[pivot] Unresolved proxy loading AS

[eclipse-ocl-bot]

| --- | --- | | Bugzilla Link | 542534 | | Status | NEW | | Importance | P3 normal | | Reported | Dec 07, 2018 15:49 EDT | | Modified | Dec 08, 2018 04:27 EDT | | Depends on | 542544 | | Reporter | Ed Willink |

Description

After adding Class.resolvedBy/resolutionFor and making DataType.behavioralClass derived, two OCLstdlib tests fail with unresolved proxies, but the tests pass in isolation. The unresolved proxies are on Ecore.oclas and are of traditional E<F-style @ownedPackages.0 style rather tha pivot LUSSIDs.

Investigating reveals that these proxies are created when Ecore.oclas is unloaded.

This is inelegant since creating proxy ids for a dead resource is a waste of time. We can improve this.

These proxies should bever be used since an unloaded *.oclas is never reloaded. Except that perhaps it is as part of the sequence of tests.

[eclipse-ocl-bot]

By Ed Willink on Dec 07, 2018 16:33

Problem is that the new Class.resolvedBy/resolutionFor have a modelld oposite so that after EBoolean is destroyed, Boolean still has a resolutionFor to it.

[eclipse-ocl-bot]

By Ed Willink on Dec 08, 2018 04:27

(In reply to Ed Willink from comment #1)

Problem is that the new Class.resolvedBy/resolutionFor have a modelld oposite

Fixing the pivot synthesis to allow the unnavigable opposite is Bug 542544.

(In reply to Ed Willink from comment #0)

This is inelegant since creating proxy ids for a dead resource is a waste of time. We can improve this.

This bug.

Ideally we never have unsound navigable oppoisites to trash.

• a) because they are not in a model
• b) because we do a fast dispose

But if we do the Ecore ids should work even if we choose to create LUSSIDs.