(self, context):
    for s in self.syms:
      s.stopRender(context)

  def usedAttributes(self):
    return []

  def clone(self):
    return RandomRenderer(self.syms)

The constructor of parent QgsFeatureRendererV2 class needs renderer name (has to be unique among renderers). symbolForFeature() method is the one that decides what symbol will be used for a particular feature. startRender() and stopRender() take care of initialization/finalization of symbol rendering. usedAttributes() method can return a list of field names that renderer expects to be present. Finally clone() function should return a copy of the renderer.

Like with symbol layers, it is possible to attach a GUI for configuration of the renderer. It has to be derived from QgsRendererV2Widget. The following sample code creates a button that allows user to set symbol of the first symbol

class RandomRendererWidget(QgsRendererV2Widget):
  def __init__(self, layer, style, renderer):
    QgsRendererV2Widget.__init__(self, layer, style)
    if renderer is None or renderer.type() != "RandomRenderer":
      self.r = RandomRenderer()
    else:
      self.r = renderer
    # setup UI
    self.btn1 = QgsColorButtonV2("Color 1")
    self.btn1.setColor(self.r.syms[0].color())
    self.vbox = QVBoxLayout()
    self.vbox.addWidget(self.btn1)
    self.setLayout(self.vbox)
    self.connect(self.btn1, SIGNAL("clicked()"), self.setColor1)

  def setColor1(self):
    color = QColorDialog.getColor(self.r.syms[0].color(), self)
    if not color.isValid(): return
    self.r.syms[0].setColor(color);
    self.btn1.setColor(self.r.syms[0].color())

  def renderer(self):
    return self.r

The constructor receives instances of the active layer (QgsVectorLayer), the global style (QgsStyleV2) and current renderer. If there is no renderer or the renderer has different type, it will be replaced with our new renderer, otherwise we will use the current renderer (which has already the type we need). The widget contents should be updated to show current state of the renderer. When the renderer dialog is accepted, widget’s renderer() method is called to get the current renderer — it will be assigned to the layer.

The last missing bit is the renderer metadata and registration in registry, otherwise loading of layers with the renderer will not work and user will not be able to select it from the list of renderers. Let us finish our RandomRenderer example

class RandomRendererMetadata(QgsRendererV2AbstractMetadata):
  def __init__(self):
    QgsRendererV2AbstractMetadata.__init__(self, "RandomRenderer", "Random renderer")

  def createRenderer(self, element):
    return RandomRenderer()
  def createRendererWidget(self, layer, style, renderer):
    return RandomRendererWidget(layer, style, renderer)

QgsRendererV2Registry.instance().addRenderer(RandomRendererMetadata())

Similarly as with symbol layers, abstract metadata constructor awaits renderer name, name visible for users and optionally name of renderer’s icon. createRenderer() method passes QDomElement instance that can be used to restore renderer’s state from DOM tree. createRendererWidget() method creates the configuration widget. It does not have to be present or can return None if the renderer does not come with GUI.

To associate an icon with the renderer you can assign it in QgsRendererV2AbstractMetadata constructor as a third (optional) argument — the base class constructor in the RandomRendererMetadata __init__() function becomes

QgsRendererV2AbstractMetadata.__init__(self,
       "RandomRenderer",
       "Random renderer",
       QIcon(QPixmap("RandomRendererIcon.png", "png")))

The icon can be associated also at any later time using setIcon() method of the metadata class. The icon can be loaded from a file (as shown above) or can be loaded from a Qt resource (PyQt4 includes .qrc compiler for Python).

Further Topics

TODO:
creating/modifying symbols working with style (QgsStyleV2) working with color ramps (QgsVectorColorRampV2) rule-based renderer (see this blogpost) exploring symbol layer and renderer registries