> I really don't see how it's "obviously incorrect" when yes only needs to make system calls for writing, whereas /dev/zero needs to make system calls for both reading and writing. Furthermore you're assuming that both are highly optimized, whereas it's entirely conceivable that /dev/zero is less optimized.

When you're "only writing", there is a counterpart that needs to read, and you have to wait for that, same as /dev/zero. It's more conceivable /dev/zero is more optimized than yes: you know, because of how pipes and streams are an essential part of a UNIX or UNIX-like system.

> If what you're trying to convey is that abstract reasoning based on fundamental principles without any consideration of real-world complications is a superior approach to determining what's true, well, I reject that.

I agree with that. But the original comment only gives us few options to check for its validity, and some of those are not feasible.

1. Actually finding all those systems, version-specific, combinations of userspace tools, setting them up and running the tests on hardware that was at OP's hand "back then": Not feasible.

2. Reasoning abstractly would tell us quickly that the original comment is plain wrong.

3. Running the tests on a modern system (some default linux install or whatever) shows that (2) is even more probably correct.