Estimating the correctness of computer programs

• Published in: Information sciences 1982-01, Vol.26 (2), p.141-153
• Main Authors: Duran, Joe W., Wiorkowski, John J.
• Format: Article
• Language: English
• ISSN: 0020-0255; 1872-6291
• DOI: 10.1016/0020-0255(82)90038-X

Program testing will for some time remain the major means by which program designers convince themselves and others of the correctness or validity of their programs. Thus it is desirable to be able to use test results to measure the degree to which a program approaches correctness. The probability of correctness of a program can sometimes be estimated according to the success of testing. A general parameter is introduced which has the property that its value is zero for a correct program and is greater than zero for an incorrect program. Under reasonable interpretations, the parameter is related, through probability density functions, to characteristics which are observable through program testing. For instance, if the parameter is interpreted as the number of errors in a program, a related, observable characteristic is the number of errors discovered through testing. Upper bounds on the parameter value are established as a function of the value of the observed characteristic and the desired level of confidence in the result. Results are presented for cases of combinations of interpretations, observable characteristics, and probability density functions.