Tool catalogue

Use this catalogue to find a tool for a question, then check whether its integration and platform support match your experiment. The Julia call below returns the inventory without running tools or installing them.

The authoritative inventory is data/tool-catalog.json. Its entries distinguish implemented collectors, existing extensions, scenario integrations and candidate or external tools. A listing does not establish availability on your machine or qualification of its results.

Start with Understand performance measurements for timing, GC, allocations and flame graphs. The check catalog explains each implemented collector before its options; the native review below explains the complementary tools and their measurement units.

using PerfChecker
tool_catalog()

julia --startup-file=no --project=. -e 'using PerfChecker; exit(perfchecker_main(ARGS))' -- tools

VS Code and Oxygen consume this same inventory. For an assessment of Julia profilers, native dependencies, Valgrind, hardware counters and platform limits, read the Julia and native profiling review.

Reuse the workload

Measure existing test items directly when whole-item timing is appropriate. Use shared scenarios for explicit preparation, operation, correctness, synchronization and cleanup boundaries.

scenario_sync(root) relates declared scenarios to literal CI matrix entries. These are proposed combinations, not proof that a job executes the scenario. Dynamic expressions and external reusable workflows remain unresolved. Input fingerprints expose drift without rewriting manually adapted scenarios.

write_scenario_workflow(path; catalog="perf/scenarios.toml", project="perf", implementations=["cpu"]) writes a new workflow and refuses to overwrite a file. Its generated matrix includes Windows, Linux and macOS; generation does not qualify those platforms. The gate checks correctness and availability, not a performance threshold. The controller must contain the selected collectors.

Static warnings and short samples may suggest irrelevant fixes. Keep diagnosis separate from timing, retain unavailable configurations and validate proposed changes against correctness oracles and unchanged measurement boundaries.