The most interesting part here is that boost::mpl::int_<sizeof(T)>(). sizeof(T) executes at compile time, so its output can be used as a template parameter. The class boost::mpl::int_<> is just an empty class that holds a compile-time value of integral type.
We need an instance of this class, which is why we have a round parentheses at the end of boost::mpl::int_<sizeof(T)>().
let's take a closer look at how the compiler will decide which process_impl function to use. First of all, the compiler tries to match functions that have a non-template second parameter . If sizeof(T) is 4, the compiler tries to search the function with signatures like process_impl(T, boost::mpl::int_<4>) and finds our 4 bytes optimized version from the detail namespace. If sizeof(T) is 34, the compiler can not find the function with a signature like process_impl(T, boost::mpl::int_<34>), and uses a template function process_impl(const T& val, Tag /ignore/).
The Boost.MPL library has several data structures for metaprogramming. In this recipe, we only scratched the tip of the iceberg. You may find the following integral constant classes from MPL useful:
- bool_
- int_
- long_
- size_t
- char_
All the Boost.MPL functions (except the for_each runtime function) are executed at compile time and won't add runtime overhead.
mkdir build
cd build
cmake ..
cmake --build .
./bin/main
0 1 2 A 4 5