Rust Config Parsing Benchmarks

This repo tries to assess Rust layered-config behavior.

We currently compare:

Name	Notes
No-op	N/A
cfg-rs
config
configure_me
confy
figment

Note: any non-performance comparison is meant to provide context for what you gain/lose with each crate's overhead. For a full comparison, see each crates docs

General expectations

• Layers
  • Arg parse layer (opt-in per field, support arguments unrelated to config)
  • Env layer (opt-in per field)
  • User config layer
  • Parent directory layer
  • Default layer
• Paths remain relative to the their source
• Schema migrations (correctly layering one field migrating to another, example)
• Isolated made
• Layers can override or append
• Report error for the correct layer
• File format is domain-specific (toml, yaml, gitconfig, etc)
• Show defaults in --help and man page
• Document env variables in --help and man page
• Support --dump-config flag, showing merged config

For a wider discussion on approaches to layered config, see the clap discussion

Results

Name	Overhead (release)	Build (debug)	Parse (release)	Downloads	Version
null	0 KiB	3s (full) 211ms (incremental)	0ms	-	v4.0.1

System: Linux 5.10.102.1-microsoft-standard-WSL2 (x86_64) w/ -j 20

rustc: rustc 1.64.0 (a55dd71d5 2022-09-19)

Notes:

• Overhead will be lower if your application shares dependencies with your config library.

Running the Benchmarks

$ ./bench.py
$ ./format.py

To be included, the crate needs meet one of the following criteria:

• 1k+ recent downloads
• Unique API design

Special Thanks

• djc for inspiration with template-benchmarks-rs
• sharkdp for hyperfine