Using CZMQ Building and Installing Linking with an Application API Summary zctx - working with ØMQ contexts zsocket - working with ØMQ sockets zsockopt - working with ØMQ socket options zstr - sending and receiving strings zfile - work with files zframe - working with single message frames zmsg - working with multipart messages zloop - event-driven reactor zthread - working with system threads zhash - expandable hash table container zlist - singly-linked list container zclock - millisecond clocks and delays
CZMQ has these goals:
- To wrap the ØMQ core API in semantics that are natural and lead to shorter, more readable applications.
- To hide the differences between versions of ØMQ, particularly 2.1 and 3.1.
- To provide a space for development of more sophisticated API semantics.
- Single API hides differences between ØMQ/2.1, and ØMQ/3.1.
- Work with messages as strings, individual frames, or multipart messages.
- Automatic closure of any open sockets at context termination.
- Automatic LINGER configuration of all sockets for context termination.
- Portable API for creating child threads and ØMQ pipes to talk to them.
- Simple reactor with one-off and repeated timers, and socket readers.
- System clock functions for sleeping and calculating timers.
- Easy API to get/set all socket options.
- Portable to Linux, UNIX, OS X, Windows (porting is not yet complete).
- Includes generic hash and list containers.
- Full selftests on all classes.
CZMQ is maintained by Pieter Hintjens and Mikko Koppanen (build system). Its other authors and contributors are listed in the AUTHORS file. It is held by the ZeroMQ organization at github.com.
The authors of CZMQ grant you use of this software under the terms of the GNU Lesser General Public License (LGPL). For details see the files
COPYING.LESSER in this directory.
CZMQ uses the C4 (Collective Code Construction Contract) process which says, "Everyone, without distinction or discrimination, SHALL have an equal right to become a Contributor under the terms of this contract".
To report an issue, use the CZMQ issue tracker at github.com.
CZMQ uses autotools for packaging. To build from git (all example commands are for Linux):
git clone git://github.com/zeromq/czmq.git cd czmq sh autogen.sh ./configure make all sudo make install sudo ldconfig
You will need the libtool and autotools packages. On FreeBSD, you may need to specify the default directories for configure:
After building, you can run the CZMQ selftests:
czmq.h in your application and link with libczmq. Here is a typical gcc link command:
gcc -lczmq -lzmq myapp.c -o myapp
The zctx class wraps ØMQ contexts. It manages open sockets in the context and automatically closes these before terminating the context. It provides a simple way to set the linger timeout on sockets, and configure contexts for number of I/O threads. Sets-up signal (interrrupt) handling for the process.
The zctx class has these main features:
Tracks all open sockets and automatically closes them before calling zmq_term(). This avoids an infinite wait on open sockets.
Automatically configures sockets with a ZMQ_LINGER timeout you can define, and which defaults to zero. The default behavior of zctx is therefore like ØMQ/2.0, immediate termination with loss of any pending messages. You can set any linger timeout you like by calling the zctx_set_linger() method.
Moves the iothreads configuration to a separate method, so that default usage is 1 I/O thread. Lets you configure this value.
Sets up signal (SIGINT and SIGTERM) handling so that blocking calls such as zmq_recv() and zmq_poll() will return when the user presses Ctrl-C.
This is the class interface:
// Create new context, returns context object, replaces zmq_init zctx_t * zctx_new (void); // Destroy context and all sockets in it, replaces zmq_term void zctx_destroy (zctx_t **self_p); // Raise default I/O threads from 1, for crazy heavy applications void zctx_set_iothreads (zctx_t *self, int iothreads); // Set msecs to flush sockets when closing them void zctx_set_linger (zctx_t *self, int linger); // Self test of this class int zctx_test (Bool verbose); // Global signal indicator, TRUE when user presses Ctrl-C or the process // gets a SIGTERM signal. extern int zctx_interrupted;
The zsocket class provides helper functions for ØMQ sockets. It doesn't wrap the ØMQ socket type, to avoid breaking all libzmq socket-related calls.
This is the class interface:
// This port range is defined by IANA for dynamic or private ports // We use this when choosing a port for dynamic binding. #define ZSOCKET_DYNFROM 0xc000 #define ZSOCKET_DYNTO 0xffff // Create a new socket within our CZMQ context, replaces zmq_socket. // Use this to get automatic management of the socket at shutdown. // Note: SUB sockets do not automatically subscribe to everything; you // must set filters explicitly. void * zsocket_new (zctx_t *self, int type); // Destroy a socket within our CZMQ context, replaces zmq_close. void zsocket_destroy (zctx_t *self, void *socket); // Bind a socket to a formatted endpoint. If the port is specified as // '*', binds to any free port from ZSOCKET_DYNFROM to ZSOCKET_DYNTO // and returns the actual port number used. Otherwise asserts that the // bind succeeded with the specified port number. Always returns the // port number if successful. int zsocket_bind (void *socket, const char *format, ...); // Connect a socket to a formatted endpoint // Returns 0 if OK, -1 if the endpoint was invalid. int zsocket_connect (void *socket, const char *format, ...); // Returns socket type as printable constant string char * zsocket_type_str (void *socket); // Self test of this class int zsocket_test (Bool verbose);
The zsockopt class provides access to the ØMQ getsockopt/setsockopt API.
This is the class interface:
#if (ZMQ_VERSION_MAJOR == 2) // Get socket options int zsocket_hwm (void *socket); int zsocket_swap (void *socket); int zsocket_affinity (void *socket); // Returns freshly allocated string, free when done char *zsocket_identity (void *socket); int zsocket_rate (void *socket); int zsocket_recovery_ivl (void *socket); int zsocket_recovery_ivl_msec (void *socket); int zsocket_mcast_loop (void *socket); int zsocket_sndbuf (void *socket); int zsocket_rcvbuf (void *socket); int zsocket_linger (void *socket); int zsocket_reconnect_ivl (void *socket); int zsocket_reconnect_ivl_max (void *socket); int zsocket_backlog (void *socket); int zsocket_type (void *socket); int zsocket_rcvmore (void *socket); int zsocket_fd (void *socket); int zsocket_events (void *socket); // Set socket options void zsocket_set_hwm (void *socket, int hwm); void zsocket_set_swap (void *socket, int swap); void zsocket_set_affinity (void *socket, int affinity); void zsocket_set_identity (void *socket, char * identity); void zsocket_set_rate (void *socket, int rate); void zsocket_set_recovery_ivl (void *socket, int recovery_ivl); void zsocket_set_recovery_ivl_msec (void *socket, int recovery_ivl_msec); void zsocket_set_mcast_loop (void *socket, int mcast_loop); void zsocket_set_sndbuf (void *socket, int sndbuf); void zsocket_set_rcvbuf (void *socket, int rcvbuf); void zsocket_set_linger (void *socket, int linger); void zsocket_set_reconnect_ivl (void *socket, int reconnect_ivl); void zsocket_set_reconnect_ivl_max (void *socket, int reconnect_ivl_max); void zsocket_set_backlog (void *socket, int backlog); void zsocket_set_subscribe (void *socket, char * subscribe); void zsocket_set_unsubscribe (void *socket, char * unsubscribe); #endif #if (ZMQ_VERSION_MAJOR == 3) // Get socket options int zsocket_type (void *socket); int zsocket_sndhwm (void *socket); int zsocket_rcvhwm (void *socket); int zsocket_affinity (void *socket); // Returns freshly allocated string, free when done char *zsocket_identity (void *socket); int zsocket_rate (void *socket); int zsocket_recovery_ivl (void *socket); int zsocket_sndbuf (void *socket); int zsocket_rcvbuf (void *socket); int zsocket_linger (void *socket); int zsocket_reconnect_ivl (void *socket); int zsocket_reconnect_ivl_max (void *socket); int zsocket_backlog (void *socket); int zsocket_maxmsgsize (void *socket); int zsocket_multicast_hops (void *socket); int zsocket_rcvtimeo (void *socket); int zsocket_sndtimeo (void *socket); int zsocket_ipv4only (void *socket); int zsocket_rcvmore (void *socket); int zsocket_fd (void *socket); int zsocket_events (void *socket); // Returns freshly allocated string, free when done char *zsocket_last_endpoint (void *socket); // Set socket options void zsocket_set_sndhwm (void *socket, int sndhwm); void zsocket_set_rcvhwm (void *socket, int rcvhwm); void zsocket_set_affinity (void *socket, int affinity); void zsocket_set_subscribe (void *socket, char * subscribe); void zsocket_set_unsubscribe (void *socket, char * unsubscribe); void zsocket_set_identity (void *socket, char * identity); void zsocket_set_rate (void *socket, int rate); void zsocket_set_recovery_ivl (void *socket, int recovery_ivl); void zsocket_set_sndbuf (void *socket, int sndbuf); void zsocket_set_rcvbuf (void *socket, int rcvbuf); void zsocket_set_linger (void *socket, int linger); void zsocket_set_reconnect_ivl (void *socket, int reconnect_ivl); void zsocket_set_reconnect_ivl_max (void *socket, int reconnect_ivl_max); void zsocket_set_backlog (void *socket, int backlog); void zsocket_set_maxmsgsize (void *socket, int maxmsgsize); void zsocket_set_multicast_hops (void *socket, int multicast_hops); void zsocket_set_rcvtimeo (void *socket, int rcvtimeo); void zsocket_set_sndtimeo (void *socket, int sndtimeo); void zsocket_set_ipv4only (void *socket, int ipv4only); void zsocket_set_fail_unroutable (void *socket, int fail_unroutable); // Emulation of widely-used 2.x socket options void zsockopt_set_hwm (void *socket, int hwm); #endif // Self test of this class int zsockopt_test (Bool verbose);
This class is generated, using the GSL code generator. See the sockopts XML file, which provides the metadata, and the sockopts.gsl template, which does the work.
The zstr class provides utility functions for sending and receiving C strings across ØMQ sockets. It sends strings without a terminating null, and appends a null byte on received strings. This class is for simple message sending.
This is the class interface:
// Receive a string off a socket, caller must free it char * zstr_recv (void *socket); // Receive a string off a socket if socket had input waiting char * zstr_recv_nowait (void *socket); // Send a string to a socket in ØMQ string format int zstr_send (void *socket, const char *string); // Send a string to a socket in ØMQ string format, with MORE flag int zstr_sendm (void *socket, const char *string); // Send a formatted string to a socket int zstr_sendf (void *socket, const char *format, ...); // Send formatted C string to socket with MORE flag int zstr_sendfm (void *socket, const char *format, ...); // Self test of this class int zstr_test (Bool verbose);
The zfile class provides methods to work with files.
This is the class interface:
// Delete file. Does not complain if the file is absent int zfile_delete (char *filename); // Make directory (maximum one level depending on OS) int zfile_mkdir (char *dirname); // Return 1 if file exists, else zero int zfile_exists (char *filename); // Return size of file, or -1 if not found ssize_t zfile_size (char *filename); // Self test of this class int zfile_test (Bool verbose);
The zframe class provides methods to send and receive single message frames across ØMQ sockets. A 'frame' corresponds to one zmq_msg_t. When you read a frame from a socket, the zframe_more() method indicates if the frame is part of an unfinished multipart message. The zframe_send method normally destroys the frame, but with the ZFRAME_REUSE flag, you can send the same frame many times. Frames are binary, and this class has no special support for text data.
This is the class interface:
#define ZFRAME_MORE 1 #define ZFRAME_REUSE 2 #define ZFRAME_DONTWAIT 4 // Callback function for zframe_free_fn method typedef void (zframe_free_fn) (void *data, void *arg); // Create a new frame with optional size, and optional data zframe_t * zframe_new (const void *data, size_t size); // Create a zero-copy frame zframe_t * zframe_new_zero_copy (void *data, size_t size, zframe_free_fn *free_fn, void *arg); // Destroy a frame void zframe_destroy (zframe_t **self_p); // Receive frame from socket, returns zframe_t object or NULL if the recv // was interrupted. Does a blocking recv, if you want to not block then use // zframe_recv_nowait(). zframe_t * zframe_recv (void *socket); // Receive a new frame off the socket. Returns newly allocated frame, or // NULL if there was no input waiting, or if the read was interrupted. zframe_t * zframe_recv_nowait (void *socket); // Send a frame to a socket, destroy frame after sending. Returns // non-zero error code on failure. int zframe_send (zframe_t **self_p, void *socket, int flags); // Return number of bytes in frame data size_t zframe_size (zframe_t *self); // Return address of frame data byte * zframe_data (zframe_t *self); // Create a new frame that duplicates an existing frame zframe_t * zframe_dup (zframe_t *self); // Return frame data encoded as printable hex string char * zframe_strhex (zframe_t *self); // Return frame data copied into freshly allocated string char * zframe_strdup (zframe_t *self); // Return TRUE if frame body is equal to string, excluding terminator Bool zframe_streq (zframe_t *self, char *string); // Return frame zero copy indicator (1 or 0) int zframe_zero_copy (zframe_t *self); // Return frame 'more' property int zframe_more (zframe_t *self); // Return TRUE if two frames have identical size and data // If either frame is NULL, equality is always false. Bool zframe_eq (zframe_t *self, zframe_t *other); // Print contents of frame to stderr void zframe_print (zframe_t *self, char *prefix); // Set new contents for frame void zframe_reset (zframe_t *self, const void *data, size_t size); // Self test of this class int zframe_test (Bool verbose);
The zmsg class provides methods to send and receive multipart messages across ØMQ sockets. This class provides a list-like container interface, with methods to work with the overall container. zmsg_t messages are composed of zero or more zframe_t frames.
This is the class interface:
// Create a new empty message object zmsg_t * zmsg_new (void); // Destroy a message object and all frames it contains void zmsg_destroy (zmsg_t **self_p); // Read 1 or more frames off the socket, into a new message object zmsg_t * zmsg_recv (void *socket); // Send a message to the socket, and then destroy it void zmsg_send (zmsg_t **self_p, void *socket); // Return number of frames in message size_t zmsg_size (zmsg_t *self); // Return combined size of all frames in message size_t zmsg_content_size (zmsg_t *self); // Push frame to front of message, before first frame int zmsg_push (zmsg_t *self, zframe_t *frame); // Pop frame off front of message, caller now owns frame zframe_t * zmsg_pop (zmsg_t *self); // Add frame to end of message, after last frame int zmsg_add (zmsg_t *self, zframe_t *frame); // Push block of memory as new frame to front of message. // Returns 0 on success, -1 on error. int zmsg_pushmem (zmsg_t *self, const void *src, size_t size); // Push block of memory as new frame to end of message. // Returns 0 on success, -1 on error. int zmsg_addmem (zmsg_t *self, const void *src, size_t size); // Push string as new frame to front of message. // Returns 0 on success, -1 on error. int zmsg_pushstr (zmsg_t *self, const char *format, ...); // Push string as new frame to end of message. // Returns 0 on success, -1 on error. int zmsg_addstr (zmsg_t *self, const char *format, ...); // Pop frame off front of message, return as fresh string char * zmsg_popstr (zmsg_t *self); // Push frame to front of message, before first frame // Pushes an empty frame in front of frame void zmsg_wrap (zmsg_t *self, zframe_t *frame); // Pop frame off front of message, caller now owns frame // If next frame is empty, pops and destroys that empty frame. zframe_t * zmsg_unwrap (zmsg_t *self); // Remove frame from message, at any position, caller owns it void zmsg_remove (zmsg_t *self, zframe_t *frame); // Return first frame in message, or null zframe_t * zmsg_first (zmsg_t *self); // Return next frame in message, or null zframe_t * zmsg_next (zmsg_t *self); // Return last frame in message, or null zframe_t * zmsg_last (zmsg_t *self); // Save message to an open file, return 0 if OK, else -1. int zmsg_save (zmsg_t *self, FILE *file); // Load/append an open file into message, create new message if // null message provided. zmsg_t * zmsg_load (zmsg_t *self, FILE *file); // Encode message to a new buffer, return buffer size size_t zmsg_encode (zmsg_t *self, byte **buffer); // Decode a buffer into a new message, returns NULL if buffer is not // properly formatted. zmsg_t * zmsg_decode (byte *buffer, size_t buffer_size); // Create copy of message, as new message object zmsg_t * zmsg_dup (zmsg_t *self); // Print message to stderr, for debugging void zmsg_dump (zmsg_t *self); // Self test of this class int zmsg_test (Bool verbose);
The zloop class provides an event-driven reactor pattern. The reactor handles zmq_pollitem_t items (pollers or writers, sockets or fds), and once-off or repeated timers. Its resolution is 1 msec. It uses a tickless timer to reduce CPU interrupts in inactive processes.
This is the class interface:
// Callback function for reactor events typedef int (zloop_fn) (zloop_t *loop, zmq_pollitem_t *item, void *arg); // Create a new zloop reactor zloop_t * zloop_new (void); // Destroy a reactor void zloop_destroy (zloop_t **self_p); // Register pollitem with the reactor. When the pollitem is ready, will call // the handler, passing the arg. Returns 0 if OK, -1 if there was an error. // If you register the pollitem more than once, each instance will invoke its // corresponding handler. int zloop_poller (zloop_t *self, zmq_pollitem_t *item, zloop_fn handler, void *arg); // Cancel a pollitem from the reactor, specified by socket or FD. If both // are specified, uses only socket. If multiple poll items exist for same // socket/FD, cancels ALL of them. void zloop_poller_end (zloop_t *self, zmq_pollitem_t *item); // Register a timer that expires after some delay and repeats some number of // times. At each expiry, will call the handler, passing the arg. To // run a timer forever, use 0 times. Returns 0 if OK, -1 if there was an // error. int zloop_timer (zloop_t *self, size_t delay, size_t times, zloop_fn handler, void *arg); // Cancel all timers for a specific argument (as provided in zloop_timer) int zloop_timer_end (zloop_t *self, void *arg); // Set verbose tracing of reactor on/off void zloop_set_verbose (zloop_t *self, Bool verbose); // Start the reactor. Takes control of the thread and returns when the ØMQ // context is terminated or the process is interrupted, or any event handler // returns -1. Event handlers may register new sockets and timers, and // cancel sockets. Returns 0 if interrupted, -1 if cancelled by a handler. int zloop_start (zloop_t *self); // Self test of this class int zloop_test (Bool verbose);
The zthread class wraps OS thread creation. It creates detached threads that look like normal OS threads, or attached threads that share the caller's ØMQ context, and get a pipe to talk back to the parent thread.
This is the class interface:
// Detached threads follow POSIX pthreads API typedef void *(zthread_detached_fn) (void *args); // Attached threads get context and pipe from parent typedef void (zthread_attached_fn) (void *args, zctx_t *ctx, void *pipe); // Create a detached thread. A detached thread operates autonomously // and is used to simulate a separate process. It gets no ctx, and no // pipe. int zthread_new (zthread_detached_fn *thread_fn, void *args); // Create an attached thread. An attached thread gets a ctx and a PAIR // pipe back to its parent. It must monitor its pipe, and exit if the // pipe becomes unreadable. Do not destroy the ctx, the thread does this // automatically when it ends. void * zthread_fork (zctx_t *ctx, zthread_attached_fn *thread_fn, void *args); // Self test of this class int zthread_test (Bool verbose);
One problem is when our application needs child threads. If we simply use pthreads_create() we're faced with several issues. First, it's not portable to legacy OSes like win32. Second, how can a child thread get access to our zctx object? If we just pass it around, we'll end up sharing the pipe socket (which we use to talk to the agent) between threads, and that will then crash ØMQ. Sockets cannot be used from more than one thread at a time.
So each child thread needs its own pipe to the agent. For the agent, this is fine, it can talk to a million threads. But how do we create those pipes in the child thread? We can't, not without help from the main thread. The solution is to wrap thread creation, like we wrap socket creation. To create a new thread, the app calls zctx_thread_new() and this method creates a dedicated zctx object, with a pipe, and then it passes that object to the newly minted child thread.
The neat thing is we can hide non-portable aspects. Windows is really a mess when it comes to threads. Three different APIs, none of which is really right, so you have to do rubbish like manually cleaning up when a thread finishes. Anyhow, it's hidden in this class so you don't need to worry.
Second neat thing about wrapping thread creation is we can make it a more enriching experience for all involved. One thing I do often is use a PAIR-PAIR pipe to talk from a thread to/from its parent. So this class will automatically create such a pair for each thread you start.
Expandable hash table container
This is the class interface:
// Callback function for zhash_foreach method typedef int (zhash_foreach_fn) (char *key, void *item, void *argument); // Callback function for zhash_freefn method typedef void (zhash_free_fn) (void *data); // Create a new, empty hash container zhash_t * zhash_new (void); // Destroy a hash container and all items in it void zhash_destroy (zhash_t **self_p); // Insert item into hash table with specified key and item. // If key is already present returns -1 and leaves existing item unchanged // Returns 0 on success. int zhash_insert (zhash_t *self, char *key, void *item); // Update item into hash table with specified key and item. // If key is already present, destroys old item and inserts new one. // Use free_fn method to ensure deallocator is properly called on item. void zhash_update (zhash_t *self, char *key, void *item); // Remove an item specified by key from the hash table. If there was no such // item, this function does nothing. void zhash_delete (zhash_t *self, char *key); // Return the item at the specified key, or null void * zhash_lookup (zhash_t *self, char *key); // Reindexes an item from an old key to a new key. If there was no such // item, does nothing. Returns 0 if successful, else -1. int zhash_rename (zhash_t *self, char *old_key, char *new_key); // Set a free function for the specified hash table item. When the item is // destroyed, the free function, if any, is called on that item. // Use this when hash items are dynamically allocated, to ensure that // you don't have memory leaks. You can pass 'free' or NULL as a free_fn. // Returns the item, or NULL if there is no such item. void * zhash_freefn (zhash_t *self, char *key, zhash_free_fn *free_fn); // Return the number of keys/items in the hash table size_t zhash_size (zhash_t *self); // Apply function to each item in the hash table. Items are iterated in no // defined order. Stops if callback function returns non-zero and returns // final return code from callback function (zero = success). int zhash_foreach (zhash_t *self, zhash_foreach_fn *callback, void *argument); // Self test of this class void zhash_test (int verbose);
Note that it's relatively slow (~50k insertions/deletes per second), so don't do inserts/updates on the critical path for message I/O. It can do ~2.5M lookups per second for 16-char keys. Timed on a 1.6GHz CPU.
Provides a generic container implementing a fast singly-linked list. You can use this to construct multi-dimensional lists, and other structures together with other generic containers like zhash.
This is the class interface:
// Create a new list container zlist_t * zlist_new (void); // Destroy a list container void zlist_destroy (zlist_t **self_p); // Return first item in the list, or null void * zlist_first (zlist_t *self); // Return last item in the list, or null void * zlist_last (zlist_t *self); // Return first item in the list, or null, leaves the cursor void * zlist_head (zlist_t *self); // Return last item in the list, or null, leaves the cursor void * zlist_tail (zlist_t *self); // Return next item in the list, or null void * zlist_next (zlist_t *self); // Append an item to the end of the list int zlist_append (zlist_t *self, void *item); // Push an item to the start of the list int zlist_push (zlist_t *self, void *item); // Pop the item off the start of the list, if any void * zlist_pop (zlist_t *self); // Remove the specified item from the list if present void zlist_remove (zlist_t *self, void *item); // Copy the entire list, return the copy zlist_t * zlist_copy (zlist_t *self); // Return number of items in the list size_t zlist_size (zlist_t *self); // Self test of this class void zlist_test (int verbose);
The zclock class provides essential sleep and system time functions, used to slow down threads for testing, and calculate timers for polling. Wraps the non-portable system calls in a simple portable API.
This is the class interface:
// Sleep for a number of milliseconds void zclock_sleep (int msecs); // Return current system clock as milliseconds int64_t zclock_time (void); // Print formatted string to stdout, prefixed by date/time and // terminated with a newline. void zclock_log (const char *format, ...); // Self test of this class int zclock_test (Bool verbose);
This class contains some small surprises. Most amazing, win32 did an API better than POSIX. The win32 Sleep() call is not only a neat 1-liner, it also sleeps for milliseconds, whereas the POSIX call asks us to think in terms of nanoseconds, which is insane. I've decided every single man page for this library will say "insane" at least once. Anyhow, milliseconds are a concept we can deal with. Seconds are too fat, nanoseconds too tiny, but milliseconds are just right for slices of time we want to work with at the ØMQ scale. zclock doesn't give you objects to work with, we like the czmq class model but we're not insane. There, got it in again. The Win32 Sleep() call defaults to 16ms resolution unless the system timer resolution is increased with a call to timeBeginPeriod() permitting 1ms granularity.
C has the significant advantage of being a small language that, if we take a little care with formatting and naming, can be easily interchanged between developers. Every C developer will use much the same 90% of the language. Larger languages like C++ provide powerful abstractions like STL containers but at the cost of interchange.
The huge problem with C is that any realistic application needs packages of functionality to bring the language up to the levels we expect for the 21st century. Much can be done by using external libraries but every additional library is a dependency that makes the resulting applications harder to build and port. While C itself is a highly portable language (and can be made more so by careful use of the preprocessor), most C libraries consider themselves part of the operating system, and as such do not attempt to be portable.
The answer to this, as we learned from building enterprise-level C applications at iMatix from 1995-2005, is to create our own fully portable, high-quality libraries of pre-packaged functionality, in C. Doing this right solves both the requirements of richness of the language, and of portability of the final applications.
C has no standard API style. It is one thing to write a useful component, but something else to provide an API that is consistent and obvious across many components. We learned from building OpenAMQ, a messaging client and server of 0.5M LoC, that a consistent model for extending C makes life for the application developer much easier.
The general model is that of a class (the source package) that provides objects (in fact C structures). The application creates objects and then works with them. When done, the application destroys the object. In C, we tend to use the same name for the object as the class, when we can, and it looks like this (to take a fictitious CZMQ class):
zregexp_t *regexp = zregexp_new (regexp_string); if (!regexp) printf ("E: invalid regular expression: %s\n", regexp_string); else if (zregexp_match (regexp, input_buffer)) printf ("I: successful match for %s\n", input buffer); zregexp_destroy (®exp);
As far as the C program is concerned, the object is a reference to a structure (not a void pointer). We pass the object reference to all methods, since this is still C. We could do weird stuff like put method addresses into the structure so that we can emulate a C++ syntax but it's not worthwhile. The goal is not to emulate an OO system, it's simply to gain consistency. The constructor returns an object reference, or NULL if it fails. The destructor nullifies the class pointer, and is idempotent.
What we aim at here is the simplest possible consistent syntax.
No model is fully consistent, and classes can define their own rules if it helps make a better result. For example:
Some classes may not be opaque. For example, we have cases of generated serialization classes that encode and decode structures to/from binary buffers. It feels clumsy to have to use methods to access the properties of these classes.
While every class has a new method that is the formal constructor, some methods may also act as constructors. For example, a "dup" method might take one object and return a second object.
While every class has a destroy method that is the formal destructor, some methods may also act as destructors. For example, a method that sends an object may also destroy the object (so that ownership of any buffers can passed to background threads). Such methods take the same "pointer to a reference" argument as the destroy method.
CZMQ aims for short, consistent names, following the theory that names we use most often should be shortest. Classes get one-word names, unless they are part of a family of classes in which case they may be two words, the first being the family name. Methods, similarly, get one-word names and we aim for consistency across classes (so a method that does something semantically similar in two classes will get the same name in both). So the canonical name for any method is:
And the reader can easily parse this without needing special syntax to separate the class name from the method name.
After a long experiment with containers, we've decided that we need exactly two containers:
- A singly-linked list.
- A hash table using text keys.
These are zlist and zhash, respectively. Both store void pointers, with no attempt to manage the details of contained objects. You can use these containers to create lists of lists, hashes of lists, hashes of hashes, etc.
We assume that at some point we'll need to switch to a doubly-linked list.
Creating a portable C application can be rewarding in terms of maintaining a single code base across many platforms, and keeping (expensive) system-specific knowledge separate from application developers. In most projects (like ØMQ core), there is no portability layer and application code does conditional compilation for all mixes of platforms. This leads to quite messy code.
CZMQ is a portability layer, similar to but thinner than libraries like the Apache Portable Runtime (APR).
These are the places a C application is subject to arbitrary system differences:
- Different compilers may offer slightly different variants of the C language, often lacking specific types or using neat non-portable names. Windows is a big culprit here. We solve this by 'patching' the language in czmq_prelude.h, e.g. defining int64_t on Windows.
- System header files are inconsistent, i.e. you need to include different files depending on the OS type and version. We solve this by pulling in all necessary header files in czmq_prelude.h. This is a proven brute-force approach that increases recompilation times but eliminates a major source of pain.
- System libraries are inconsistent, i.e. you need to link with different libraries depending on the OS type and version. We solve this with an external compilation tool, 'C', which detects the OS type and version (at runtime) and builds the necessary link commands.
- System functions are inconsistent, i.e. you need to call different functions depending, again, on OS type and version. We solve this by building small abstract classes that handle specific areas of functionality, and doing conditional compilation in these.
An example of the last:
#if (defined (__UNIX__)) pid = GetCurrentProcessId(); #elif (defined (__WINDOWS__)) pid = getpid (); #else pid = 0; #endif
CZMQ uses the GNU autotools system, so non-portable code can use the macros this defines. It can also use macros defined by the czmq_prelude.h header file.
- Thread safety: the use of opaque structures is thread safe, though ØMQ applications should not share state between threads in any case.
Name spaces: we prefix class names with
z, which ensures that all exported functions are globally safe.
- Library versioning: we don't make any attempt to version the library at this stage. Classes are in our experience highly stable once they are built and tested, the only changes typically being added methods.
- Performance: for critical path processing, you may want to avoid creating and destroying classes. However on modern Linux systems the heap allocator is very fast. Individual classes can choose whether or not to nullify their data on allocation.
Self-testing: every class has a
selftestmethod that runs through the methods of the class. In theory, calling all selftest functions of all classes does a full unit test of the library. The
czmq_selftestapplication does this.
- Memory management: CZMQ classes do not use any special memory management techiques to detect leaks. We've done this in the past but it makes the code relatively complex. Instead, we do memory leak testing using tools like valgrind.
If you define a new CZMQ class
myclass you need to:
- Write the
zmyclass.hsource files, in
- Add the myclass header and test call to
- Add a reference documentation to 'doc/zmyclass.txt'.
- Add myclass to 'src/Makefile.am
bin/newclass.sh shell script will automate these steps for you.
In general the zctx class defines the style for the whole library. The overriding rules for coding style are consistency, clarity, and ease of maintenance. We use the C99 standard for syntax including principally:
- The // comment style.
- Variables definitions placed in or before the code that uses them.
So while ANSI C code might say:
zblob_t *file_buffer; /* Buffer for our file */ ... (100 lines of code) file_buffer = zblob_new (); ...
The style in CZMQ would be:
zblob_t *file_buffer = zblob_new ();
We use assertions heavily to catch bad argument values. The CZMQ classes do not attempt to validate arguments and report errors; bad arguments are treated as fatal application programming errors.
We also use assertions heavily on calls to system functions that are never supposed to fail, where failure is to be treated as a fatal non-recoverable error (e.g. running out of memory).
Assertion code should always take this form:
int rc = some_function (arguments); assert (rc == 0);
Rather than the side-effect form:
assert (some_function (arguments) == 0);
Since assertions may be removed by an optimizing compiler.
Man pages are generated from the class header and source files via the doc/mkman tool, and similar functionality in the gitdown tool (http://github.com/imatix/gitdown). The header file for a class must wrap its interface as follows (example is from include/zclock.h):
// @interface // Sleep for a number of milliseconds void zclock_sleep (int msecs); // Return current system clock as milliseconds int64_t zclock_time (void); // Self test of this class int zclock_test (Bool verbose); // @end
The source file for a class must provide documentation as follows:
/* @header ...short explanation of class... @discuss ...longer discussion of how it works... @end */
The source file for a class then provides the self test example as follows:
// @selftest int64_t start = zclock_time (); zclock_sleep (10); assert ((zclock_time () - start) >= 10); // @end
The template for man pages is in doc/mkman.
CZMQ is developed through a test-driven process that guarantees no memory violations or leaks in the code:
- Modify a class or method.
- Update the test method for that class.
- Run the 'selftest' script, which uses the Valgrind memcheck tool.
- Repeat until perfect.
When you try CZMQ on an OS that it's not been used on (ever, or for a while), you will hit code that does not compile. In some cases the patches are trivial, in other cases (usually when porting to Windows), the work needed to build equivalent functionality may be non-trivial. In any case, the benefit is that once ported, the functionality is available to all applications.
Before attempting to patch code for portability, please read the
czmq_prelude.h header file. There are several typical types of changes you may need to make to get functionality working on a specific operating system:
- Defining typedefs which are missing on that specific compiler: do this in czmq_prelude.h.
- Defining macros that rename exotic library functions to more conventional names: do this in czmq_prelude.h.
- Reimplementing specific methods to use a non-standard API: this is typically needed on Windows. Do this in the relevant class, using #ifdefs to properly differentiate code for different platforms.
The canonical 'standard operating system' for all CZMQ code is Linux, gcc, POSIX. The canonical 'weird operating system' for CZMQ is Windows.
We generate the zsockopt class using the mysterious but powerful GSL code generator. It's actually cool, since about 30 lines of XML are sufficient to generate 700 lines of code. Better, since many of the option data types changed in ØMQ/3.1, it's possible to completely hide the differences. To regenerate the zsockopt class, build and install GSL from https://github.com/imatix/gsl, and then:
You may also enjoy using this same technique if you're writing bindings in other languages. See the sockopts.gsl file, this can be easily modified to produce code in whatever language interests you.
This document is originally at README.txt and is built using gitdown.