Repository navigation
Conversation
krafczyk
marked this pull request as ready for review
October 4, 2026 22:42
Add detached fixed identity and occurrence results with collision-checked graph identity and bounded provenance diagnostics. Keep existing query entry points intact for the staged V3 cutover.\n\nVerified with focused identity and legacy result tests in big_env (35 passed).
Introduce private Store and Repo inventory cuts with explicit record-family coverage, separate stored and metadata authority, and direct relationship facts. Keep existing public query behavior unchanged.\n\nVerified in big_env with focused source, multistore, and metadata tests (25 passed).
Add private immutable selector, reference, metadata, and authority restrictions while retaining existing public query entry points. Preserve validation failures and reject unsupported indexed-only claims; keep routine coverage representative rather than rerunning the whole save-backed matrix.\n\nVerified in big_env with 29 new identity/source/restriction tests and 17 existing selector/metadata tests.
Add private closure and strict non-empty occurrence traversal with literal edge policies and direct existential projections. Preserve typed paths and bounded fixed-input evidence without changing authoritative GraphPath storage.\n\nVerified with focused traversal, path, and fixed-result tests (82 and 13 passed).
Restart whole composed terminals when metadata demand expands or a Store cut changes, and keep exact selective reads consistent with later inventory. Add scalar and bounded sinks with collision-only ordering work and fail closed on repeated instability. Curate low-cost routine tests.\n\nVerified in big_env with focused U1-U5 tests (43 passed) and the existing federation/performance boundary (96 passed before the final cut fix).
Store graph-exact typed identity and direct relationship candidates in derived memory and schema-8 SQLite views. Keep V3 indexed-only unsupported until authority coverage is proven; maintain transactional row-set witnesses using only incremental row deltas rather than rescanning history on clean saves.\n\nVerified in big_env with 99 focused SQLite/metadata/index tests and 6 V3 index tests after incremental changes.
Use the same V3 identity and occurrence query models for detached fixed results, preserving captured membership, typed path refinement, bounds, and contributor evidence. Characterize non-query Repo get/item/load contracts before the public producer cutover.\n\nVerified with focused fixed-result, traversal, and lookup tests in big_env (17 passed).
Build one exact holder map per Store cut and reuse snapshot reads for current, lineage, and snapshot scopes. Avoid repeated aggregate scans and duplicate current-metadata reads while retaining failure facts for later validation.\n\nVerified in big_env with 44 focused V3 restriction/execution and legacy metadata tests.
Bind cache checks to an explicit Repo tier and retain cache-only exact StateRef receipts on selective Repo queries. Snapshot selected cache knowledge per terminal without adding members to Store- or collection-backed inputs.\n\nVerified in big_env with 41 focused restriction, source, and execution tests.
Conjoin target selectors and scoped metadata on live and fixed occurrence queries while sharing a terminal source cut. Validate eligible target facts before scalar early stops, restart on late-demand mutation, and preserve direct existential projections.\n\nVerified with 85 focused V3 traversal/restriction/execution and legacy nested tests in big_env.
Precompute complete typed stored-root, object, and state membership once per detached Store cut. Avoid rebuilding all root definitions for each candidate in broad stored() queries while retaining graph-distinct identity.\n\nVerified with 25 focused source and restriction tests in big_env.
Use typed root/path/target keys for OccurrenceSet containment instead of dataclass equality, which can collapse graph-distinct CDef roots.\n\nVerified in big_env with 21 fixed identity and traversal tests.
Use validated digest-addressed stored-root authority for positive graph-exact CDef selections without broad hydration. Retain complete inventory fallback for alias/declaration/snapshot/main-only authority and reject unsupported no-scan requests rather than report false absence.\n\nVerified in big_env with 34 V3 execution/source and existing Repo performance tests.
Pin StateSelectorRef leaves through selector(value, scope=repo) before V3 composition, share the existing legacy resolver kernel, and reject hidden live-Object selection. Preserve exact Generator support and witness budgets while bounding selector-failure diagnostics.\n\nVerified in big_env with focused V3 selectors (26 passed) and existing structural/template selector cases (107 passed before final diagnostic tightening).
Make identity take(n) an explicit fixed-result terminal and add canonical bounded occurrence retrieval without collecting every raw path. Preserve visible requested limits, zero-limit no-read behavior, and collision-only ordering ties.\n\nVerified in big_env with 71 focused execution, traversal, and legacy nested tests.
Preserve get/item root and explicit-cache candidate selection, direct exact stored-root reads, callable strong-cache filtering, and structural loading errors without depending on the legacy public query builder. Validate complete root inventory for broad reads.\n\nVerified in big_env with 64 affected Repo query, result, save/load, and lookup tests.
Allow deferred union and intersection with captured IdentitySet members while preserving boundedness and conservative authority scope. Fixed-only algebra remains source-free.\n\nVerified in big_env with 32 focused execution and identity tests.
Retain exact ObjectRef members and their existing StateRefs under ObjectRef selection without projecting new identities. Add top-level union/intersection delegation that keeps fixed-only operations source-free and preserves conservative scope.\n\nVerified in big_env with 57 focused V3 restriction/execution and legacy reference tests.
Use verified direct root markers across the selected Repo Stores when all contribute positive authority; otherwise fall back to a complete inventory without losing alias-only roots or source evidence. Count selective fenced reads in terminal diagnostics.\n\nVerified in big_env with 61 focused V3 source/restriction/execution and existing Repo performance tests.
Replace the remaining public Query V2 builders and result sets with Query V3 identity queries, fixed result algebra, and bounded execution semantics. Harden authoritative/indexed execution, stored-root paging, refresh behavior, diagnostics, and immutable identity caching. Update migration and backend documentation and consolidate coverage around the supported V3 contract. Verification: good-enough (3415 passed, 30 skipped); focused Query V3, API, index, import, and dispatch tests.
Stop source-path injection from changing pinned worker identity and remove the retired Query V2 reference module from package artifact requirements. Verification: 3 focused exhaustive-delta regressions passed.
This was referenced Oct 6, 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
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
Completes the Query V3 public cutover.
Repo.query()andStore.query()now expose unified identity queries; the retired Query V2 builders, result classes, and maintained callers are removed rather than preserved behind a compatibility wrapper.Behavior
Repo.get()andRepo.__getitem__()lookup results.Verification
good-enough: 3,415 passed, 30 skipped.good-enoughdelta: 1,103 passed, 301 skipped across exactly 1,404 maintained nodes; no full exhaustive rerun.main: 21 passed.good-enough, Ubuntu lightweight/notebook, and Windows Python 3.10-3.14.git diff --checkpasses.Boundaries
No Query V2 compatibility API is shipped. Authoritative Store contents remain the source of truth; SQLite query projections remain derived and rebuildable.