This crate offers a two-byte serial number with wraparound.

Release notes

A serial number is an identifier assigned incrementally to an item. In many cases, you can use a u32 or u64 and call it a day, without having to worry about overflow. The niche benefit of this type is that it only uses the space of a u16, with the problem of overflow solved by wraparound.

One scenario where this is useful, is when signing network packets with a serial number, and being able to respond with an ack packet that contains a larger amount of serial numbers than when using u32 or u64. Especially with a protocol like UDP, the more numbers you can fit into that ack packet, the more redundancy you get, and the more likely is it that all received packets are also successfully acknowledged.

Due to the limited number space, recently allocated serial numbers may duplicate very old serial numbers, but not other recently allocated serial numbers. To avoid ambiguity with these non-unique numbers, RFC 1982 "Serial Number Arithmetic", defines special rules for calculations involving these kinds of serial numbers.

[dependencies]
serial-num = "0.10"

# or with additional features:
[dependencies]
serial-num = { version = "0.10", features = ["serde"] }

Feature Flags

The following feature flags implement additional traits for the Serial type:

• arbitrary: derives arbitrary's Arbitrary (⚠️ requires std)
• bincode: derives bincode's Decode/Encode
• bitcode: derives bitcode's Decode/Encode (⚠️ requires std)
• borsh: derives borsh's BorshDeserialize/BorshSerialize
• bytemuck: derives bytemuck's Pod/Zeroable
• postcard: derives postcard's Schema/MaxSize
• rkyv: derives rkyv's Archive/Deserialize/Serialize
• rkyv-safe: additionally enables rkyv’s safe API
• serde: derives serde's Deserialize/Serialize
• speedy: derives speedy's Readable/Writable (⚠️ requires std)

Usage

Simple example

use serial_num::Serial;

// the default is a reference point - not serial number "zero"
let mut a = Serial::default();
let mut b = Serial::default();
let mut c = Serial::default();

// three ways to increase
let x = a.increase_get(); // increase, then copy
let y = b.get_increase(); // copy, then increase
c.increase();

assert!(y.precedes(x));
assert_eq!(-1_i16, y.diff(x)); // "diff()" is signed
assert_eq!(1_u16, y.dist(x)); // "dist()" is unsigned

// addition is the same as calling "increase()" n times
assert_eq!(y + 1_u16, x);

Wraparound example

use serial_num::Serial;

// a serial number can be increased indefinitely
let mut x = Serial::default();
for _ in 0..u16::MAX {
    x.increase();
}
let x = x + u16::MAX + u16::MAX + u16::MAX;

// comparison is trivial as long as two serial numbers have
// a distance of less than half of our number space (32767).
let a = Serial::default() + 5;
let b = Serial::default() + 32000;
assert!(a.precedes(b)); // 5th successor < 32000th successor

// but: the comparison flips if the distance is larger
let a = Serial::default() + 5;
let b = Serial::default() + 65000;
assert!(a.succeeds(b)); // 5th successor > 65000th successor

// this means that you get the right ordering as long as
// you compare one serial number at most with one that
// is its 32767th successor:
let num = Serial::default();
assert!(num.precedes(num + 32767));     //  0 < 32767 (still intuitive)
assert!(num.succeeds(num + 32768));     //  0 > 32768 (flip #1)
assert!(num.succeeds(num + 65534));     //  0 > 65534
assert!(num == num + 65535);            // 0 == 65535 (due to same internal representation)
assert!(num.precedes(num + 65535 + 1)); //  0 < 65536 (flip #2)

The NAN value

use serial_num::Serial;

// "NAN" exists to have value representing "no serial number",
// since it saves encoding space vs wrapping Serial in an Option.
let nan = Serial::NAN;
let num = Serial::default();

// you can check whether a serial number is NAN
assert!(nan.is_nan());

// NAN cannot be increased
assert_eq!(Serial::NAN, nan + 1_u16);

// distance between two NAN values is zero
assert_eq!(0_u16, nan.dist(nan));
assert_eq!(0_i16, nan.diff(nan));

// distance and difference of non-NAN to NAN is the maximum distance
assert_eq!(32_767_u16, num.dist(nan));
assert_eq!(32_767_u16, nan.dist(num));
assert_eq!(32_767_i16, num.diff(nan));
assert_eq!(32_767_i16, nan.diff(num));

// partial ordering does not include the NAN value
assert_eq!(None, nan.partial_cmp(num));
assert!(!nan.precedes(num) && !nan.succeeds(num));