Lambda the Ultimate

In connection with an effort to reason about numerical programs we recently ran into what seems to be strange decisions in the design of the floating-point standard, IEEE 754. I discuss the issues in a blog post ("Reflexivity and other pillars of civilization"), showing in particular that the convention for equality regarding NaN (Not A Number) seem wrong; the reverse convention would be more in line with basic mathematical properties such as the reflexivity of equality.

We encountered the problem in connection with the work on revising the ISO standard for Eiffel, when we realized that transposing the IEEE 754 rule to the language would mean that we cannot put a value into an array and then assume that the value that we put is there! Neither can we assign a value v to a variable x and assume that afterward the value of x is equal to v.

See the detailed explanation at the page linked to above. I would be interested in the experience of people having run into similar issues.