Re: Bug in select()?

[Roland McGrath]

That's what I was expecting to see.  This is a fundamental issue with the
design of the io_select interface.  You could try the hack of setting the
minimum timeout to 1ms (or maybe 10ms, since that's the time slice), but I
don't know how effective that will be.  

The trouble is that the Hurdish notion of "polling" is that you don't block
on any server, while a "polling select" is really intended to positively
check that there is or isn't data that can be read.  There is no way to
both guarantee that you have really gone into the server's data structures
and checked for pending data, while at the same time not being susceptible
to getting hung up a malicious or buggy server just blocking you.