Feature/multiinstance refactor #292

essweine · 2023-02-14T16:50:55Z

This is a complete refactor of how we handle multiinstance tasks. We no longer:

use dynamically generated classes
copy tasks and insert them into the task tree manually during predict
modify the workflow spec while the workflow is in progress (except in the Camunda version)
rely on Camnuda extensions in the base package

During parsing, a multiinstance task is added to the spec, with a reference to the original task spec. When the MI task is reached, it changes state to waiting and creates children from the original referenced spec as needed, merges the data from the children into the original task as they are completed, and when all children are complete, moves to ready/complete.

This GREATLY simplifies dealing with multiinstance tasks.

jbirddog

Minor comments, lratm (looks really awesome to me).

jbirddog · 2023-02-15T12:23:30Z

SpiffWorkflow/bpmn/serializer/migration/version_1_1.py

+    waiting = []
+    for sp in subprocesses:
+        completed = sorted(
+            [t for t in subprocesses[sp]['tasks'].values() if t['state'] in [32, 64] ],


Safer/more obvious to use the finished mask?

I am going to leave these as ints, since that's what's in the serialied version.

jbirddog · 2023-02-15T12:24:38Z

SpiffWorkflow/bpmn/serializer/migration/version_1_1.py

+    for sp in [dct] + list(subprocesses.values()):
+        for task_id, task in sp['tasks'].items():
+            if task_id in waiting:
+                task['state'] = 8
