A memcpy is eliminated when receiving data on a ZMQ_STREAM socket. Instead
of receiving into a static buffer and then copying the data into the
buffer malloced in msg_t::init_size, the raw_decoder allocates the memory
for together with the reference-counter and creates a msg_t object
on top of that memory. This saves the memcpy operation.
For small messages, data is still copied and the receive buffer is reused.
Solution: increased to 4096 by default for all MSVC builds, for MinGW,
and for CMake.
Note: this is a speculative change, it needs confirmation before we
can keep it. Particularly, there is some doubt that changing this in
libzmq will affect upstream applications using libzmq.dll.
fork() support is optional and its availability is correctly detected at
contfigure time.
But test_fork was all always built, preventing build for targets that do
not provide fork() from building successfully.
This pacth fixes the CMakeLists.txt on this point.
The decision about the poller mechanism to use (select, poll, ...)
was done twice: once by the build system and once by the code in
poller.hpp. As the build-system can actually detect the mechanisms
available, prefer that result to the hard coded defaults in
poller.hpp.
At the same time, remove the duplicate detection of select() vs.
poll()-variant from proxy.cpp, signaler.cpp and zmq.cpp.
This patch has not been tested on many build platforms: especially
the cmake build needs testing / patching. For the other builds,
hard code the result as these these are all Windows platforms.
The ${libdir} was getting replaced/removed by configure_file() making pkg-config give bad flags: -L -lzmq
My fix was to add @ONLY to configure_file() so ${} style pkg-config substitutions are left alone.
In addition, I put the other typical ${} substitutions back into the libzmq.pc, since its now safe.
