wrapped-bivariate-to-array test fails

[Symbolics]

This test fails under fiveam, but passes under clunit. Investigation shows that the two function results being compared:

(funcall op a b)
(funcall op (funcall convert a) b)

produce different results the first and second time being called. This leads to what appears to be a false negative in the test because the fiveam reason-arg, which is supposed to print a helpful diagnostic message, calls the functions a second time, producing correct results. The source code for the test contains experiments and debugging examples that demonstrate this.

The test code will not win any 'most readable code of the year' award, and the answer is buried below several layers of macro expansions, funcalls and currying. This one is not for beginners.