Jonathan de Boyne Pollard wrote in a message to Mario Semo:

 JdBP>   Interesting.  I disagree with their putting user code in
 JdBP> _DLL_InitTerm,   however.  User code is better off driven by
 JdBP> the constructor and   destructor of a class instance with
 JdBP> static storage duration.  That way,   it is more
 JdBP> exception-proof, and the programmer is assured that at the  
 JdBP> time that his initialisation/termination code is run, the
 JdBP> C++ RTL is   fully available.

 JdBP>   In their code that you quoted, placing initialisation code
 JdBP> at the point   of the "user code goes here" comments would
 JdBP> restrict the programmer to   not using any C++ RTL functions
 JdBP> during DLL initialisation, since the   call to _CRT_Init
 JdBP> would not have yet occurred at that point. 

More to the point, their way of doing this would disable several features
of the C++ language, particularly exception handling.  This is
compiler-dependent but would still fail in all actual implementations I
know about.  While most programmers probably do not make direct use of the
exception facilities of C++, many important libraries do, including STL and
whatever ICLUI is called now.  Your approach avoids this problem.

 JdBP>   I suppose, in a way, mine is a minimalist approach.  DLL 
 JdBP> initialisation   and termination is such a compiler-dependent 
 JdBP> thing that I advocate   putting as little user code as possible 
 JdBP> in the actual _DLL_InitTerm/  _dllmain/__dll_initialize 
 JdBP> function, and using more "conventional"   mechanisms (like 
 JdBP> constructors and destructors) for user initialisation   and 
 JdBP> termination.

I don't think this is a close call.  Your approach is clearly better.
 
-- Mike


---