I've finally gotten round to changing from the pre-release version of Boost.Locale (v2.92) to the version now in Boost, and I was quite surprised by one change. Previously, I could do the following: void function_takes_wide_string(std::wstring message); function_takes_wide_string(boost::locale::translate("My translation key")); However, with the new version, the key must be passed in as a wide string if I am to be able to get a wide string out. This means changing `translate("blah")` to `translate(L"blah")` throughout. The change itself is no big deal, but the problem is that it feels like a change for the worse. The character width of the key is now coupled to the character width of the output, when the two are separate concepts and should be allowed to vary independently. The old class `message` coeerced the string to whichever width was required. The new class `basic_message<T>` coerces only to `basic_string<T>` and, worse than that, doesn't provide a method to explicitly convert to a different width string. My question, I suppose, is why was this change made? What makes the new API better? Alex -- Swish - Easy SFTP for Windows Explorer (http://www.swish-sftp.org)