initial commit

Makefile.am

+include_HEADERS = include/zs.h include/zs.hpp
+
+SUBDIRS = src examples
+DIST_SUBDIRS = src examples

autogen.sh

+#!/bin/sh
+#   Copyright (c) 2007 FastMQ Inc.
+#
+#   This file is part of 0MQ.
+#
+#   0MQ is free software; you can redistribute it and/or modify
+#   it under the terms of the GNU General Public License as published by
+#   the Free Software Foundation; either version 3 of the License, or
+#   (at your option) any later version.
+
+#   0MQ is distributed in the hope that it will be useful,
+#   but WITHOUT ANY WARRANTY; without even the implied warranty of
+#   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+#   GNU General Public License for more details.
+
+#   You should have received a copy of the GNU General Public License
+#   along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+# Script to generate all required files from fresh svn checkout.
+
+
+
+autoreconf --install --force --verbose -I config
+
+if [ $? -ne 0 ]; then
+    echo
+    echo "Could not run autoreconf, check autotools installation."
+    echo
+fi

configure.in

+#                                               -*- Autoconf -*-
+# Process this file with autoconf to produce a configure script.
+AC_PREREQ(2.61)
+AC_INIT([zsock],[dev])
+AC_CONFIG_AUX_DIR(config)
+AM_CONFIG_HEADER(src/platform.hpp)
+AM_INIT_AUTOMAKE(AC_PACKAGE_NAME, AC_PACKAGE_VERSION)
+
+AM_PROG_CC_C_O
+
+# Checks for programs.
+AC_PROG_CXX
+AC_PROG_LIBTOOL
+
+# Checks for libraries.
+AC_CHECK_LIB(pthread, pthread_create)
+
+# Host speciffic checks
+AC_CANONICAL_HOST
+
+case "${host_os}" in
+    *linux*)
+        AC_DEFINE(ZS_HAVE_LINUX, 1, [Have Linux OS])
+        CPPFLAGS="-D_REENTRANT $CPPFLAGS"
+        sed < libtool > libtool-2 \
+        's/^hardcode_libdir_flag_spec.*$'/'hardcode_libdir_flag_spec=" "/'
+        mv libtool-2 libtool
+        chmod 755 libtool
+        AC_CHECK_LIB(uuid, uuid_generate)
+        ;;
+    *solaris*)
+        AC_DEFINE(ZS_HAVE_SOLARIS, 1, [Have Solaris OS])
+        AC_CHECK_LIB(socket, main)
+        AC_CHECK_LIB(nsl, main)
+        AC_CHECK_LIB(rt, main)
+        CPPFLAGS="-D_REENTRANT -D_PTHREADS $CPPFLAGS"
+        AC_MSG_CHECKING([wheter atomic operations can be used])
+        AC_COMPILE_IFELSE([AC_LANG_PROGRAM(
+            [[#include <atomic.h>]],
+            [[uint32_t value;
+              atomic_cas_32 (&value, 0, 0);
+              return 0;]])],
+            [solaris_has_atomic=yes],
+            [solaris_has_atomic=no])
+        AC_MSG_RESULT([$solaris_has_atomic])
+        # Solaris 8 does not have atomic operations exported to user space.
+        if test "x$solaris_has_atomic" = "xno"; then
+            AC_DEFINE(ZS_FORCE_MUTEXES, 1, [Force to use mutexes])
+        fi
+        ;;
+    *freebsd*)
+        AC_DEFINE(ZS_HAVE_FREEBSD, 1, [Have FreeBSD OS])
+        CPPFLAGS="-D_THREAD_SAFE $CPPFLAGS"
+        LIBS="-pthread"
+        ;;
+    *darwin*)
+        AC_DEFINE(ZS_HAVE_OSX, 1, [Have DarwinOSX OS])
+        LIBS="-pthread"
+        ZS_EXTRA_CXXFLAGS+="-Wno-uninitialized"
+        ;;
+    *openbsd*)
+        AC_DEFINE(ZS_HAVE_OPENBSD, 1, [Have OpenBSD OS])
+        CPPFLAGS="-pthread $CPPFLAGS"
+        LIBS="-pthread"
+        ;;
+    *nto-qnx*)
+        AC_DEFINE(ZS_HAVE_QNXNTO, 1, [Have QNX Neutrino OS])
+        CPPFLAGS="-D_THREAD_SAFE $CPPFLAGS"
+        AC_CHECK_LIB(socket,main)
+        ;;
+    *aix*)
+        AC_DEFINE(ZS_HAVE_AIX, 1, [Have AIX OS])
+        if test "x$GXX" = "xyes"; then
+          CPPFLAGS="-D_THREAD_SAFE $CPPFLAGS"
+        fi
+        ;;
+    *hpux*)
+        AC_DEFINE(ZS_HAVE_HPUX, 1, [Have HPUX OS])
+        if test "x$GXX" = "xyes"; then
+          CPPFLAGS="-D_THREAD_SAFE $CPPFLAGS"
+        fi
+        AC_CHECK_LIB(rt, main)
+        sed < libtool > libtool-2 \
+        's/^hardcode_libdir_flag_spec.*$'/'hardcode_libdir_flag_spec=" "/'
+        mv libtool-2 libtool
+        chmod 755 libtool
+        ;;
+    *mingw32*)
+        AC_DEFINE(ZS_HAVE_WINDOWS, 1, [Have Windows OS])
+        AC_DEFINE(ZS_HAVE_MINGW32, 1, [Have MinGW32])
+        AC_CHECK_HEADERS(windows.h)
+        LIBS="-lwsock32 -lws2_32 -no-undefined"
+        CFLAGS="-std=c99"
+        install_man="no"
+        ;;
+    *)
+        AC_MSG_ERROR([Not supported os: $host.])
+        ;;
+esac
+
+# Check if we are running at sparc harware
+AC_MSG_CHECKING([wheter __sparc__ is defined])
+AC_COMPILE_IFELSE([AC_LANG_PROGRAM(
+    [[#if defined __sparc__
+        //OK we are on sparc
+    #else
+        error: we are not on sparc
+    #endif
+    ]])],
+        [sparc=yes],
+        [sparc=no])
+
+AC_MSG_RESULT([$sparc])
+        
+if test "x$sparc" = "xyes"; then
+    CPPFLAGS="$CPPFLAGS -mcpu=v9"
+fi
+
+# Checks for header files.
+AC_HEADER_STDC
+AC_CHECK_HEADERS(errno.h arpa/inet.h netinet/tcp.h netinet/in.h stddef.h \
+stdlib.h string.h sys/socket.h sys/time.h unistd.h limits.h)
+
+# Check if we have eventfd.h header file.
+AC_CHECK_HEADERS(sys/eventfd.h, [AC_DEFINE(ZS_HAVE_EVENTFD, 1, [Have eventfd extension.])])
+
+# Check if we have ifaddrs.h header file.
+AC_CHECK_HEADERS(ifaddrs.h, [AC_DEFINE(ZS_HAVE_IFADDRS, 1, [Have ifaddrs.h header.])])
+
+# Use c++ in subsequent tests
+AC_LANG(C++)
+
+# Optional stuff
+AC_CHECK_PROG(have_pkg_config, pkg-config, yes, no)
+
+if test "x$have_pkg_config" != "xno"; then
+    #  First instance of PKG_CHECK_ has to be executed
+    PKG_CHECK_EXISTS([dummy_pkg], [], [])
+fi
+
+# Checks for typedefs, structures, and compiler characteristics.
+AC_HEADER_STDBOOL
+AC_C_CONST
+AC_C_INLINE
+AC_TYPE_SIZE_T
+AC_TYPE_SSIZE_T
+AC_HEADER_TIME
+AC_TYPE_UINT32_T
+AC_C_VOLATILE
+
+# Substs
+stdint="0"
+if test "x$HAVE_STDINT_H" = "xyes"; then
+    stdint="1"
+fi
+
+inttypes="0"
+if test "x$HAVE_INTTYPES_H" = "xyes"; then
+    inttypes="1"
+fi
+
+AC_SUBST(stdint)
+AC_SUBST(inttypes)
+
+# Subst ZS_EXTRA_CXXFLAGS
+AC_SUBST(ZS_EXTRA_CXXFLAGS)
+
+
+# Checks for library functions.
+AC_FUNC_MALLOC
+AC_TYPE_SIGNAL
+AC_CHECK_FUNCS(perror gettimeofday memset socket getifaddrs freeifaddrs)
+
+AC_OUTPUT(Makefile src/Makefile examples/Makefile examples/chat/Makefile)
+
+AC_MSG_RESULT([])
+AC_MSG_RESULT([ ******************************************************** ])
+AC_MSG_RESULT([                         0SOCKETS     ])
+AC_MSG_RESULT([ ******************************************************** ])
+AC_MSG_RESULT([  This software is distributed under the terms and ])
+AC_MSG_RESULT([  conditions of the LESSER GNU GENERAL PUBLIC LICENSE. ])
+AC_MSG_RESULT([  See the file COPYING and COPYING.LESSER for the full ])
+AC_MSG_RESULT([  license text. ])
+AC_MSG_RESULT([ ******************************************************** ])
+AC_MSG_RESULT([])
+AC_MSG_RESULT([ zsock install dir: $prefix])
+AC_MSG_RESULT([])
+

examples/Makefile.am

+SUBDIRS = chat
+DIST_SUBDIRS = chat

examples/chat/Makefile.am

+INCLUDES = -I$(top_builddir) -I$(top_builddir)/include
+
+noinst_PROGRAMS = chatroom display prompt
+
+chatroom_SOURCES = chatroom.cpp 
+chatroom_LDADD = $(top_builddir)/src/libzs.la
+chatroom_CXXFLAGS = -Wall -pedantic -Werror
+
+display_SOURCES = display.cpp
+display_LDADD = $(top_builddir)/src/libzs.la
+display_CXXFLAGS = -Wall -pedantic -Werror
+
+prompt_SOURCES = prompt.cpp 
+prompt_LDADD = $(top_builddir)/src/libzs.la
+prompt_CXXFLAGS = -Wall -pedantic -Werror

examples/chat/chatroom.cpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include <time.h>
+#include <string.h>
+#include <iostream>
+
+using namespace std;
+
+#include <zs.hpp>
+
+int main (int argc, const char *argv [])
+{
+    //  Check the command line syntax
+    if (argc != 3) {
+        cerr << "usage: chatroom <in-interface> <out-interface>" << endl;
+        return 1;
+    }
+
+    //  Retrieve command line arguments
+    const char *in_interface = argv [1];
+    const char *out_interface = argv [2];
+
+    //  Initialise 0MQ infrastructure
+    zs::context_t ctx (1, 1);
+
+    //  Create two sockets. One for receiving messages from 'propmt'
+    //  applications, one for sending messages to 'display' applications
+    zs::socket_t in_socket (ctx, ZS_SUB);
+    in_socket.bind (in_interface);
+    zs::socket_t out_socket (ctx, ZS_PUB);
+    out_socket.bind (out_interface);
+
+    while (true) {
+
+        //  Get a message
+        zs::message_t in_message;
+        in_socket.recv (&in_message);
+
+        //  Get the current time. Replace the newline character at the end
+        //  by space character.
+        char timebuf [256];
+        time_t current_time;
+        time (&current_time);
+        snprintf (timebuf, 256, "%s", ctime (&current_time));
+        timebuf [strlen (timebuf) - 1] = ' ';
+
+        //  Create and fill in the message
+        zs::message_t out_message (strlen (timebuf) + in_message.size ());
+        char *data = (char*) out_message.data ();
+        memcpy (data, timebuf, strlen (timebuf));
+        data += strlen (timebuf);
+        memcpy (data, in_message.data (), in_message.size ());
+
+        //  Send the message
+        out_socket.send (out_message);
+    }
+}

examples/chat/display.cpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include <string.h>
+#include <string>
+#include <iostream>
+
+using namespace std;
+
+#include <zs.hpp>
+
+int main (int argc, const char *argv [])
+{
+    //  Check the command line syntax.
+    if (argc != 2) {
+        cerr << "usage: display <chatroom-out-address>" << endl;
+        return 1;
+    }
+
+    //  Retrieve command line arguments
+    const char *chatroom_out_address = argv [1];
+
+    //  Initialise 0MQ infrastructure, connect to the chatroom and ask for all
+    //  messages and gap notifications.
+    zs::context_t ctx (1, 1);
+    zs::socket_t s (ctx, ZS_SUB);
+    s.connect (chatroom_out_address);
+    s.subscribe ("*");
+    
+    while (true) {
+
+        //  Get a message and print it to the console.
+        zs::message_t message;
+        s.recv (&message);
+        if (message.type () == zs::message_gap)
+            cout << "Problems connecting to the chatroom..." << endl;
+        else
+            cout << (char*) message.data () << flush;
+    }
+}

examples/chat/prompt.cpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include <string.h>
+#include <string>
+#include <iostream>
+
+using namespace std;
+
+#include <zs.hpp>
+
+int main (int argc, const char *argv [])
+{
+    //  Check the command line syntax.
+    if (argc != 3) {
+        cerr << "usage: prompt <chatroom-in-address> <user name>" << endl;
+        return 1;
+    }
+
+    //  Retrieve command line arguments
+    const char *chatroom_in_address = argv [1];
+    const char *user_name = argv [2];
+
+    //  Initialise 0MQ infrastructure and connect to the chatroom.
+    zs::context_t ctx (1, 1);
+    zs::socket_t s (ctx, ZS_PUB);
+    s.connect (chatroom_in_address);
+
+    while (true) {
+
+        //  Allow user to input the message text. Prepend it by user name.
+        char textbuf [1024];
+        char *rcc = fgets (textbuf, sizeof (textbuf), stdin);
+        assert (rcc);
+        string text (user_name);
+        text = text + ": " + textbuf;
+
+        //  Create the message (terminating zero is part of the message)
+        zs::message_t message (text.size () + 1);
+        memcpy (message.data (), text.c_str (), text.size () + 1);
+
+        //  Send the message
+        s.send (message);
+    }
+}

include/zs.h

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef __ZSOCKETS_H_INCLUDED__
+#define __ZSOCKETS_H_INCLUDED__
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <stddef.h>
+#include <stdint.h>
+
+#if defined MSC_VER && defined ZS_BUILDING_LIBZS
+#define ZS_EXPORT __declspec(dllexport)
+#else
+#define ZS_EXPORT
+#endif
+
+//  Maximal size of "Very Small Message". VSMs are passed by value
+//  to avoid excessive memory allocation/deallocation.
+#define ZS_MAX_VSM_SIZE 30
+
+//  Message & notification types.
+#define ZS_GAP 1
+#define ZS_DELIMITER 31
+#define ZS_VSM 32
+
+//  The operation should be performed in non-blocking mode. I.e. if it cannot
+//  be processed immediately, error should be returned with errno set to EAGAIN.
+#define ZS_NOBLOCK 1
+
+//  zs_send should not flush the message downstream immediately. Instead, it
+//  should batch ZS_NOFLUSH messages and send them downstream only when zs_flush
+//  is invoked. This is an optimisation for cases where several messages are
+//  sent in a single business transaction. However, the effect is measurable
+//  only in extremely high-perf scenarios (million messages a second or so).
+//  If that's not your case, use standard flushing send instead. See exchange
+//  example for illustration of ZS_NOFLUSH functionality.
+#define ZS_NOFLUSH 2
+
+//  Socket to communicate with a single peer. Allows for a singe connect or a
+//  single accept. There's no message routing or message filtering involved.
+#define ZS_P2P 0
+
+//  Socket to distribute data. Recv fuction is not implemeted for this socket
+//  type. Messages are distributed in fanout fashion to all peers.
+#define ZS_PUB 1
+
+//  Socket to subscribe to distributed data. Send function is not implemented
+//  for this socket type. However, subscribe function can be used to modify the
+//  message filter.
+#define ZS_SUB 2
+
+//  Socket to send requests on and receive replies from. Requests are
+//  load-balanced among all the peers. This socket type doesn't allow for more
+//  recv's that there were send's.
+#define ZS_REQ 3
+
+//  Socket to receive requests from and send replies to. This socket type allows
+//  only an alternated sequence of recv's and send's. Each send is routed to
+//  the peer that the previous recv delivered message from.
+#define ZS_REP 4
+
+//  Prototype for the message body deallocation functions.
+//  It is deliberately defined in the way to comply with standard C free.
+typedef void (zs_free_fn) (void *data);
+
+//  A message. If 'shared' is true, message content pointed to by 'content'
+//  is shared, i.e. reference counting is used to manage its lifetime
+//  rather than straighforward malloc/free. struct zs_msg_content is
+//  not declared in the API.
+struct zs_msg
+{
+    struct zs_msg_content *content;
+    unsigned char shared;
+    uint16_t vsm_size;
+    unsigned char vsm_data [ZS_MAX_VSM_SIZE];
+};
+
+//  TODO: Different options...
+struct zs_opts
+{
+    uint64_t hwm;
+    uint64_t lwm;
+    uint64_t swap;
+    uint64_t mask;
+    uint64_t taskset;
+    const char *identity;
+    const char *args;
+};
+
+//  Initialise an empty message (zero bytes long).
+ZS_EXPORT int zs_msg_init (zs_msg *msg);
+
+//  Initialise a message 'size' bytes long.
+//
+//  Errors: ENOMEM - the size is too large to allocate.
+ZS_EXPORT int zs_msg_init_size (zs_msg *msg, size_t size);
+
+//  Initialise a message from an existing buffer. Message isn't copied,
+//  instead 0SOCKETS infrastructure take ownership of the buffer and call
+//  deallocation functio (ffn) once it's not needed anymore.
+ZS_EXPORT int zs_msg_init_data (zs_msg *msg, void *data, size_t size,
+    zs_free_fn *ffn);
+
+//  Deallocate the message.
+ZS_EXPORT int zs_msg_close (zs_msg *msg);
+
+//  Move the content of the message from 'src' to 'dest'. The content isn't
+//  copied, just moved. 'src' is an empty message after the call. Original
+//  content of 'dest' message is deallocated.
+ZS_EXPORT int zs_msg_move (zs_msg *dest, zs_msg *src);
+
+//  Copy the 'src' message to 'dest'. The content isn't copied, instead
+//  reference count is increased. Don't modify the message data after the
+//  call as they are shared between two messages. Original content of 'dest'
+//  message is deallocated.
+ZS_EXPORT int zs_msg_copy (zs_msg *dest, zs_msg *src);
+
+//  Returns pointer to message data.
+ZS_EXPORT void *zs_msg_data (zs_msg *msg);
+
+//  Return size of message data (in bytes).
+ZS_EXPORT size_t zs_msg_size (zs_msg *msg);
+
+//  Returns type of the message.
+ZS_EXPORT int zs_msg_type (zs_msg *msg);
+
+//  Initialise 0SOCKETS context. 'app_threads' specifies maximal number
+//  of application threads that can have open sockets at the same time.
+//  'io_threads' specifies the size of thread pool to handle I/O operations.
+//
+//  Errors: EINVAL - one of the arguments is less than zero or there are no
+//                   threads declared at all.
+ZS_EXPORT void *zs_init (int app_threads, int io_threads);
+
+//  Deinitialise 0SOCKETS context including all the open sockets. Closing
+//  sockets after zs_term has been called will result in undefined behaviour.
+ZS_EXPORT int zs_term (void *context);
+
+//  Open a socket.
+//
+//  Errors: EINVAL - invalid socket type.
+//          EMFILE - the number of application threads entitled to hold open
+//                   sockets at the same time was exceeded.
+ZS_EXPORT void *zs_socket (void *context, int type);
+
+//  Close the socket.
+ZS_EXPORT int zs_close (void *s);
+
+//  Bind the socket to a particular address.
+ZS_EXPORT int zs_bind (void *s, const char *addr, zs_opts *opts);
+
+//  Connect the socket to a particular address.
+ZS_EXPORT int zs_connect (void *s, const char *addr, zs_opts *opts);
+
+//  Subscribe for the subset of messages identified by 'criteria' argument.
+ZS_EXPORT int zs_subscribe (void *s, const char *criteria);
+
+//  Send the message 'msg' to the socket 's'. 'flags' argument can be
+//  combination of following values:
+//  ZS_NOBLOCK - if message cannot be sent, return immediately.
+//  ZS_NOFLUSH - message won't be sent immediately. It'll be sent with either
+//               subsequent flushing send or explicit call to zs_flush function.
+//
+//  Errors: EAGAIN - message cannot be sent at the moment (applies only to
+//                   non-blocking send).
+//          ENOTSUP - function isn't supported by particular socket type.
+ZS_EXPORT int zs_send (void *s, zs_msg *msg, int flags);
+
+//  Flush the messages that were send using ZS_NOFLUSH flag down the stream.
+//
+//  Errors: ENOTSUP - function isn't supported by particular socket type.
+ZS_EXPORT int zs_flush (void *s);
+
+//  Send a message from the socket 's'. 'flags' argument can be combination
+//  of following values:
+//  ZS_NOBLOCK - if message cannot be received, return immediately.
+//
+//  Errors: EAGAIN - message cannot be received at the moment (applies only to
+//                   non-blocking receive).
+//          ENOTSUP - function isn't supported by particular socket type.
+ZS_EXPORT int zs_recv (void *s, zs_msg *msg, int flags);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

include/zs.hpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef __ZSOCKETS_HPP_INCLUDED__
+#define __ZSOCKETS_HPP_INCLUDED__
+
+#include "zs.h"
+
+#include <assert.h>
+
+namespace zs
+{
+
+    typedef zs_free_fn free_fn;
+
+    enum message_type_t
+    {
+        message_data = 1 << 0,
+        message_gap = 1 << ZS_GAP,
+        message_delimiter = 1 << ZS_DELIMITER
+    };
+
+    //  A message. Caution: Don't change the body of the message once you've
+    //  copied it - the behaviour is undefined. Don't change the body of the
+    //  received message either - other threads may be accessing it in parallel.
+
+    class message_t : private zs_msg
+    {
+        friend class socket_t;
+
+    public:
+
+        //  Creates message size_ bytes long.
+        inline message_t (size_t size_ = 0)
+        {
+            int rc = zs_msg_init_size (this, size_);
+            assert (rc == 0);
+        }
+
+        //  Creates message from the supplied buffer. 0MQ takes care of
+        //  deallocating the buffer once it is not needed. The deallocation
+        //  function is supplied in ffn_ parameter. If ffn_ is NULL, no
+        //  deallocation happens - this is useful for sending static buffers.
+        inline message_t (void *data_, size_t size_, 
+            free_fn *ffn_)
+        {
+            int rc = zs_msg_init_data (this, data_, size_, ffn_);
+            assert (rc == 0);
+        }
+
+        //  Destroys the message.
+        inline ~message_t ()
+        {
+            int rc = zs_msg_close (this);
+            assert (rc == 0);
+        }
+
+        //  Destroys old content of the message and allocates buffer for the
+        //  new message body. Having this as a separate function allows user
+        //  to reuse once-allocated message for multiple times.
+        inline void rebuild (size_t size_)
+        {
+            int rc = zs_msg_close (this);
+            assert (rc == 0);
+            rc = zs_msg_init_size (this, size_);
+            assert (rc == 0);
+        }
+
+        //  Same as above, however, the message is rebuilt from the supplied
+        //  buffer. See appropriate constructor for discussion of buffer
+        //  deallocation mechanism.
+        inline void rebuild (void *data_, size_t size_, free_fn *ffn_)
+        {
+            int rc = zs_msg_close (this);
+            assert (rc == 0);
+            rc = zs_msg_init_data (this, data_, size_, ffn_);
+            assert (rc == 0);
+        }
+
+        //  Moves the message content from one message to the another. If the
+        //  destination message have contained data prior to the operation
+        //  these get deallocated. The source message will contain 0 bytes
+        //  of data after the operation.
+        inline void move_to (message_t *msg_)
+        {
+            int rc = zs_msg_move (this, (zs_msg*) msg_);
+            assert (rc == 0);
+        }
+
+        //  Copies the message content from one message to the another. If the
+        //  destination message have contained data prior to the operation
+        //  these get deallocated.
+        inline void copy_to (message_t *msg_)
+        {
+            int rc = zs_msg_copy (this, (zs_msg*) msg_);
+            assert (rc == 0);
+        }
+
+        //  Returns message type.
+        inline message_type_t type ()
+        {
+            return (message_type_t) (1 << zs_msg_type (this));
+        }
+
+        //  Returns pointer to message's data buffer.
+        inline void *data ()
+        {
+            return zs_msg_data (this);
+        }
+
+        //  Returns the size of message data buffer.
+        inline size_t size ()
+        {
+            return zs_msg_size (this);
+        }
+
+    private:
+
+        //  Disable implicit message copying, so that users won't use shared
+        //  messages (less efficient) without being aware of the fact.
+        message_t (const message_t&);
+        void operator = (const message_t&);
+    };
+
+    class context_t
+    {
+        friend class socket_t;
+
+    public:
+
+        inline context_t (int app_threads_, int io_threads_)
+        {
+            ptr = zs_init (app_threads_, io_threads_);
+            assert (ptr);
+        }
+
+        inline ~context_t ()
+        {
+            int rc = zs_term (ptr);
+            assert (rc == 0);
+        }
+
+    private:
+
+        void *ptr;
+
+        //  Disable copying.
+        context_t (const context_t&);
+        void operator = (const context_t&);
+    };
+
+    class socket_t
+    {
+    public:
+
+        inline socket_t (context_t &context_, int type_ = 0)
+        {
+            ptr = zs_socket (context_.ptr, type_);
+            assert (ptr);
+        }
+
+        inline ~socket_t ()
+        {
+            int rc = zs_close (ptr);
+            assert (rc == 0);
+        }
+
+        inline void bind (const char *addr_, zs_opts *opts_ = NULL)
+        {
+            int rc = zs_bind (ptr, addr_, opts_);
+            assert (rc == 0);
+        }
+
+        inline void connect (const char *addr_, zs_opts *opts_ = NULL)
+        {
+            int rc = zs_connect (ptr, addr_, opts_);
+            assert (rc == 0);
+        }
+
+        inline void subscribe (const char *criteria_)
+        {
+            int rc = zs_subscribe (ptr, criteria_);
+            assert (rc == 0);
+        }
+
+        inline void send (message_t &msg_, int flags_ = 0)
+        {
+            int rc = zs_send (ptr, &msg_, flags_);
+            assert (rc == 0);
+        }
+
+        inline void flush ()
+        {
+            int rc = zs_flush (ptr);
+            assert (rc == 0);
+        }
+
+        inline void recv (message_t *msg_, int flags_ = 0)
+        {
+            int rc = zs_recv (ptr, msg_, flags_);
+            assert (rc == 0);
+        }
+
+    private:
+
+        void *ptr;
+
+        //  Disable copying.
+        socket_t (const socket_t&);
+        void operator = (const socket_t&);
+    };
+
+}
+
+#endif

src/Makefile.am

+lib_LTLIBRARIES = libzs.la
+
+libzs_la_SOURCES = \
+    app_thread.hpp \
+    atomic_bitmap.hpp \
+    atomic_counter.hpp \
+    atomic_ptr.hpp \
+    command.hpp \
+    config.hpp \
+    connecter.hpp \
+    data_distributor.hpp \
+    decoder.hpp \
+    devpoll.hpp \
+    dispatcher.hpp \
+    dummy_aggregator.hpp \
+    dummy_distributor.hpp \
+    encoder.hpp \
+    epoll.hpp \
+    err.hpp \
+    fair_aggregator.hpp \
+    fd.hpp \
+    fd_signaler.hpp \
+    io_object.hpp \
+    io_thread.hpp \
+    ip.hpp \
+    i_api.hpp \
+    i_demux.hpp \
+    i_mux.hpp \
+    i_poller.hpp \
+    i_poll_events.hpp \
+    i_session.hpp \
+    i_signaler.hpp \
+    i_engine.hpp \
+    i_thread.hpp \
+    listener.hpp \
+    kqueue.hpp \
+    load_balancer.hpp \
+    msg.hpp \
+    mutex.hpp \
+    object.hpp \
+    p2p.hpp \
+    pipe.hpp \
+    pipe_reader.hpp \
+    pipe_writer.hpp \
+    platform.hpp \
+    poll.hpp \
+    pub.hpp \
+    rep.hpp \
+    req.hpp \
+    safe_object.hpp \
+    select.hpp \
+    session.hpp \
+    session_stub.hpp \
+    simple_semaphore.hpp \
+    socket_base.hpp \
+    sub.hpp \
+    stdint.hpp \
+    tcp_connecter.hpp \
+    tcp_listener.hpp \
+    tcp_socket.hpp \
+    thread.hpp \
+    uuid.hpp \
+    windows.hpp \
+    wire.hpp \
+    ypipe.hpp \
+    ypollset.hpp \
+    yqueue.hpp \
+    zmq_decoder.hpp \
+    zmq_encoder.hpp \
+    zmq_tcp_engine.hpp \
+    app_thread.cpp \
+    connecter.cpp \
+    data_distributor.cpp \
+    devpoll.hpp \
+    dispatcher.cpp \
+    dummy_aggregator.cpp \
+    dummy_distributor.cpp \
+    epoll.cpp \
+    err.cpp \
+    fair_aggregator.cpp \
+    fd_signaler.cpp \
+    io_object.cpp \
+    io_thread.cpp \
+    ip.cpp \
+    kqueue.cpp \
+    listener.cpp \
+    load_balancer.cpp \
+    object.cpp \
+    p2p.cpp \
+    pipe.cpp \
+    pipe_reader.cpp \
+    pipe_writer.cpp \
+    poll.cpp \
+    pub.cpp \
+    rep.cpp \
+    req.cpp \
+    safe_object.cpp \
+    select.cpp \
+    session.cpp \
+    session_stub.cpp \
+    socket_base.cpp \
+    sub.cpp \
+    tcp_connecter.cpp \
+    tcp_listener.cpp \
+    tcp_socket.cpp \
+    thread.cpp \
+    uuid.cpp \
+    ypollset.cpp \
+    zmq_decoder.cpp \
+    zmq_encoder.cpp \
+    zmq_tcp_engine.cpp \
+    zs.cpp
+
+libzs_la_LDFLAGS = -version-info 0:0:0
+libzs_la_CXXFLAGS = -Wall -pedantic -Werror @ZS_EXTRA_CXXFLAGS@
+
+dist-hook:
+		-rm $(distdir)/src/platform.hpp
+
+

src/app_thread.cpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "../include/zs.h"
+
+#if defined ZS_HAVE_WINDOWS
+#include "windows.hpp"
+#else
+#include <unistd.h>
+#endif
+
+#include "app_thread.hpp"
+#include "dispatcher.hpp"
+#include "err.hpp"
+#include "session.hpp"
+#include "pipe.hpp"
+#include "config.hpp"
+#include "i_api.hpp"
+#include "dummy_aggregator.hpp"
+#include "fair_aggregator.hpp"
+#include "dummy_distributor.hpp"
+#include "data_distributor.hpp"
+#include "load_balancer.hpp"
+#include "p2p.hpp"
+#include "pub.hpp"
+#include "sub.hpp"
+#include "req.hpp"
+#include "rep.hpp"
+
+//  If the RDTSC is available we use it to prevent excessive
+//  polling for commands. The nice thing here is that it will work on any
+//  system with x86 architecture and gcc or MSVC compiler.
+#if (defined __GNUC__ && (defined __i386__ || defined __x86_64__)) ||\
+    (defined _MSC_VER && (defined _M_IX86 || defined _M_X64))
+#define ZS_DELAY_COMMANDS
+#endif
+
+zs::app_thread_t::app_thread_t (dispatcher_t *dispatcher_, int thread_slot_) :
+    object_t (dispatcher_, thread_slot_),
+    tid (0),
+    last_processing_time (0)
+{
+}
+
+void zs::app_thread_t::shutdown ()
+{
+    //  Deallocate all the sessions associated with the thread.
+    while (!sessions.empty ())
+        sessions [0]->shutdown ();
+
+    delete this;
+}
+
+zs::app_thread_t::~app_thread_t ()
+{
+}
+
+void zs::app_thread_t::attach_session (session_t *session_)
+{
+    session_->set_index (sessions.size ());
+    sessions.push_back (session_); 
+}
+
+void zs::app_thread_t::detach_session (session_t *session_)
+{
+    //  O(1) removal of the session from the list.
+    sessions_t::size_type i = session_->get_index ();
+    sessions [i] = sessions [sessions.size () - 1];
+    sessions [i]->set_index (i);
+    sessions.pop_back ();
+}
+
+zs::i_poller *zs::app_thread_t::get_poller ()
+{
+    zs_assert (false);
+}
+
+zs::i_signaler *zs::app_thread_t::get_signaler ()
+{
+    return &pollset;
+}
+
+bool zs::app_thread_t::is_current ()
+{
+    return !sessions.empty () && tid == getpid ();
+}
+
+bool zs::app_thread_t::make_current ()
+{
+    //  If there are object managed by this slot we cannot assign the slot
+    //  to a different thread.
+    if (!sessions.empty ())
+        return false;
+
+    tid = getpid ();
+    return true;
+}
+
+zs::i_api *zs::app_thread_t::create_socket (int type_)
+{
+    i_mux *mux = NULL;
+    i_demux *demux = NULL;
+    session_t *session = NULL;
+    i_api *api = NULL;
+
+    switch (type_) {
+    case ZS_P2P:
+        mux = new dummy_aggregator_t;
+        zs_assert (mux);
+        demux = new dummy_distributor_t;
+        zs_assert (demux);
+        session = new session_t (this, this, mux, demux, true, false);
+        zs_assert (session);
+        api = new p2p_t (this, session);
+        zs_assert (api);
+        break;
+    case ZS_PUB:
+        demux = new data_distributor_t;
+        zs_assert (demux);
+        session = new session_t (this, this, mux, demux, true, false);
+        zs_assert (session);
+        api = new pub_t (this, session);
+        zs_assert (api);
+        break;
+    case ZS_SUB:
+        mux = new fair_aggregator_t;
+        zs_assert (mux);
+        session = new session_t (this, this, mux, demux, true, false);
+        zs_assert (session);
+        api = new sub_t (this, session);
+        zs_assert (api);
+        break;
+    case ZS_REQ:
+        //  TODO
+        zs_assert (false);
+        api = new req_t (this, session);
+        zs_assert (api);
+        break;
+    case ZS_REP:
+        //  TODO
+        zs_assert (false);
+        api = new rep_t (this, session);
+        zs_assert (api);
+        break;
+    default:
+        errno = EINVAL;
+        return NULL;
+    }
+
+    attach_session (session);
+    
+    return api;
+}
+
+void zs::app_thread_t::process_commands (bool block_)
+{
+    ypollset_t::signals_t signals;
+    if (block_)
+        signals = pollset.poll ();
+    else {
+
+#if defined ZS_DELAY_COMMANDS
+        //  Optimised version of command processing - it doesn't have to check
+        //  for incoming commands each time. It does so only if certain time
+        //  elapsed since last command processing. Command delay varies
+        //  depending on CPU speed: It's ~1ms on 3GHz CPU, ~2ms on 1.5GHz CPU
+        //  etc. The optimisation makes sense only on platforms where getting
+        //  a timestamp is a very cheap operation (tens of nanoseconds).
+
+        //  Get timestamp counter.
+#if defined __GNUC__
+        uint32_t low;
+        uint32_t high;
+        __asm__ volatile ("rdtsc" : "=a" (low), "=d" (high));
+        uint64_t current_time = (uint64_t) high << 32 | low;
+#elif defined _MSC_VER
+        uint64_t current_time = __rdtsc ();
+#else
+#error
+#endif
+
+        //  Check whether certain time have elapsed since last command
+        //  processing.
+        if (current_time - last_processing_time <= max_command_delay)
+            return;
+        last_processing_time = current_time;
+#endif
+
+        //  Check whether there are any commands pending for this thread.
+        signals = pollset.check ();
+    }
+
+    if (signals) {
+
+        //  Traverse all the possible sources of commands and process
+        //  all the commands from all of them.
+        for (int i = 0; i != thread_slot_count (); i++) {
+            if (signals & (ypollset_t::signals_t (1) << i)) {
+                command_t cmd;
+                while (dispatcher->read (i, get_thread_slot (), &cmd))
+                    cmd.destination->process_command (cmd);
+            }
+        }
+    }
+}

src/app_thread.hpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef __ZS_APP_THREAD_HPP_INCLUDED__
+#define __ZS_APP_THREAD_HPP_INCLUDED__
+
+#include <vector>
+
+#include "i_thread.hpp"
+#include "stdint.hpp"
+#include "object.hpp"
+#include "ypollset.hpp"
+
+namespace zs
+{
+
+    class app_thread_t : public object_t, public i_thread
+    {
+    public:
+
+        app_thread_t (class dispatcher_t *dispatcher_, int thread_slot_);
+
+        //  To be called when the whole infrastrucure is being closed (zs_term).
+        void shutdown ();
+
+        //  Returns signaler associated with this application thread.
+        i_signaler *get_signaler ();
+
+        //  Create socket engine in this thread. Return false if the calling
+        //  thread doesn't match the thread handled by this app thread object.
+        struct i_api *create_socket (int type_);
+
+        //  Nota bene: The following two functions are accessed from different
+        //  threads. The caller (dispatcher) is responsible for synchronisation
+        //  of accesses.
+
+        //  Returns true is current thread is associated with the app thread.
+        bool is_current ();
+
+        //  Tries to associate current thread with the app thread object.
+        //  Returns true is successfull, false otherwise.
+        bool make_current ();
+
+        //  Processes commands sent to this thread (if any). If 'block' is
+        //  set to true, returns only after at least one command was processed.
+        void process_commands (bool block_);
+
+        //  i_thread implementation.
+        void attach_session (class session_t *session_);
+        void detach_session (class session_t *session_);
+        struct i_poller *get_poller ();
+
+    private:
+
+        //  Clean-up.
+        ~app_thread_t ();
+
+        //  Thread ID associated with this slot.
+        //  TODO: Virtualise pid_t!
+        //  TODO: Check whether getpid returns unique ID for each thread.
+        int tid;
+
+        //  Vector of all sessionss associated with this app thread.
+        typedef std::vector <class session_t*> sessions_t;
+        sessions_t sessions;
+
+        //  App thread's signaler object.
+        ypollset_t pollset;
+
+        //  Timestamp of when commands were processed the last time.
+        uint64_t last_processing_time;
+
+        app_thread_t (const app_thread_t&);
+        void operator = (const app_thread_t&);
+    };
+
+}
+
+#endif

src/atomic.hpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef __ZS_ATOMIC_HPP_INCLUDED__
+#define __ZS_ATOMIC_HPP_INCLUDED__
+
+#include "stdint.hpp"
+
+#if defined ZS_FORCE_MUTEXES
+#define ZS_ATOMIC_MUTEX
+#elif (defined __i386__ || defined __x86_64__) && defined __GNUC__
+#define ZS_ATOMIC_X86
+#elif defined ZMQ_HAVE_WINDOWS
+#define ZS_ATOMIC_WINDOWS
+#elif defined ZMQ_HAVE_SOLARIS
+#define ZS_ATOMIC_SOLARIS
+#else
+#define ZS_ATOMIC_MUTEX
+#endif
+
+namespace zs
+{
+
+    //  Atomic assignement.
+    inline void atomic_uint32_set (volatile uint32_t *p_, uint32_t value_)
+    {
+        *p_ = value_;
+        //  StoreLoad memory barrier should go here on platforms with
+        //  memory models that require it.
+    }
+
+    //  Atomic retrieval of an integer.
+    inline uint32_t atomic_uint32_get (volatile uint32_t *p_)
+    {
+        //  StoreLoad memory barrier should go here on platforms with
+        //  memory models that require it.
+        return *p_;
+    }
+
+    //  Atomic addition. Returns the old value.
+    inline uint32_t atomic_uint32_add (volatile uint32_t *p_, uint32_t delta_)
+    {
+#if defined ZS_ATOMIC_WINDOWS
+        return InterlockedExchangeAdd ((LONG*) &value, increment_);
+#elif defined ZS_ATOMIC_SOLARIS
+        return atomic_add_32_nv (&value, increment_) - delta_;
+#elif defined ZS_ATOMIC_X86
+        uint32_t old;
+        __asm__ volatile (
+            "lock; xadd %0, %1\n\t"
+            : "=r" (old), "=m" (*p_)
+            : "0" (delta_), "m" (*p_)
+            : "cc", "memory");
+        return old;
+#else
+#error // TODO: 
+        sync.lock ();
+        uint32_t old = *p_;
+        *p_ += delta_;
+        sync.unlock ();
+#endif
+    }
+
+    //  Atomic subtraction. Returns the old value.
+    inline uint32_t atomic_uint32_sub (volatile uint32_t *p_, uint32_t delta_)
+    {
+#if defined ZS_ATOMIC_WINDOWS
+        LONG delta = - ((LONG) delta_);
+        return InterlockedExchangeAdd ((LONG*) &value, delta);
+#elif defined ZS_ATOMIC_SOLARIS
+        int32_t delta = - ((int32_t) delta_);
+        return atomic_add_32_nv (&value, delta) + delta_;
+#elif defined ZS_ATOMIC_X86
+        uint32_t old = -delta_;
+        __asm__ volatile ("lock; xaddl %0,%1"
+            : "=r" (old), "=m" (*p_)
+            : "0" (old), "m" (*p_)
+            : "cc");
+        return old;
+#else
+#error // TODO: 
+        sync.lock ();
+        uint32_t old = *p_;
+        *p_ -= delta_;
+        sync.unlock ();
+        return old;
+#endif
+    }
+
+    //  Atomic assignement.
+    template <typename T>
+    inline void atomic_ptr_set (volatile T **p_, T *value_)
+    {
+        *p_ = value_;
+        //  StoreLoad memory barrier should go here on platforms with
+        //  memory models that require it.
+    }
+
+    //  Perform atomic 'exchange pointers' operation. Old value is returned.
+    template <typename T>
+    inline void *atomic_ptr_xchg (volatile T **p_, T *value_)
+    {
+#if defined ZS_ATOMIC_WINDOWS
+        return InterlockedExchangePointer (p_, value_);
+#elif defined ZS_ATOMIC_SOLARIS
+        return atomic_swap_ptr (p_, value_);
+#elif defined ZS_ATOMIC_X86
+        void *old;
+        __asm__ volatile (
+            "lock; xchg %0, %2"
+            : "=r" (old), "=m" (*p_)
+            : "m" (*p_), "0" (value_));
+        return old;
+#else
+#error //  TODO:
+        sync.lock ();
+        void *old = *p_;
+        *p_ = value_;
+        sync.unlock ();
+        return old;
+#endif
+    }
+
+    //  Perform atomic 'compare and swap' operation on the pointer.
+    //  The pointer is compared to 'cmp' argument and if they are
+    //  equal, its value is set to 'value'. Old value of the pointer
+    //  is returned.
+    template <typename T>
+    inline void *atomic_ptr_cas (volatile T **p_, T *cmp_, T *value_)
+    {
+#if defined ZS_ATOMIC_WINDOWS
+        return InterlockedCompareExchangePointer (p_, value_, cmp_);
+#elif defined ZS_ATOMIC_SOLARIS
+        return atomic_cas_ptr (p_, cmp_, value_);
+#elif defined ZS_ATOMIC_X86
+        void *old;
+        __asm__ volatile (
+            "lock; cmpxchg %2, %3"
+            : "=a" (old), "=m" (*p_)
+            : "r" (value_), "m" (*p_), "0" (cmp_)
+            : "cc");
+        return old;
+#else
+#error //  TODO:
+        sync.lock ();
+        void *old = *p_;
+        if (old == cmp_)
+            *p_ = value_;
+        sync.unlock ();
+        return old;
+#endif
+    }
+
+#if defined ZS_ATOMIC_X86 && defined __x86_64__
+    typedef uint64_t atomic_bitmap_t;
+#else
+    typedef uint32_t atomic_bitmap_t;
+#endif
+
+    //  Atomic assignement.
+    inline void atomic_bitmap_set (volatile atomic_bitmap_t *p_,
+        atomic_bitmap_t value_)
+    {
+        *p_ = value_;
+        //  StoreLoad memory barrier should go here on platforms with
+        //  memory models that require it.
+    }
+
+    //  Bit-test-set-and-reset. Sets one bit of the value and resets
+    //  another one. Returns the original value of the reset bit.
+    inline bool atomic_bitmap_btsr (volatile atomic_bitmap_t *p_,
+        int set_index_, int reset_index_)
+    {
+#if defined ZS_ATOMIC_WINDOWS
+        while (true) {
+            atomic_bitmap_t oldval = *p_;
+            atomic_bitmap_t newval = (oldval | (atomic_bitmap_t (1) <<
+                set_index_)) & ~(integer_t (1) << reset_index_);
+            if (InterlockedCompareExchange ((volatile LONG*) p_, newval,
+                  oldval) == (LONG) oldval)
+                return (oldval & (atomic_bitmap_t (1) << reset_index_)) ?
+                    true : false; 
+        }
+#elif defined ZS_ATOMIC_SOLARIS
+        while (true) {
+            atomic_bitmap_t oldval = *p_;
+            atomic_bitmap_t newval = (oldval | (atomic_bitmap_t (1) <<
+                set_index_)) & ~(integer_t (1) << reset_index_);
+            if (atomic_cas_32 (p_, oldval, newval) == oldval)
+                return (oldval & (atomic_bitmap_t (1) << reset_index_)) ?
+                    true : false; 
+        }
+#elif defined ZS_ATOMIC_X86
+        atomic_bitmap_t oldval, dummy;
+        __asm__ volatile (
+            "mov %0, %1\n\t"
+            "1:"
+            "mov %1, %2\n\t"
+            "bts %3, %2\n\t"
+            "btr %4, %2\n\t"
+            "lock cmpxchg %2, %0\n\t"
+            "jnz 1b\n\t"
+            : "+m" (*p_), "=&a" (oldval), "=&r" (dummy)
+            : "r" (atomic_bitmap_t (set_index_)),
+                "r" (atomic_bitmap_t (reset_index_))
+            : "cc");
+        return (bool) (oldval & (atomic_bitmap_t (1) << reset_index_)); 
+#else
+#error // TODO:
+        sync.lock ();
+        atomic_bitmap_t oldval = *p_;
+        *p_ = (oldval | (atomic_bitmap_t (1) << set_index_)) &
+            ~(atomic_bitmap_t (1) << reset_index_);
+        sync.unlock ();
+        return (oldval & (atomic_bitmap_t (1) << reset_index_)) ? true : false;
+#endif
+    }
+
+    //  Sets value to newval. Returns the original value.
+    inline atomic_bitmap_t atomic_bitmap_xchg (volatile atomic_bitmap_t *p_,
+        atomic_bitmap_t newval_)
+    {
+#if defined ZS_ATOMIC_WINDOWS
+        return InterlockedExchange ((volatile LONG*) p_, newval_);
+#elif defined ZS_ATOMIC_SOLARIS
+        return atomic_swap_32 (p_, newval_);
+#elif defined ZS_ATOMIC_X86
+        atomic_bitmap_t oldval = newval_;
+        __asm__ volatile (
+            "lock; xchg %0, %1"
+            : "=r" (oldval)
+            : "m" (*p_), "0" (oldval)
+            : "memory");
+        return oldval; 
+#else
+#error //  TODO:
+        sync.lock ();
+        atomic_bitmap_t oldval = *p_;
+        *p_ = newval_;
+        sync.unlock ();
+#endif
+    }
+
+    //  izte is "if-zero-then-else" atomic operation - if the value is zero
+    //  it substitutes it by 'thenval' else it rewrites it by 'elseval'.
+    //  Original value of the integer is returned from this function.
+    inline atomic_bitmap_t atomic_bitmap_izte (volatile atomic_bitmap_t *p_,
+            atomic_bitmap_t thenval_,  atomic_bitmap_t elseval_)
+    {
+#if defined ZS_ATOMIC_WINDOWS
+        while (true) {
+            atomic_bitmap_t oldval = *p_;
+            atomic_bitmap_t newval = (oldval ? elseval_ : thenval_); 
+            if (InterlockedCompareExchange ((volatile LONG*) p_, newval,
+                  oldval) == (LONG) oldval)
+                return oldval;
+        }
+#elif defined ZS_ATOMIC_SOLARIS
+        while (true) {
+            atomic_bitmap_t oldval = *p_;
+            atomic_bitmap_t newval = (oldval ? elseval_ : thenval_); 
+            if (atomic_cas_32 (p_, oldval, newval) == oldval)
+                return oldval;
+        }
+#elif defined ZS_ATOMIC_X86
+        atomic_bitmap_t oldval;
+        atomic_bitmap_t dummy;
+        __asm__ volatile (
+            "mov %0, %1\n\t"
+            "1:"
+            "mov %3, %2\n\t"
+            "test %1, %1\n\t"
+            "jz 2f\n\t"
+            "mov %4, %2\n\t"
+            "2:"
+            "lock cmpxchg %2, %0\n\t"
+            "jnz 1b\n\t"
+            : "+m" (*p_), "=&a" (oldval), "=&r" (dummy)
+            : "r" (thenval_), "r" (elseval_)
+            : "cc");
+        return oldval;
+#else
+#error //  TODO:
+        sync.lock ();
+        atomic_bitmap_t oldval = *p_;
+        *p_ = oldval ? elseval_ : thenval_;
+        sync.unlock ();
+        return oldval;
+#endif
+    }
+
+}
+
+#endif

src/atomic_bitmap.hpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef __ZS_ATOMIC_BITMAP_HPP_INCLUDED__
+#define __ZS_ATOMIC_BITMAP_HPP_INCLUDED__
+
+#include "stdint.hpp"
+#include "platform.hpp"
+
+//  These are the conditions to choose between different implementations
+//  of atomic_bitmap.
+
+#if defined ZS_FORCE_MUTEXES
+#define ZS_ATOMIC_BITMAP_MUTEX
+#elif (defined __i386__ || defined __x86_64__) && defined __GNUC__
+#define ZS_ATOMIC_BITMAP_X86
+#elif 0 && defined __sparc__ && defined __GNUC__
+#define ZS_ATOMIC_BITMAP_SPARC
+#elif defined ZS_HAVE_WINDOWS
+#define ZS_ATOMIC_BITMAP_WINDOWS
+#elif defined ZS_HAVE_SOLARIS
+#define ZS_ATOMIC_BITMAP_SOLARIS
+#else
+#define ZS_ATOMIC_BITMAP_MUTEX
+#endif
+
+#if defined ZS_ATOMIC_BITMAP_MUTEX
+#include "mutex.hpp"
+#elif defined ZS_ATOMIC_BITMAP_WINDOWS
+#include "windows.hpp"
+#elif defined ZS_ATOMIC_BITMAP_SOLARIS
+#include <atomic.h>
+#endif
+
+namespace zs
+{
+
+    //  This class encapuslates several bitwise atomic operations on unsigned
+    //  integer. Selection of operations is driven specifically by the needs
+    //  of ypollset implementation.
+
+    class atomic_bitmap_t
+    {
+    public:
+
+#if (defined ZMQ_ATOMIC_BITMAP_X86 || defined ZMQ_FORCE_MUTEXES) \
+    && defined __x86_64__
+        typedef uint64_t bitmap_t;
+#else
+        typedef uint32_t bitmap_t;
+#endif
+
+        inline atomic_bitmap_t (bitmap_t value_ = 0) :
+            value (value_)
+        {
+        }
+
+        inline ~atomic_bitmap_t ()
+        {
+        }
+
+        //  Bit-test-set-and-reset. Sets one bit of the value and resets
+        //  another one. Returns the original value of the reset bit.
+        inline bool btsr (int set_index_, int reset_index_)
+        {
+#if defined ZS_ATOMIC_BITMAP_WINDOWS
+            while (true) {
+                bitmap_t oldval = value;
+                bitmap_t newval = (oldval | (bitmap_t (1) << set_index_)) &
+                    ~(bitmap_t (1) << reset_index_);
+                if (InterlockedCompareExchange ((volatile LONG*) &value, newval,
+                      oldval) == (LONG) oldval)
+                    return (oldval & (bitmap_t (1) << reset_index_)) ?
+                        true : false;
+            }
+#elif defined ZS_ATOMIC_BITMAP_SOLARIS
+            while (true) {
+                bitmap_t oldval = value;
+                bitmap_t newval = (oldval | (bitmap_t (1) << set_index_)) &
+                    ~(bitmap_t (1) << reset_index_);
+                if (atomic_cas_32 (&value, oldval, newval) == oldval)
+                    return (oldval & (bitmap_t (1) << reset_index_)) ?
+                        true : false;
+            }
+#elif defined ZS_ATOMIC_BITMAP_X86
+            bitmap_t oldval, dummy;
+            __asm__ volatile (
+                "mov %0, %1\n\t"
+                "1:"
+                "mov %1, %2\n\t"
+                "bts %3, %2\n\t"
+                "btr %4, %2\n\t"
+                "lock cmpxchg %2, %0\n\t"
+                "jnz 1b\n\t"
+                : "+m" (value), "=&a" (oldval), "=&r" (dummy)
+                : "r" (bitmap_t(set_index_)), "r" (bitmap_t(reset_index_))
+                : "cc");
+            return (bool) (oldval & (bitmap_t(1) << reset_index_));
+#elif defined ZS_ATOMIC_BITMAP_SPARC
+            volatile bitmap_t* valptr = &value;
+            bitmap_t set_val = bitmap_t(1) << set_index_;
+            bitmap_t reset_val = ~(bitmap_t(1) << reset_index_);
+            bitmap_t tmp;
+            bitmap_t oldval;
+            __asm__ volatile(
+                "ld       [%5], %2       \n\t"
+                "1:                      \n\t"
+                "or       %2, %0, %3     \n\t"
+                "and      %3, %1, %3     \n\t"
+                "cas      [%5], %2, %3   \n\t"
+                "cmp      %2, %3         \n\t"
+                "bne,a,pn %%icc, 1b      \n\t"
+                "mov      %3, %2         \n\t"
+                : "+r" (set_val), "+r" (reset_val), "=&r" (tmp),
+                  "=&r" (oldval), "+m" (*valptr)
+                : "r" (valptr)
+                : "cc");
+            return oldval;
+#elif defined ZS_ATOMIC_BITMAP_MUTEX
+            sync.lock ();
+            bitmap_t oldval = value;
+            value = (oldval | (bitmap_t (1) << set_index_)) &
+                ~(bitmap_t (1) << reset_index_);
+            sync.unlock ();
+            return (oldval & (bitmap_t (1) << reset_index_)) ? true : false;
+#else
+#error
+#endif
+        }
+
+        //  Sets value to newval. Returns the original value.
+        inline bitmap_t xchg (bitmap_t newval_)
+        {
+            bitmap_t oldval;
+#if defined ZS_ATOMIC_BITMAP_WINDOWS
+            oldval = InterlockedExchange ((volatile LONG*) &value, newval_);
+#elif defined ZS_ATOMIC_BITMAP_SOLARIS
+            oldval = atomic_swap_32 (&value, newval_);
+#elif defined ZS_ATOMIC_BITMAP_X86
+            oldval = newval_;
+            __asm__ volatile (
+                "lock; xchg %0, %1"
+                : "=r" (oldval)
+                : "m" (value), "0" (oldval)
+                : "memory");
+#elif defined ZS_ATOMIC_BITMAP_SPARC
+            oldval = value;
+            volatile bitmap_t* ptrin = &value;
+            bitmap_t tmp;
+            bitmap_t prev;
+            __asm__ __volatile__(
+                "ld [%4], %1\n\t"
+                "1:\n\t"
+                "mov %0, %2\n\t"
+                "cas [%4], %1, %2\n\t"
+                "cmp %1, %2\n\t"
+                "bne,a,pn %%icc, 1b\n\t"
+                "mov %2, %1\n\t"
+                : "+r" (newval_), "=&r" (tmp), "=&r" (prev), "+m" (*ptrin)
+                : "r" (ptrin)
+                : "cc");
+            return prev;
+#elif defined ZS_ATOMIC_BITMAP_MUTEX
+            sync.lock ();
+            oldval = value;
+            value = newval_;
+            sync.unlock ();
+#else
+#error
+#endif
+            return oldval;
+        }
+
+        //  izte is "if-zero-then-else" atomic operation - if the value is zero
+        //  it substitutes it by 'thenval' else it rewrites it by 'elseval'.
+        //  Original value of the integer is returned from this function.
+        inline bitmap_t izte (bitmap_t thenval_,
+            bitmap_t elseval_)
+        {
+#if defined ZS_ATOMIC_BITMAP_WINDOWS
+            while (true) {
+                bitmap_t oldval = value;
+                bitmap_t newval = oldval == 0 ? thenval_ : elseval_;
+                if (InterlockedCompareExchange ((volatile LONG*) &value,
+                      newval, oldval) == (LONG) oldval)
+                    return oldval;
+            }
+#elif defined ZS_ATOMIC_BITMAP_SOLARIS
+            while (true) {
+                bitmap_t oldval = value;
+                bitmap_t newval = oldval == 0 ? thenval_ : elseval_;
+                if (atomic_cas_32 (&value, oldval, newval) == oldval)
+                    return oldval;
+            }
+#elif defined ZS_ATOMIC_BITMAP_X86
+            bitmap_t oldval;
+            bitmap_t dummy;
+            __asm__ volatile (
+                "mov %0, %1\n\t"
+                "1:"
+                "mov %3, %2\n\t"
+                "test %1, %1\n\t"
+                "jz 2f\n\t"
+                "mov %4, %2\n\t"
+                "2:"
+                "lock cmpxchg %2, %0\n\t"
+                "jnz 1b\n\t"
+                : "+m" (value), "=&a" (oldval), "=&r" (dummy)
+                : "r" (thenval_), "r" (elseval_)
+                : "cc");
+            return oldval;
+#elif defined ZS_ATOMIC_BITMAP_SPARC
+            volatile bitmap_t* ptrin = &value;
+            bitmap_t tmp;
+            bitmap_t prev;
+            __asm__ __volatile__(
+                "ld      [%3], %0       \n\t"
+                "mov     0,    %1       \n\t"
+                "cas     [%3], %1, %4   \n\t"
+                "cmp     %0,   %1       \n\t"
+                "be,a,pn %%icc,1f       \n\t"
+                "ld      [%3], %0       \n\t"
+                "cas     [%3], %0, %5   \n\t"
+                "1:                     \n\t"
+                : "=&r" (tmp), "=&r" (prev), "+m" (*ptrin)
+                : "r" (ptrin), "r" (thenval_), "r" (elseval_)
+                : "cc");
+            return prev;
+#elif defined ZS_ATOMIC_BITMAP_MUTEX
+            sync.lock ();
+            bitmap_t oldval = value;
+            value = oldval ? elseval_ : thenval_;
+            sync.unlock ();
+            return oldval;
+#else
+#error
+#endif
+        }
+
+    private:
+
+        volatile bitmap_t value;
+#if defined ZS_ATOMIC_BITMAP_MUTEX
+        mutex_t sync;
+#endif
+
+        atomic_bitmap_t (const atomic_bitmap_t&);
+        void operator = (const atomic_bitmap_t&);
+    };
+
+}
+
+//  Remove macros local to this file.
+#if defined ZS_ATOMIC_BITMAP_WINDOWS
+#undef ZS_ATOMIC_BITMAP_WINDOWS
+#endif
+#if defined ZS_ATOMIC_BITMAP_SOLARIS
+#undef ZS_ATOMIC_BITMAP_SOLARIS
+#endif
+#if defined ZS_ATOMIC_BITMAP_X86
+#undef ZS_ATOMIC_BITMAP_X86
+#endif
+#if defined ZS_ATOMIC_BITMAP_SPARC
+#undef ZS_ATOMIC_BITMAP_SPARC
+#endif
+#if defined ZS_ATOMIC_BITMAP_MUTEX
+#undef ZS_ATOMIC_BITMAP_MUTEX
+#endif
+
+#endif

src/atomic_counter.hpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+
+#ifndef __ZS_ATOMIC_COUNTER_HPP_INCLUDED__
+#define __ZS_ATOMIC_COUNTER_HPP_INCLUDED__
+
+#include "stdint.hpp"
+#include "platform.hpp"
+
+#if defined ZS_FORCE_MUTEXES
+#define ZS_ATOMIC_COUNTER_MUTEX
+#elif (defined __i386__ || defined __x86_64__) && defined __GNUC__
+#define ZS_ATOMIC_COUNTER_X86
+#elif 0 && defined __sparc__ && defined __GNUC__
+#define ZS_ATOMIC_COUNTER_SPARC
+#elif defined ZS_HAVE_WINDOWS
+#define ZS_ATOMIC_COUNTER_WINDOWS
+#elif defined ZS_HAVE_SOLARIS
+#define ZS_ATOMIC_COUNTER_SOLARIS
+#else
+#define ZS_ATOMIC_COUNTER_MUTEX
+#endif
+
+#if defined ZS_ATOMIC_COUNTER_MUTEX
+#include "mutex.hpp"
+#elif defined ZS_ATOMIC_COUNTER_WINDOWS
+#include "windows.hpp"
+#elif defined ZS_ATOMIC_COUNTER_SOLARIS
+#include <atomic.h>
+#endif
+
+namespace zs
+{
+
+    //  This class represents an integer that can be incremented/decremented
+    //  in atomic fashion.
+
+    class atomic_counter_t
+    {
+    public:
+
+        typedef uint32_t integer_t;
+
+        inline atomic_counter_t (integer_t value_ = 0) :
+            value (value_)
+        {
+        }
+
+        inline ~atomic_counter_t ()
+        {
+        }
+
+        //  Set counter value (not thread-safe).
+        inline void set (integer_t value_)
+        {
+            value = value_;
+        }
+
+        //  Atomic addition. Returns the old value.
+        inline integer_t add (integer_t increment_)
+        {
+            integer_t old_value;
+
+#if defined ZS_ATOMIC_COUNTER_WINDOWS
+            old_value = InterlockedExchangeAdd ((LONG*) &value, increment_);
+#elif defined ZS_ATOMIC_COUNTER_SOLARIS
+            integer_t new_value = atomic_add_32_nv (&value, increment_);
+            old_value = new_value - increment_;
+#elif defined ZS_ATOMIC_COUNTER_X86
+            __asm__ volatile (
+                "lock; xadd %0, %1          \n\t"
+                : "=r" (old_value), "=m" (value)
+                : "0" (increment_), "m" (value)
+                : "cc", "memory");
+#elif defined ZS_ATOMIC_COUNTER_SPARC
+            integer_t tmp;
+            __asm__ volatile (
+                "ld [%4], %0                \n\t"
+                "1:                         \n\t"
+                "add %0, %3, %1             \n\t"
+                "cas [%4], %0, %1           \n\t"
+                "cmp %0, %1                 \n\t"
+                "bne,a,pn %%icc, 1b         \n\t"
+                "mov %1, %0                 \n\t"
+                : "=&r" (old_value), "=&r" (tmp), "=m" (value)
+                : "r" (increment_), "r" (&value)
+                : "cc", "memory");
+#elif defined ZS_ATOMIC_COUNTER_MUTEX
+            sync.lock ();
+            old_value = value;
+            value += increment_;
+            sync.unlock ();
+#else
+#error
+#endif
+            return old_value;
+        }
+
+        //  Atomic subtraction. Returns false if the counter drops to zero.
+        inline bool sub (integer_t decrement)
+        {
+#if defined ZS_ATOMIC_COUNTER_WINDOWS
+            LONG delta = - ((LONG) decrement);
+            integer_t old = InterlockedExchangeAdd ((LONG*) &value, delta);
+            return old - decrement != 0;
+#elif defined ZS_ATOMIC_COUNTER_SOLARIS
+            int32_t delta = - ((int32_t) decrement);
+            integer_t nv = atomic_add_32_nv (&value, delta);
+            return nv != 0;
+#elif defined ZS_ATOMIC_COUNTER_X86
+            integer_t oldval = -decrement;
+            volatile integer_t *val = &value;
+            __asm__ volatile ("lock; xaddl %0,%1"
+                : "=r" (oldval), "=m" (*val)
+                : "0" (oldval), "m" (*val)
+                : "cc");
+            return oldval != decrement;
+#elif defined ZS_ATOMIC_COUNTER_SPARC
+            volatile integer_t *val = &value;
+            integer_t tmp;
+            integer_t result;
+            __asm__ volatile(
+                "ld [%4], %1\n\t"
+                "1:\n\t"
+                "add %1, %0, %2\n\t"
+                "cas [%4], %1, %2\n\t"
+                "cmp %1, %2\n\t"
+                "bne,a,pn %%icc, 1b\n\t"
+                "mov %2, %1\n\t"
+                : "+r" (-decrement), "=&r" (tmp), "=&r" (result), "+m" (*val)
+                : "r" (val)
+                : "cc");
+            return result <= decrement;
+#elif defined ZS_ATOMIC_COUNTER_MUTEX
+            sync.lock ();
+            value -= decrement;
+            bool result = value ? true : false;
+            sync.unlock ();
+            return result;
+#else
+#error
+#endif
+        }
+
+        inline integer_t get ()
+        {
+            return value;
+        }
+
+    private:
+
+        volatile integer_t value;
+#if defined ZS_ATOMIC_COUNTER_MUTEX
+        mutex_t sync;
+#endif
+
+        atomic_counter_t (const atomic_counter_t&);
+        void operator = (const atomic_counter_t&);
+    };
+
+}
+
+//  Remove macros local to this file.
+#if defined ZS_ATOMIC_COUNTER_WINDOWS
+#undef ZS_ATOMIC_COUNTER_WINDOWS
+#endif
+#if defined ZS_ATOMIC_COUNTER_SOLARIS
+#undef ZS_ATOMIC_COUNTER_SOLARIS
+#endif
+#if defined ZS_ATOMIC_COUNTER_X86
+#undef ZS_ATOMIC_COUNTER_X86
+#endif
+#if defined ZS_ATOMIC_COUNTER_SPARC
+#undef ZS_ATOMIC_COUNTER_SPARC
+#endif
+#if defined ZS_ATOMIC_COUNTER_MUTEX
+#undef ZS_ATOMIC_COUNTER_MUTEX
+#endif
+
+#endif

src/atomic_ptr.hpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+
+#ifndef __ZS_ATOMIC_PTR_HPP_INCLUDED__
+#define __ZS_ATOMIC_PTR_HPP_INCLUDED__
+
+#include "platform.hpp"
+
+#if defined ZS_FORCE_MUTEXES
+#define ZS_ATOMIC_PTR_MUTEX
+#elif (defined __i386__ || defined __x86_64__) && defined __GNUC__
+#define ZS_ATOMIC_PTR_X86
+#elif 0 && defined __sparc__ && defined __GNUC__
+#define ZS_ATOMIC_PTR_SPARC
+#elif defined ZS_HAVE_WINDOWS
+#define ZS_ATOMIC_PTR_WINDOWS
+#elif defined ZS_HAVE_SOLARIS
+#define ZS_ATOMIC_PTR_SOLARIS
+#else
+#define ZS_ATOMIC_PTR_MUTEX
+#endif
+
+#if defined ZS_ATOMIC_PTR_MUTEX
+#include "mutex.hpp"
+#elif defined ZS_ATOMIC_PTR_WINDOWS
+#include "windows.hpp"
+#elif defined ZS_ATOMIC_PTR_SOLARIS
+#include <atomic.h>
+#endif
+
+namespace zs
+{
+
+    //  This class encapsulates several atomic operations on pointers.
+
+    template <typename T> class atomic_ptr_t
+    {
+    public:
+
+        //  Initialise atomic pointer
+        inline atomic_ptr_t ()
+        {
+            ptr = NULL;
+        }
+
+        //  Destroy atomic pointer
+        inline ~atomic_ptr_t ()
+        {
+        }
+
+        //  Set value of atomic pointer in a non-threadsafe way
+        //  Use this function only when you are sure that at most one
+        //  thread is accessing the pointer at the moment.
+        inline void set (T *ptr_)
+        {
+            this->ptr = ptr_;
+        }
+
+        //  Perform atomic 'exchange pointers' operation. Pointer is set
+        //  to the 'val' value. Old value is returned.
+        inline T *xchg (T *val_)
+        {
+#if defined ZS_ATOMIC_PTR_WINDOWS
+            return (T*) InterlockedExchangePointer (&ptr, val_);
+#elif defined ZS_ATOMIC_PTR_SOLARIS
+            return (T*) atomic_swap_ptr (&ptr, val_);
+#elif defined ZS_ATOMIC_PTR_X86
+            T *old;
+            __asm__ volatile (
+                "lock; xchg %0, %2"
+                : "=r" (old), "=m" (ptr)
+                : "m" (ptr), "0" (val_));
+            return old;
+#elif defined ZS_ATOMIC_PTR_SPARC
+            T* newptr = val_;
+            volatile T** ptrin = &ptr;
+            T* tmp;
+            T* prev;
+            __asm__ __volatile__(
+                "ld [%4], %1\n\t"
+                "1:\n\t"
+                "mov %0, %2\n\t"
+                "cas [%4], %1, %2\n\t"
+                "cmp %1, %2\n\t"
+                "bne,a,pn %%icc, 1b\n\t"
+                "mov %2, %1\n\t"
+                : "+r" (newptr), "=&r" (tmp), "=&r" (prev), "+m" (*ptrin)
+                : "r" (ptrin)
+                : "cc");
+            return prev;
+#elif defined ZS_ATOMIC_PTR_MUTEX
+            sync.lock ();
+            T *old = (T*) ptr;
+            ptr = val_;
+            sync.unlock ();
+            return old;
+#else
+#error
+#endif
+        }
+
+        //  Perform atomic 'compare and swap' operation on the pointer.
+        //  The pointer is compared to 'cmp' argument and if they are
+        //  equal, its value is set to 'val'. Old value of the pointer
+        //  is returned.
+        inline T *cas (T *cmp_, T *val_)
+        {
+#if defined ZS_ATOMIC_PTR_WINDOWS
+            return (T*) InterlockedCompareExchangePointer (
+                (volatile PVOID*) &ptr, val_, cmp_);
+#elif defined ZS_ATOMIC_PTR_SOLARIS
+            return (T*) atomic_cas_ptr (&ptr, cmp_, val_);
+#elif defined ZS_ATOMIC_PTR_X86
+            T *old;
+            __asm__ volatile (
+                "lock; cmpxchg %2, %3"
+                : "=a" (old), "=m" (ptr)
+                : "r" (val_), "m" (ptr), "0" (cmp_)
+                : "cc");
+            return old;
+#elif defined ZS_ATOMIC_PTR_SPARC
+            volatile T** ptrin = &ptr;
+            volatile T* prev = ptr;
+            __asm__ __volatile__(
+                "cas [%3], %1, %2\n\t"
+                : "+m" (*ptrin)
+                : "r" (cmp_), "r" (val_), "r" (ptrin)
+                : "cc");
+            return prev;
+#elif defined ZS_ATOMIC_PTR_MUTEX
+            sync.lock ();
+            T *old = (T*) ptr;
+            if (ptr == cmp_)
+                ptr = val_;
+            sync.unlock ();
+            return old;
+#else
+#error
+#endif
+        }
+
+    private:
+
+        volatile T *ptr;
+#if defined ZS_ATOMIC_PTR_MUTEX
+        mutex_t sync;
+#endif
+
+        atomic_ptr_t (const atomic_ptr_t&);
+        void operator = (const atomic_ptr_t&);
+    };
+
+}
+
+//  Remove macros local to this file.
+#if defined ZS_ATOMIC_PTR_WINDOWS
+#undef ZS_ATOMIC_PTR_WINDOWS
+#endif
+#if defined ZS_ATOMIC_PTR_SOLARIS
+#undef ZS_ATOMIC_PTR_SOLARIS
+#endif
+#if defined ZS_ATOMIC_PTR_X86
+#undef ZS_ATOMIC_PTR_X86
+#endif
+#if defined ZS_ATOMIC_PTR_SPARC
+#undef ZS_ATOMIC_PTR_SPARC
+#endif
+#if defined ZS_ATOMIC_PTR_MUTEX
+#undef ZS_ATOMIC_PTR_MUTEX
+#endif
+
+#endif

src/command.hpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef __ZS_COMMAND_HPP_INCLUDED__
+#define __ZS_COMMAND_HPP_INCLUDED__
+
+#include "stdint.hpp"
+
+namespace zs
+{
+
+    //  This structure defines the commands that can be sent between threads.
+
+    struct command_t
+    {
+        //  Object to process the command.
+        class object_t *destination;
+
+        enum type_t
+        {
+            stop,
+            bind,
+            head,
+            tail,
+            reg,
+            reg_and_bind,
+            unreg,
+            engine,
+            terminate,
+            terminate_ack
+        } type;
+
+        union {
+
+            struct {
+            } stop;
+
+            struct {
+                class pipe_reader_t *reader;
+                class session_t *peer;
+            } bind;
+
+            struct {
+                uint64_t bytes;
+            } tail;
+
+            struct {
+                uint64_t bytes;
+            } head;
+
+            struct {
+                class simple_semaphore_t *smph;
+            } reg;
+
+            struct {
+                class session_t *peer;
+                bool flow_in;
+                bool flow_out;
+            } reg_and_bind;
+
+            struct {
+                class simple_semaphore_t *smph;
+            } unreg;
+
+            //  TODO: Engine object won't be deallocated on terminal shutdown
+            //  while the command is still on the fly!
+            struct {
+                class i_engine *engine;
+            } engine;
+
+            struct {
+            } terminate;
+
+            struct {
+            } terminate_ack;
+
+        } args;
+    };
+
+}    
+
+#endif

src/config.hpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef __ZS_CONFIG_HPP_INCLUDED__
+#define __ZS_CONFIG_HPP_INCLUDED__
+
+namespace zs
+{
+
+    //  Compile-time settings.
+
+    enum 
+    {
+
+        //  Number of new messages in message pipe needed to trigger new memory
+        //  allocation. Setting this parameter to 256 decreases the impact of
+        //  memory allocation by approximately 99.6%
+        message_pipe_granularity = 256,
+
+        //  Number of new commands in command pipe needed to trigger new memory
+        //  allocation. The number should  be kept low to decrease the memory
+        //  footprint of dispatcher.
+        command_pipe_granularity = 4,
+
+        //  Maximal batching size for engines with receiving functionality.
+        //  So, if there are 10 messages that fit into the batch size, all of
+        //  them may be read by a single 'recv' system call, thus avoiding
+        //  unnecessary network stack traversals.
+        in_batch_size = 8192,
+
+        //  Maximal batching size for engines with sending functionality.
+        //  So, if there are 10 messages that fit into the batch size, all of
+        //  them may be written by a single 'send' system call, thus avoiding
+        //  unnecessary network stack traversals.
+        out_batch_size = 8192,        
+
+        //  Maximum number of events the I/O thread can process in one go.
+        max_io_events = 256,
+
+        //  Maximal wait time for a timer (milliseconds).
+        max_timer_period = 100,
+
+        //  Maximal delay to process command in API thread (in CPU ticks).
+        //  3,000,000 ticks equals to 1 - 2 milliseconds on current CPUs.
+        max_command_delay = 3000000,
+
+        //  Maximal number of non-accepted connections that can be held by
+        //  TCP listener object.
+        tcp_connection_backlog = 10
+
+    };
+
+}
+
+#endif

src/connecter.cpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "connecter.hpp"
+#include "io_thread.hpp"
+#include "session.hpp"
+#include "err.hpp"
+#include "simple_semaphore.hpp"
+#include "zmq_tcp_engine.hpp"
+
+zs::connecter_t::connecter_t (io_thread_t *thread_, const char *addr_,
+      session_t *session_) :
+    io_object_t (thread_),
+    state (idle),
+    poller (NULL),
+    session (session_),
+    addr (addr_),
+    identity ("abcde"),
+    engine (NULL)
+{
+}
+
+void zs::connecter_t::terminate ()
+{
+    delete this;
+}
+
+void zs::connecter_t::shutdown ()
+{
+    delete this;
+}
+
+zs::connecter_t::~connecter_t ()
+{
+}
+
+void zs::connecter_t::process_reg (simple_semaphore_t *smph_)
+{
+    //  Fet poller pointer for further use.
+    zs_assert (!poller);
+    poller = get_poller ();
+
+    //  Ask the session to register itself with the I/O thread. Note that
+    //  the session is living in the same I/O thread, thus this results
+    //  in a synchronous call.
+    session->inc_seqnum ();
+    send_reg (session, NULL);
+
+    //  Unlock the application thread that created the connecter.
+    if (smph_)
+        smph_->post ();
+
+    //  Manually trigger timer event which will launch asynchronous connect.
+    state = waiting;
+    timer_event ();
+}
+
+void zs::connecter_t::process_unreg (simple_semaphore_t *smph_)
+{
+    //  Unregister connecter/engine from the poller.
+    zs_assert (poller);
+    if (state == connecting)
+        poller->rm_fd (handle);
+    else if (state == waiting)
+        poller->cancel_timer (this);
+    else if (state == sending)
+        engine->terminate ();
+
+    //  Unlock the application thread closing the connecter.
+    if (smph_)
+        smph_->post ();
+}
+
+void zs::connecter_t::in_event ()
+{
+    //  Error occured in asynchronous connect. Retry to connect after a while.
+    if (state == connecting) {
+        fd_t fd = tcp_connecter.connect ();
+        zs_assert (fd == retired_fd);
+        poller->rm_fd (handle);
+        poller->add_timer (this);
+        state = waiting;
+        return;
+    }
+
+    zs_assert (false);
+}
+
+void zs::connecter_t::out_event ()
+{
+    if (state == connecting) {
+
+        fd_t fd = tcp_connecter.connect ();
+        if (fd == retired_fd) {
+            poller->rm_fd (handle);
+            poller->add_timer (this);
+            state = waiting;
+            return;
+        }
+
+        poller->rm_fd (handle);
+        engine = new zmq_tcp_engine_t (fd);
+        zs_assert (engine);
+        engine->attach (poller, this);
+        state = sending;
+        return;
+    }
+
+    zs_assert (false);
+}
+
+void zs::connecter_t::timer_event ()
+{
+    zs_assert (state == waiting);
+
+    //  Initiate async connect and start polling for its completion. If async
+    //  connect fails instantly, try to reconnect after a while.
+    int rc = tcp_connecter.open (addr.c_str ());
+    if (rc == 0) {
+        state = connecting;
+        in_event ();
+    }
+    else if (rc == 1) {
+        handle = poller->add_fd (tcp_connecter.get_fd (), this);
+        poller->set_pollout (handle);
+        state = connecting;
+    }
+    else {
+        poller->add_timer (this);
+        state = waiting;
+    }
+}
+
+void zs::connecter_t::set_engine (struct i_engine *engine_)
+{
+    engine = engine_;
+}
+
+bool zs::connecter_t::read (zs_msg *msg_)
+{
+    zs_assert (state == sending);
+
+    //  Deallocate old content of the message just in case.
+    zs_msg_close (msg_);
+
+    //  Send the identity.
+    zs_msg_init_size (msg_, identity.size ());
+    memcpy (zs_msg_data (msg_), identity.c_str (), identity.size ());
+
+    //  Ask engine to unregister from the poller.
+    i_engine *e = engine;
+    engine->detach ();
+
+    //  Attach the engine to the session. (Note that this is actually
+    //  a synchronous call.
+    session->inc_seqnum ();
+    send_engine (session, e);
+
+    state = idle;
+
+    return true;    
+}
+
+bool zs::connecter_t::write (struct zs_msg *msg_)
+{
+    //  No incoming messages are accepted till identity is sent.
+    return false;
+}
+
+void zs::connecter_t::flush ()
+{
+    //  No incoming messages are accepted till identity is sent.
+}

src/connecter.hpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef __ZS_CONNECTER_HPP_INCLUDED__
+#define __ZS_CONNECTER_HPP_INCLUDED__
+
+#include <string>
+
+#include "../include/zs.h"
+
+#include "i_poller.hpp"
+#include "io_object.hpp"
+#include "i_poll_events.hpp"
+#include "i_session.hpp"
+#include "tcp_connecter.hpp"
+
+namespace zs
+{
+
+    class connecter_t : public io_object_t, public i_poll_events,
+        public i_session
+    {
+    public:
+
+        connecter_t (class io_thread_t *thread_, const char *addr_,
+            class session_t *session_);
+
+        void terminate ();
+        void shutdown ();
+
+        void process_reg (class simple_semaphore_t *smph_);
+        void process_unreg (class simple_semaphore_t *smph_);
+
+        //  i_poll_events implementation.
+        void in_event ();
+        void out_event ();
+        void timer_event ();
+
+        //  i_session implementation
+        void set_engine (struct i_engine *engine_);
+        //  void shutdown ();
+        bool read (struct zs_msg *msg_);
+        bool write (struct zs_msg *msg_);
+        void flush ();
+
+    private:
+
+        //  Clean-up.
+        ~connecter_t ();
+
+         enum {
+             idle,
+             waiting,
+             connecting,
+             sending
+         } state;
+
+        //  Cached pointer to the poller.
+        struct i_poller *poller;
+
+        //  Handle of the connecting socket.
+        handle_t handle;
+
+        //  Associated session. It lives in the same I/O thread.
+        class session_t *session;
+
+        //  Address to connect to.
+        std::string addr;
+
+        //  Identity of the connection.
+        std::string identity;
+
+        tcp_connecter_t tcp_connecter;
+
+        struct i_engine *engine;
+
+        connecter_t (const connecter_t&);
+        void operator = (const connecter_t&);
+    };
+
+}
+
+#endif

src/data_distributor.cpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "../include/zs.h"
+
+#include "data_distributor.hpp"
+#include "pipe_writer.hpp"
+#include "err.hpp"
+#include "session.hpp"
+#include "msg.hpp"
+
+zs::data_distributor_t::data_distributor_t () :
+    session (NULL)
+{
+}
+
+void zs::data_distributor_t::set_session (session_t *session_)
+{
+    zs_assert (!session);
+    session = session_;
+}
+
+void zs::data_distributor_t::shutdown ()
+{
+    //  No need to deallocate pipes here. They'll be deallocated during the
+    //  shutdown of the dispatcher.
+    delete this;
+}
+
+void zs::data_distributor_t::terminate ()
+{
+    //  Pipe unregisters itself during the call to terminate, so the pipes
+    //  list shinks by one in each iteration.
+    while (!pipes.empty ())
+        pipes [0]->terminate ();
+
+   delete this;
+}
+
+zs::data_distributor_t::~data_distributor_t ()
+{
+}
+
+void zs::data_distributor_t::attach_pipe (pipe_writer_t *pipe_)
+{
+    //  Associate demux with a new pipe.
+    pipe_->set_demux (this);
+    pipe_->set_index (pipes.size ());
+    pipes.push_back (pipe_);
+}
+
+void zs::data_distributor_t::detach_pipe (pipe_writer_t *pipe_)
+{
+    //  Release the reference to the pipe.
+    int index = pipe_->get_index ();
+    pipe_->set_index (-1);
+    pipes [index] = pipes.back ();
+    pipes [index]->set_index (index);
+    pipes.pop_back ();
+}
+
+bool zs::data_distributor_t::empty ()
+{
+    return pipes.empty ();
+}
+
+bool zs::data_distributor_t::send (zs_msg *msg_)
+{
+    int pipes_count = pipes.size ();
+
+    //  If there are no pipes available, simply drop the message.
+    if (pipes_count == 0) {
+        zs_msg_close (msg_);
+        zs_msg_init (msg_);
+        return true;
+    }
+
+    //  TODO: ???
+    //  First check whether all pipes are available for writing.
+//    for (pipes_t::iterator it = pipes.begin (); it != pipes.end (); it ++)
+//        if (!(*it)->check_write (msg_))
+//            return false;
+
+    //  For VSMs the copying is straighforward.
+    if (msg_->content == (zs_msg_content*) ZS_VSM) {
+        for (pipes_t::iterator it = pipes.begin (); it != pipes.end (); it ++)
+            write_to_pipe (*it, msg_);
+        zs_msg_init (msg_);
+        return true;
+    }
+
+    //  Optimisation for the case when there's only a single pipe
+    //  to send the message to - no refcount adjustment (i.e. atomic
+    //  operations) needed.
+    if (pipes_count == 1) {
+        write_to_pipe (*pipes.begin (), msg_);
+        zs_msg_init (msg_);
+        return true;
+    }
+
+    //  There are at least 2 destinations for the message. That means we have
+    //  to deal with reference counting. First add N-1 references to
+    //  the content (we are holding one reference anyway, that's why the -1).
+    if (msg_->shared)
+        msg_->content->refcnt.add (pipes_count - 1);
+    else {
+        msg_->shared = true;
+        // TODO: Add memory barrier here.
+        msg_->content->refcnt.set (pipes_count);
+    }
+
+    //  Push the message to all destinations.
+    for (pipes_t::iterator it = pipes.begin (); it != pipes.end (); it ++)
+        write_to_pipe (*it, msg_);
+
+    //  Detach the original message from the data buffer.
+    zs_msg_init (msg_);
+
+    return true;
+}
+
+void zs::data_distributor_t::flush ()
+{
+    //  Flush all pipes. If there's large number of pipes, it can be pretty
+    //  inefficient (especially if there's new message only in a single pipe).
+    //  Can it be improved?
+    for (pipes_t::iterator it = pipes.begin (); it != pipes.end (); it ++)
+        (*it)->flush ();
+}
+
+void zs::data_distributor_t::write_to_pipe (class pipe_writer_t *pipe_,
+    struct zs_msg *msg_)
+{
+    if (!pipe_->write (msg_)) {
+        //  TODO: Push gap notification to the pipe.
+        zs_assert (false);
+    }
+}
+

src/data_distributor.hpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef __ZS_DATA_DISTRIBUTOR_HPP_INCLUDED__
+#define __ZS_DATA_DISTRIBUTOR_HPP_INCLUDED__
+
+#include <vector>
+
+#include <i_demux.hpp>
+
+namespace zs
+{
+
+    //  Object to distribute messages to outbound pipes.
+
+    class data_distributor_t : public i_demux
+    {
+    public:
+
+        data_distributor_t ();
+
+        //  i_demux implementation.
+        void set_session (class session_t *session_);
+        void shutdown ();
+        void terminate ();
+        void attach_pipe (class pipe_writer_t *pipe_);
+        void detach_pipe (class pipe_writer_t *pipe_);
+        bool empty ();
+        bool send (struct zs_msg *msg_);
+        void flush ();
+
+    private:
+
+        //  Clean-up.
+        ~data_distributor_t ();
+
+        //  Reference to the owner session object.
+        class session_t *session;
+
+        //  Writes the message to the pipe if possible. If it isn't, writes
+        //  a gap notification to the pipe.
+        void write_to_pipe (class pipe_writer_t *pipe_, struct zs_msg *msg_);
+
+        //  The list of outbound pipes.
+        typedef std::vector <class pipe_writer_t*> pipes_t;
+        pipes_t pipes;
+
+        data_distributor_t (const data_distributor_t&);
+        void operator = (const data_distributor_t&);
+    };
+
+}
+
+#endif

src/decoder.hpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef __ZS_DECODER_HPP_INCLUDED__
+#define __ZS_DECODER_HPP_INCLUDED__
+
+#include <stddef.h>
+#include <string.h>
+#include <algorithm>
+
+namespace zs
+{
+
+    //  Helper base class for decoders that know the amount of data to read
+    //  in advance at any moment. Knowing the amount in advance is a property
+    //  of the protocol used. Both AMQP and backend protocol are based on
+    //  size-prefixed paradigm, therefore they are using decoder_t to parse
+    //  the messages. On the other hand, XML-based transports (like XMPP or
+    //  SOAP) don't allow for knowing the size of data to read in advance and
+    //  should use different decoding algorithms.
+    //
+    //  Decoder implements the state machine that parses the incoming buffer.
+    //  Derived class should implement individual state machine actions.
+
+    template <typename T> class decoder_t
+    {
+    public:
+
+        inline decoder_t () :
+            read_ptr (NULL),
+            to_read (0),
+            next (NULL)
+        {
+        }
+
+        //  Push the binary data to the decoder. Returns number of bytes
+        // actually parsed.
+        inline size_t write (unsigned char *data_, size_t size_)
+        {
+            size_t pos = 0;
+            while (true) {
+                size_t to_copy = std::min (to_read, size_ - pos);
+                if (read_ptr) {
+                    memcpy (read_ptr, data_ + pos, to_copy);
+                    read_ptr += to_copy;
+                }
+                pos += to_copy;
+                to_read -= to_copy;
+                while (!to_read)
+                    if (!(static_cast <T*> (this)->*next) ())
+                        return pos;
+                if (pos == size_)
+                    return pos;
+            }
+        }
+
+    protected:
+
+        //  Prototype of state machine action. Action should return false if
+        //  it is unable to push the data to the system.
+        typedef bool (T::*step_t) ();
+
+        //  This function should be called from derived class to read data
+        //  from the buffer and schedule next state machine action.
+        inline void next_step (void *read_ptr_, size_t to_read_,
+            step_t next_)
+        {
+            read_ptr = (unsigned char*) read_ptr_;
+            to_read = to_read_;
+            next = next_;
+        }
+
+    private:
+
+        unsigned char *read_ptr;
+        size_t to_read;
+        step_t next;
+
+        decoder_t (const decoder_t&);
+        void operator = (const decoder_t&);
+    };
+
+}
+
+#endif

src/devpoll.cpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "platform.hpp"
+
+#if defined ZS_HAVE_SOLARIS || defined ZS_HAVE_HPUX
+
+#include <sys/devpoll.h>
+#include <sys/time.h>
+#include <sys/resource.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/ioctl.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <limits.h>
+
+#include "devpoll.hpp"
+#include "err.hpp"
+#include "config.hpp"
+
+zs::devpoll_t::devpoll_t ()
+{
+    //  Get limit on open files
+    struct rlimit rl;
+    int rc = getrlimit (RLIMIT_NOFILE, &rl);
+    errno_assert (rc != -1);
+    fd_table.resize (rl.rlim_cur);
+
+    for (rlim_t i = 0; i < rl.rlim_cur; i ++)
+        fd_table [i].valid = false;
+
+    devpoll_fd = open ("/dev/poll", O_RDWR);
+    errno_assert (devpoll_fd != -1);
+}
+
+zs::devpoll_t::~devpoll_t ()
+{
+    close (devpoll_fd);
+}
+
+void zs::devpoll_t::devpoll_ctl (fd_t fd_, short events_)
+{
+    struct pollfd pfd = {fd_, events_, 0};
+    ssize_t rc = write (devpoll_fd, &pfd, sizeof pfd);
+    zs_assert (rc == sizeof pfd);
+}
+
+zs::handle_t zs::devpoll_t::add_fd (fd_t fd_, i_poll_events *reactor_)
+{
+    assert (!fd_table [fd_].valid);
+
+    fd_table [fd_].events = 0;
+    fd_table [fd_].reactor = reactor_;
+    fd_table [fd_].valid = true;
+    fd_table [fd_].accepted = false;
+
+    devpoll_ctl (fd_, 0);
+    pending_list.push_back (fd_);
+
+    //  Increase the load metric of the thread.
+    load.add (1);
+
+    handle_t handle;
+    handle.fd = fd_;
+    return handle;
+}
+
+void zs::devpoll_t::rm_fd (handle_t handle_)
+{
+    assert (fd_table [handle_.fd].valid);
+
+    devpoll_ctl (handle_.fd, POLLREMOVE);
+    fd_table [handle_.fd].valid = false;
+
+    //  Decrease the load metric of the thread.
+    load.sub (1);
+}
+
+void zs::devpoll_t::set_pollin (handle_t handle_)
+{
+    fd_t fd = handle_.fd;
+    devpoll_ctl (fd, POLLREMOVE);
+    fd_table [fd].events |= POLLIN;
+    devpoll_ctl (fd, fd_table [fd].events);
+}
+
+void zs::devpoll_t::reset_pollin (handle_t handle_)
+{
+    fd_t fd = handle_.fd;
+    devpoll_ctl (fd, POLLREMOVE);
+    fd_table [fd].events &= ~((short) POLLIN);
+    devpoll_ctl (fd, fd_table [fd].events);
+}
+
+void zs::devpoll_t::set_pollout (handle_t handle_)
+{
+    fd_t fd = handle_.fd;
+    devpoll_ctl (fd, POLLREMOVE);
+    fd_table [fd].events |= POLLOUT;
+    devpoll_ctl (fd, fd_table [fd].events);
+}
+
+void zs::devpoll_t::reset_pollout (handle_t handle_)
+{
+    fd_t fd = handle_.fd;
+    devpoll_ctl (fd, POLLREMOVE);
+    fd_table [fd].events &= ~((short) POLLOUT);
+    devpoll_ctl (fd, fd_table [fd].events);
+}
+
+void zs::devpoll_t::add_timer (i_poll_events *events_)
+{
+     timers.push_back (events_);
+}
+
+void zs::devpoll_t::cancel_timer (i_poll_events *events_)
+{
+    timers_t::iterator it = std::find (timers.begin (), timers.end (), events_);
+    if (it != timers.end ())
+        timers.erase (it);
+}
+
+int zs::devpoll_t::get_load ()
+{
+    return load.get ();
+}
+
+void zs::devpoll_t::start ()
+{
+    worker.start (worker_routine, this);
+}
+
+void zs::devpoll_t::stop ()
+{
+    stopping = true;
+}
+
+void zs::devpoll_t::join ()
+{
+    worker.stop ();
+}
+
+bool zs::devpoll_t::loop ()
+{
+    //  According to the poll(7d) man page, we can retrieve
+    //  no more then (OPEN_MAX - 1) events.
+    int nfds = std::min (max_io_events, OPEN_MAX - 1);
+
+    while (!stopping) {
+
+        struct pollfd ev_buf [max_io_events];
+        struct dvpoll poll_req;
+
+        for (pending_list_t::size_type i = 0; i < pending_list.size (); i ++)
+            fd_table [pending_list [i]].accepted = true;
+        pending_list.clear ();
+
+        poll_req.dp_fds = &ev_buf [0];
+        poll_req.dp_nfds = nfds;
+        poll_req.dp_timeout = timers.empty () ? -1 : max_timer_period;
+
+        //  Wait for events.
+        int n = ioctl (devpoll_fd, DP_POLL, &poll_req);
+        if (n == -1 && errno == EINTR)
+            continue;
+        errno_assert (n != -1);
+
+        //  Handle timer.
+        if (!n) {
+
+            //  Use local list of timers as timer handlers may fill new timers
+            //  into the original array.
+            timers_t t;
+            std::swap (timers, t);
+
+            //  Trigger all the timers.
+            for (timers_t::iterator it = t.begin (); it != t.end (); it ++)
+                (*it)->timer_event ();
+
+            continue;
+        }
+
+        for (int i = 0; i < n; i ++) {
+
+            fd_entry_t *fd_ptr = &fd_table [ev_buf [i].fd];
+            if (!fd_ptr->valid || !fd_ptr->accepted)
+                continue;
+            if (ev_buf [i].revents & (POLLERR | POLLHUP))
+                fd_ptr->reactor->in_event ();
+            if (!fd_ptr->valid || !fd_ptr->accepted)
+                continue;
+            if (ev_buf [i].revents & POLLOUT)
+                fd_ptr->reactor->out_event ();
+            if (!fd_ptr->valid || !fd_ptr->accepted)
+                continue;
+            if (ev_buf [i].revents & POLLIN)
+                fd_ptr->reactor->in_event ();
+        }
+    }
+}
+
+void zs::devpoll_t::worker_routine (void *arg_)
+{
+    ((devpoll_t*) arg_)->loop ();
+}
+
+#endif

src/devpoll.hpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef __ZS_DEVPOLL_HPP_INCLUDED__
+#define __ZS_DEVPOLL_HPP_INCLUDED__
+
+#include "platform.hpp"
+
+#if defined ZS_HAVE_SOLARIS || ZS_HAVE_HPUX
+
+#include <vector>
+
+#include "i_poller.hpp"
+#include "fd.hpp"
+#include "thread.hpp"
+#include "atomic_counter.hpp"
+
+namespace zs
+{
+
+    //  Implements socket polling mechanism using the Solaris-specific
+    //  "/dev/poll" interface.
+
+    class devpoll_t : public i_poller
+    {
+    public:
+
+        devpoll_t ();
+        virtual ~devpoll_t ();
+
+        //  i_poller implementation.
+        handle_t add_fd (fd_t fd_, i_poll_events *events_);
+        void rm_fd (handle_t handle_);
+        void set_pollin (handle_t handle_);
+        void reset_pollin (handle_t handle_);
+        void set_pollout (handle_t handle_);
+        void reset_pollout (handle_t handle_);
+        void add_timer (i_poll_events *events_);
+        void cancel_timer (i_poll_events *events_);
+        int get_load ();
+        void start ();
+        void stop ();
+        void join ();
+
+    private:
+
+        //  Main worker thread routine.
+        static void worker_routine (void *arg_);
+
+        //  Main event loop.
+        void loop ();
+
+        //  File descriptor referring to "/dev/poll" pseudo-device.
+        fd_t devpoll_fd;
+
+        struct fd_entry_t
+        {
+            short events;
+            struct i_poll_events *reactor;
+            bool valid;
+            bool accepted;
+        };
+
+        std::vector <fd_entry_t> fd_table;
+
+        typedef std::vector <fd_t> pending_list_t;
+        pending_list_t pending_list;
+
+        //  Pollset manipulation function.
+        void devpoll_ctl (fd_t fd_, short events_);
+
+        //  List of all the engines waiting for the timer event.
+        typedef std::vector <struct i_poll_events*> timers_t;
+        timers_t timers;
+
+        //  If true, thread is in the process of shutting down.
+        bool stopping;
+
+        //  Handle of the physical thread doing the I/O work.
+        thread_t worker;
+
+        //  Load of the poller. Currently number of file descriptors
+        //  registered with the poller.
+        atomic_counter_t load;
+
+        devpoll_t (const devpoll_t&);
+        void operator = (const devpoll_t&);
+    };
+
+}
+
+#endif
+
+#endif

src/dispatcher.cpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "../include/zs.h"
+
+#include "dispatcher.hpp"
+#include "app_thread.hpp"
+#include "io_thread.hpp"
+#include "platform.hpp"
+#include "err.hpp"
+#include "pipe.hpp"
+#include "pipe_reader.hpp"
+#include "pipe_writer.hpp"
+#include "session.hpp"
+#include "i_api.hpp"
+
+#if defined ZS_HAVE_WINDOWS
+#include "windows.h"
+#endif
+
+zs::dispatcher_t::dispatcher_t (int app_threads_, int io_threads_)
+{
+#ifdef ZS_HAVE_WINDOWS
+    //  Intialise Windows sockets. Note that WSAStartup can be called multiple
+    //  times given that WSACleanup will be called for each WSAStartup.
+    WORD version_requested = MAKEWORD (2, 2);
+    WSADATA wsa_data;
+    int rc = WSAStartup (version_requested, &wsa_data);
+    zs_assert (rc == 0);
+    zs_assert (LOBYTE (wsa_data.wVersion) == 2 &&
+        HIBYTE (wsa_data.wVersion) == 2);
+#endif
+
+    //  Create application thread proxies.
+    for (int i = 0; i != app_threads_; i++) {
+        app_thread_t *app_thread = new app_thread_t (this, i);
+        zs_assert (app_thread);
+        app_threads.push_back (app_thread);
+        signalers.push_back (app_thread->get_signaler ());
+    }
+
+    //  Create I/O thread objects.
+    for (int i = 0; i != io_threads_; i++) {
+        io_thread_t *io_thread = new io_thread_t (this, i + app_threads_);
+        zs_assert (io_thread);
+        io_threads.push_back (io_thread);
+        signalers.push_back (io_thread->get_signaler ());
+    }
+
+    //  Create command pipe matrix.
+    command_pipes = new command_pipe_t [signalers.size () * signalers.size ()];
+    zs_assert (command_pipes);
+
+    //  Launch I/O threads.
+    for (int i = 0; i != io_threads_; i++)
+        io_threads [i]->start ();
+}
+
+void zs::dispatcher_t::shutdown ()
+{
+     delete this;
+}
+
+zs::dispatcher_t::~dispatcher_t ()
+{
+    //  Ask I/O threads to terminate.
+    for (io_threads_t::size_type i = 0; i != io_threads.size (); i++)
+        io_threads [i]->stop ();
+
+    //  Wait till I/O threads actually terminate.
+    for (io_threads_t::size_type i = 0; i != io_threads.size (); i++)
+        io_threads [i]->join ();
+
+    //  At this point the current thread is the only thread with access to
+    //  our internal data. Deallocation will be done exclusively in this thread.
+    for (app_threads_t::size_type i = 0; i != app_threads.size (); i++)
+        app_threads [i]->shutdown ();
+    for (io_threads_t::size_type i = 0; i != io_threads.size (); i++)
+        io_threads [i]->shutdown ();
+
+    delete [] command_pipes;
+
+    //  Deallocate all the pipes, pipe readers and pipe writers.
+    for (pipes_t::iterator it = pipes.begin (); it != pipes.end (); it++) {
+        delete it->pipe;
+        delete it->reader;
+        delete it->writer;
+    }
+
+#ifdef ZS_HAVE_WINDOWS
+    //  On Windows, uninitialise socket layer.
+    int rc = WSACleanup ();
+    wsa_assert (rc != SOCKET_ERROR);
+#endif
+}
+
+int zs::dispatcher_t::thread_slot_count ()
+{
+    return signalers.size ();
+}
+
+zs::i_api *zs::dispatcher_t::create_socket (int type_)
+{
+    threads_sync.lock ();
+    app_thread_t *thread = choose_app_thread ();
+    if (!thread) {
+        threads_sync.unlock ();
+        return NULL;
+    }
+    i_api *s = thread->create_socket (type_);
+    threads_sync.unlock ();
+    return s;
+}
+
+zs::app_thread_t *zs::dispatcher_t::choose_app_thread ()
+{
+    //  Check whether thread ID is already assigned. If so, return it.
+    for (app_threads_t::size_type i = 0; i != app_threads.size (); i++)
+        if (app_threads [i]->is_current ())
+            return app_threads [i];
+
+    //  Check whether there's an unused thread slot in the dispatcher.
+    for (app_threads_t::size_type i = 0; i != app_threads.size (); i++)
+        if (app_threads [i]->make_current ())
+            return app_threads [i];
+
+    //  Thread limit was exceeded.
+    errno = EMFILE;
+    return NULL;
+}
+
+zs::io_thread_t *zs::dispatcher_t::choose_io_thread (uint64_t taskset_)
+{
+    zs_assert (io_threads.size () > 0);
+
+    //  Find the I/O thread with minimum load.
+    int min_load = io_threads [0]->get_load ();
+    io_threads_t::size_type result = 0;
+    for (io_threads_t::size_type i = 1; i != io_threads.size (); i++) {
+        if (!taskset_ || (taskset_ & (uint64_t (1) << i))) {
+            int load = io_threads [i]->get_load ();
+            if (load < min_load) {
+                min_load = load;
+                result = i;
+            }
+        }
+    }
+
+    return io_threads [result];
+}
+
+void zs::dispatcher_t::create_pipe (object_t *reader_parent_,
+    object_t *writer_parent_, uint64_t hwm_, uint64_t lwm_,
+    pipe_reader_t **reader_, pipe_writer_t **writer_)
+{
+    //  Create the pipe, reader & writer triple.
+    pipe_t *pipe = new pipe_t;
+    zs_assert (pipe);
+    pipe_reader_t *reader = new pipe_reader_t (reader_parent_, pipe,
+        hwm_, lwm_);
+    zs_assert (reader);
+    pipe_writer_t *writer = new pipe_writer_t (writer_parent_, pipe, reader,
+        hwm_, lwm_);
+    zs_assert (writer);
+    reader->set_peer (writer);
+
+    //  Store the pipe in the repository.
+    pipe_info_t info = {pipe, reader, writer};
+    pipes_sync.lock ();
+    pipe->set_index (pipes.size ());
+    pipes.push_back (info);
+    pipes_sync.unlock ();
+
+    *reader_ = reader;
+    *writer_ = writer;
+}
+
+void zs::dispatcher_t::destroy_pipe (pipe_t *pipe_)
+{
+    //  Remove the pipe from the repository.
+    pipe_info_t info;
+    pipes_sync.lock ();
+    pipes_t::size_type i = pipe_->get_index ();
+    info = pipes [i];
+    pipes [i] = pipes.back ();
+    pipes.pop_back ();
+    pipes_sync.unlock ();
+
+    //  Deallocate the pipe and associated pipe reader & pipe writer.
+    zs_assert (info.pipe == pipe_);
+    delete info.pipe;
+    delete info.reader;
+    delete info.writer;
+}
+
+int zs::dispatcher_t::register_inproc_endpoint (const char *endpoint_,
+    session_t *session_)
+{
+   inproc_endpoint_sync.lock ();
+   inproc_endpoints_t::iterator it = inproc_endpoints.find (endpoint_);
+
+   if (it != inproc_endpoints.end ()) {
+       inproc_endpoint_sync.unlock ();
+       errno = EADDRINUSE;
+       return -1;
+   }
+
+   inproc_endpoints.insert (std::make_pair (endpoint_, session_));
+
+   inproc_endpoint_sync.unlock ();
+   return 0;
+}
+
+zs::object_t *zs::dispatcher_t::get_inproc_endpoint (const char *endpoint_)
+{
+    inproc_endpoint_sync.lock ();
+    inproc_endpoints_t::iterator it = inproc_endpoints.find (endpoint_);
+
+    if (it == inproc_endpoints.end ()) {
+        inproc_endpoint_sync.unlock ();
+        errno = EADDRNOTAVAIL;
+        return NULL;
+    }
+
+    it->second->inc_seqnum ();
+    object_t *session = it->second;
+
+    inproc_endpoint_sync.unlock ();
+    return session;
+}
+
+void zs::dispatcher_t::unregister_inproc_endpoints (session_t *session_)
+{
+    inproc_endpoint_sync.lock ();
+
+    //  Remove the connection from the repository.
+    //  TODO: Yes, the algorithm has O(n^2) complexity. Should be O(log n).
+    for (inproc_endpoints_t::iterator it = inproc_endpoints.begin ();
+          it != inproc_endpoints.end ();) {
+        if (it->second == session_) {
+            inproc_endpoints.erase (it);
+            it = inproc_endpoints.begin ();
+        }
+        else
+            it++;
+    }
+        
+    inproc_endpoint_sync.unlock ();
+}
+

src/dispatcher.hpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef __ZS_DISPATCHER_HPP_INCLUDED__
+#define __ZS_DISPATCHER_HPP_INCLUDED__
+
+#include <vector>
+#include <map>
+#include <string>
+
+#include "i_signaler.hpp"
+#include "ypipe.hpp"
+#include "command.hpp"
+#include "config.hpp"
+#include "mutex.hpp"
+#include "stdint.hpp"
+
+namespace zs
+{
+
+    //  Dispatcher implements bidirectional thread-safe passing of commands
+    //  between N threads. It consists of a ypipes to pass commands and
+    //  signalers to wake up the receiver thread when new commands are
+    //  available. Note that dispatcher is inefficient for passing messages
+    //  within a thread (sender thread = receiver thread). The optimisation is
+    //  not part of the class and should be implemented by individual threads
+    //  (presumably by calling the command handling function directly).
+    
+    class dispatcher_t
+    {
+    public:
+
+        //  Create the dispatcher object. Matrix of pipes to communicate between
+        //  each socket and each I/O thread is created along with appropriate
+        //  signalers.
+        dispatcher_t (int app_threads_, int io_threads_);
+
+        //  To be called to terminate the whole infrastructure (zs_term).
+        void shutdown ();
+
+        //  Create a socket engine.
+        struct i_api *create_socket (int type_);
+
+        //  Returns number of thread slots in the dispatcher. To be used by
+        //  individual threads to find out how many distinct signals can be
+        //  received.
+        int thread_slot_count ();
+
+        //  Write command to the dispatcher.
+        inline void write (int source_, int destination_,
+            const command_t &command_)
+        {
+            command_pipe_t &pipe =
+                command_pipes [source_ * signalers.size () + destination_];
+            pipe.write (command_);
+            if (!pipe.flush ())
+                signalers [destination_]->signal (source_);
+        }
+
+        //  Read command from the dispatcher. Returns false if there is no
+        //  command available.
+        inline bool read (int source_,  int destination_, command_t *command_)
+        {
+            return command_pipes [source_ * signalers.size () +
+                destination_].read (command_);
+        }
+
+        //  Creates new pipe.
+        void create_pipe (class object_t *reader_parent_,
+            class object_t *writer_parent_, uint64_t hwm_, uint64_t lwm_,
+            class pipe_reader_t **reader_, class pipe_writer_t **writer_);
+
+        //  Deallocates the pipe.
+        void destroy_pipe (class pipe_t *pipe_);
+
+        //  Registers existing session object as an inproc endpoint.
+        int register_inproc_endpoint (const char *endpoint_,
+            class session_t *session_);
+
+        //  Retrieves an inproc endpoint. Increments the command sequence number
+        //  of the object by one. Caller is thus bound to send the command
+        //  to the connection after invoking this function. Returns NULL if
+        //  the endpoint doesn't exist.
+        class object_t *get_inproc_endpoint (const char *endpoint_);
+
+        //  Removes all the inproc endpoints associated with the given session
+        //  object from the global repository.
+        void unregister_inproc_endpoints (class session_t *session_);
+
+        //  Returns the I/O thread that is the least busy at the moment.
+        //  Taskset specifies which I/O threads are eligible (0 = all).
+        class io_thread_t *choose_io_thread (uint64_t taskset_);
+
+    private:
+
+        //  Clean-up.
+        ~dispatcher_t ();
+
+        //  Returns the app thread associated with the current thread.
+        //  NULL if we are out of app thread slots.
+        class app_thread_t *choose_app_thread ();
+
+        //  Application threads.
+        typedef std::vector <class app_thread_t*> app_threads_t;
+        app_threads_t app_threads;
+
+        //  I/O threads.
+        typedef std::vector <class io_thread_t*> io_threads_t;
+        io_threads_t io_threads;
+
+        //  Signalers for both application and I/O threads.
+        std::vector <i_signaler*> signalers;
+
+        //  Pipe to hold the commands.
+        typedef ypipe_t <command_t, true,
+            command_pipe_granularity> command_pipe_t;
+
+        //  NxN matrix of command pipes.
+        command_pipe_t *command_pipes;
+
+        //  Synchronisation of accesses to shared thread data.
+        mutex_t threads_sync;
+
+        //  Global repository of pipes. It's used only on terminal shutdown
+        //  to deallocate all the pipes irrespective of whether they are
+        //  referenced from pipe_reader, pipe_writer or both.
+        struct pipe_info_t
+        {
+            class pipe_t *pipe;
+            class pipe_reader_t *reader;
+            class pipe_writer_t *writer;
+        };
+        typedef std::vector <pipe_info_t> pipes_t;
+        pipes_t pipes;
+
+        //  Synchronisation of access to global repository of pipes.
+        mutex_t pipes_sync;
+
+        //  Global repository of available inproc endpoints.
+        typedef std::map <std::string, class session_t*> inproc_endpoints_t;
+        inproc_endpoints_t inproc_endpoints;
+
+        //  Synchronisation of access to the global repository
+        //  of inproc endpoints.
+        mutex_t inproc_endpoint_sync;
+
+        dispatcher_t (const dispatcher_t&);
+        void operator = (const dispatcher_t&);
+    };
+    
+}
+
+#endif
+

src/dummy_aggregator.cpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "../include/zs.h"
+
+#include "dummy_aggregator.hpp"
+#include "err.hpp"
+#include "pipe_reader.hpp"
+#include "session.hpp"
+
+//  Swaps pipes at specified indices. 
+#define swap_pipes(i1_, i2_) \
+    std::swap (pipes [i1_], pipes [i2_]);\
+    pipes [i1_]->set_index (i1_);\
+    pipes [i2_]->set_index (i2_);
+
+zs::dummy_aggregator_t::dummy_aggregator_t () :
+    session (NULL),
+    pipe (NULL),
+    active (false)
+{
+}
+
+void zs::dummy_aggregator_t::set_session (session_t *session_)
+{
+    zs_assert (!session);
+    session = session_;
+}
+
+void zs::dummy_aggregator_t::shutdown ()
+{
+    //  No need to deallocate the pipe here. It'll be deallocated during the
+    //  shutdown of the dispatcher.
+    delete this;
+}
+
+void zs::dummy_aggregator_t::terminate ()
+{
+    if (pipe)
+        pipe->terminate ();
+
+   delete this;
+}
+
+zs::dummy_aggregator_t::~dummy_aggregator_t ()
+{
+}
+
+void zs::dummy_aggregator_t::attach_pipe (pipe_reader_t *pipe_)
+{
+    zs_assert (!pipe);
+    pipe = pipe_;
+    active = true;
+
+    //  Associate new pipe with the mux object.
+    pipe_->set_mux (this);
+    session->revive ();
+}
+
+void zs::dummy_aggregator_t::detach_pipe (pipe_reader_t *pipe_)
+{
+    zs_assert (pipe == pipe_);
+    deactivate (pipe_);
+    pipe = NULL;
+}
+
+bool zs::dummy_aggregator_t::empty ()
+{
+    return pipe == NULL;
+}
+
+bool zs::dummy_aggregator_t::recv (zs_msg *msg_)
+{
+    //  Deallocate old content of the message.
+    zs_msg_close (msg_);
+        
+    //  Try to read from the pipe.
+    if (pipe && pipe->read (msg_))
+        return true;
+
+    //  No message is available. Initialise the output parameter
+    //  to be a 0-byte message.
+    zs_msg_init (msg_);
+    return false;
+}
+
+void zs::dummy_aggregator_t::deactivate (pipe_reader_t *pipe_)
+{
+    active = false;
+}
+
+void zs::dummy_aggregator_t::reactivate (pipe_reader_t *pipe_)
+{
+    active = true;
+}

src/dummy_aggregator.hpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef __ZS_DUMMY_AGGREGATOR_HPP_INCLUDED__
+#define __ZS_DUMMY_AGGREGATOR_HPP_INCLUDED__
+
+#include <vector>
+
+#include "i_mux.hpp"
+
+namespace zs
+{
+
+    //  Fake message aggregator. There can be at most one pipe bound to it,
+    //  so there's no real aggregation going on. However, it is more efficient
+    //  than a real aggregator. It's intended to be used in the contexts
+    //  where business logic ensures there'll be at most one pipe bound.
+
+    class dummy_aggregator_t : public i_mux
+    {
+    public:
+
+        dummy_aggregator_t ();
+
+        //  i_mux interface implementation.
+        void set_session (session_t *session_);
+        void shutdown ();
+        void terminate ();
+        void attach_pipe (class pipe_reader_t *pipe_);
+        void detach_pipe (class pipe_reader_t *pipe_);
+        bool empty ();
+        void deactivate (class pipe_reader_t *pipe_);
+        void reactivate (class pipe_reader_t *pipe_);
+        bool recv (struct zs_msg *msg_);
+
+
+    private:
+
+        //  Clean-up.
+        ~dummy_aggregator_t ();
+
+        //  Reference to the owner session object.
+        class session_t *session;
+
+        //  The single pipe bound.
+        class pipe_reader_t *pipe;
+
+        //  If true, the pipe is active.
+        bool active;
+
+        dummy_aggregator_t (const dummy_aggregator_t&);
+        void operator = (const dummy_aggregator_t&);
+    };
+
+}
+
+#endif

src/dummy_distributor.cpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "../include/zs.h"
+
+#include "dummy_distributor.hpp"
+#include "pipe_writer.hpp"
+#include "err.hpp"
+#include "session.hpp"
+#include "msg.hpp"
+
+zs::dummy_distributor_t::dummy_distributor_t () :
+    session (NULL)
+{
+}
+
+void zs::dummy_distributor_t::set_session (session_t *session_)
+{
+    zs_assert (!session);
+    session = session_;
+}
+
+void zs::dummy_distributor_t::shutdown ()
+{
+    //  No need to deallocate pipe here. It'll be deallocated during the
+    //  shutdown of the dispatcher.
+    delete this;
+}
+
+void zs::dummy_distributor_t::terminate ()
+{
+   if (pipe)
+       pipe->terminate ();
+
+   delete this;
+}
+
+zs::dummy_distributor_t::~dummy_distributor_t ()
+{
+}
+
+void zs::dummy_distributor_t::attach_pipe (pipe_writer_t *pipe_)
+{
+    zs_assert (!pipe);
+    pipe = pipe_;
+}
+
+void zs::dummy_distributor_t::detach_pipe (pipe_writer_t *pipe_)
+{
+    zs_assert (pipe == pipe_);
+    pipe = NULL;
+}
+
+bool zs::dummy_distributor_t::empty ()
+{
+    return pipe == NULL;
+}
+
+bool zs::dummy_distributor_t::send (zs_msg *msg_)
+{
+    return pipe && pipe->write (msg_);
+}
+
+void zs::dummy_distributor_t::flush ()
+{
+    if (pipe)
+        pipe->flush ();
+}
+

src/dummy_distributor.hpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef __ZS_DUMMY_DISTRIBUTOR_HPP_INCLUDED__
+#define __ZS_DUMMY_DISTRIBUTOR_HPP_INCLUDED__
+
+#include <vector>
+
+#include <i_demux.hpp>
+
+namespace zs
+{
+
+    //  Fake message distributor. There can be only one pipe bound to it
+    //  so there no real distribution going on. However, it is more efficient
+    //  than a real distributor and should be used where business logic
+    //  ensures there'll be at most one pipe bound.
+
+    class dummy_distributor_t : public i_demux
+    {
+    public:
+
+        dummy_distributor_t ();
+
+        //  i_demux implementation.
+        void set_session (class session_t *session_);
+        void shutdown ();
+        void terminate ();
+        void attach_pipe (class pipe_writer_t *pipe_);
+        void detach_pipe (class pipe_writer_t *pipe_);
+        bool empty ();
+        bool send (struct zs_msg *msg_);
+        void flush ();
+
+    private:
+
+        //  Clean-up.
+        ~dummy_distributor_t ();
+
+        //  Reference to the owner session object.
+        class session_t *session;
+
+        //  The bound pipe.
+        class pipe_writer_t *pipe;
+
+        dummy_distributor_t (const dummy_distributor_t&);
+        void operator = (const dummy_distributor_t&);
+    };
+
+}
+
+#endif

src/encoder.hpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef __ZS_ENCODER_HPP_INCLUDED__
+#define __ZS_ENCODER_HPP_INCLUDED__
+
+#include <stddef.h>
+#include <string.h>
+#include <algorithm>
+
+namespace zs
+{
+
+    //  Helper base class for encoders. It implements the state machine that
+    //  fills the outgoing buffer. Derived classes should implement individual
+    //  state machine actions.
+
+    template <typename T> class encoder_t
+    {
+    public:
+
+        inline encoder_t ()
+        {
+        }
+
+        //  The function tries to fill the supplied chunk by binary data.
+        //  Returns the size of data actually filled in. If offset is not
+        //  NULL, it is filled by offset of the first message in the batch.
+        //  If there's no beginning of a message in the batch, offset is
+        //  set to -1.
+        inline size_t read (unsigned char *data_, size_t size_,
+            int *offset_ = NULL)
+        {
+            int offset = -1;
+            size_t pos = 0;
+
+            while (pos < size_) {
+                if (to_write) {
+
+                    size_t to_copy = std::min (to_write, size_ - pos);
+                    memcpy (data_ + pos, write_pos, to_copy);
+                    pos += to_copy;
+                    write_pos += to_copy;
+                    to_write -= to_copy;
+                }
+                else {
+                    bool more = (static_cast <T*> (this)->*next) ();
+                    if (beginning && offset == -1) {
+                        offset = pos;
+                        beginning = false;
+                    }
+                    if (!more)
+                        break;
+                }
+            }
+
+            if (offset_)
+                *offset_ = offset;
+
+            return pos;
+        }
+    protected:
+
+        //  Prototype of state machine action.
+        typedef bool (T::*step_t) ();
+
+        //  This function should be called from derived class to write the data
+        //  to the buffer and schedule next state machine action. Set beginning
+        //  to true when you are writing first byte of a message.
+        inline void next_step (void *write_pos_, size_t to_write_,
+            step_t next_, bool beginning_)
+        {
+            write_pos = (unsigned char*) write_pos_;
+            to_write = to_write_;
+            next = next_;
+            beginning = beginning_;
+        }
+
+    private:
+
+        unsigned char *write_pos;
+        size_t to_write;
+        step_t next;
+        bool beginning;
+
+        encoder_t (const encoder_t&);
+        void operator = (const encoder_t&);
+    };
+
+}
+
+#endif

src/epoll.cpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "platform.hpp"
+
+#ifdef ZS_HAVE_LINUX
+
+#include <sys/epoll.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+#include <algorithm>
+
+#include "epoll.hpp"
+#include "err.hpp"
+#include "config.hpp"
+#include "i_poll_events.hpp"
+
+zs::epoll_t::epoll_t () :
+    stopping (false)
+{
+    epoll_fd = epoll_create (1);
+    errno_assert (epoll_fd != -1);
+}
+
+zs::epoll_t::~epoll_t ()
+{
+    close (epoll_fd);
+
+    for (retired_t::iterator it = retired.begin (); it != retired.end (); it ++)
+        delete *it;
+}
+
+zs::handle_t zs::epoll_t::add_fd (fd_t fd_, i_poll_events *events_)
+{
+    poll_entry_t *pe = new poll_entry_t;
+    zs_assert (pe != NULL);
+
+    //  The memset is not actually needed. It's here to prevent debugging
+    //  tools to complain about using uninitialised memory.
+    memset (pe, 0, sizeof (poll_entry_t));
+
+    pe->fd = fd_;
+    pe->ev.events = 0;
+    pe->ev.data.ptr = pe;
+    pe->events = events_;
+
+    int rc = epoll_ctl (epoll_fd, EPOLL_CTL_ADD, fd_, &pe->ev);
+    errno_assert (rc != -1);
+
+    //  Increase the load metric of the thread.
+    load.add (1);
+
+    handle_t handle;
+    handle.ptr = pe;
+    return handle;
+}
+
+void zs::epoll_t::rm_fd (handle_t handle_)
+{
+    poll_entry_t *pe = (poll_entry_t*) handle_.ptr;
+    int rc = epoll_ctl (epoll_fd, EPOLL_CTL_DEL, pe->fd, &pe->ev);
+    errno_assert (rc != -1);
+    pe->fd = retired_fd;
+    retired.push_back (pe);
+
+    //  Decrease the load metric of the thread.
+    load.sub (1);
+}
+
+void zs::epoll_t::set_pollin (handle_t handle_)
+{
+    poll_entry_t *pe = (poll_entry_t*) handle_.ptr;
+    pe->ev.events |= EPOLLIN;
+    int rc = epoll_ctl (epoll_fd, EPOLL_CTL_MOD, pe->fd, &pe->ev);
+    errno_assert (rc != -1);
+}
+
+void zs::epoll_t::reset_pollin (handle_t handle_)
+{
+    poll_entry_t *pe = (poll_entry_t*) handle_.ptr;
+    pe->ev.events &= ~((short) EPOLLIN);
+    int rc = epoll_ctl (epoll_fd, EPOLL_CTL_MOD, pe->fd, &pe->ev);
+    errno_assert (rc != -1);
+}
+
+void zs::epoll_t::set_pollout (handle_t handle_)
+{
+    poll_entry_t *pe = (poll_entry_t*) handle_.ptr;
+    pe->ev.events |= EPOLLOUT;
+    int rc = epoll_ctl (epoll_fd, EPOLL_CTL_MOD, pe->fd, &pe->ev);
+    errno_assert (rc != -1);
+}
+
+void zs::epoll_t::reset_pollout (handle_t handle_)
+{
+    poll_entry_t *pe = (poll_entry_t*) handle_.ptr;
+    pe->ev.events &= ~((short) EPOLLOUT);
+    int rc = epoll_ctl (epoll_fd, EPOLL_CTL_MOD, pe->fd, &pe->ev);
+    errno_assert (rc != -1);
+}
+
+void zs::epoll_t::add_timer (i_poll_events *events_)
+{
+     timers.push_back (events_);
+}
+
+void zs::epoll_t::cancel_timer (i_poll_events *events_)
+{
+    timers_t::iterator it = std::find (timers.begin (), timers.end (), events_);
+    if (it == timers.end ())
+        return;
+    timers.erase (it);
+}
+
+int zs::epoll_t::get_load ()
+{
+    return load.get ();
+}
+
+void zs::epoll_t::start ()
+{
+    worker.start (worker_routine, this);
+}
+
+void zs::epoll_t::stop ()
+{
+    stopping = true;
+}
+
+void zs::epoll_t::join ()
+{
+    worker.stop ();
+}
+
+void zs::epoll_t::loop ()
+{
+    epoll_event ev_buf [max_io_events];
+
+    while (!stopping) {
+
+        //  Wait for events.
+        int n;
+        while (true) {
+            n = epoll_wait (epoll_fd, &ev_buf [0], max_io_events,
+                timers.empty () ? -1 : max_timer_period);
+            if (!(n == -1 && errno == EINTR)) {
+                errno_assert (n != -1);
+                break;
+            }
+        }
+
+        //  Handle timer.
+        if (!n) {
+
+            //  Use local list of timers as timer handlers may fill new timers
+            //  into the original array.
+            timers_t t;
+            std::swap (timers, t);
+
+            //  Trigger all the timers.
+            for (timers_t::iterator it = t.begin (); it != t.end (); it ++)
+                (*it)->timer_event ();
+
+            continue;
+        }
+
+        for (int i = 0; i < n; i ++) {
+            poll_entry_t *pe = ((poll_entry_t*) ev_buf [i].data.ptr);
+
+            if (pe->fd == retired_fd)
+                continue;
+            if (ev_buf [i].events & (EPOLLERR | EPOLLHUP))
+                pe->events->in_event ();
+            if (pe->fd == retired_fd)
+               continue;
+            if (ev_buf [i].events & EPOLLOUT)
+                pe->events->out_event ();
+            if (pe->fd == retired_fd)
+                continue;
+            if (ev_buf [i].events & EPOLLIN)
+                pe->events->in_event ();
+        }
+
+        //  Destroy retired event sources.
+        for (retired_t::iterator it = retired.begin (); it != retired.end ();
+              it ++)
+            delete *it;
+        retired.clear ();
+    }
+}
+
+void zs::epoll_t::worker_routine (void *arg_)
+{
+    ((epoll_t*) arg_)->loop ();
+}
+
+#endif

src/epoll.hpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef __ZS_EPOLL_HPP_INCLUDED__
+#define __ZS_EPOLL_HPP_INCLUDED__
+
+#include "platform.hpp"
+
+#ifdef ZS_HAVE_LINUX
+
+#include <vector>
+#include <sys/epoll.h>
+
+#include "i_poller.hpp"
+//#include "i_poll_events.hpp"
+#include "fd.hpp"
+#include "thread.hpp"
+#include "atomic_counter.hpp"
+
+namespace zs
+{
+
+    //  This class implements socket polling mechanism using the Linux-specific
+    //  epoll mechanism.
+
+    class epoll_t : public i_poller
+    {
+    public:
+
+        epoll_t ();
+        virtual ~epoll_t ();
+
+        //  i_poller implementation.
+        handle_t add_fd (fd_t fd_, i_poll_events *events_);
+        void rm_fd (handle_t handle_);
+        void set_pollin (handle_t handle_);
+        void reset_pollin (handle_t handle_);
+        void set_pollout (handle_t handle_);
+        void reset_pollout (handle_t handle_);
+        void add_timer (i_poll_events *events_);
+        void cancel_timer (i_poll_events *events_);
+        int get_load ();
+        void start ();
+        void stop ();
+        void join ();
+
+    private:
+
+        //  Main worker thread routine.
+        static void worker_routine (void *arg_);
+
+        //  Main event loop.
+        void loop ();
+
+        //  Main epoll file descriptor
+        fd_t epoll_fd;
+
+        struct poll_entry_t
+        {
+            fd_t fd;
+            epoll_event ev;
+            struct i_poll_events *events;
+        };
+
+        //  List of retired event sources.
+        typedef std::vector <poll_entry_t*> retired_t;
+        retired_t retired;
+
+        //  List of all the engines waiting for the timer event.
+        typedef std::vector <struct i_poll_events*> timers_t;
+        timers_t timers;
+
+        //  If true, thread is in the process of shutting down.
+        bool stopping;
+
+        //  Handle of the physical thread doing the I/O work.
+        thread_t worker;
+
+        //  Load of the poller. Currently number of file descriptors
+        //  registered with the poller.
+        atomic_counter_t load;
+
+        epoll_t (const epoll_t&);
+        void operator = (const epoll_t&);
+    };
+
+}
+
+#endif
+
+#endif

src/err.cpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "err.hpp"
+#include "platform.hpp"
+
+#ifdef ZS_HAVE_WINDOWS
+
+const char *zs::wsa_error()
+{
+    int errcode = WSAGetLastError ();
+    //  TODO: This is not a generic way to handle this...
+    if (errcode == WSAEWOULDBLOCK)
+        return NULL;
+
+    return
+        (errcode == WSABASEERR) ?
+            "No Error" : 
+        (errcode == WSAEINTR) ?
+            "Interrupted system call" : 
+        (errcode == WSAEBADF) ?
+            "Bad file number" : 
+        (errcode == WSAEACCES) ?
+            "Permission denied" : 
+        (errcode == WSAEFAULT) ?
+            "Bad address" : 
+        (errcode == WSAEINVAL) ?
+            "Invalid argument" : 
+        (errcode == WSAEMFILE) ?
+            "Too many open files" : 
+        (errcode == WSAEWOULDBLOCK) ?
+            "Operation would block" : 
+        (errcode == WSAEINPROGRESS) ?
+            "Operation now in progress" : 
+        (errcode == WSAEALREADY) ?
+            "Operation already in progress" : 
+        (errcode == WSAENOTSOCK) ?
+            "Socket operation on non-socket" : 
+        (errcode == WSAEDESTADDRREQ) ?
+            "Destination address required" : 
+        (errcode == WSAEMSGSIZE) ?
+            "Message too long" : 
+        (errcode == WSAEPROTOTYPE) ?
+            "Protocol wrong type for socket" : 
+        (errcode == WSAENOPROTOOPT) ?
+            "Bad protocol option" : 
+        (errcode == WSAEPROTONOSUPPORT) ?
+            "Protocol not supported" : 
+        (errcode == WSAESOCKTNOSUPPORT) ?
+            "Socket type not supported" : 
+        (errcode == WSAEOPNOTSUPP) ?
+            "Operation not supported on socket" : 
+        (errcode == WSAEPFNOSUPPORT) ?
+            "Protocol family not supported" : 
+        (errcode == WSAEAFNOSUPPORT) ?
+            "Address family not supported by protocol family" : 
+        (errcode == WSAEADDRINUSE) ?
+            "Address already in use" : 
+        (errcode == WSAEADDRNOTAVAIL) ?
+            "Can't assign requested address" : 
+        (errcode == WSAENETDOWN) ?
+            "Network is down" : 
+        (errcode == WSAENETUNREACH) ?
+            "Network is unreachable" : 
+        (errcode == WSAENETRESET) ?
+            "Net dropped connection or reset" : 
+        (errcode == WSAECONNABORTED) ?
+            "Software caused connection abort" : 
+        (errcode == WSAECONNRESET) ?
+            "Connection reset by peer" : 
+        (errcode == WSAENOBUFS) ?
+            "No buffer space available" : 
+        (errcode == WSAEISCONN) ?
+            "Socket is already connected" : 
+        (errcode == WSAENOTCONN) ?
+            "Socket is not connected" : 
+        (errcode == WSAESHUTDOWN) ?
+            "Can't send after socket shutdown" : 
+        (errcode == WSAETOOMANYREFS) ?
+            "Too many references can't splice" : 
+        (errcode == WSAETIMEDOUT) ?
+            "Connection timed out" : 
+        (errcode == WSAECONNREFUSED) ?
+            "Connection refused" : 
+        (errcode == WSAELOOP) ?
+            "Too many levels of symbolic links" : 
+        (errcode == WSAENAMETOOLONG) ?
+            "File name too long" : 
+        (errcode == WSAEHOSTDOWN) ?
+            "Host is down" : 
+        (errcode == WSAEHOSTUNREACH) ?
+            "No Route to Host" : 
+        (errcode == WSAENOTEMPTY) ?
+            "Directory not empty" : 
+        (errcode == WSAEPROCLIM) ?
+            "Too many processes" : 
+        (errcode == WSAEUSERS) ?
+            "Too many users" : 
+        (errcode == WSAEDQUOT) ?
+            "Disc Quota Exceeded" : 
+        (errcode == WSAESTALE) ?
+            "Stale NFS file handle" : 
+        (errcode == WSAEREMOTE) ?
+            "Too many levels of remote in path" : 
+        (errcode == WSASYSNOTREADY) ?
+            "Network SubSystem is unavailable" : 
+        (errcode == WSAVERNOTSUPPORTED) ?
+            "WINSOCK DLL Version out of range" : 
+        (errcode == WSANOTINITIALISED) ?
+            "Successful WSASTARTUP not yet performed" : 
+        (errcode == WSAHOST_NOT_FOUND) ?
+            "Host not found" : 
+        (errcode == WSATRY_AGAIN) ?
+            "Non-Authoritative Host not found" : 
+        (errcode == WSANO_RECOVERY) ?
+            "Non-Recoverable errors: FORMERR REFUSED NOTIMP" : 
+        (errcode == WSANO_DATA) ?
+            "Valid name no data record of requested" :
+        "error not defined"; 
+}
+void zs::win_error (char *buffer_, size_t buffer_size_)
+{
+    DWORD errcode = GetLastError ();
+    DWORD rc = FormatMessageA (FORMAT_MESSAGE_FROM_SYSTEM |
+        FORMAT_MESSAGE_IGNORE_INSERTS, NULL, errcode, MAKELANGID(LANG_NEUTRAL,
+        SUBLANG_DEFAULT), buffer_, buffer_size_, NULL );
+    zs_assert (rc);
+}
+
+#endif

src/err.hpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef __ZS_ERR_HPP_INCLUDED__
+#define __ZS_ERR_HPP_INCLUDED__
+
+#include <assert.h>
+#include <errno.h>
+#include <stdlib.h>
+#include <stdio.h>
+
+#include "platform.hpp"
+
+#ifdef ZS_HAVE_WINDOWS
+#include "windows.hpp"
+#else
+#include <netdb.h>
+#endif
+
+#ifdef ZS_HAVE_WINDOWS
+
+namespace zs
+{
+
+    const char *wsa_error ();
+    void win_error (char *buffer_, size_t buffer_size_);
+  
+}
+
+//  Provides convenient way to check WSA-style errors on Windows.
+#define wsa_assert(x) do { if (!(x)){\
+    const char *errstr = zs::wsa_error ();\
+    if (errstr != NULL) {\
+    fprintf (stderr, "Assertion failed: %s (%s:%d)\n", errstr, \
+        __FILE__, __LINE__);\
+        abort ();\
+    }\
+}} while (false)
+
+//  Provides convenient way to check GetLastError-style errors on Windows.
+#define win_assert(x) do { if (!(x)) {\
+    char errstr [256];\
+    zs::win_error (errstr, 256);\
+    fprintf (stderr, "Assertion failed: %s (%s:%d)\n", errstr, \
+        __FILE__, __LINE__);\
+    abort ();\
+}} while (false)
+
+#endif
+
+//  This macro works in exactly the same way as the normal assert. It is used
+//  in its stead because standard assert on Win32 in broken - it prints nothing
+//  when used within the scope of JNI library.
+#define zs_assert(x) do { if (!(x)){\
+    fprintf (stderr, "Assertion failed: %s (%s:%d)\n", #x, \
+        __FILE__, __LINE__);\
+        abort ();\
+}} while (false) 
+
+//  Provides convenient way to check for errno-style errors.
+#define errno_assert(x) do { if (!(x)) {\
+    perror (NULL);\
+    fprintf (stderr, "%s (%s:%d)\n", #x, __FILE__, __LINE__);\
+    abort ();\
+}} while (false)
+
+// Provides convenient way to check for errors from getaddrinfo.
+#define gai_assert(x) do { if (x) {\
+    const char *errstr = gai_strerror (x);\
+    fprintf (stderr, "%s (%s:%d)\n", errstr, __FILE__, __LINE__);\
+    abort ();\
+}} while (false)
+
+#endif

src/fair_aggregator.cpp

+/*
+    Copyright (c) 2007-2009 FastMQ Inc.
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the Lesser GNU General Public License as published by
+    the Free Software Foundation; either version 3 of the License, or
+    (at your option) any later version.
+
+    0MQ is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Lesser GNU General Public License for more details.
+
+    You should have received a copy of the Lesser GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "../include/zs.h"
+
+#include "fair_aggregator.hpp"
+#include "err.hpp"
+#include "pipe_reader.hpp"
+#include "session.hpp"
+
+//  Swaps pipes at specified indices. 
+#define swap_pipes(i1_, i2_) \
+    std::swap (pipes [i1_], pipes [i2_]);\
+    pipes [i1_]->set_index (i1_);\
+    pipes [i2_]->set_index (i2_);
+
+zs::fair_aggregator_t::fair_aggregator_t () :
+    session (NULL),
+    active (0),
+    current (0)
+{
+}
+
+void zs::fair_aggregator_t::set_session (session_t *session_)
+{
+    zs_assert (!session);
+    session = session_;
+}
+
+void zs::fair_aggregator_t::shutdown ()
+{
+    //  No need to deallocate pipes here. They'll be deallocated during the
+    //  shutdown of the dispatcher.
+    delete this;
+}
+
+void zs::fair_aggregator_t::terminate ()
+{
+    //  Pipe unregisters itself during the call to terminate, so the pipes
+    //  list shinks by one in each iteration.
+    while (!pipes.empty ())
+        pipes [0]->terminate ();
+
+   delete this;
+}
+
+zs::fair_aggregator_t::~fair_aggregator_t ()
+{
+}
+
+void zs::fair_aggregator_t::attach_pipe (pipe_reader_t *pipe_)
+{
+    //  Associate new pipe with the mux object.
+    pipe_->set_mux (this);
+    pipes.push_back (pipe_);
+    active++;
+    if (pipes.size () > active)
+        swap_pipes (pipes.size () - 1, active - 1);
+    if (active == 1)
+        session->revive ();
+}
+
+void zs::fair_aggregator_t::detach_pipe (pipe_reader_t *pipe_)
+{
+    //  Move the pipe from the list of active pipes to the list of idle pipes.
+    deactivate (pipe_);
+            
+    //  Move the pipe to the end of the idle list and remove it.
+    swap_pipes (pipe_->get_index (), pipes.size () - 1);
+    pipes.pop_back ();
+}
+
+bool zs::fair_aggregator_t::empty ()
+{
+    return pipes.empty ();
+}
+
+bool zs::fair_aggregator_t::recv (zs_msg *msg_)
+{
+    //  Deallocate old content of the message.
+    zs_msg_close (msg_);
+
+    //  O(1) fair queueing. Round-robin over the active pipes to get
+    //  next message.
+    for (pipes_t::size_type i = active; i != 0; i--) {
+
+        //  Update current.
+        current = (current + 1) % active;
+        
+        //  Try to read from current.
+        if (pipes [current]->read (msg_))
+            return true;
+    }
+
+    //  No message is available. Initialise the output parameter
+    //  to be a 0-byte message.
+    zs_msg_init (msg_);
+    return false;
+}
+
+void zs::fair_aggregator_t::deactivate (pipe_reader_t *pipe_)
+{
+    int index = pipe_->get_index ();
+
+    //  Suspend an active pipe.
+    swap_pipes (index, active - 1);
+    active--;
+
+    //  If the deactiveted pipe is the current one, shift the current one pipe
+    //  backwards so that the pipe that replaced the deactiveted one will be
+    //  processed immediately rather than skipped.
+    if (index == (int) current) {
+        index--;
+        if (index == -1)
+            index = active - 1;
+        current = index;
+    }
+}
+
+void zs::fair_aggregator_t::reactivate (pipe_reader_t *pipe_)
+{
+    //  Revive an idle pipe.
+    swap_pipes (pipe_->get_index (), active);
+    active++;
+    if (active == 1)
+        session->revive ();
+}