The funny thing is that you've got that drilled into your head even though it's not true. Userspace breaks all the time, it's just that some of them matter to Linus and some don't, and he's never been clear about which is which.
Several months ago at work I found a vendor app that was calling listen(0), i.e., listen for up to 0 connections, and worked fine on older kernels because they rounded to a minimum of 8. Newer kernels stopped doing that. Nobody cared. Nobody got criticized, constructively or unconstructively. (By the time we noticed it was several kernel releases upstream so bringing it up would just be asking for another change to userspace.)
And that's not counting all the things that aren't part of the stable interface like procfs and sysfs nodes, device names, etc. There is a clear version of what Linus is trying to say. He's just never said it.
Several months ago at work I found a vendor app that was calling listen(0), i.e., listen for up to 0 connections, and worked fine on older kernels because they rounded to a minimum of 8. Newer kernels stopped doing that. Nobody cared. Nobody got criticized, constructively or unconstructively. (By the time we noticed it was several kernel releases upstream so bringing it up would just be asking for another change to userspace.)
Linus himself broke userspace once in the hope nobody would notice. Spoiler: they did. https://lkml.org/lkml/2017/5/29/541
And that's not counting all the things that aren't part of the stable interface like procfs and sysfs nodes, device names, etc. There is a clear version of what Linus is trying to say. He's just never said it.