At one point I wrote my own STL-replacement container classes, with careful benchmarks to assess memory/cpu performance in various common use cases. My alternate implementations of list and map came out better for my needs, but std::vector is a beast (at least the MSVC impl). I could only ever best its performance in edge cases, never in real-world situations.
At some point I'm interested in implementing a container/string-unifying library based around ranges (a la python or D) instead of iterators. The work there, though, lies mostly in writing a range-centric <algorithm> equivalent. The end result would be container-based code would more concisely describe what I mean in higher-level terms, with the added benefit of unifying the interface of classes that, in STL, are distinct.
All that being said, I'm a tech-head, and I like writing shiny new toys for myself. At work, though, my engine code is driven by a desire to express my games as closely as possible to how I conceive of their behavior.
Developer
