DMN improvements for demo

[joeribekker]

Some improvements to the DMN UI

• When selecting a decision definition/table to evaluate, the dependencies in the DRD do not seem to be taken into account
• Consequently, only the inputs of a single table, but not as a whole DRD are taken into account
• Check if we can directly provide the necessary inputs instead of a required decision
• Using a + b expressions in an input expression is... uncommon. We don't necessarily need to support this right now.
• Invoking a table for each item in an editgrid (repeating group)?
• Try out -> and relay back error message

Useful resource: https://consulting.camunda.com/dmn-simulator/

Tasks

• Update DMN parser in django-camunda to correctly resolve requirements and their inputs - this doesn't seem to go right now
• Add JS library (https://github.com/nikku/feelin/blob/main/src/parser.ts) to parse the FEEL expressions client-side to extract the remaining input parameters.
• Ensure that we can configure variable mappings with implicit parameters (e.g. input with a + b FEEL expression -> vars a and b)

[sergei-maertens]

Findings of research done for this ticket:

• Camunda 7.20 implements a part of DMN 1.3
  • It only supports two kinds of boxed expressions: decision table and literal expression. DMN spec itself supports more.
  • It does not support business knowledge models, which is much more advanced and can be used for advanced decision modelling. This could be a direction to deal with repeating groups.
  • It does seem to support contexts per decision, but this doesn't appear to be explicitly exposed in the modeller. Untested.
• When a decision has (one or more other) required decisions, then the input parameters of that need to be provided.
  • It does not seem possible in Camunda to evaluate a leaf decision by providing all of its inputs directly, thereby skipping the requirement.
  • Likely evaluating the requirement by itself is possible.
  • Camunda only returns the output of the requested decision, not the outputs of all intermediates (not sure how their playground works then...)
• The DROOLS DMN engine seems more full-featured and can be a nice reference. Way out of scope for here though!
• The Camunda modeler only supports primitive types for inputs, so we have a problem with repeating groups (putting that out of scope for now, it may be solvable in OF rather than at the Camunda level)
• FEEL expression language is part of the standard and the default. We only support FEEL (Camunda supports JS, Python... too)

[sergei-maertens]

While these are enhancements, they are critical enhancements to be backported because in its current state, DMN integration is not usable.