Playing with this today, the issues discussed in PR7406 seem to stem from
two causes:
1. AttrNonNullChecker doesn't expect the value of a nonnull argument to be
unknown when it's checked. Patch here:
http://llvm.org/bugs/attachment.cgi?id=5067
2. The additive folding code in SimpleSValuator and the constraint manager
doesn't handle the case where the constants are of different types
(unsigned, long long, etc). Patch here:
http://llvm.org/bugs/attachment.cgi?id=5069
#2 in particular could use a review, since it's a nontrivial fix (a number
of type conversions). And of course it was me who didn't get this right the
first time around. Ted? Zhongxing?
Jordy