In article <199511292207.RAA11392@glacier.MIT.EDU>,
Greg Hudson  <ghudson@MIT.EDU> wrote:
> If we're going to abandon the current select() interface, the right
> answer is to the use the SVR4 poll(), one of the few things System V
> did much better than BSD.  (There's the minor weakness that it uses a
> millisecond timeout, but that's much less of a flaw than those of the
> select() interface.)

So, why not introduce a "upoll()", and make poll() a simple libc
wrapper? Or even put it in the kernel, since emulations will probably
want it anyway.

I agree poll() is cleaner from an API standpoint. However, I fear that
the changes to implement it in the kernel might be significant and
far-reaching.

If it does look doable, before tackiling the kernel you could do this as
a proof-of-concept: start by implementing upoll() as a user-space
wrapper on select(), and the poll() on top of it.  This will be useful
in porting NetBSD code to other OSes which have select() but not
poll(), if NetBSD is converted.  Then implement select() as a wrapper on
upoll(), which will definately be needed if NetBSD converts.  Then put
the emulated-select()-on-emulated-upoll()-on-syscall-select() in your
libc and run your system this way for a while.  That way you can get
all the bugs ironed out of the userspace, and even get a start on
converting NetBSD's userspace, all before the kernel is touched.