This paper presents one facet of the SDI software debate--the problem of residual errors and the associated dilemma that “we can’t fix it after it fails for the first time.” The quotation is from David Parnas, whose cogent arguments against SDI software are summarized in the first half of the paper.
The paper reviews the data on software error rates, noting that operational software typically contains 1–3 errors per thousand lines. Thus, 20 million lines of SDI software could contain some 40,000 errors. Parnas’s question, therefore, is the key: “[How can] we believe, with any confidence, that we have succeeded?”
The paper also summarizes the report of the SDI Organization’s Panel on Computing in Support of Battle Management (Eastport Study Group, from which Parnas resigned). This panel accepted the reality that “all systems of useful complexity contain software errors.” Addressing that matter requires research in many fields, including software architecture, computing power, testing, fault tolerance, and program management.
There is considerable editorializing here: banner headlines, sidebars by different authors, lots of quotes, and a personal conclusion. As such, the paper probably represents more of the state-of-the-art in the debate over SDI than the state-of-the-art in SDI software.