The Linux kernel itself is a remarkably stable system as far as ABI/API is concerned
You're joking, right? There have been cases where ABIs have been broken by security patches, and users have been told that they have to choose between having a secure system and being able to use their third-party binary drivers.
(And yes, I understand the arguments about how binary drivers are evil; but I don't think users should be used as pawns in that fight.)
As other people have said, userland ABI/API is solid. I should have clarified that.
There is no reason the internal API would be as to be honest it's the Kernel's job to make the drivers sit behind that API and stay there regardless of the problems. That is the abstraction boundary.
The issue is where the abstraction boundary is. In NetworkManager for example, which is an example I'm probably just picking on, there are many more abstractions above the kernel's abstraction which are unstable.
Again, comparing to Windows, there have been several driver model revisions but the userland API+ABI is stable between them.
> The Linux kernel itself is a remarkably stable system as far as ABI/API is concerned
If we're talking about userland applications, the Linux kernel A[BP]Is have IME historically been more stable than the Glibc interfaces (which are supposedly the "supported" interfaces) layered atop them.
In-kernel interfaces for binary-only modules are a different kettle of fish altogether.
He was probably talking of the user-space ABI/API. Drivers are internal to the kernel (or should be) and Linux never made a claim that those APIs where going to be stable.
You're joking, right? There have been cases where ABIs have been broken by security patches, and users have been told that they have to choose between having a secure system and being able to use their third-party binary drivers.
(And yes, I understand the arguments about how binary drivers are evil; but I don't think users should be used as pawns in that fight.)