kevy-config

Zero-dependency TOML subset parser + schema for the kevy server.

#![forbid(unsafe_code)] · 0 crates.io deps · pure Rust · works on x86_64-unknown-linux-gnu, aarch64-unknown-linux-gnu, *-apple-darwin, and wasm32-unknown-unknown (via embedded use).

TOML subset

Supports just what kevy's config file needs — nothing more, nothing less.

[section] table headers (one level deep)
key = value with value ∈ {string, integer, boolean, size literal}
String literals: "double-quoted" (with escapes) and 'single-quoted' (raw)
Integers: signed decimal, underscore separators (1_000_000)
Booleans: true / false
# comment to end of line

Size literals are a kevy extension: "64mb" / "2gb" / "512kb" parsed via [parse_size] when a schema field expects bytes. Suffix is case-insensitive; multipliers are binary (1 KB = 1024 bytes) to match Redis convention.

Intentionally unsupported (full TOML features kevy doesn't need): dotted keys, multi-line strings, arrays / arrays of tables, inline tables, datetime literals.

Precedence chain (top wins)

CLI flags via [Config::merge_cli] (e.g. --bind 0.0.0.0)
Environment via [Config::merge_env] (KEVY_BIND, KEVY_PORT, …)
TOML file via [Config::load] — explicit path or auto-detect ($KEVY_DIR/kevy.toml, then ./kevy.toml, then /etc/kevy/kevy.toml)
[Config::default()]

Quick example

use kevy_config::{Config, CliOverrides};

// Load from auto-detect path, or fall back to defaults.
let mut cfg = Config::load(None).expect("config error");

// Overlay env vars (caller-controlled — easy to fixture in tests).
cfg.merge_env(std::env::vars()).expect("bad env value");

// Overlay CLI overrides (parsed by the caller).
cfg.merge_cli(CliOverrides {
    bind: Some([0, 0, 0, 0]),
    ..CliOverrides::default()
}).expect("bad cli value");

// Hand off to kevy::serve(...).
println!("listening on {:?}:{}", cfg.server.bind, cfg.server.port);

Schema reference

See [Config] for the full struct shape and per-field defaults. Quick summary by section:

Section	Fields
`[server]`	`bind` `port` `threads` `data_dir`
`[persistence]`	`aof` `appendfsync` `auto_aof_rewrite_percentage` `auto_aof_rewrite_min_size`
`[memory]`	`maxmemory` `maxmemory_policy`
`[expiry]`	`hz` `sample`
`[log]`	`level` `output`

A fully-annotated sample lives at crates/kevy/kevy.toml.example.

Errors

[ConfigError] has three variants:

IoOpen — couldn't open the file
Parse — tokenizer / parser error with (line, col) + message
Schema — value rejected by the schema (unknown enum, out-of-range integer, wrong type) with (line, field, msg)

Parser failures are never silently downgraded; startup must fail loudly so misconfigurations surface immediately.

Why custom TOML

kevy's zero-dependency policy forbids crates.io dependencies. Writing a focused TOML subset parser (~600 LOC including lexer + parser + tests) was cheaper than introducing the toml crate (~10 K LOC, two-phase parse, datetime types we don't need) and lets the parser stay forbid(unsafe_code) + miri-clean.

License

MIT OR Apache-2.0, at your option.

kevy-config 1.13.0