Andrew Dunstan <andrew(at)dunslane(dot)net> writes:
> This might all be OK - I just noticed it as I was looking at the
> problem, so I though I'd mention it.
I'm inclined to leave that code alone unless someone can point to a
platform where setting POLLERR in events actually causes a problem.
The pgstat code was modeled on libpq, which has been passing POLLERR
ever since its poll() support was written (for 7.4), so I think we
have lots of evidence that this way works everywhere, and none that
the other one does.
regards, tom lane