Bug with reductions when ckJustMigrated is implemented in user code

[kavithachandrasekar]

When user code implements ckJustMigrated to track migrations after migrateMe, reductions do not work. Instead of calling ArrayElement::ckJustMigrated, the runtime calls the user's ckJustMigrated. This results in the object count on PEs for reductions not being correctly updated when objects migrate in. Future reduction contributions result in a hang. The issue can be reproduced by calling migrateMe and implementing ckJustMigrated in user code and doing a few reduction calls after ckJustMigrated.

[olawlor]

I believe you need to call ArrayElement::ckJustMigrated() from your ckJustMigrated method if you override that virtual method.

This is one representative of the "didn't call the superclass" bugs that can hit constructors and virtual methods in OOP C++.

(20+ years ago when I was writing this stuff, I really should have made all the user-facing virtual override methods have empty default implementations, and used inaccessible system methods for the critical system plumbing, but here we are!)