Explicitly handle listen sockets and connection sockets differently in Linux and Mac OS eventhandler

Explicitly handle listen sockets and connection sockets differently in Linux and Mac OS eventhandler

Previously the assumption was that an POLLIN event with no data available was from a listen socket with pending accepts. However a POLLIN event from a connection socket can have no data available if the data has been read before the event is processed.

R=ager@google.com

BUG=
TEST=

Committed: https://code.google.com/p/dart/source/detail?r=878

[Mads Ager (google), 2011-10-28 08:33:23]

lgtm

http://codereview.chromium.org/8399039/diff/2001/runtime/bin/eventhandler.h
File runtime/bin/eventhandler.h (right):

http://codereview.chromium.org/8399039/diff/2001/runtime/bin/eventhandler.h#n...
runtime/bin/eventhandler.h:30: kListenSocket = 16,
The naming here confuses me. 4 Events, 1 Command and something that is neither?
Can we streamline that? Or maybe the enum shouldn't be called Message?

http://codereview.chromium.org/8399039/diff/2001/runtime/bin/eventhandler_lin...
File runtime/bin/eventhandler_linux.cc (right):

http://codereview.chromium.org/8399039/diff/2001/runtime/bin/eventhandler_lin...
runtime/bin/eventhandler_linux.cc:100: return (port_map_[fd].mask & (1 <<
kListenSocket)) != 0;
kListeningSocket?

http://codereview.chromium.org/8399039/diff/2001/runtime/bin/eventhandler_lin...
runtime/bin/eventhandler_linux.cc:222: if ((pollfd->revents & POLLIN) != 0)
event_mask |= (1 << kInEvent);
Should we assert that there is no overlap in these? Can we ever get more than
one of these and what does that mean?

[Søren Gjesse, 2011-10-28 09:24:18]

http://codereview.chromium.org/8399039/diff/2001/runtime/bin/eventhandler.h
File runtime/bin/eventhandler.h (right):

http://codereview.chromium.org/8399039/diff/2001/runtime/bin/eventhandler.h#n...
runtime/bin/eventhandler.h:30: kListenSocket = 16,
On 2011/10/28 08:33:23, Mads Ager wrote:
> The naming here confuses me. 4 Events, 1 Command and something that is
neither?
> Can we streamline that? Or maybe the enum shouldn't be called Message? 

So this is the bits that we send to the event loop when stuff changes for a file
descriptor. I will rename to MessageFlags and provide a better comment.

http://codereview.chromium.org/8399039/diff/2001/runtime/bin/eventhandler_lin...
File runtime/bin/eventhandler_linux.cc (right):

http://codereview.chromium.org/8399039/diff/2001/runtime/bin/eventhandler_lin...
runtime/bin/eventhandler_linux.cc:100: return (port_map_[fd].mask & (1 <<
kListenSocket)) != 0;
On 2011/10/28 08:33:23, Mads Ager wrote:
> kListeningSocket?

Done.

http://codereview.chromium.org/8399039/diff/2001/runtime/bin/eventhandler_lin...
runtime/bin/eventhandler_linux.cc:222: if ((pollfd->revents & POLLIN) != 0)
event_mask |= (1 << kInEvent);
On 2011/10/28 08:33:23, Mads Ager wrote:
> Should we assert that there is no overlap in these? Can we ever get more than
> one of these and what does that mean?

I would be reluctant to try to assert anything here. However reading the
documentation once more both POLLHUP and POLLERR are always provided together
with POLLIN, so POLLIN should only mean "connections ready for accept" if
provided alone.