Jeff Garzik (jgarzik@cc.gatech.edu)
Thu, 28 Sep 1995 00:34:38 -0400 (EDT)

I had an idea a while back and I wonder if anyone has discussed this
before -- regression tests for the kernel.

A kernel is a hairy beast, and sometimes changes cause or reveal bugs
in other portions of the code. User-level testing programs --
programs that stress-test features based on their usage in the "real
world" -- for each little slice of the code would, I'm sure, be an
immense help to future kernel developers everywhere.

I have a couple stress testing programs written for my tmpfs project
that I simply cannot live without, and I know I have found at least
two hard-to-spot bugs because of the tests. Adding regression tests
would be a great diagnostic tool, an opportunity perhaps to move some
kernel debug code into separate test programs, etc...