History.txt in ffi-rzmq-0.9.3 vs History.txt in ffi-rzmq-0.9.6

@@ -1,5 +1,54 @@
+== 0.9.6 / 20120808
+  * Never released 0.9.5 as a gem. It was available via github only.
+  
+  * Improved error message when DLL loading fails on Windows.
+  
+  * Added support for 0mq 2.2. Support for 2.1 might be getting shakey...
+    patches to make it fully support 2.1 (assuming it's even broken at
+    all) are welcome.
+  
+  * Added support for 0mq 3.2 (no support for 3.0 or 3.1). Not all methods
+    are exposed yet. For example, setting values on the context after it
+    has been created is not supported; instead, pass the correct keys
+    (:io_threads and :max_sockets) to the call to Context.create or
+    Context#new.
+    
+  * Reduced spec running time from 30+ seconds to under 1 by eliminating
+    most uses of "sleep." It now polls sockets to wait for message
+    delivery. It also uses a technique of binding an inproc transport and
+    busy-looping on the connect side until it succeeds. These techniques
+    both allowed me to eliminate most uses of sleep.
+    
+  * Some changes to support usage on Win7x64 with a 64-bit libzmq DLL.
+
+== 0.9.5 / 20120119
+  * BROKE THE API.
+    In 0mq 2.x, there were two functions zmq_send() and zmq_recv().
+    As of 3.x, those functions were renamed zmq_sendmsg() and
+    zmq_recvmsg(). As everyone starts moving to 0mq 3.x, it doesn't
+    make sense to make the code break with 2.x. So, I'm breaking this
+    binding so that it always uses sendmsg/recvmsg and eliminates the
+    original send/recv methods.
+    Sorry!
+    This is likely to be the last non-backward-compatible API breakage.
+    Release 1.0 is around the corner and the API will be stable (I follow
+    semantic versioning).
+    
+  * Introduced ZMQ::NonBlocking. This flag returns the correct value to set
+    a socket in non-blocking mode when sending/receiving. This hides the
+    differences between 0mq 2.x and 3.x since the constant names have
+    changed.
+
+== 0.9.4 / 20120102
+  * Fixed bug in Poller#delete. Added specs to catch a regression.
+    In short, a socket that was deleted from the Poller set wasn't
+    always actually *removed* from the array. This led to a closed
+    socket being part of the pollset which would return errno 38.
+    This took about 4 days to find. <sigh>
+    
+
 == 0.9.3 / 20111214
   * Performance optimizations for #getsockopt.
   
   * Fixed Message#copy and Message#move. They didn't work before.
   
@@ -13,9 +62,11 @@
     the arguments that are expected to be passed up the chain. By
     explicitly listing the args and using parentheses, the runtime
     can avoid that work and dispatch directly. This effects all
     Ruby runtimes, but it was through the work of Evan Phoenix that
     I figured this out. Results in a 2-5% speedup on method dispatch.
+    
+  
 
 == 0.9.2 / 20111115
   * Removed all references to the version4 API.
   
   * Dropped support for 3.0.x and added support for 3.1.x. The 0mq