base62

A fast, zero-dependency base62 encoder/decoder library for Rust, typically used in URL shorteners. It supports both standard [0-9A-Za-z] and alternative [0-9a-zA-Z] variants.

Features

• no_std compatible with optional alloc and std support
• Encodes integers up to u128
• Zero-copy decoding
• Efficient string handling
• Two encoding variants:
  • Standard [0-9A-Za-z]
  • Alternative [0-9a-zA-Z]

Usage

Add this to your Cargo.toml:

[dependencies]
base62 = "2"

Basic Example

use base62;

// Encoding
let encoded = base62::encode(1234567890);
assert_eq!(encoded, "1LY7VK");

// Decoding
let decoded = base62::decode("1LY7VK").unwrap();
assert_eq!(decoded, 1234567890);

No-std Usage

The crate works in no_std environments by default:

#![no_std]
use base62;

// Encode into a fixed buffer
let mut buf = [0u8; 22];  // Maximum size needed for u128
let len = base62::encode_bytes(1234567890, &mut buf).unwrap();
assert_eq!(&buf[..len], b"1LY7VK");

// Decode from bytes
let decoded = base62::decode(&buf[..len]).unwrap();
assert_eq!(decoded, 1234567890);

Feature Flags

• alloc: Enables String allocation support (enabled by default)
• std: Enables std::io traits support

Performance

The library is optimized for both encoding and decoding performance:

• Zero-copy decoding
• Efficient buffer management
• Direct string manipulation for optimal performance when appending
• Pair-at-a-time encoding, with fixed-width chunks for larger integers
• Byte-indexed decoding tables and a short-input path without wide arithmetic
• Separate encoding and decoding tables, so decode-only binaries can discard encoding tables

Comparing performance

The by_digits Criterion group measures both alphabets at every width from 1 to 22 digits, plus mixed-length inputs. Each iteration processes 128 seeded, precomputed inputs. Criterion reports time per 128-value batch and throughput in values per second; divide the batch time by 128 to obtain time per value. Random-number generation, allocation, and input destruction are outside these core-operation timings. Both runs must use identical corpus generation; regenerate baselines whenever benchmark inputs change.

Record a baseline before changing the implementation, then compare against it:

cargo +stable bench --bench base62 -- by_digits --save-baseline before
# After changing the implementation:
cargo +stable bench --bench base62 -- by_digits --baseline before

Use the same compiler, build flags, and CPU affinity for both runs, without other CPU-intensive jobs. On Linux, taskset -c <cpu> pins the command to one logical CPU. The existing encode and decode groups also cover allocation and fixed inputs.

License

Licensed under the MIT license. See LICENSE for details.