Description
The example used is sqrt() which needs a non-negative input. If passed nan or infinity we should presumably return NaN/infinity. So that's the first question - should that be stated in expects? if we agree that NaN is a defined result, a naive Expects( x >= 0 ) will fail. That's one issue.
Another is that not all preconditions can be so easily stated. Geometrical routines may be undefined for degenerate inputs. So if we are asked whether two "lines" which consist of two co-incident points intersect, is that meant to fail an "Expects"? That's a simple example, you can make the invalid input arbitrarily complex. At what point should the routine treat the input as normal processing and return an error flag / something (you can just decide arbitrarily whether two co-incident points intersect or not), throw an exception?