Repository navigation
refbench v0.1.0: one-core, load-gated, interleaved reference benchmarks - #1
Merged
Merged
Conversation
Library and cobra command for wave W0 of the go-scikit performance campaign (go-scikit/roadmap#27): - Case registry naming the public function measured; Run alternates the order of the two sides per round, gates on the 1-minute load before every round (fleettools/loadavg, wait or refuse), records the load before and after each round, runs Go under GOMAXPROCS=1 and references under single-thread env, stops on a reference error. - Median/quartile verdict (slower, tie, faster); perf.json (schema 1, strict reader, fuzzed) and perf-table.md writers; deferred reports for allocation-only recording. - JSON-lines worker, embedded Python worker run through `uv run --no-project --python 3.12.12 --with-requirements refs.txt`, with refs.txt pinned with hashes; BuildC for C workers without cgo. - CI mode: allocations equal perf.json, table equals what perf.json generates, every public function has a case or a named exemption. - refbench table|check|issues|load; issues drafts, never files. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…ndows clock test - rbtest.Perf runs the CI mode, or a measurement or an allocation-only recording with flags, so no repository repeats the glue. A closed load gate skips with the reason and writes nothing. - CIConfig.Paths exposes the default file locations. - The alternation test used a Go body that did not scale with iterations, which a coarse clock (windows) could not calibrate. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Opening a path below a file is "not exist" on Windows and "not a directory" on Unix, so the 100% gate failed on Windows only. The open is now replaceable. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
A body that cycles through records allocated only on the first one in the count, which hid the exact-stage allocations of the degenerate records. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
First version of the shared reference-benchmark tool (wave W0 of the performance campaign, go-scikit/roadmap#27; design in
docs/performance-campaign.md, section 3).docs/perf.json(schema 1, strict reader, fuzzed) anddocs/perf-table.md.uv run --no-project --python 3.12.12 --with-requirements refs.txt,refs.txtpinned with hashes (numpy 2.5.4, scipy 1.18.1). The real worker was run once through uv locally (REFBENCH_UV), versions and protocol checked; no timing was taken (load was above the gate).refbench table|check|issues|load;issuesdrafts only.Reuse:
loadavgis imported from go-fleettools/fleettools v0.81.0 (it belongs in a module of its own; noted in the README). abbench's alternation rule and stats' JSON-lines worker are re-implemented in the library's shape, not copied.🤖 Generated with Claude Code