Right, but the We Do Not Break User Space rule has often been explained (somewhat emphatically on occasion) as depending on whether anyone's code actually gets broken, not on whether the thing they were relying on was intended to be, or advertised as, a stable interface.

I think WDNBUS is only respected for the syscall api.

Here's a relatively mild rant where the breaking change in question was in /proc/cpuinfo:

«Really. Shut up.

The whole "no regressions" thing is very much about the fact that we don't waste users time.»

https://linux.kernel.narkive.com/7hMPh8zX/patch-revert-9fc21...

Of course in an ideal world the extent of the intended promise would be clearly written down in some obvious place. But we don't live in an ideal world.

I found this statement from Linus in that thread;

"It's not that the kernel ABI (with all the system calls, all the /proc files, all the ioctl's, etc) is set in stone and "sacred". Absolutely anything can be changed, wildly.

But if it turns out that applications (or hardware) that people use end up breaking noticeably, then that is a regression."

What about /proc ?

/proc has surely killed some user apps. If you were expected to see a fixed number of lines, a fixed number of columns or the fields to mean the same thing, you're in for suffering. At least I know of meminfo changes, stuff in /proc/net...

But to be honest, so have ifconfig, hostname, df...

At least in procfs they add things (columns, lines) so apart from meminfo I don't know of changing the meaning of fields.

Linus has been pretty strict in not unnecessarily breaking userspace even if something is not strictly ‘part of the ABI’. It’s a judgement call though.