History notes . . .

Must Time and Distance Obscure Truth?

Conrad Weisert, October 14, 2011^*
ŠInformation Disciplines, Inc.

NOTE: This document may be circulated or quoted from freely, as long as the copyright credit is included.

^*—This article was written back in July, but was held for web publication pending confirmations by various sources.

"Niklaus Wirth, the designer of Pascal (and other languages since then), states that a 'programming language is as good as its compiler'. The first compilers for PL/1 were so buggy that the language's use was affected. When Ada was designed, the design team, remembering the problems with PL/1, designed validation requirements that a compiler would have to pass to be considered a legal Ada compiler " —Worcester Polytechnic Institute, course material

Students in that course will no doubt come away quite sure that a major cause of PL/1's failure to gain sufficient market share for long-term survival was the poor quality of early compilers, mainly from IBM. There's just one problem with that:   It's untrue!

Twenty years ago such an assertion would quickly have provoked responses from people who actually used IBM's PL/1 compilers in the late 1960s. Today, however, it goes unchallenged, because many of the people who had first-hand knowledge have now:

• died,
• retired,
• forgotten,
• or lost interest.

Tossing in a famous authority is helpful. The writer didn't actually claim that Niklaus Wirth thought that early PL/I compilers were full of bugs. but the proximity of the citation in the paragraph could easily convey that impression to a naïve student.

Academic Rigor

If the paragraph at the top of this article had appeared in a student's research paper, we can easily imagine a professor's challenges:

The first compilers for PL/1 were so buggy that the language's use was affected.	according to whom? Citation needed!
When Ada was designed, the design team, remembering the problems with PL/1, . . .	Who? Reference?

Only an usually confident student would dare raise such questions about an instructor's course handout. But most of the mature faculty I know would welcome being challenged in the interest of furnishing fair and accurate information to their students.

What really happened?

This article isn't about the quality of early PL/1 compilers or the reasons behind PL/1's failure to attain market share. It's mainly about the growing frequency of spreading folklore and wishful thinking as historical fact in Computer Science.

Since I brought it up as an example, however, I shall try to set the record straight on PL/1:

• The early IBM F-level compiler was very slow, due to a combination of the language's complexity and IBM's decision that it had to run in 64K memory.

• The generated code, too, could be inefficient, especially if the programmer made unwise data-representation choices.

• Some essential features were deferred until the third or fifth release of the compiler, including RECORD I-O, needed for most business applications.

• There was a good bit of correspondence about violations of the "Law of Least Astonishment", a user-group complaint, not about the compiler, but about the language specifications. Non-integer fixed-point scaling did yield some astonishing odd cases, but if you didn't use non-integer fixed-point numbers¹ you didn't encounter them.

• Despite complaints about inefficiency and language specifications, the F-level compiler was extremely reliable. I recall warning students and colleagues not to suspect a compiler bug when they encountered a hard-to-diagnose problem, and I recall only one occasion when we actually found one. That compared favorably with the IBM COBOL products of the time.

• Eventually a new generation of IBM compilers alleviated the inefficiencies of the F-level compiler, and compilers from other vendors followed suit. I'm not aware of any compiler that supported the full language and was said to be "buggy".

I've checked the above with a half-dozen colleagues who were active in using PL/I daily in the late 1960s and early 1970s, and they concur. If you know of counter examples, let me know, and I'll revise this article accordingly.

Last modified October 14, 2011

Return to IDI home page
Historical notes.