Description of problem: We use BPMN2 xml files for displaying process runtime path in our monitoring console. The BPMN2 files with subprocesses containing shapes and edges are displayed correctly. Shapes and edges were moved outside of the subprocess boundary. Edges were disconnected from the shapes too. We tried to open the BPMN2 file in other tool like https://bpmn.io/ and received the same view. Debugging showed us to classes Bpmn2JsonUnmarshaller and Bpmn2JsonMarshaller. The root cause of the problem seems to be variable coordianteManipulation in Bpmn2JsonMarshaller class. Is there a reason for transforming the coordinates when loading XML to JSON and transforming again when saving JSON to XML? When I disable the transformation at both ends (set coordianteManipulation to false and comment out methods updateShapeBoundsInSubprocess and updateEdgeBoundsInContainers) processes display correctly. Version-Release number of selected component (if applicable): 6.3.0 How reproducible: 1. create BPMN2 process with subprocess and tasks and connections within subprocess in business-central. 2. go to View Process Sources in business-central. 3. visualize XML given in BPMN2 tab. Actual results: Connectors and tasks are misplaced. Expected results: Connectors and tasks are aligned. Additional info: Follow up on Redhat support case 01604255.
Yes there is a reason for the transformation. In the Designer JSON nested node coordinates are represented as relative to the top-left corder of their container(s). The transformation changes them to absolute which is required by the BPMN2 spec (and also to work as it does in our eclipse editor). Maybe the tools you are opening the bpmn2 with expect relative coordinates for nested nodes? In this case those tools should probably be update to follow spec as well.
Hi, Thank you for your reply. We tried to display the diagram in bpmn.io (http://demo.bpmn.io/). I looked into BPMN 2.0 specification and it says:"All BPMNShape elements are owned directly by a BPMNPlane (that is the root element in a BPMNDiagram), i.e., shapes are not nested within each other in the BPMN DI model although they may appear that way when depicted. The bounds of a BPMNShape are always relative to that plane's origin point and are REQUIRED to be positive coordinates. Note that the bounds’ x and y coordinates are the position of the upper left corner of the shape (relative to the upper left corner of the plane).", chapter 12.2.3.5 and respectively for Edges... Radovan
@Radovam, yes and the eclipse modeller and web designer conform to that. I tried importing/exporting a couple of small diagrams with designer and http://demo.bpmn.io/new and at first glance it seems to me that http://demo.bpmn.io/new is doing some weird things during the import but we do not control that source so I can't tell what's going on and IMO would be not good to just assume it is 100% correct. We can however compare and control the code of the web designer and the eclipse modeler so testing with those 2 tools would be preferred and if you find any issues there please report them.
Hi Tihomir, You are right. It boils down to the way the processes are drawn inside our monitoring application. Thank you very much, Radovan
I was looking into the issue bit more. Seems like the update of coordinates for edges is broken. Method updateEdgeBoundsInContainers in the class Bpmn2JsonUnmarshaller is adding coordinates of shapes to the end points instead coordinates of a container in our case of a subprocess. Another issue is that it does not transform vertexes and only end points. Reverse operation in the class Bpmn2JsonMarshaller is fixing that. Again it is transforming end points separately in the method marshallSequenceFlow. It transforms also vertexes though. IMHO it would be sufficient to transform coordinates of all vertexes of an edge in a simple loop instead of looking for end points etc...