== Quote from Andrei Alexandrescu ([email protected])'s article > dsimcha wrote: > > I'm thinking it might be useful to have a library type that allows for > > arbitrary size integers <= 64 bits (128 if ucent is ever implemented) to be > > packed into an array. This might be nice when you want a huge array of > > them, > > the size you actually need is somewhere in between two native sizes, and you > > care more about space/cache efficiency than raw CPU cycles. For example, > > let's say you wanted an array of N 19-bit integers. You'd make an array of > > uints (32-bit) with ceil(N * 19 / 32) elements, and when it was indexed, > > you'd > > do a bunch of modulus and bit shifting to get the right 19 bits out of that > > block of memory, and return a uint. It would be slow as molasses, but > > really > > space efficient, which would in certain cases be a reasonable tradeoff. > > > > Has anyone implemented, or considered implementing this before? If not, if > > I > > implemented it well and submitted it as an enhancement, would it be a > > reasonable thing to include in std.array? > I think that would be a nice addition. We haven't established some > general rules for containers. Values or references? In the latter case: > Classes or structs? If classes: hierarchy or largely independent classes? > Andrei
I'd go with struct value type to make it as similar feeling to a regular array as possible. On the other hand, regular arrays are going to be overhauled anyhow, so that makes me not so sure.
