Obtaining maximum confidence from unrevealing tests [software testing]

Software testing cannot guarantee that a program is at least correct with respect to the inputs over which the program was tested. This is a problem when testing concurrent programs, or could also be due to the possibility of a mistake in evaluating the results of a test of a sequential program. It is shown that, even if such evaluation mistakes are eliminated, testing still cannot normally guarantee program correctness, even with respect to a single input, given most programming language implementations. Reasonably unrestrictive conditions on the language implementation are proposed to resolve this problem