I have no experience with this `slotmap` thing you refer to, but it may be
relevant to point out @jackhftang's
[lrucache](https://github.com/jackhftang/lrucache.nim). (That duplicates key
storage. A linked-list hash collision strategy is the usual workaround for
that.)
Anyway, I don't mean to sound judgemental. If one is being very picky, then the
entire, broad exact problem is usually required to fully constrain best data
structs/algo choice. If one is not being picky, then it often matters how picky
one is being { and about what dimension (memory, time, robustness, etc.) }. And
I put "should not be obscure" in quotes because, well, who knows, really? Scare
quoting that was an attempt to emphasize its subjectivity.
