packed object / bitfields

[timotheecour]

TODO: write proper RFC

proposal

add Packed, which generalizes builtin set by packing several set into 1 bit sequence, allowing one to represent objects compactly as a single bit sequence (eg equivalent to an int)

Example

import std/packed
type Bar = enum b0, b1, b2, b3
type A = Packed(switch: bool, val: 10..12, bar: Bar)
assert A.sizeof == nextPow2((2 * 3 * 4) div 8) = 4 # fits in an int32
let a = A.init(switch = true, bar = b2)
assert a.matches(switch = {true}, bar = {b0, b2})

links

• What languages have bit struct / field constructs? : ProgrammingLanguages

[juancarlospaco]

Hows this different from a set of enum ?, or is it the metaprogramming part that has the magic ?. 🤔

[timotheecour]

a packed object is intended to replace simple objects in cases like this:

type Bar = enum b0, b1, b2, b3

type A0 = object
  switch: bool
  val: range[10..12]
  bar: Bar

type A1 = Packed(switch: bool, val: 10..12, bar: Bar)

assert A0.sizeof == 24
assert A1.sizeof == 4

beside the large saving (and likely more efficient operations), you get syntax sugar like matches which generalizes set APIs (not just sugar; it allows more efficient operations, just like set does)

set[Bar] doesn't help with that and serves other purposes (note that a packed object could hold a set[Bar] too, as a field)

[juancarlospaco]

So like an object implemented on top of a set ?, all attributes must be ordinal right ?, sounds interesting... 🤔

[timotheecour]

all attributes must be ordinal right ?,

correct, you can't have these types as fields:
string, seq, ptr, ref, object, float

you can have these:
Ordinal

that's the restriction
(whether you can have a set[Bar] as field, I'm not sure yet; we can disallow it for now)

[juancarlospaco]

Write the RFC.

Offtopic: I wish stdlib had a fixed point.