Version 2 of S

S and Unix

Those of us involved in statistical computing at Bell Labs had been aware of Unix from the beginning, without at first thinking it had much direct relevance. Actually, I had briefly been a user of the very earliest version of that system. The level of tolerance for my playing around was not too high; I recall once very early on discovering that it was possible to clear memory by some version of dividing by zero. Since this brought the machine to a halt, it was not viewed favorably.

A marriage of Unix with statistical computing was not initially looked on very warmly by either side. Unix began on a 16-bit machine with relatively primitive facilities for numerical computation. The originators of the system felt (probably correctly) that statisticians and numerical analysts didn't have enough rapport with their approach to be appropriate guests on a crowded machine.

Aside from technical issues, there was a definitely different mindset at the time between the two groups.

All this started to change when experiments began on making Unix portable, and in particular on porting it to 32-bit machines. It was almost immediately clear to the S folks that here was the sort of environment we needed to pursue portability for S.


John Chambers<jmc@research.bell-labs.com>
Last modified: Thu Apr 1 14:33:09 EST 1999