On Jun 14, 2012, at 3:05 PM, Jed Brown wrote:

In the case of a matrix, shouldn't these checks go the first place we encounter them (MatSetValues)?

Could be.

Running with -fp_trap is really good at finding these kind of mistakes.

On Wed, Jun 13, 2012 at 8:50 AM, Barry Smith <[email protected]> wrote:

On return from user FormFunction and FormJacobian in debug mode should we check if all user output is sane? That is numerical values in the matrix and vector are sane, no 1.e-307 or 1.e85 and stuff like that?

Barry