FORTRAN arrived without instructions

In early April 1957, IBM’s FORTRAN team decided its compiler was finally reliable enough to distribute. The plan was to make 30 or 40 copies. Each copy was a deck of roughly 2,000 punched cards.

The copying failed. Producing binary cards was so error-prone that after working through the night, the team had made only one or two usable decks. One of them was sent to Westinghouse’s Bettis installation without a label or directions.

The recipients had ordered an IBM 704 and were waiting for FORTRAN. Faced with a box of unexplained cards, Herbert Bright’s group guessed that this might be it. They loaded the deck and gave it a small test program.

The mystery software replied that a particular statement was missing a comma.

This was an excellent identification label. Random cards would crash, do nothing or produce nonsense. A precise complaint about the program’s punctuation strongly suggested that the long-promised formula translator had arrived and understood what it was looking at.

After the comma was inserted, the compiler produced a program that printed 28 pages of correct results, apart from some formatting errors. The date was April 20, 1957. John Backus’s later history of FORTRAN preserves the episode, while his oral history adds that it was the first executable compiler outside his group—and that it did not work so obligingly for long afterward.

FORTRAN was meant to free programmers from managing the machine at its lowest level. Its first distribution still depended on an all-night card-punching session, a physical shipment and some educated guesswork at the destination.

Before it could translate a useful program, the compiler had to explain what was wrong with one. The diagnostic did more than report an error. It introduced the software.