[tests] Rationalize GlobalStateMemento for Attribution.REGISTRY

[eclipse-ocl-bot]

| --- | --- |
| Bugzilla Link | 583509 |
| Status | NEW |
| Importance | P3 normal |
| Reported | Jul 30, 2024 03:32 EDT |
| Modified | Sep 06, 2024 11:02 EDT |
| Reporter | Ed Willink |

Description

The local copy of org.eclipse.xtext.testing.GlobalRegistries resolves EMF issues but is still inadequate. Perversely it is only used for testCompleteOCLRegistry_Access. Should be a full rteplacement of the PivotTestCase copy.

The standalone version of testUMLXCompiler_Forward2Reverse_CG identified a propblem whereby

registry.put(EssentialOCLCSPackage.Literals.SHADOW_PART_CS, ShadowPartCSAttribution.INSTANCE);

from EssentialOCLScoping has an initialization order dependency wrt

registry.put(EssentialOCLCSPackage.Literals.SHADOW_PART_CS, QVTimperativeShadowPartCSAttribution.INSTANCE);

from QVTimperativeScoping.

The immediate problem was worked around by an extra QVTimperativeScoping.init().

Attribution.REGISTRY should be part of the global state.

registry.put should perhaps be overloaded to detect a paut that is not a derived class of a prevailing init.

More generally makeCopyOfGlobalState() should perhaps diagnose unexpected global state that results from failure to observe the need to initialize pre-setup so that tearDown restores.

[eclipse-ocl-bot]

By Ed Willink on Jul 30, 2024 04:07

(In reply to Ed Willink from comment #0)

Attribution.REGISTRY should be part of the global state.

Yes, but clearing it in PicotStandaloneSetup.init forces a full re-init.

registry.put should perhaps be overloaded to detect a paut that is not a
derived class of a prevailing init.

Done.

[eclipse-ocl-bot]

By Ed Willink on Sep 06, 2024 11:02

a) Enforcing the doXXXStandaloneSetup before JUnit setUp() policy
b) abstracting AbstractPivotTestCAse with a TEstHelper to unifi OCL/QVTd
c) using a single GlobalStateMemento
improves general behaviour.

Since XXXScoping only happens when XXX is initialized, a thorough tearDown is needed to support GC testing.

There are no conflicting Attributions, e.g a refined QVTr Attribution extends and tests for the unrefined case. The AttributionRegistry can therefore just grow and grow with the no insyallation of anti-refined test.