The way I do ordered tables in `lptabz` is "only" 2x worse than the current tables for giant/out of cache tables with as mentioned a very (optimally?) packed index level memory use-wise. I have a feeling some people would rebel against that 2x.
Most of this goes to what is "most user-friendly" \-- often a hard & subjective question. E.g., the stdlib has (deprecated) multi-tables, but not multi-sets. I tried to keep a full Cartesian product of features in `lptabz`. Some people think _any_ feature/capability "confusing"..even any unfamiliar aspect of any language/runtime/tool. Normalizing for learning impatience over the many dimensions in play is more art than a science. Rather than try to do that, I just provide "all of the above"/the pile of basic features that is easy-ish to implement. That makes it "only" a training problem (not to minimize that problem...). It's totally fine with me/understandable if Araq wants to support less. He has a lot on his plate with the compiler. Nim, like C or C++, is far less reliant upon the stdlib being "state of the art" than say, Python (without Cython or PyPy).
