For atomic_counter and atomic_ptr classes, detect the Tile architecture
using #if defined __tile__ matching ARM and Solaris and then use the
Tile atomic instructions. Without this change, the default Mutex
implementation is used, which is slower.
The problem is that other threads might still be in mailbox::send() when
it is destroyed. So as a workaround, we just acquire the mutex in the
destructor. Therefore the running send will finish before the mailbox is
destroyed.
See also the fix for LIBZMQ-281 in zeromq2-x.
Signed-off-by: Mika Fischer <mika.fischer@zoopnet.de>
Copyrights had become ads for Sustrik's corporate sponsors, going against the original
agreement to share copyrights with the community (that agreement was: one line stating
iMatix copyright + one reference to AUTHORS file). The proliferation of corporate ads
is also unfair to the many individual authors. I've removed ALL corporate title from
the source files so the copyright statements can now be centralized in AUTHORS and
source files can be properly updated on an annual basis.
- Windows CE does not manage security attributes (no SetSecurityDescriptorDacl(), SetEvent (NULL, xxx) ...)
- Windows CE does not inheritance of sockets for child process (SetHandleInformation ((HANDLE) *w_, HANDLE_FLAG_INHERIT...)
- see comments about story "Porting ZeroMQ to Windows Mobile" on webpage http://www.zeromq.org/story:5
- when compiling for Windows CE, a C++ project must define the preprocessor definitions: UNDER_CE=$(CEVersion),_WIN32_WCE=$(CEVersion)
- choosing the "_WIN32_CE" form for uniformization with "_WIN32" and "_WIN32_WINNT" already used in libzmq (boost is using both forms)
- see http://msdn.microsoft.com/en-us/library/ee479161(v=winembedded.60).aspx
When we send a large message, the message can be splitted into two chunks.
One is in the encoder buffer and the other is the zero-copy pointer.
The session could get the term before the last chunk is sent.
Until now, zmq_poll always allocates the poll items on the heap.
Now, small item sets, up to ZMQ_POLLITEMS_DFLT, are stack allocated
and only larger sets are allocated on the heap.
zmq::xpub_t::xread_activated() – change to process messages without 0
or 1 prefix, but without affecting subscriptions
zmq::xsub_t::xsend() – change to send rather than discard messages
without 0 or 1 prefix, but without affecting subscriptions
Update documentation
