Correct-by-Construction Collections

Non-empty variants of the standard collections.

Non-emptiness can be a powerful guarantee. If your main use of Vec is as an Iterator, then you may not need to distinguish on emptiness. But there are indeed times when the Vec you receive as a function argument needs to be non-empty or your function can't proceed. Similarly, there are times when the Vec you return to a calling user needs to promise it actually contains something.

With NEVec, you're freed from the boilerplate of constantly needing to check is_empty() or pattern matching before proceeding, or erroring if you can't. So overall, code, type signatures, and logic become cleaner.

Consider that unlike Vec, NEVec::first and NEVec::last don't return in Option; they always succeed.

Alongside NEVec are its cousins NESlice, NEMap, and NESet, which are all guaranteed to contain at least one item.

Examples

The simplest way to construct these non-empty collections is via their macros: nev!, nes!, and nem!:

use nonempty_collections::*;

let v: NEVec<u32> = nev![1, 2, 3];
let s: NESet<u32> = nes![1, 2, 2, 3]; // 1 2 3
let m: NEMap<&str, bool> = nem!["a" => true, "b" => false];
assert_eq!(1, v.head);
assert_eq!(3, s.len().get());
assert!(m.get("a").unwrap());

Unlike the familiar vec! macro, nev! and friends require at least one element:

use nonempty_collections::nev;

let v = nev![1];

// Doesn't compile!
// let v = nev![];

Like Vec, you can also construct a NEVec the old fashioned way with NEVec::new or its constructor:

use nonempty_collections::NEVec;

let mut l = NEVec { head: 42, tail: vec![36, 58] };
assert_eq!(l.head, 42);

l.push(9001);
assert_eq!(l.last(), &9001);

And if necessary, you're free to convert to and from Vec:

use nonempty_collections::{NEVec, nev};

let l: NEVec<u32> = nev![42, 36, 58, 9001];
let v: Vec<u32> = l.into();
assert_eq!(v, vec![42, 36, 58, 9001]);

let u: Option<NEVec<u32>> = NEVec::from_vec(v);
assert_eq!(Some(nev![42, 36, 58, 9001]), u);

Iterators

This library extends the notion of non-emptiness to Iterators, and provides the NonEmptyIterator trait. This has some interesting consequences:

• Functions like map preserve non-emptiness.
• Functions like max always have a result.
• A non-empty Iterator chain can be collected back into a non-empty structure.
• You can chain many operations together without having to double-check for emptiness.

use nonempty_collections::*;

let v: NEVec<_> = nev![1, 2, 3].into_nonempty_iter().map(|n| n + 1).collect();
assert_eq!(2, v.head);

Consider also IteratorExt::to_nonempty_iter for converting any given Iterator into a non-empty one, if it contains at least one item.

Arrays

Since fixed-size arrays are by definition already not empty, they aren't given a special wrapper type like crate::NEVec. Instead, we enable them to be easily iterated over in a compatible way:

use nonempty_collections::*;

let a: [u32; 4] = [1, 2, 3, 4];
let v: NEVec<_> = a.into_nonempty_iter().map(|n| n + 1).collect();
assert_eq!(nev![2, 3, 4, 5], v);

See NonEmptyArrayExt for more conversions.

Caveats

Since NEVec, NEMap, and NESet must have a least one element, it is not possible to implement the FromIterator trait for them. We can't know, in general, if any given standard-library Iterator actually contains something.

Features

• serde: serde support.
• indexmap: support for non-empty IndexMap