I know the codebase for Boost.Thread is in the process of being restructured and some parts are being rewritten, but since the new code won't be available for the 1.34 release, it would probably be good to patch some of the minor WINTHREADS issues in the current code. Most of these relate to inconsistent use of narrow vs. wide versions of the Win32 API functions. There are also a couple of other small patches specifically related to using Boost.Thread under WinCE (I encountered most of these issues while building the thread library for WinCE 5 under MSVC 8). - libs\thread\src\exceptions.cpp: system_message() explicitly calls FormatMessageA. On some platforms like WinCE, this function does not exist. I replaced this with a call to FormatMessage and some code to convert the result to a narrow string for BOOST_NO_ANSI_APIS builds. - libs\thread\src\mutex.inl: In init_TryEnterCriticalSection, the GetProcAddress call is passed "TryEnterCriticalSection". Passing it TEXT("TryEnterCriticalSection") instead allows this statement to compile correctly on wide character builds (and still work on narrow builds as well). In new_mutex, I replaced the ATL conversion macro with some Win32 code to do the conversion. - libs\thread\src\tss_hooks.cpp: WinCE headers do not define the TLS_OUT_OF_INDEXES constant. Instead TlsAlloc() just returns 0xFFFFFFFF when it fails. Added some code to define TLS_OUT_OF_INDEXES as 0xFFFFFFF if it is not present. - libs\thread\src\tss_dll.cpp: Inexplicably, WinCE defines DllMain as receiving a HANDLE for its first parameter unlike HINSTANCE that normal Win32 uses. I believe many of these items have been mentioned in previous discussions, but I thought they might have fallen out of memory before the patches actually made it into the code. If there were important reasons why the changes were not made before, forgive me for missing the details of those earlier discussions. Thanks, -Dave