feat(hash): generate the README quality tables from JSON + a measured bundle; measurements as a dev package - #256
Merged
Conversation
…ts table - Drop the nested mbo/hash/measurements module and its .bazelignore entry so it is an ordinary dev package (built/tested in dev + CI). It is still stripped from release archives by release_prep.sh, so nothing shipped depends on it. - Add //mbo/hash/measurements:quality_sh_test, a bashtest that runs `hash_benchmark_report.py quality --check` (README Results table in sync with the curated data), plus drift-detection and idempotence checks. - Add a no-deps-on-measurements pre-commit guard: no .bazel/.bzl file outside the dir may reference //mbo/hash/measurements, so the release strip cannot leave a dangling reference. - Export //mbo/hash:README.md (public) for the test; refresh measurements/README and the hash README Methodology (how to measure on another machine).
The intro credited only mumbo/jumbo with passing and appended "dumbo is the compact companion", which read as if dumbo did not pass. All three in-house hashes pass 188/188; dumbo is a compact single-lane companion with a different profile, not a lower-quality one.
helly25
enabled auto-merge (squash)
July 12, 2026 23:20
…red bundle Replace the curated _SMH_RESULTS_ROWS with a clean separation of concerns: - hash_algorithms.json holds ONLY manual/editorial per-instance columns (role, notice, seeded, streaming, starlark, available_via, smhasher name), one entry per (algorithm, width) so both tables separate 64/128 consistently (murmur3 is 128-only - its GetHash64 folds the 128; xxh3 is native at both). - Measured verdict/score/failures are re-parsed from a data bundle's logs (_measured_from_bundle; the stored smhasher.json predates parser fixes) and joined by smhasher name; a _MISSING_MEASURED stopgap covers FNV-1a-64/ MurmurHash3-128 (invalid-name logs) until a bundle carries them, warned loudly. - now renders BOTH the Algorithm-overview (SMHasher3 PASS/FAIL derived from the measurement) and the Results table (measured score + families) between their own marker regions; gates both. - New command: all bundles at the same source SHA must report identical measurements (verdicts are machine-independent). Verified: the two committed bundles agree. - quality_sh_test reads the LFS bundles (data dep); the test CI job fetches them via git lfs pull (actions/checkout leaves pointers).
helly25
changed the title
Fab-Cat
approved these changes
Jul 13, 2026
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
2 participants
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.
Makes both SMHasher3 quality tables in
mbo/hash/README.mdgenerated (not hand-edited), with a clean separation of manual vs measured data, and makes the measurement tooling a normal dev package that CI gates without leaking into releases.Generated quality tables
hash_algorithms.json— the manual/editorial SOT: per (algorithm, width) instance, only the non-measured columns (role,available_via,notice,seeded,streaming,starlark, and thesmhasherregistration name). One entry per width, so both tables separate 64/128 consistently (murmur3 is 128-only — itsGetHash64folds the native 128; xxh3 is native at both)._measured_from_bundlere-parses the bundle's per-algorithm logs with the current parser (the storedsmhasher.jsonpredates parser fixes and is score-less). A hardcoded_MISSING_MEASUREDstopgap coversFNV-1a-64/MurmurHash3-128(their committed logs are the invalid-name stub) until a bundle carries them — warned loudly on every render.quality --bundle <tgz>renders BOTH the "Algorithm overview" (itsSMHasher3PASS/FAIL derived from the measurement, so it cannot disagree with Results) and the SMHasher3 "Results" table (measured score + failing families) between their own<!-- BEGIN/END -->marker regions.quality --checkgates them.consistency— all bundles at the same source git SHA must report identical measurements (a verdict is a property of the algorithm, not the machine). The two committed bundles agree.Dev package + CI
mbo/hash/measurementsmodule and its.bazelignoreentry so it's an ordinary dev package; still stripped from release archives (release_prep.sh).//mbo/hash/measurements:quality_sh_testrunsquality --check+consistency+ a drift/idempotence guard over the committed bundles.git lfs pull) —actions/checkoutleaves LFS pointers, so the quality gate needs them materialized.no-deps-on-measurementspre-commit guard: nothing outside the release-stripped dir may reference it.Docs
shexample for producing measurements on another machine; intro clarified that all three in-house hashes pass 188/188.Verification (local, all green)
quality --checkmatches the committed tables; the drift guard fails on a changed table; regeneration is idempotent;consistencypasses;quality_sh_testpasses under bazel; trunk clean.Follow-up (separate): the AMD/linux bundle's FNV-1a 64-byte latency is a gcc-15 measurement artifact (chain elided → 0.27 ns); needs a re-measure or a latency-benchmark hardening. Not a quality-table issue.