Using QGIS 2 processing model in QGIS 3

Question

Is it somehow possible to convert a QGIS 2.x processing model (*.model) to a *.model3 so I can run them in QGIS 3.0 without having to redo mine from scratch?

Answer 1

I fear there is no simple way to convert geoprocessing models from QGIS 2 to 3. Simply converting it from json to xml can be done with services like http://www.utilities-online.info/xmltojson/ but the structure of the result is completely different between the two versions:

Converted from QGIS 2 model json (example):

<?xml version="1.0" encoding="UTF-8" ?>
    <values>
        <inputs>
            <network>
                <values>
                    <pos>
                        <values>
                            <y>458</y>
                            <x>296</x>
                        </values>
                        <class>point</class>
                    </pos>
                    <param>
                        <values>
                            <isAdvanced>false</isAdvanced>

                            <name>network</name>
                            <shapetype>1</shapetype>
                            <default />
                            <value />
                            <exported />
                            <hidden>false</hidden>
                            <optional>false</optional>
                            <description>network</description>
                        </values>
                           <class>processing.core.parameters.ParameterVector</class>
                </param>
            </values>
            <class>processing.modeler.ModelerAlgorithm.ModelerParameter</class>
        </network>

QGIS 3 model (native xml)

<!DOCTYPE model>
<Option type="Map">
  <Option name="children" type="Map">
    <Option name="native:buffer_1" type="Map">
      <Option value="true" name="active" type="bool"/>
      <Option name="alg_config"/>
      <Option value="native:buffer" name="alg_id" type="QString"/>
      <Option value="Buffer" name="component_description" type="QString"/>
      <Option value="382" name="component_pos_x" type="double"/>
      <Option value="447" name="component_pos_y" type="double"/>
      <Option name="dependencies"/>
      <Option value="native:buffer_1" name="id" type="QString"/>
      <Option name="outputs" type="Map">
        <Option name="result" type="Map">
          <Option value="native:buffer_1" name="child_id" type="QString"/>
          <Option value="result" name="component_description" type="QString"/>
          <Option value="582" name="component_pos_x" type="double"/>
          <Option value="492" name="component_pos_y" type="double"/>
          <Option value="" name="default_value" type="Unknown"/>
          <Option value="false" name="mandatory" type="bool"/>
          <Option value="result" name="name" type="QString"/>
          <Option value="OUTPUT" name="output_name" type="QString"/>
        </Option>
      </Option><!DOCTYPE model>

Apart from the fact that the two structures have very little in common, the processing environment and the modeler have undergone extensive changes (additional/changed input parameters and algorithms). So my conclusion is that we have (sad but true...) to manually convert our QGIS 2 models to the new environment.