[appleskeptic]

“Never break userland” is more of a strong guideline than a hard rule, even if Linus has sometimes made it sound otherwise during a heated gamer moment. The fact is, kernel devs break userland code all the time, sometimes because the alternative is worse, or sometimes because a bugfix broke something and no one noticed until much later and it was decided not to revert the fix because it would now break other things to do that. Try to run an entire large userland from 2001 on a modern kernel and tell me how that goes.

[tux3]

Mistakes are human. Certainly a lot of bugs slip through, and that's understandable, software is hard.

That being said, I find that people tend to take the no regression rule pretty seriously, if you bring a valid complaint. It makes good sense. We don't want to give people a reason to stay on old kernels even more than they do today (this post written on an ancient Android kernel, chock full of out of tree drivers). We can at least save people from pain when it comes to in-tree drivers.

I think we really all benefit from that policy. Even if it sometimes keeps some less than perfect workarounds in the code

[dfc]

I always assumed it was a rule. I don't know of any kernel regressions that broke something I used daily. Are there any examples you can think of?