Investigate NaN error catching in compiler

[permcody]

None

[permcody]

There doesn't seem to be a portable way to do this. On the Mac, one can use the Intel intrinsics to set the FP control register to throw an exception when a NaN is used in a calculation but despite my best attempts, I've not been able to recreate a good stack after catching the signal.

For now we are using an explicit check in debug mode for catching nans.

[permcody]

Explore the PETSC -fp_trap option to see if we can get a stack trace from it.

[permcody]

David is working on a solution similar to the one I was working on but has disabled the Petsc error handler as well allowing for a stack trace to be caught at least on Mac/GCC.

[andrsd]

This is implemented. It works only in debug mode - where we get the full stack trace. It does not work in opt mode, since the compiler optimizes out stuff and we were getting false alarms.

Thus, closing this ticket.