```
$ clang --version
Apple LLVM version 5.0 (clang-500.2.79) (based on LLVM 3.3svn)
Target: x86_64-apple-darwin13.0.0
Thread model: posix
In file included from /Users/drew/Dropbox/Code/caffeine-ios/ext/msgpack/src/msgpack/pack.h:120:
/Users/drew/Dropbox/Code/caffeine-ios/ext/msgpack/src/msgpack/pack_template.h:773:57: warning: implicit conversion loses integer precision: 'size_t' (aka 'unsigned long') to 'unsigned int' [-Wshorten-64-to-32]
msgpack_pack_append_buffer(x, (const unsigned char*)b, l);
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~
In file included from /Users/drew/Dropbox/Code/caffeine-ios/ext/msgpack/src/objectc.c:19:
/Users/drew/Dropbox/Code/caffeine-ios/ext/msgpack/src/msgpack/pack.h:118:61: note: expanded from macro 'msgpack_pack_append_buffer'
return (*(user)->callback)((user)->data, (const char*)buf, len)
^
1 warning generated.
```
This patch is provided under the Apache License version 2.
The memcpy approach is a workaround for gcc's bug. The speed performance on gcc is the same between both approach. But on clang, the memcpy approach is 10% slower than an assignment approach. Hence I added the switching approach code using compiler checked macro.
Note: __GNUC__ is defined both gcc and clang. So I use __GNUC__ && !__clang__
When doing streaming deserialization, valgrind complains about mismatched free()...
It's caused by calls to auto_ptr::free for zones that are null pointers.
The patch add serialization operator for the given enum to avoid the use
of a cast which leads to strict-aliasing warnings and the test to fail.
Solution suggested by qehgt.
Nobuyuki Kubota