Patches from sub-forward branch incorporated

Signed-off-by: Martin Sustrik <sustrik@250bpm.com>

src/Makefile.am

@@ -34,6 +34,7 @@ libzmq_la_SOURCES = \
     likely.hpp \
     mailbox.hpp \
     msg.hpp \
+    mtrie.hpp \
     mutex.hpp \
     named_session.hpp \
     object.hpp \
@@ -97,6 +98,7 @@ libzmq_la_SOURCES = \
     lb.cpp \
     mailbox.cpp \
     msg.cpp \
+    mtrie.cpp \
     named_session.cpp \
     object.cpp \
     options.cpp \

src/command.hpp

@@ -42,6 +42,7 @@ namespace zmq
             bind,
             activate_read,
             activate_write,
+            hiccup,
             pipe_term,
             pipe_term_ack,
             term_req,
@@ -95,6 +96,13 @@ namespace zmq
                 uint64_t msgs_read;
             } activate_write;
 
+            //  Sent by pipe reader to writer after creating a new inpipe.
+            //  The parameter is actually of type pipe_t::upipe_t, however,
+            //  its definition is private so we'll have to do with void*.
+            struct {
+                void *pipe;
+            } hiccup;
+
             //  Sent by pipe reader to pipe writer to ask it to terminate
             //  its end of the pipe.
             struct {

src/connect_session.cpp

@@ -111,7 +111,7 @@ void zmq::connect_session_t::start_connecting (bool wait_)
     zmq_assert (false);
 }
 
-bool zmq::connect_session_t::attached (const blob_t &peer_identity_)
+bool zmq::connect_session_t::xattached (const blob_t &peer_identity_)
 {
     //  If there was no previous connection...
     if (!connected) {
@@ -153,9 +153,12 @@ bool zmq::connect_session_t::attached (const blob_t &peer_identity_)
     return true;
 }
 
-void zmq::connect_session_t::detached ()
+bool zmq::connect_session_t::xdetached ()
 {
     //  Reconnect.
     start_connecting (true);
+
+    //  Don't tear the session down.
+    return true;
 }
 

src/connect_session.hpp

@@ -44,8 +44,8 @@ namespace zmq
     private:
 
         //  Handlers for events from session base class.
-        bool attached (const blob_t &peer_identity_);
-        void detached ();
+        bool xattached (const blob_t &peer_identity_);
+        bool xdetached ();
 
         //  Start the connection process.
         void start_connecting (bool wait_);

src/fq.cpp

@@ -62,6 +62,11 @@ void zmq::fq_t::activated (pipe_t *pipe_)
 }
 
 int zmq::fq_t::recv (msg_t *msg_, int flags_)
+{
+    return recvpipe (msg_, flags_, NULL);
+}
+
+int zmq::fq_t::recvpipe (msg_t *msg_, int flags_, pipe_t **pipe_)
 {
     //  Deallocate old content of the message.
     int rc = msg_->close ();
@@ -83,6 +88,8 @@ int zmq::fq_t::recv (msg_t *msg_, int flags_)
         //  and replaced by another active pipe. Thus we don't have to increase
         //  the 'current' pointer.
         if (fetched) {
+            if (pipe_)
+                *pipe_ = pipes [current];
             more = msg_->flags () & msg_t::more;
             if (!more) {
                 current++;

src/fq.hpp

@@ -29,8 +29,9 @@ namespace zmq
 {
 
     //  Class manages a set of inbound pipes. On receive it performs fair
-    //  queueing (RFC970) so that senders gone berserk won't cause denial of
+    //  queueing so that senders gone berserk won't cause denial of
     //  service for decent senders.
+
     class fq_t
     {
     public:
@@ -43,6 +44,7 @@ namespace zmq
         void terminated (pipe_t *pipe_);
 
         int recv (msg_t *msg_, int flags_);
+        int recvpipe (msg_t *msg_, int flags_, pipe_t **pipe_);
         bool has_in ();
 
     private:

src/mtrie.cpp

+/*
+    Copyright (c) 2007-2011 iMatix Corporation
+    Copyright (c) 2007-2011 Other contributors as noted in the AUTHORS file
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the GNU Lesser 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 Lesser General Public License for more details.
+
+    You should have received a copy of the GNU Lesser General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include <stdlib.h>
+
+#include <new>
+#include <algorithm>
+
+#include "platform.hpp"
+#if defined ZMQ_HAVE_WINDOWS
+#include "windows.hpp"
+#endif
+
+#include "err.hpp"
+#include "pipe.hpp"
+#include "mtrie.hpp"
+
+zmq::mtrie_t::mtrie_t () :
+    min (0),
+    count (0)
+{
+}
+
+zmq::mtrie_t::~mtrie_t ()
+{
+    if (count == 1)
+        delete next.node;
+    else if (count > 1) {
+        for (unsigned short i = 0; i != count; ++i)
+            if (next.table [i])
+                delete next.table [i];
+        free (next.table);
+    }
+}
+
+bool zmq::mtrie_t::add (unsigned char *prefix_, size_t size_, pipe_t *pipe_)
+{
+    //  We are at the node corresponding to the prefix. We are done.
+    if (!size_) {
+        bool result = pipes.empty ();
+        pipes.insert (pipe_);
+        return result;
+    }
+
+    unsigned char c = *prefix_;
+    if (c < min || c >= min + count) {
+
+        //  The character is out of range of currently handled
+        //  charcters. We have to extend the table.
+        if (!count) {
+            min = c;
+            count = 1;
+            next.node = NULL;
+        }
+        else if (count == 1) {
+            unsigned char oldc = min;
+            mtrie_t *oldp = next.node;
+            count = (min < c ? c - min : min - c) + 1;
+            next.table = (mtrie_t**)
+                malloc (sizeof (mtrie_t*) * count);
+            zmq_assert (next.table);
+            for (unsigned short i = 0; i != count; ++i)
+                next.table [i] = 0;
+            min = std::min (min, c);
+            next.table [oldc - min] = oldp;
+        }
+        else if (min < c) {
+
+            //  The new character is above the current character range.
+            unsigned short old_count = count;
+            count = c - min + 1;
+            next.table = (mtrie_t**) realloc ((void*) next.table,
+                sizeof (mtrie_t*) * count);
+            zmq_assert (next.table);
+            for (unsigned short i = old_count; i != count; i++)
+                next.table [i] = NULL;
+        }
+        else {
+
+            //  The new character is below the current character range.
+            unsigned short old_count = count;
+            count = (min + old_count) - c;
+            next.table = (mtrie_t**) realloc ((void*) next.table,
+                sizeof (mtrie_t*) * count);
+            zmq_assert (next.table);
+            memmove (next.table + min - c, next.table,
+                old_count * sizeof (mtrie_t*));
+            for (unsigned short i = 0; i != min - c; i++)
+                next.table [i] = NULL;
+            min = c;
+        }
+    }
+
+    //  If next node does not exist, create one.
+    if (count == 1) {
+        if (!next.node) {
+            next.node = new (std::nothrow) mtrie_t;
+            zmq_assert (next.node);
+        }
+        return next.node->add (prefix_ + 1, size_ - 1, pipe_);
+    }
+    else {
+        if (!next.table [c - min]) {
+            next.table [c - min] = new (std::nothrow) mtrie_t;
+            zmq_assert (next.table [c - min]);
+        }
+        return next.table [c - min]->add (prefix_ + 1, size_ - 1, pipe_);
+    }
+}
+
+
+void zmq::mtrie_t::rm (pipe_t *pipe_,
+    void (*func_) (unsigned char *data_, size_t size_, void *arg_),
+    void *arg_)
+{
+    unsigned char *buff = NULL;
+    rm_helper (pipe_, &buff, 0, 0, func_, arg_);
+    free (buff);
+}
+
+void zmq::mtrie_t::rm_helper (pipe_t *pipe_, unsigned char **buff_,
+    size_t buffsize_, size_t maxbuffsize_,
+    void (*func_) (unsigned char *data_, size_t size_, void *arg_),
+    void *arg_)
+{
+    //  Remove the subscription from this node.
+    if (pipes.erase (pipe_) && pipes.empty ())
+        func_ (*buff_, buffsize_, arg_);
+
+    //  Adjust the buffer.
+    if (buffsize_ >= maxbuffsize_) {
+        maxbuffsize_ = buffsize_ + 256;
+        *buff_ = (unsigned char*) realloc (*buff_, maxbuffsize_);
+        alloc_assert (*buff_);
+    }
+
+    //  If there are no subnodes in the trie, return.
+    if (count == 0)
+        return;
+
+    //  If there's one subnode (optimisation).
+    if (count == 1) {
+        (*buff_) [buffsize_] = min;
+        buffsize_++;
+        next.node->rm_helper (pipe_, buff_, buffsize_, maxbuffsize_,
+            func_, arg_);
+        return;
+    }
+
+    //  If there are multiple subnodes.
+    for (unsigned char c = 0; c != count; c++) {
+        (*buff_) [buffsize_] = min + c;
+        if (next.table [c])
+            next.table [c]->rm_helper (pipe_, buff_, buffsize_ + 1,
+                maxbuffsize_, func_, arg_);
+    }  
+}
+
+bool zmq::mtrie_t::rm (unsigned char *prefix_, size_t size_, pipe_t *pipe_)
+{
+     if (!size_) {
+         pipes_t::size_type erased = pipes.erase (pipe_);
+         zmq_assert (erased == 1);
+         return pipes.empty ();
+     }
+
+     unsigned char c = *prefix_;
+     if (!count || c < min || c >= min + count)
+         return false;
+
+     mtrie_t *next_node =
+         count == 1 ? next.node : next.table [c - min];
+
+     if (!next_node)
+         return false;
+
+     return next_node->rm (prefix_ + 1, size_ - 1, pipe_);
+}
+
+void zmq::mtrie_t::match (unsigned char *data_, size_t size_, pipes_t &pipes_)
+{
+    //  Merge the subscriptions from this node to the resultset.
+    pipes_.insert (pipes.begin (), pipes.end ());
+
+    //  If there are no subnodes in the trie, return.
+    if (count == 0)
+        return;
+
+    //  If there's one subnode (optimisation).
+    if (count == 1) {
+        next.node->match (data_ + 1, size_ - 1, pipes_);
+        return;
+    }
+
+    //  If there are multiple subnodes.
+    for (unsigned char c = 0; c != count; c++) {
+        if (next.table [c])
+            next.table [c]->match (data_ + 1, size_ - 1, pipes_);
+    }   
+}
+

src/mtrie.hpp

+/*
+    Copyright (c) 2007-2011 iMatix Corporation
+    Copyright (c) 2007-2011 Other contributors as noted in the AUTHORS file
+
+    This file is part of 0MQ.
+
+    0MQ is free software; you can redistribute it and/or modify it under
+    the terms of the GNU Lesser 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 Lesser General Public License for more details.
+
+    You should have received a copy of the GNU Lesser General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef __ZMQ_MTRIE_HPP_INCLUDED__
+#define __ZMQ_MTRIE_HPP_INCLUDED__
+
+#include <stddef.h>
+#include <set>
+
+#include "stdint.hpp"
+
+namespace zmq
+{
+
+    //  Multi-trie. Each node in the trie is a set of pointers to pipes.
+
+    class mtrie_t
+    {
+    public:
+
+        typedef std::set <class pipe_t*> pipes_t;
+
+        mtrie_t ();
+        ~mtrie_t ();
+
+        //  Add key to the trie. Returns true if it's a new subscription
+        //  rather than a duplicate.
+        bool add (unsigned char *prefix_, size_t size_, class pipe_t *pipe_);
+
+        //  Remove all subscriptions for a specific peer from the trie.
+        //  If there are no subscriptions left on some topics, invoke the
+        //  supplied callback function.
+        void rm (class pipe_t *pipe_,
+            void (*func_) (unsigned char *data_, size_t size_, void *arg_),
+            void *arg_);
+
+        //  Remove specific subscription from the trie. Return true is it was
+        //  actually removed rather than de-duplicated.
+        bool rm (unsigned char *prefix_, size_t size_, class pipe_t *pipe_);
+
+        //  Get all matching pipes.
+        void match (unsigned char *data_, size_t size_, pipes_t &pipes_);
+
+    private:
+
+        void rm_helper (class pipe_t *pipe_, unsigned char **buff_,
+            size_t buffsize_, size_t maxbuffsize_,
+            void (*func_) (unsigned char *data_, size_t size_, void *arg_),
+            void *arg_);
+
+        pipes_t pipes;
+        unsigned char min;
+        unsigned short count;
+        union {
+            class mtrie_t *node;
+            class mtrie_t **table;
+        } next;
+
+        mtrie_t (const mtrie_t&);
+        const mtrie_t &operator = (const mtrie_t&);
+    };
+
+}
+
+#endif
+

src/named_session.cpp

@@ -45,7 +45,7 @@ zmq::named_session_t::~named_session_t ()
     unregister_session (peer_identity);
 }
 
-bool zmq::named_session_t::attached (const blob_t &peer_identity_)
+bool zmq::named_session_t::xattached (const blob_t &peer_identity_)
 {
     //  Double check that identities match.
     zmq_assert (peer_identity == peer_identity_);
@@ -58,9 +58,10 @@ bool zmq::named_session_t::attached (const blob_t &peer_identity_)
     return true;
 }
 
-void zmq::named_session_t::detached ()
+bool zmq::named_session_t::xdetached ()
 {
     //  Do nothing. Named sessions are never destroyed because of disconnection.
     //  Neither they have to actively reconnect.
+    return true;
 }
 

src/named_session.hpp

@@ -40,8 +40,8 @@ namespace zmq
         ~named_session_t ();
 
         //  Handlers for events from session base class.
-        bool attached (const blob_t &peer_identity_);
-        void detached ();
+        bool xattached (const blob_t &peer_identity_);
+        bool xdetached ();
 
     private:
 

src/object.cpp

@@ -74,7 +74,7 @@ void zmq::object_t::process_command (command_t &cmd_)
     case command_t::plug:
         process_plug ();
         process_seqnum ();
-        return;
+        break;
 
     case command_t::own:
         process_own (cmd_.args.own.object);
@@ -96,9 +96,13 @@ void zmq::object_t::process_command (command_t &cmd_)
         process_seqnum ();
         break;
 
+    case command_t::hiccup:
+        process_hiccup (cmd_.args.hiccup.pipe);
+        break;
+
     case command_t::pipe_term:
         process_pipe_term ();
-        return;
+        break;
 
     case command_t::pipe_term_ack:
         process_pipe_term_ack ();
@@ -290,6 +294,18 @@ void zmq::object_t::send_activate_write (pipe_t *destination_,
     send_command (cmd);
 }
 
+void zmq::object_t::send_hiccup (pipe_t *destination_, void *pipe_)
+{
+    command_t cmd;
+#if defined ZMQ_MAKE_VALGRIND_HAPPY
+    memset (&cmd, 0, sizeof (cmd));
+#endif
+    cmd.destination = destination_;
+    cmd.type = command_t::hiccup;
+    cmd.args.hiccup.pipe = pipe_;
+    send_command (cmd);
+}
+
 void zmq::object_t::send_pipe_term (pipe_t *destination_)
 {
     command_t cmd;
@@ -418,6 +434,11 @@ void zmq::object_t::process_activate_write (uint64_t msgs_read_)
     zmq_assert (false);
 }
 
+void zmq::object_t::process_hiccup (void *pipe_)
+{
+    zmq_assert (false);
+}
+
 void zmq::object_t::process_pipe_term ()
 {
     zmq_assert (false);

src/object.hpp

@@ -71,6 +71,7 @@ namespace zmq
         void send_activate_read (class pipe_t *destination_);
         void send_activate_write (class pipe_t *destination_,
              uint64_t msgs_read_);
+        void send_hiccup (class pipe_t *destination_, void *pipe_);
         void send_pipe_term (class pipe_t *destination_);
         void send_pipe_term_ack (class pipe_t *destination_);
         void send_term_req (class own_t *destination_,
@@ -92,6 +93,7 @@ namespace zmq
             const blob_t &peer_identity_);
         virtual void process_activate_read ();
         virtual void process_activate_write (uint64_t msgs_read_);
+        virtual void process_hiccup (void *pipe_);
         virtual void process_pipe_term ();
         virtual void process_pipe_term_ack ();
         virtual void process_term_req (class own_t *object_);

src/pipe.cpp

@@ -205,6 +205,29 @@ void zmq::pipe_t::process_activate_write (uint64_t msgs_read_)
     }
 }
 
+void zmq::pipe_t::process_hiccup (void *pipe_)
+{
+    //  Destroy old outpipe. Note that the read end of the pipe was already
+    //  migrated to this thread.
+    zmq_assert (outpipe);
+    outpipe->flush ();
+    msg_t msg;
+    while (outpipe->read (&msg)) {
+       int rc = msg.close ();
+       errno_assert (rc == 0);
+    }
+    delete outpipe;
+
+    //  Plug in the new outpipe.
+    zmq_assert (pipe_);
+    outpipe = (upipe_t*) pipe_;
+    out_active = true;
+
+    //  If appropriate, notify the user about the hiccup.
+    if (state == active)
+        sink->hiccuped (this);
+}
+
 void zmq::pipe_t::process_pipe_term ()
 {
     //  This is the simple case of peer-induced termination. If there are no
@@ -379,3 +402,23 @@ void zmq::pipe_t::delimit ()
     //  Delimiter in any other state is invalid.
     zmq_assert (false);
 }
+
+void zmq::pipe_t::hiccup ()
+{
+    //  If termination is already under way do nothing.
+    if (state != active)
+        return;
+
+    //  We'll drop the pointer to the inpipe. From now on, the peer is
+    //  responsible for deallocating it.
+    inpipe = NULL;
+
+    //  Create new inpipe.
+    inpipe = new (std::nothrow) pipe_t::upipe_t ();
+    alloc_assert (inpipe);
+    in_active = true;
+
+    //  Notify the peer about the hiccup.
+    send_hiccup (peer, (void*) inpipe);
+}
+

src/pipe.hpp

@@ -44,6 +44,7 @@ namespace zmq
 
         virtual void read_activated (class pipe_t *pipe_) = 0;
         virtual void write_activated (class pipe_t *pipe_) = 0;
+        virtual void hiccuped (class pipe_t *pipe_) = 0;
         virtual void terminated (class pipe_t *pipe_) = 0;
     };
 
@@ -86,6 +87,11 @@ namespace zmq
         //  Flush the messages downsteam.
         void flush ();
 
+        //  Temporaraily disconnects the inbound message stream and drops
+        //  all the messages on the fly. Causes 'hiccuped' event to be generated
+        //  in the peer.
+        void hiccup ();
+
         //  Ask pipe to terminate. The termination will happen asynchronously
         //  and user will be notified about actual deallocation by 'terminated'
         //  event.
@@ -93,18 +99,19 @@ namespace zmq
 
     private:
 
+        //  Type of the underlying lock-free pipe.
+        typedef ypipe_t <msg_t, message_pipe_granularity> upipe_t;
+
         //  Command handlers.
         void process_activate_read ();
         void process_activate_write (uint64_t msgs_read_);
+        void process_hiccup (void *pipe_);
         void process_pipe_term ();
         void process_pipe_term_ack ();
 
         //  Handler for delimiter read from the pipe.
         void delimit ();
 
-        //  Type of the underlying lock-free pipe.
-        typedef ypipe_t <msg_t, message_pipe_granularity> upipe_t;
-
         //  Constructor is private. Pipe can only be created using
         //  pipepair function.
         pipe_t (object_t *parent_, upipe_t *inpipe_, upipe_t *outpipe_,

src/pub.cpp

@@ -30,3 +30,16 @@ zmq::pub_t::pub_t (class ctx_t *parent_, uint32_t tid_) :
 zmq::pub_t::~pub_t ()
 {
 }
+
+int zmq::pub_t::xrecv (class msg_t *msg_, int flags_)
+{
+    //  Messages cannot be received from PUB socket.
+    errno = ENOTSUP;
+    return -1;
+}
+
+bool zmq::pub_t::xhas_in ()
+{
+    return false;
+}
+

src/pub.hpp

@@ -33,6 +33,10 @@ namespace zmq
         pub_t (class ctx_t *parent_, uint32_t tid_);
         ~pub_t ();
 
+        //  Implementations of virtual functions from socket_base_t.
+        int xrecv (class msg_t *msg_, int flags_);
+        bool xhas_in ();
+
     private:
 
         pub_t (const pub_t&);

src/session.cpp

@@ -146,6 +146,13 @@ void zmq::session_t::write_activated (pipe_t *pipe_)
         engine->activate_in ();
 }
 
+void zmq::session_t::hiccuped (pipe_t *pipe_)
+{
+    //  Hiccups are always sent from session to socket, not the other
+    //  way round.
+    zmq_assert (false);
+}
+
 void zmq::session_t::process_plug ()
 {
 }
@@ -287,4 +294,25 @@ void zmq::session_t::unregister_session (const blob_t &name_)
     socket->unregister_session (name_);
 }
 
+bool zmq::session_t::attached (const blob_t &peer_identity_)
+{
+    return xattached (peer_identity_);
+}
+
+void zmq::session_t::detached ()
+{
+    if (!xdetached ()) {
+
+        //  Derived session type have asked for session termination.
+        terminate ();
+        return;
+    }
+
+    //  For subscriber sockets we hiccup the inbound pipe, which will cause
+    //  the socket object to resend all the subscriptions.
+    if (pipe && (options.type == ZMQ_SUB || options.type == ZMQ_XSUB))
+        pipe->hiccup ();  
+}
+
+
 

src/session.hpp

@@ -44,9 +44,7 @@ namespace zmq
         //  To be used once only, when creating the session.
         void attach_pipe (class pipe_t *pipe_);
 
-        //  i_inout interface implementation. Note that detach method is not
-        //  implemented by generic session. Different session types may handle
-        //  engine disconnection in different ways.
+        //  i_inout interface implementation.
         bool read (msg_t *msg_);
         bool write (msg_t *msg_);
         void flush ();
@@ -55,17 +53,19 @@ namespace zmq
         //  i_pipe_events interface implementation.
         void read_activated (class pipe_t *pipe_);
         void write_activated (class pipe_t *pipe_);
+        void hiccuped (class pipe_t *pipe_);
         void terminated (class pipe_t *pipe_);
 
     protected:
 
-        //  Two events for the derived session type. Attached is triggered
-        //  when session is attached to a peer. The function can reject the new
-        //  peer by returning false. Detached is triggered at the beginning of
+        //  Events from the engine. Attached is triggered when session is
+        //  attached to a peer. The function can reject the new peer by
+        //  returning false. Detached is triggered at the beginning of
         //  the termination process when session is about to be detached from
-        //  the peer.
-        virtual bool attached (const blob_t &peer_identity_) = 0;
-        virtual void detached () = 0;
+        //  the peer. If it returns false, session will be terminated.
+        //  To be overloaded by the derived session type.
+        virtual bool xattached (const blob_t &peer_identity_) = 0;
+        virtual bool xdetached () = 0;
 
         //  Returns true if there is an engine attached to the session.
         bool has_engine ();
@@ -78,6 +78,9 @@ namespace zmq
 
     private:
 
+        bool attached (const blob_t &peer_identity_);
+        void detached ();
+
         //  Handlers for incoming commands.
         void process_plug ();
         void process_attach (struct i_engine *engine_,

src/socket_base.cpp

@@ -764,7 +764,7 @@ bool zmq::socket_base_t::xhas_out ()
     return false;
 }
 
-int zmq::socket_base_t::xsend (msg_t *msg_, int options_)
+int zmq::socket_base_t::xsend (msg_t *msg_, int flags_)
 {
     errno = ENOTSUP;
     return -1;
@@ -775,7 +775,7 @@ bool zmq::socket_base_t::xhas_in ()
     return false;
 }
 
-int zmq::socket_base_t::xrecv (msg_t *msg_, int options_)
+int zmq::socket_base_t::xrecv (msg_t *msg_, int flags_)
 {
     errno = ENOTSUP;
     return -1;
@@ -790,6 +790,11 @@ void zmq::socket_base_t::xwrite_activated (pipe_t *pipe_)
     zmq_assert (false);
 }
 
+void zmq::socket_base_t::xhiccuped (pipe_t *pipe_)
+{
+    zmq_assert (false);
+}
+
 void zmq::socket_base_t::in_event ()
 {
     //  Process any commands from other threads/sockets that may be available
@@ -837,6 +842,11 @@ void zmq::socket_base_t::write_activated (pipe_t *pipe_)
     xwrite_activated (pipe_);
 }
 
+void zmq::socket_base_t::hiccuped (pipe_t *pipe_)
+{
+    xhiccuped (pipe_);
+}
+
 void zmq::socket_base_t::terminated (pipe_t *pipe_)
 {
     //  Notify the specific socket type about the pipe termination.

src/socket_base.hpp

@@ -96,6 +96,7 @@ namespace zmq
         //  i_pipe_events interface implementation.
         void read_activated (pipe_t *pipe_);
         void write_activated (pipe_t *pipe_);
+        void hiccuped (pipe_t *pipe_);
         void terminated (pipe_t *pipe_);
 
     protected:
@@ -116,15 +117,16 @@ namespace zmq
 
         //  The default implementation assumes that send is not supported.
         virtual bool xhas_out ();
-        virtual int xsend (class msg_t *msg_, int options_);
+        virtual int xsend (class msg_t *msg_, int flags_);
 
         //  The default implementation assumes that recv in not supported.
         virtual bool xhas_in ();
-        virtual int xrecv (class msg_t *msg_, int options_);
+        virtual int xrecv (class msg_t *msg_, int flags_);
 
         //  i_pipe_events will be forwarded to these functions.
         virtual void xread_activated (pipe_t *pipe_);
         virtual void xwrite_activated (pipe_t *pipe_);
+        virtual void xhiccuped (pipe_t *pipe_);
         virtual void xterminated (pipe_t *pipe_) = 0;
 
         //  Delay actual destruction of the socket.

src/sub.cpp

@@ -22,43 +22,18 @@
 #include "msg.hpp"
 
 zmq::sub_t::sub_t (class ctx_t *parent_, uint32_t tid_) :
-    xsub_t (parent_, tid_),
-    has_message (false),
-    more (false)
+    xsub_t (parent_, tid_)
 {
     options.type = ZMQ_SUB;
-    int rc = message.init ();
-    errno_assert (rc == 0);
 }
 
 zmq::sub_t::~sub_t ()
 {
-    int rc = message.close ();
-    errno_assert (rc == 0);
 }
 
 int zmq::sub_t::xsetsockopt (int option_, const void *optval_,
     size_t optvallen_)
 {
-    //  Process a subscription.
-    if (option_ == ZMQ_SUBSCRIBE)
-        subscriptions.add ((unsigned char*) optval_, optvallen_);
-
-    //  Process an unsubscription. Return error if there is no corresponding
-    //  subscription.
-    else if (option_ == ZMQ_UNSUBSCRIBE) {
-        if (!subscriptions.rm ((unsigned char*) optval_, optvallen_)) {
-            errno = EINVAL;
-            return -1;
-        }
-    }
-
-    //  Unknow option.
-    else {
-        errno = EINVAL;
-        return -1;
-    }
-
     //  Create the subscription message.
     msg_t msg;
     int rc = msg.init_size (optvallen_ + 1);
@@ -82,7 +57,7 @@ int zmq::sub_t::xsetsockopt (int option_, const void *optval_,
     return rc;
 }
 
-int zmq::sub_t::xsend (msg_t *msg_, int options_)
+int zmq::sub_t::xsend (msg_t *msg_, int flags_)
 {
     //  Overload the XSUB's send.
     errno = ENOTSUP;
@@ -95,89 +70,3 @@ bool zmq::sub_t::xhas_out ()
     return false;
 }
 
-int zmq::sub_t::xrecv (msg_t *msg_, int flags_)
-{
-    //  If there's already a message prepared by a previous call to zmq_poll,
-    //  return it straight ahead.
-    if (has_message) {
-        int rc = msg_->move (message);
-        errno_assert (rc == 0);
-        has_message = false;
-        more = msg_->flags () & msg_t::more;
-        return 0;
-    }
-
-    //  TODO: This can result in infinite loop in the case of continuous
-    //  stream of non-matching messages which breaks the non-blocking recv
-    //  semantics.
-    while (true) {
-
-        //  Get a message using fair queueing algorithm.
-        int rc = xsub_t::xrecv (msg_, flags_);
-
-        //  If there's no message available, return immediately.
-        //  The same when error occurs.
-        if (rc != 0)
-            return -1;
-
-        //  Check whether the message matches at least one subscription.
-        //  Non-initial parts of the message are passed 
-        if (more || match (msg_)) {
-            more = msg_->flags () & msg_t::more;
-            return 0;
-        }
-
-        //  Message doesn't match. Pop any remaining parts of the message
-        //  from the pipe.
-        while (msg_->flags () & msg_t::more) {
-            rc = xsub_t::xrecv (msg_, ZMQ_DONTWAIT);
-            zmq_assert (rc == 0);
-        }
-    }
-}
-
-bool zmq::sub_t::xhas_in ()
-{
-    //  There are subsequent parts of the partly-read message available.
-    if (more)
-        return true;
-
-    //  If there's already a message prepared by a previous call to zmq_poll,
-    //  return straight ahead.
-    if (has_message)
-        return true;
-
-    //  TODO: This can result in infinite loop in the case of continuous
-    //  stream of non-matching messages.
-    while (true) {
-
-        //  Get a message using fair queueing algorithm.
-        int rc = xsub_t::xrecv (&message, ZMQ_DONTWAIT);
-
-        //  If there's no message available, return immediately.
-        //  The same when error occurs.
-        if (rc != 0) {
-            zmq_assert (errno == EAGAIN);
-            return false;
-        }
-
-        //  Check whether the message matches at least one subscription.
-        if (match (&message)) {
-            has_message = true;
-            return true;
-        }
-
-        //  Message doesn't match. Pop any remaining parts of the message
-        //  from the pipe.
-        while (message.flags () & msg_t::more) {
-            rc = xsub_t::xrecv (&message, ZMQ_DONTWAIT);
-            zmq_assert (rc == 0);
-        }
-    }
-}
-
-bool zmq::sub_t::match (msg_t *msg_)
-{
-    return subscriptions.check ((unsigned char*) msg_->data (), msg_->size ());
-}
-

src/sub.hpp

@@ -22,8 +22,6 @@
 #define __ZMQ_SUB_HPP_INCLUDED__
 
 #include "xsub.hpp"
-#include "trie.hpp"
-#include "msg.hpp"
 
 namespace zmq
 {
@@ -38,28 +36,11 @@ namespace zmq
     protected:
 
         int xsetsockopt (int option_, const void *optval_, size_t optvallen_);
-        int xsend (class msg_t *msg_, int options_);
+        int xsend (class msg_t *msg_, int flags_);
         bool xhas_out ();
-        int xrecv (class msg_t *msg_, int flags_);
-        bool xhas_in ();
 
     private:
 
-        //  Check whether the message matches at least one subscription.
-        bool match (class msg_t *msg_);
-
-        //  The repository of subscriptions.
-        trie_t subscriptions;
-
-        //  If true, 'message' contains a matching message to return on the
-        //  next recv call.
-        bool has_message;
-        msg_t message;
-
-        //  If true, part of a multipart message was already received, but
-        //  there are following parts still waiting.
-        bool more;
-
         sub_t (const sub_t&);
         const sub_t &operator = (const sub_t&);
     };

src/transient_session.cpp

@@ -30,14 +30,14 @@ zmq::transient_session_t::~transient_session_t ()
 {
 }
 
-bool zmq::transient_session_t::attached (const blob_t &peer_identity_)
+bool zmq::transient_session_t::xattached (const blob_t &peer_identity_)
 {
     //  Transient session is always valid.
     return true;
 }
 
-void zmq::transient_session_t::detached ()
+bool zmq::transient_session_t::xdetached ()
 {
     //  There's no way to reestablish a transient session. Tear it down.
-    terminate ();
+    return false;
 }

src/transient_session.hpp

@@ -40,8 +40,8 @@ namespace zmq
     private:
 
         //  Handlers for events from session base class.
-        bool attached (const blob_t &peer_identity_);
-        void detached ();
+        bool xattached (const blob_t &peer_identity_);
+        bool xdetached ();
 
         transient_session_t (const transient_session_t&);
         const transient_session_t &operator = (const transient_session_t&);

src/trie.cpp

@@ -50,12 +50,12 @@ zmq::trie_t::~trie_t ()
     }
 }
 
-void zmq::trie_t::add (unsigned char *prefix_, size_t size_)
+bool zmq::trie_t::add (unsigned char *prefix_, size_t size_)
 {
     //  We are at the node corresponding to the prefix. We are done.
     if (!size_) {
         ++refcnt;
-        return;
+        return refcnt == 1;
     }
 
     unsigned char c = *prefix_;
@@ -74,7 +74,7 @@ void zmq::trie_t::add (unsigned char *prefix_, size_t size_)
             count = (min < c ? c - min : min - c) + 1;
             next.table = (trie_t**)
                 malloc (sizeof (trie_t*) * count);
-            alloc_assert (next.table);
+            zmq_assert (next.table);
             for (unsigned short i = 0; i != count; ++i)
                 next.table [i] = 0;
             min = std::min (min, c);
@@ -111,26 +111,28 @@ void zmq::trie_t::add (unsigned char *prefix_, size_t size_)
     if (count == 1) {
         if (!next.node) {
             next.node = new (std::nothrow) trie_t;
-            alloc_assert (next.node);
+            zmq_assert (next.node);
         }
-        next.node->add (prefix_ + 1, size_ - 1);
+        return next.node->add (prefix_ + 1, size_ - 1);
     }
     else {
         if (!next.table [c - min]) {
             next.table [c - min] = new (std::nothrow) trie_t;
-            alloc_assert (next.table [c - min]);
+            zmq_assert (next.table [c - min]);
         }
-        next.table [c - min]->add (prefix_ + 1, size_ - 1);
+        return next.table [c - min]->add (prefix_ + 1, size_ - 1);
     }
 }
 
 bool zmq::trie_t::rm (unsigned char *prefix_, size_t size_)
 {
+     //  TODO: Shouldn't an error be reported if the key does not exist?
+
      if (!size_) {
          if (!refcnt)
              return false;
          refcnt--;
-         return true;
+         return refcnt == 0;
      }
 
      unsigned char c = *prefix_;
@@ -179,3 +181,48 @@ bool zmq::trie_t::check (unsigned char *data_, size_t size_)
         size_--;
     }
 }
+
+void zmq::trie_t::apply (void (*func_) (unsigned char *data_, size_t size_,
+    void *arg_), void *arg_)
+{
+    unsigned char *buff = NULL;
+    apply_helper (&buff, 0, 0, func_, arg_);
+    free (buff);
+}
+
+void zmq::trie_t::apply_helper (
+    unsigned char **buff_, size_t buffsize_, size_t maxbuffsize_,
+    void (*func_) (unsigned char *data_, size_t size_, void *arg_), void *arg_)
+{
+    //  If this node is a subscription, apply the function.
+    if (refcnt)
+        func_ (*buff_, buffsize_, arg_);
+
+    //  Adjust the buffer.
+    if (buffsize_ >= maxbuffsize_) {
+        maxbuffsize_ = buffsize_ + 256;
+        *buff_ = (unsigned char*) realloc (*buff_, maxbuffsize_);
+        zmq_assert (*buff_);
+    }
+
+    //  If there are no subnodes in the trie, return.
+    if (count == 0)
+        return;
+
+    //  If there's one subnode (optimisation).
+    if (count == 1) {
+        (*buff_) [buffsize_] = min;
+        buffsize_++;
+        next.node->apply_helper (buff_, buffsize_, maxbuffsize_, func_, arg_);
+        return;
+    }
+
+    //  If there are multiple subnodes.
+    for (unsigned char c = 0; c != count; c++) {
+        (*buff_) [buffsize_] = min + c;
+        if (next.table [c])
+            next.table [c]->apply_helper (buff_, buffsize_ + 1, maxbuffsize_,
+                func_, arg_);
+    }   
+}
+

src/trie.hpp

@@ -35,12 +35,28 @@ namespace zmq
         trie_t ();
         ~trie_t ();
 
-        void add (unsigned char *prefix_, size_t size_);
+        //  Add key to the trie. Returns true if this is a new item in the trie
+        //  rather than a duplicate.
+        bool add (unsigned char *prefix_, size_t size_);
+
+        //  Remove key from the trie. Returns true if the item is actually
+        //  removed from the trie.
         bool rm (unsigned char *prefix_, size_t size_);
+
+        //  Check whether particular key is in the trie.
         bool check (unsigned char *data_, size_t size_);
 
+        //  Apply the function supplied to each subscription in the trie.
+        void apply (void (*func_) (unsigned char *data_, size_t size_,
+            void *arg_), void *arg_);
+
     private:
 
+        void apply_helper (
+            unsigned char **buff_, size_t buffsize_, size_t maxbuffsize_,
+            void (*func_) (unsigned char *data_, size_t size_, void *arg_),
+            void *arg_);
+
         uint32_t refcnt;
         unsigned char min;
         unsigned short count;

src/xpub.cpp

@@ -18,6 +18,8 @@
     along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
+#include <string.h>
+
 #include "xpub.hpp"
 #include "pipe.hpp"
 #include "err.hpp"
@@ -37,6 +39,12 @@ void zmq::xpub_t::xattach_pipe (pipe_t *pipe_, const blob_t &peer_identity_)
 {
     zmq_assert (pipe_);
     dist.attach (pipe_);
+    fq.attach (pipe_);
+}
+
+void zmq::xpub_t::xread_activated (pipe_t *pipe_)
+{
+    fq.activated (pipe_);
 }
 
 void zmq::xpub_t::xwrite_activated (pipe_t *pipe_)
@@ -46,11 +54,37 @@ void zmq::xpub_t::xwrite_activated (pipe_t *pipe_)
 
 void zmq::xpub_t::xterminated (pipe_t *pipe_)
 {
+    //  Remove the pipe from the trie. If there are topics that nobody
+    //  is interested in anymore, send corresponding unsubscriptions
+    //  upstream.
+    subscriptions.rm (pipe_, send_unsubscription, this);
+
     dist.terminated (pipe_);
+    fq.terminated (pipe_);
 }
 
 int zmq::xpub_t::xsend (msg_t *msg_, int flags_)
 {
+    //  First, process any (un)subscriptions from downstream.
+    msg_t sub;
+    sub.init ();
+    while (true) {
+
+        //  Grab next subscription.
+        pipe_t *pipe;
+        int rc = fq.recvpipe (&sub, 0, &pipe);
+        if (rc != 0 && errno == EAGAIN)
+            break;
+        errno_assert (rc == 0);
+
+        //  Apply the subscription to the trie. If it's not a duplicate,
+        //  store it so that it can be passed to used on next recv call.
+        if (apply_subscription (&sub, pipe) && options.type != ZMQ_PUB)
+            pending.push_back (blob_t ((unsigned char*) sub.data (),
+                sub.size ()));
+    }
+    sub.close ();
+    
     return dist.send (msg_, flags_);
 }
 
@@ -61,12 +95,92 @@ bool zmq::xpub_t::xhas_out ()
 
 int zmq::xpub_t::xrecv (msg_t *msg_, int flags_)
 {
+    //  If there is at least one 
+    if (!pending.empty ()) {
+        int rc = msg_->close ();
+        errno_assert (rc == 0);
+        rc = msg_->init_size (pending.front ().size ());
+        errno_assert (rc == 0);
+        memcpy (msg_->data (), pending.front ().data (),
+            pending.front ().size ());
+        pending.pop_front ();
+        return 0;
+    }
+
+    //  Grab and apply next subscription.
+    pipe_t *pipe;
+    int rc = fq.recvpipe (msg_, 0, &pipe);
+    if (rc != 0)
+        return -1;
+    if (!apply_subscription (msg_, pipe)) {
+//  TODO: This should be a loop rather!
+        msg_->close ();
+        msg_->init ();
         errno = EAGAIN;
         return -1;
+    }
+    return 0;
 }
 
 bool zmq::xpub_t::xhas_in ()
 {
+    if (!pending.empty ())
+        return true;
+
+    //  Even if there are subscriptions in the fair-queuer they may be
+    //  duplicates. Thus, we have to check by hand wheter there is any
+    //  subscription available to pass upstream.
+    //  First, process any (un)subscriptions from downstream.
+    msg_t sub;
+    sub.init ();
+    while (true) {
+
+        //  Grab next subscription.
+        pipe_t *pipe;
+        int rc = fq.recvpipe (&sub, 0, &pipe);
+        if (rc != 0 && errno == EAGAIN) {
+            sub.close ();
             return false;
+        }
+        errno_assert (rc == 0);
+
+        //  Apply the subscription to the trie. If it's not a duplicate store
+        //  it so that it can be passed to used on next recv call.
+        if (apply_subscription (&sub, pipe) && options.type != ZMQ_PUB) {
+            pending.push_back (blob_t ((unsigned char*) sub.data (),
+                sub.size ()));
+            sub.close ();
+            return true;
+        }
+    }
+}
+
+bool zmq::xpub_t::apply_subscription (msg_t *sub_, pipe_t *pipe_)
+{
+    unsigned char *data = (unsigned char*) sub_->data ();
+    size_t size = sub_->size ();
+    zmq_assert (size > 0 && (*data == 0 || *data == 1));
+
+    if (*data == 0)
+        return subscriptions.rm (data + 1, size - 1, pipe_);
+    else
+        return subscriptions.add (data + 1, size - 1, pipe_);
+}
+
+void zmq::xpub_t::send_unsubscription (unsigned char *data_, size_t size_,
+    void *arg_)
+{
+    xpub_t *self = (xpub_t*) arg_;
+
+    if (self->options.type != ZMQ_PUB) {
+
+		//  Place the unsubscription to the queue of pending (un)sunscriptions
+		//  to be retrived by the user later on.
+		xpub_t *self = (xpub_t*) arg_;
+		blob_t unsub (size_ + 1, 0);
+		unsub [0] = 0;
+		memcpy (&unsub [1], data_, size_);
+		self->pending.push_back (unsub);
+    }
 }
 

src/xpub.hpp

@@ -21,9 +21,14 @@
 #ifndef __ZMQ_XPUB_HPP_INCLUDED__
 #define __ZMQ_XPUB_HPP_INCLUDED__
 
+#include <deque>
+
 #include "socket_base.hpp"
+#include "mtrie.hpp"
 #include "array.hpp"
+#include "blob.hpp"
 #include "dist.hpp"
+#include "fq.hpp"
 
 namespace zmq
 {
@@ -42,14 +47,35 @@ namespace zmq
         bool xhas_out ();
         int xrecv (class msg_t *msg_, int flags_);
         bool xhas_in ();
+        void xread_activated (class pipe_t *pipe_);
         void xwrite_activated (class pipe_t *pipe_);
         void xterminated (class pipe_t *pipe_);
 
     private:
 
+        //  Applies the subscription to the trie. Return false if it is a
+        //  duplicate.
+        bool apply_subscription (class msg_t *sub_, class pipe_t *pipe_);
+
+        //  Function to be applied to the trie to send all the subsciptions
+        //  upstream.
+        static void send_unsubscription (unsigned char *data_, size_t size_,
+            void *arg_);
+
+        //  List of all subscriptions mapped to corresponding pipes.
+        mtrie_t subscriptions;
+
         //  Distributor of messages holding the list of outbound pipes.
         dist_t dist;
 
+        //  Object to fair-queue the subscription requests.
+        fq_t fq;
+
+        //  List of pending (un)subscriptions, ie. those that were already
+        //  applied to the trie, but not yet received by the user.
+        typedef std::deque <blob_t> pending_t;
+        pending_t pending;
+
         xpub_t (const xpub_t&);
         const xpub_t &operator = (const xpub_t&);
     };

src/xsub.cpp

@@ -24,19 +24,30 @@
 #include "err.hpp"
 
 zmq::xsub_t::xsub_t (class ctx_t *parent_, uint32_t tid_) :
-    socket_base_t (parent_, tid_)
+    socket_base_t (parent_, tid_),
+    has_message (false),
+    more (false)
 {
     options.type = ZMQ_XSUB;
+    int rc = message.init ();
+    errno_assert (rc == 0);
 }
 
 zmq::xsub_t::~xsub_t ()
 {
+    int rc = message.close ();
+    errno_assert (rc == 0);
 }
 
 void zmq::xsub_t::xattach_pipe (pipe_t *pipe_, const blob_t &peer_identity_)
 {
     zmq_assert (pipe_);
     fq.attach (pipe_);
+    dist.attach (pipe_);
+
+    //  Send all the cached subscriptions to the new upstream peer.
+    subscriptions.apply (send_subscription, pipe_);
+    pipe_->flush ();
 }
 
 void zmq::xsub_t::xread_activated (pipe_t *pipe_)
@@ -44,20 +55,49 @@ void zmq::xsub_t::xread_activated (pipe_t *pipe_)
     fq.activated (pipe_);
 }
 
+void zmq::xsub_t::xwrite_activated (pipe_t *pipe_)
+{
+    dist.activated (pipe_);
+}
+
 void zmq::xsub_t::xterminated (pipe_t *pipe_)
 {
     fq.terminated (pipe_);
+    dist.terminated (pipe_);
 }
 
-int zmq::xsub_t::xsend (msg_t *msg_, int options_)
+void zmq::xsub_t::xhiccuped (pipe_t *pipe_)
 {
-    //  TODO: Once we'll send the subscription upstream here. For now
-    //  just empty the message.
-    int rc = msg_->close ();
-    errno_assert (rc == 0);
-    rc = msg_->init ();
-    errno_assert (rc == 0);
+    //  Send all the cached subscriptions to the hiccuped pipe.
+    subscriptions.apply (send_subscription, pipe_);
+    pipe_->flush ();
+}
+
+int zmq::xsub_t::xsend (msg_t *msg_, int flags_)
+{
+    size_t size = msg_->size ();
+    unsigned char *data = (unsigned char*) msg_->data ();
+
+    // Malformed subscriptions.
+    if (size < 1 || (*data != 0 && *data != 1)) {
+        errno = EINVAL;
+        return -1;
+    }
+
+    // Process the subscription.
+    if (*data == 1) {
+        if (subscriptions.add (data + 1, size - 1))
+            return dist.send (msg_, flags_);
+        else
             return 0;
+    }
+    else if (*data == 0) {
+        if (subscriptions.rm (data + 1, size - 1))
+            return dist.send (msg_, flags_);
+        else
+            return 0;
+    }
+    zmq_assert (false);
 }
 
 bool zmq::xsub_t::xhas_out ()
@@ -66,13 +106,109 @@ bool zmq::xsub_t::xhas_out ()
     return true;
 }
 
-int zmq::xsub_t::xrecv (class msg_t *msg_, int flags_)
+int zmq::xsub_t::xrecv (msg_t *msg_, int flags_)
 {
-    return fq.recv (msg_, flags_);
+    //  If there's already a message prepared by a previous call to zmq_poll,
+    //  return it straight ahead.
+    if (has_message) {
+        int rc = msg_->move (message);
+        errno_assert (rc == 0);
+        has_message = false;
+        more = msg_->flags () & msg_t::more;
+        return 0;
+    }
+
+    //  TODO: This can result in infinite loop in the case of continuous
+    //  stream of non-matching messages which breaks the non-blocking recv
+    //  semantics.
+    while (true) {
+
+        //  Get a message using fair queueing algorithm.
+        int rc = fq.recv (msg_, flags_);
+
+        //  If there's no message available, return immediately.
+        //  The same when error occurs.
+        if (rc != 0)
+            return -1;
+
+        //  Check whether the message matches at least one subscription.
+        //  Non-initial parts of the message are passed 
+        if (more || match (msg_)) {
+            more = msg_->flags () & msg_t::more;
+            return 0;
+        }
+
+        //  Message doesn't match. Pop any remaining parts of the message
+        //  from the pipe.
+        while (msg_->flags () & msg_t::more) {
+            rc = fq.recv (msg_, ZMQ_DONTWAIT);
+            zmq_assert (rc == 0);
+        }
+    }
 }
 
 bool zmq::xsub_t::xhas_in ()
 {
-    return fq.has_in ();
+    //  There are subsequent parts of the partly-read message available.
+    if (more)
+        return true;
+
+    //  If there's already a message prepared by a previous call to zmq_poll,
+    //  return straight ahead.
+    if (has_message)
+        return true;
+
+    //  TODO: This can result in infinite loop in the case of continuous
+    //  stream of non-matching messages.
+    while (true) {
+
+        //  Get a message using fair queueing algorithm.
+        int rc = fq.recv (&message, ZMQ_DONTWAIT);
+
+        //  If there's no message available, return immediately.
+        //  The same when error occurs.
+        if (rc != 0) {
+            zmq_assert (errno == EAGAIN);
+            return false;
+        }
+
+        //  Check whether the message matches at least one subscription.
+        if (match (&message)) {
+            has_message = true;
+            return true;
+        }
+
+        //  Message doesn't match. Pop any remaining parts of the message
+        //  from the pipe.
+        while (message.flags () & msg_t::more) {
+            rc = fq.recv (&message, ZMQ_DONTWAIT);
+            zmq_assert (rc == 0);
+        }
+    }
+}
+
+bool zmq::xsub_t::match (msg_t *msg_)
+{
+    return subscriptions.check ((unsigned char*) msg_->data (), msg_->size ());
 }
 
+void zmq::xsub_t::send_subscription (unsigned char *data_, size_t size_,
+    void *arg_)
+{
+    pipe_t *pipe = (pipe_t*) arg_;
+
+    //  Create the subsctription message.
+    msg_t msg;
+    int rc = msg.init_size (size_ + 1);
+    zmq_assert (rc == 0);
+    unsigned char *data = (unsigned char*) msg.data ();
+    data [0] = 1;
+    memcpy (data + 1, data_, size_);
+
+    //  Send it to the pipe.
+    bool sent = pipe->write (&msg);
+    zmq_assert (sent);
+    msg.close ();
+}
+
+

src/xsub.hpp

@@ -22,7 +22,10 @@
 #define __ZMQ_XSUB_HPP_INCLUDED__
 
 #include "socket_base.hpp"
+#include "dist.hpp"
 #include "fq.hpp"
+#include "trie.hpp"
+#include "msg.hpp"
 
 namespace zmq
 {
@@ -39,18 +42,43 @@ namespace zmq
 
         //  Overloads of functions from socket_base_t.
         void xattach_pipe (class pipe_t *pipe_, const blob_t &peer_identity_);
-        int xsend (class msg_t *msg_, int options_);
+        int xsend (class msg_t *msg_, int flags_);
         bool xhas_out ();
         int xrecv (class msg_t *msg_, int flags_);
         bool xhas_in ();
         void xread_activated (class pipe_t *pipe_);
+        void xwrite_activated (class pipe_t *pipe_);
+        void xhiccuped (pipe_t *pipe_);
         void xterminated (class pipe_t *pipe_);
 
     private:
 
+        //  Check whether the message matches at least one subscription.
+        bool match (class msg_t *msg_);
+
+        //  Function to be applied to the trie to send all the subsciptions
+        //  upstream.
+        static void send_subscription (unsigned char *data_, size_t size_,
+            void *arg_);
+
         //  Fair queueing object for inbound pipes.
         fq_t fq;
 
+        //  Object for distributing the subscriptions upstream.
+        dist_t dist;
+
+        //  The repository of subscriptions.
+        trie_t subscriptions;
+
+        //  If true, 'message' contains a matching message to return on the
+        //  next recv call.
+        bool has_message;
+        msg_t message;
+
+        //  If true, part of a multipart message was already received, but
+        //  there are following parts still waiting.
+        bool more;
+
         xsub_t (const xsub_t&);
         const xsub_t &operator = (const xsub_t&);
     };