As mentioned in #607, we want to try cranking up the precision to quad precision and make sure the failures we see with threads are merely floating point accumulation rather than something more nefarious. We'll probably need to resolve #608 first for that specific instance, but I imagine trying to enable quad precision support is going to bring out the (probably many) places where we haven't been systematic on leveraging the libMesh numeric types (libMesh::Real, libMesh::Number, etc.). In retrospect, I wish I'd just started out by wrapping those. :-/
As mentioned in #607, we want to try cranking up the precision to quad precision and make sure the failures we see with threads are merely floating point accumulation rather than something more nefarious. We'll probably need to resolve #608 first for that specific instance, but I imagine trying to enable quad precision support is going to bring out the (probably many) places where we haven't been systematic on leveraging the libMesh numeric types (libMesh::Real, libMesh::Number, etc.). In retrospect, I wish I'd just started out by wrapping those. :-/