It's a stunningly difficult and surprisingly low-value proposition to create abstractions that don't leak. That's not the goal anyway: The goal is to leverage abstractions to build things. The implementation details frequently do matter. Even if your abstraction is airtight, it has to run on a real machine. When possible, it's preferable to design abstractions that have a clear implementation in terms of the lower level's abstractions. This way, when something goes wrong - and it will - you can understand and fix it.