mbo/hash/measurements: decouple measure/publish; labeled multi-machine README from data bundles - #250
Merged
Merged
Conversation
…multi-machine README The measurement run now writes ONLY a data bundle - no committed charts/tables into the tree - so it stays authoritative and per-compiler/per-machine runs go back-to-back (fixes the "dirty=True" non-authoritative warning caused by writing SVGs up front). A new `publish` command renders the README perf section from a SELECTED set of committed bundles: one labeled `###` block per machine (its charts + curated tables), replacing a marker region it cuts and re-appends. Charts carry the machine identifier as a subtitle (cpu-brand · arch · cores · compiler · sha), and each block heading covers both charts and tables. `verify` (no args) re-renders every featured bundle from the manifest publish leaves in the README and diffs the committed charts. Single measurement, not two: the benchmark emits its README size subset (kReadmeSizes) as `readme_sizes` in the dataset context, so the curated ~22-row tables are EXTRACTED from the one FULL run - no separate fast run, and no second size list to drift (the C++ list is the sole source, emitted into the data and read back). Fast mode stays only as a quick dev loop.
The bundle filename already carries the full machine identity (<slug>_<cores>c_<compiler>_<sha>_<stamp>.tgz), so the per-machine subdirectory was redundant and only added a level to fight with. Bundles now live flat under data/; .gitignore/.gitattributes and the docs follow.
Fab-Cat
approved these changes
Jul 11, 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Overhauls how measurement results become the README, so results from several machines can be shown from the committed data packages, and so the measurement run no longer dirties the tree (which was silently marking runs non-authoritative).
Decouple measure from publish
run_measurements.pynow writes only a data bundle - no charts/tables into the tree. So the run stays authoritative and you can do per-compiler / per-machine runs back-to-back.publishcommand renders the hash README's perf section from a selected set of committed bundles: one labeled###block per machine (its charts + curated tables), by cutting and re-appending a marker region.--bundlesis the selection.verify(no args) re-renders every featured bundle from the manifestpublishleaves in the README and diffs the committed charts - published figures are provably from committed data.One measurement, not two
The benchmark emits its README size subset (
kReadmeSizes) asreadme_sizesin the dataset context, so the curated ~22-row tables are extracted from the single FULL run - no separate fast run, and no second size list to drift (the C++ list is the sole source, emitted into the data and read back). Fast mode stays only as a quick dev loop.Flat data layout
Bundle filenames already carry the full machine identity, so bundles live flat under
data/(no redundant per-machine subdir).End-to-end tested locally: run → bundle → publish (multi-machine) → verify all green; markers matched, charts labeled, tables filtered to the 22-size subset.
Workflow: measure on N machines (each commits one bundle), then a single
publish --bundles ...renders the whole page.