The authors use side-effects in order to infer how different compilers treat identical programs. In this case, the programs are in FORTRAN (66) and use DO-loops. Compilers are shown to behave very differently and some do very strange things in the name of optimization. This paper is interesting at an anecdotal level, but is hardly profound.
The authors display a certain amount of misplaced outrage that the nonstandard extensions of FORTRAN in the various compilers actually have different behavior. This outrage is mixed in with a general annoyance at the insufficient documentation available. While their complaints are no doubt accurate, they are not exactly technically enlightening. The substance of this 14-page paper could have been better presented in a one- or two-page correspondence.