[PATCH] Prevent MDNode's RAUW from introducing a reference to a void Value.

[David Blaikie]

>>! In D5828#10, @friss wrote:
> Seems I was wrong about being able to extend the testcase. Changing the scope of the fix didn't change the fact that I need a constant bitcast inside a call to trigger that code path. And I'm not able to produce this without resorting to UB at the C level. 
> 
> Unless you have another idea, I think I'll commit the test I have.

Nope, no particular ideas short of mucking with the assertion (perhaps so it triggers on anything other than this case) & running over a selfhost or other corpus to see if anything more realistic/reasonable/future-proof crops up.

Not a requirement, just a desire/preference if it were possible/practical.

http://reviews.llvm.org/D5828