Environmental Effects on the Detection of Errors in Software Systems

Abstract

It is very important for software developers to thoroughly test their software. Since programs are tested in a specific environment, however, some errors are difficult, if not impossible, to detect by traditional testing methods such as path analysis. A program may contain errors that are masked by the environment in which it was developed and tested. One such element in a program’s environment is the value to which memory is initialized. Mistakes in initialization would not appear until the software is ported to an environment in which memory has a different initialization value, when the software would mysteriously stop working correctly.

To prevent “mystery errors” and increase software portability, a programmer should be aware of the types of errors that are undetectable for a particular machine and system. These errors must be looked for specifically, realizing that test data will never expose them. Systems programmers also should be aware of the masking affect of different memory initialization values and use values that hide the fewest types of errors. This paper examines errors that programmers tend to make with FORTRAN to determine what values in memory could mask each. A “best” initialization value is proposed for FORTRAN, and findings are extrapolated to other programming languages such as Ada.