Crate krabka_parse_benches 

Parses Criterion benchmark output in bencher format into structured JSON summaries, and judges repeated reference/candidate measurements against variance-calibrated budgets.

Two lanes share one set of statistics. The criterion lane compares BenchmarkSummary files with compare_summaries. The broker performance lane judges a record::PerformanceRecord against a contract::Contract with budget::evaluate. Both take medians over repeated runs and allow the larger of a fixed floor and a multiple of the relative median absolute deviation: calibrated_tolerance.

Each criterion summary also names the bench executables its pass ran, as executables verified them, so a verdict says which binaries produced its numbers.

Note: If a shared krabka-tools repository is established across the krabka-io organization, this crate can be migrated there as a common benchmark utility for krabka-broker, krabka-protocol, and krabka-client-rs.

Modules

budget

The budget evaluator: judge a PerformanceRecord against the Contract it was measured under.

contract

The regression contract: the workload matrix and the budgets that a performance comparison is judged by.

executables

Which compiled benchmark executables a measurement pass ran.

record

The performance record: every raw trial of one comparison, with the provenance needed to reproduce it.

render

Render a record and its verdict as the Markdown that is published.

Structs

Args

CLI configuration arguments for parsing benchmarks.

BenchmarkComparison

One benchmark’s variance-calibrated reference/candidate decision.

BenchmarkMetric

A single benchmark measurement in nanoseconds per iteration.

BenchmarkSummary

Aggregated benchmark summary document.

BenchmarkVerdict

Machine-readable verdict over repeated same-host benchmark runs.

NamedMetric

A benchmark measurement together with the name of the benchmark it belongs to, as read from one line of bencher output.

Enums

Command

The commands besides parsing and comparing criterion output.

ParseBenchesError

Errors encountered while parsing benchmark output.

Functions

calibrated_tolerance

The relative change a comparison allows: multiplier times the larger of the two sides’ spreads, but never less than floor.

compare_summaries

Compare at least three repeated reference and candidate summaries.

format_rfc3339_utc

Formats a SystemTime as an ISO8601 / RFC3339 UTC string (YYYY-MM-DDTHH:MM:SSZ).

generate_summary

Generates a BenchmarkSummary from the provided command-line arguments.

median

The median of values; an even count averages the middle two. An empty values has no median, and gives NaN.

parse_bencher_file

Parses all benchmark metrics from a .txt file into the target map.

parse_bencher_line

Parses a single line of Criterion bencher output.

parse_benchmark_dir

Scans the given directory for all *.txt files and parses their benchmark metrics.

read_summaries

Read every JSON summary in directory, sorted by filename.

relative_mad

The median absolute deviation of values around center, as a ratio of center.

resolve_commit_sha

Resolves commit SHA from an optional CLI argument, falling back to an optional environment SHA.

run_command

Run one subcommand and say what it found.

run_from_args

Executes benchmark parsing and writes the resulting JSON summary to disk.