POSIX doesn't specify the internal layout of the fd_set object, so let's
not assume it is just a bit mask. This increases compatibility with
systems that have a different layout.
The assumption is also worthless as we already refuse to use file
descriptors over FD_SETSIZE anyway. Meaning that the default size of
fd_set is quite sufficient.
If the underlying system's select() returns bits that were not in the
request set, our ppoll() implementation can return revents for events
not requested, which can apparently cause a hang. Only return revents
for activity in the requested event set. bz#3416, analysis and fix by
yaroslav.kuzmin at vmssoftware com, ok djm@
Correct handling of select(2) exceptfds. These should only be consulted
for POLLPRI flagged pfds and not unconditionally converted to POLLERR.
with and ok dtucker@
Darwin's poll(2) implementation is broken. For character-special
devices like /dev/null, it returns POLLNVAL when polled with
POLLIN.
Apparently this is Apple bug 3710161, which is AFAIK not public,
but a websearch will find other OSS projects rediscovering it
periodically since it was first identified in 2005 (!!)
configure (and there's not much point, as openssh won't work without it)
so HAVE_SELECT is not defined and the poll(2) compat code doesn't get
built in. Remove HAVE_SELECT so we can build on platforms without poll.
openbsd-compat/bsd-poll.{c,h} openbsd-compat/openbsd-compat.h]
Add an implementation of poll() built on top of select(2). Code from
OpenNTPD with changes suggested by djm. ok djm@
Pierre Ossman
Darren Tucker
Damien Miller
Darren Tucker