Thomas M
We could discuss issues of the Khronos C++ bindings and your wrapper to no end, but my main point remains: The Khronos version is official and established; an alternative version intended for wider use must be clearly superior and give people strong reasons to migrate (I am not only referring to programming itself, but also people writing books on OpenCL, blogs etc.).
[snipping exemplary comparison of OpenCL C, OpenCL C++ and Boost.Compute] I'd like to offer some more general thoughts on this: Isn't there a precedent here? If we look at MPI and Boost.MPI we have something similar. There is a standard C implementation and a crude (but official and established) C++ wrapper around that (that was removed in MPI3 btw). And as far as I know Boost.MPI builds on top the C implementation of MPI and just does a much better job than the official C++ wrapper. I think Boost.Compute is the better wrapper for OpenCL C. The weight of "official" and "established" should not play such a significant role if we're trying to think ahead. Sebastian