In a prior post, I described an example of a side effect of a feature that our users came to rely on as a feature in its own right.

While reading Raymond Chen’s The Old New Thing: Practical Development Throughout the Evolution of Windows, I came across what I consider the mother of such side effects. In an essay entitled “The hunt for a faster syscall trap”, Raymond describes how an Intel representative was perplexed when a Windows engineer requested speeding up the fault when an invalid instruction is executed. This would seem to imply that the Windows code was buggy in some sort of way. The story goes on to relate that executing an invalid instruction was intentional – as a way to get into the CPU kernel mode.

A version of the essay can be found online here.