Sparse Set Container

A container based on a sparse set.

It is useful if you want a container with performance close to Vec but also to safely store the indexes to the elements (so that they are not invalidated on removals).
E.g. you have a list of elements in UI that the user can add and remove, but you want to refer to the elements of that list from somewhere else.

Usage

Add this to your Cargo.toml:

[dependencies]
sparse_set_container = "1.2"

Description

An array-like container based on sparse set implementation that allows O(1) access to elements without hashing and allows cache-friendly iterations.

Operation	SparseSet	Vec
push	O(1)	O(1)
lookup	O(1)	O(1)
len	O(1)	O(1)
remove	O(n)	O(n)
swap_remove	O(1)	O(1)

For iterating over the elements SparseSet exposes an iterator over a tightly packed slice with values, which is as efficient as iterating over a Vec.

Differences to Vec:

Instead of using indexes, when adding an element, it returns a lightweight key structure that can be used to access the element later
- The key is not invalidated when elements are removed from the container
- If the pointed-at element was removed, the key will not be pointing to any other elements, even if new elements are inserted
There is a slight overhead in insertion/lookup/removal operations compared to Vec
Consumes more memory:
- for each value 4*sizeof(usize) bytes on top of the size of the element itself
  - (e.g. 32 bytes per element on 64-bit systems)
- per each 2^(sizeof(usize)*8) removals the memory consumption will also grow by 2*sizeof(usize)
  - (e.g. 16 bytes per 18446744073709551616 elements removed on 64-bit systems)
Many Vec operations are not supported (create an issue on github if you want to request one)

Examples

extern crate sparse_set_container;
use sparse_set_container::SparseSet;

fn main() {
    let mut elements = SparseSet::new();
    elements.push("1");
    let key2 = elements.push("2");
    elements.push("3");

    elements.remove(key2);
    elements.push("4");

    if !elements.contains(key2) {
        println!("Value 2 is not in the container");
    }

    // Prints 1 3 4
    for v in elements.values() {
        print!("{} ", v);
    }

    // Prints 1 3 4
    for k in elements.keys() {
        print!("{} ", elements.get(k).unwrap());
    }
}

Benchmarks

The values captured illustrate the difference between this SparseSet container implementation, Vec, and standard HashMap, as well as comparing to other libraries with similar functionality.

Benchmark	`SparseSet<String>`	`Vec<String>`	`HashMap<i32, String>`	`thunderdome::Arena<String>`	`generational_arena::Arena<String>`	`slotmap::SlotMap<_, String>`	`slotmap::DenseSlotMap<_, String>`
Create empty	0 ns ±0	0 ns ±0	2 ns ±0	0 ns ±0	15 ns ±0	8 ns ±0	9 ns ±0
Create with capacity (1000)	20 ns ±0	20 ns ±0	37 ns ±0	20 ns ±0	686 ns ±3	20 ns ±0	54 ns ±0
Push 100 elements	2,849 ns ±11	2,738 ns ±12	4,714 ns ±24	2,870 ns ±15	2,919 ns ±10	2,802 ns ±17	3,380 ns ±14
With capacity push 100	2,724 ns ±14	2,637 ns ±14	3,644 ns ±19	2,737 ns ±14	2,740 ns ±11	2,690 ns ±10	2,859 ns ±12
Lookup 100 elements	94 ns ±0	39 ns ±3	474 ns ±26	82 ns ±0	81 ns ±1	68 ns ±0	90 ns ±4
Iterate over 100 elements	32 ns ±0	32 ns ±0	44 ns ±0	98 ns ±1	73 ns ±0	38 ns ±0	33 ns ±0
Clone with 100 elements	2,654 ns ±11	2,574 ns ±10	1,663 ns ±34	2,620 ns ±20	2,666 ns ±17	2,655 ns ±20	2,691 ns ±14
Clone 100 and remove 10	3,334 ns ±69	2,646 ns ±35	1,839 ns ±139	2,788 ns ±91	2,823 ns ±64	2,805 ns ±83	2,774 ns ±80
Clone 100 and swap_remove 10	2,749 ns ±91	2,489 ns ±61	N/A	N/A	N/A	N/A	N/A

To run the benchmark on your machine, execute cargo run --example bench --release

Or to build this table you can run python tools/collect_benchmark_table.py and then find the results in bench_table.md

License