Expand description
§kael_perf
Versioned benchmark data and a small command-line profiler for Kael performance
tests. kael_util_macros::perf emits the matching test and metadata functions
when its perf-enabled feature is enabled.
Application binaries do not need this crate. It is published because Kael’s optional performance macros use its protocol types, and because framework and application maintainers can use the profiler to catch regressions in their own performance-sensitive tests.
§Profiling
Install Hyperfine, build a test binary
with kael_util_macros/perf-enabled, then pass that binary to the profiler:
kael_perf path/to/test-binary --important
kael_perf path/to/test-binary --json=currentThe importance flags are --critical, --important, --average, --iffy, and
--fluff. --quiet suppresses progress output. A JSON run identifier may use
ASCII letters, digits, -, and _; saved runs are bounded and written beneath
the invocation directory’s private .perf-runs folder. The profiler rejects
symlinked run directories, malformed or unpaired generated tests, and oversized
subprocess output. Run the command from your workspace root so profile and
comparison commands share the same data.
Compare a new run against a baseline with:
kael_perf compare current baseline
kael_perf compare --save=report.md current baselinePositive arrows mean the new run has higher throughput. Only tests present in both runs and in the same importance category contribute to a comparison.
The Rust API exposes the versioned metadata, timing, and report types used by the profiler and macro crate. See the API documentation and the Kael guide for the wider framework.
§License
Licensed under the Apache License, Version 2.0.
Modules§
- consts
- Preset identifiers and constants so that the profiler and proc macro agree on their communication protocol.
Structs§
- Output
- Aggregate output of all tests run by this handler.
- Perf
Report - Shim type for reporting all performance deltas across importance categories.
- Test
Mdata - Information about a given perf test.
- Timings
- The actual timings of a test, as measured by Hyperfine.
Enums§
- Fail
Kind - Why or when did this test fail?
- Importance
- How relevant a benchmark is.