perf: reuse stored items on republish and drop stale benchmark campaigns - #7
Merged
Merged
Conversation
Publishing bytes whose content address already exists re-read the whole item, hashed it again and compared it with the input. On a 32 MiB stream that second pass cost 115 ms and 33 MB of memory for a check every read already makes. An existing regular file of the right length is now reused and touched; a file of the wrong length is an interrupted publication and is replaced. Same-length damage is still rejected on read. Measured with bench/performance.py, macOS arm64, 30 repetitions: 32 MiB warm 287 -> 172 ms, 3 MB log warm 36 -> 25 ms, repository query warm 29 -> 21 ms, outputs byte-identical on every case.
…nary The Luna, Haiku and Fable pilots against Caveman, Ponytail, RTK and Headroom measured Scopelet 0.1.x-0.3.x with the manual skill. The engine, the default presentation and the installation model have all changed since, so their figures, protocols, harness, exporters and reports are removed rather than kept as stale evidence. What remains is reproducible offline on the binary that ships: bench/content.py (facts kept, bytes round-tripped, cold and warm timings) and bench/performance.py (wall time and peak memory, one or more arms with byte-identical output enforced). Fresh 0.5.0 reports are published under bench/results, the content gate is pinned to them, and the README speed table and verification page describe only what is measured now.
maxgfr
changed the title
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
bench/content.py(facts kept, bytes round-tripped, cold and warm timings) andbench/performance.py(wall time and peak memory, one or more arms with byte-identical output enforced). Fresh reports are published underbench/results/, the content gate is pinned to them, and the README speed table,docs/verification.mdandbench/README.mddescribe only what is measured now.Measurements (macOS arm64, 30 repetitions, outputs byte-identical on every case)
Verification
cargo fmt --check,cargo clippy --all-targets -- -D warnings,cargo test(new tests for reuse, repair and same-length rejection),cargo run -- bench,scripts/check_skill.py --pack, bench unit tests,npm test,scripts/check_readme.py,scripts/check_adapters.py. The CLI, the hook adapters for all three hosts and the installer were also exercised end to end in isolated configurations on realistic outputs (test logs, diffs, ANSI progress, malformed JSONL, binary input, threshold sizes, exit-status propagation) with no anomaly.