Performance
Every row invokes the product's direct CLI executable; TomlSmith uses the release-mode Rust tomlsmith binary. Package-manager launchers and wrapper processes are excluded from the comparison.
Reproduction commands, product setup, and the measurement harness are maintained in the tomlsmith/benchmark repository.
Check · TOML 1.0
median latency · logarithmic scale · lower is betterCheck · TOML 1.1
median latency · logarithmic scale · lower is betterFormat · TOML 1.0
median latency · logarithmic scale · lower is betterFormat · TOML 1.1
median latency · logarithmic scale · lower is better* tomll is a semantic re-printer: it discards comments and rewrites literal styles, so it does less work than the comment-preserving formatters.
How to read the results
- Every number is the median cold-start end-to-end latency for the selected synthetic document; lower is better, and every timed sample starts a new process.
- TOML 1.0 and 1.1, and
checkandformat, are four independent lanes. Compare products only within one panel. - Differences within a few percent can be session drift and should not be read as a stable ordering.
tomlldiscards comments and rewrites literal styles, so it performs less work than the comment-preserving formatters.
Measurement contract
This comparison measures product CLIs, not parser libraries:
- TomlSmith is built in release mode and invoked directly as the native Rust CLI, matching the single-process boundary used by Taplo, dprint, and the Go tools.
- Every timed sample starts a fresh process and uses stdin/stdout/stderr.
- TOML 1.0 and 1.1, and
checkandformat, are separate lanes. - Correctness is checked before timing. Format output must reparse under the same version, preserve decoded content, and remain unchanged on a second format pass.
- Peak RSS is measured in separate fresh processes so the resource meter does not affect latency samples.
On this page