Repository navigation
Reduce exact Fano census memory with compact singleton storage - #3312
Merged
Mister-Meeseeks merged 1 commit intoOct 17, 2026
Merged
Conversation
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The exact Fano census allocates a member list and tuple for almost every one of its 1.38 million singleton product classes. This opt-in package stores one base-n word code per singleton and allocates lists only for actual collisions, preserving full matrix keys, arbitrary class sizes, every ordered relator pair and the complete original replay.
Four alternating fresh-process repetitions on CPython 3.12.15 measured mean peak RSS 392.4 → 223.8 MiB (43.0% lower) with canonical multiplication, and 380.3 → 211.4 MiB (44.4% lower) when composed with Charan's pinned PR3219 row tables. Mean wall times were 29.12 → 27.97 seconds and 11.65 → 10.32 seconds. Timing ranges overlap for the row-table pair; these are descriptive shared-host measurements. All 16 complete receipts and full census digests match. Raw samples, commands, source hashes and ranges are retained.
Validation: the complete registered CPython 3.12.15 replay passes, including all four complete verifier modes and 17 isolated optimized oracle, mutation, authentication, restoration and benchmark controls. The controls also pass normally. Reproduce:
This is verifier engineering, with no new mathematical result or supplier admission. Canonical multiplication/packing and every original source/Fano/control check remain unchanged in the core pair. The optional comparison retains PR3219's exact pending source and credit; baseline/compact workers do not read it. No canonical verifier, selected record or frozen source changes. Selected cold replay and native/all-width obligations are outside scope.
Coordination: Discussion3062 comment19004550. Base canonical main/fork
463113d675af7410ac91d53a4fed51ec6494eb9a, priority revision 6. Retain Sudarsh Chaturvedi/PR3168, Gonzalo Mier/PR3097, Charan/PR3219 (209d20d9da025c90110465f9a1f18186a6160147) and the full chart-swap/selected-source lineage in the package. Substantial OpenAI Codex assistance; separate AI executions are not independent human review. Apache-2.0.