An interesting (and yet to be fully explored) property of the described design is that Boost.Convert is not limited to string-to-type and type-to-string conversions. The boost::convert() interface is type-agnostic and the plugged-in converter ultimately dictates what type transformations are available. Consequently, a wide range of conversion/transformation-related tasks can be addressed and deployed uniformly by plugging-in special-purpose converters.

As an experiment, the code below (taken from test/encryption.cpp) does not do type conversion. Instead, it applies a string transformation:

string encrypted = boost::convert<string>("ABC", my_cypher).value();
string decrypted = boost::convert<string>(encrypted, my_cypher).value();

BOOST_ASSERT(encrypted == "123");
BOOST_ASSERT(decrypted == "ABC");

The original "ABC" string is "encrypted" as "123" first and then "123" is "decrypted" back to its original "ABC" form.

Similarly, I personally do not immediately see as objectionable string-transformations like:

std::u8string utf8 = boost::convert<std::u8string>(utf32_str, cnv);
std::u8string utf8 = boost::convert<std::u8string>(mbcs_str, cnv);