Good things about Boost.Credible "certification" of library quality and completness through formal review processCredible "certification" of implementation through formal testing, release process.Boost isn't scaling well.Testing time takes longer and longer to complete..It's harder for testing to be comprehensive and include less well known compilersCurrent testing can't handle all the possible combinations of compiler settings (rtti on/off, debug/release, stl library, etc.installation becomes more onerous and fragile as more libraries are added.Hence, the "cost" or "overhead" of using boost increases with time even though new users will start using only one library. But there is no way to test/deploy just the libraries one is going to use.Identity crisis - Is BoostA single (now large) library of core utilities? i) Tools test, release and deployment of boost is one operation. ii) The test, deployment and release of a subset of libraries has been considered necessary. ii) There is one release number for all libraries.Or a group (now large) of independent, decoupled libraries? i) Many or most libraries depend only on a subset of the total set of other boost libraries.My argument is that it looks like boost started out as the former and is evolving towards the latter.It's been a great 10 years for Boost. But to continue to be succeesful it's going to have to evolve. This presentation describes how I think boost has to change.Download slides Less