Re: [PATCH v3 0/1] pipe: only enable the extra wake_up(rd_wait) for edge-triggered consumers

[Oleg Nesterov <oleg@redhat.com>]

On 07/30, Linus Torvalds wrote:
>
> I like this patch mostly for the renaming, not because I think it
> matters. I think "poll usage" was a mistake in naming and doesn't
> explain the issue. That said, I'd go even further, and make it clear
> that it's not about "poll" itself - which is fine, it's about "epoll",
> which has that broken crazy bug where it calls something "edge
> triggered" but then actually wants effectively level-triggered
> behavior - wakeups when nothing actually changed, which is the
> *opposite* of an edge.

I am not sure this is a "crazy bug". To me, epoll with EPOLLET works
"as documented". Perhaps I am wrong, this predates the git history.

I'd say the very idea of EPOLLET was wrong, but this doesn't matter:
we have what we have.

And this patch (mostly) tries to document what we have.

> Pure garbage.
>
> So the real name should be something like "epoll_pseudo_edgetrigger".
> Because "et" isn't really helpful either.

I'd agree with "pseudo" simply because I can hardly say how could we
define "edgetrigger" in this particular case.

But then we should name it "epoll_or_io_uring_pseudo_edgetrigger".
See the test-cases, they demonstrate the same pattern.

Oleg.