The downside to that is that %useValueEquality delegates to the C++ operator== and just negates the result when checking for inequality. By also defining an operator!= in C++ we could quite easily end up with different semantics in C++ and Python, which seems unfortunate.

(Of course, if somebody wants to define an operator!= which isn't just the complement of operator== they should probably think again, but...)

I suppose we could essentially require the existence of operator!= in C++ by having %useValueEquality delegate to that for __ne__.

I watched an entertaining rant by Alex Stepanov over the weekend, to the effect that it's basically always a design flaw if !(thing == other) is not equivalent to thing != other. I'm persuaded that I like that solution best, iff it doesn't introduce too much overhead – I'm not sure how many places we're using %useValueEquality, and I don't think it's worth spending a long time checking them down. I made the change as suggested, fixed a couple of places where I know it'll break, and submitted to buildbot – let's see what fails.

It turns out that it took only a few lines of changes in utils and afw to delegate Python's __ne__ to C++ and then provide operator!= everywhere that it's required. Given that, I think this is the approach I prefer.

Jim, the changes are on the tickets/DM-2347 branch on utils and afw. Please check that you're happy not just with those changes but with the approach adopted. Thanks!

Looks good to me. I think I agree that's the best approach (I certainly agree that everything that implements == should implement != in C++; I'm ambivalent as to how we enforce that).