v0.6.32: the truth cut — RDFC-1.0 graph_digest, VALUES-on-graph-var refuses, staged refuses in txn blocks (#117, #109, #123) - #126
Merged
Conversation
VALUES-on-graph-var refuses Two silent failure modes become refusals, both measured before coded: #123 - staged workers commit per phase; inside a caller transaction the coordinator waited on workers that waited on the caller's locks - 31 min at 0% CPU, and statement_timeout NEVER fired (the wait sits in wait_for_shutdown, which does not observe interrupts - that deeper uncancellability is measured and goes up as its own issue). Guard at load_turtle_staged_run entry raises pgRDF#123 with the rewrite before any slot is taken; load_turtle's auto-dispatch adds !IsTransactionBlock to staged eligibility so the sniff defers to the standard parser. #109 - re-confirmed live on 0.6.31: VALUES binding a graph variable contributed nothing and the query answered over EVERY graph. refuse_values_on_graph_variable runs beside the #114 sibling on BOTH assembly paths, unconditionally (the repro has no UNION), increments filter_clauses_dropped, and names the rewrite. Control pinned: VALUES on a plain variable keeps applying. Error texts are exact-match test contracts (pgrx compares ==, verified in framework.rs:174 - the lock-module precedent followed).
Graph identity that survives reload: canonical blank-node relabelling per W3C RDFC-1.0, canonical N-Triples serialization, sha256. Algorithm label rdfc-1.0-sha256 (per the sealed interface contract - values are NOT comparable with first-degree structural pins, by design). Asserted triples only: inferred is a check value, never content (I13). Implementation is in-engine over pgRDF's own quad rows - no second RDF term model. Reuses the shacl.rs graph-read join, the sha2 dep that arrived with #118, and the term_type constants. First-degree hashing, related-hash, and the full hash-n-degree gossip-path tie-breaker with temporary issuers and lexicographic early-abort. Three budgets RAISE pgRDF#117 rather than degrade (calls 10k, permutation group 7, depth 32) - and the W3C poison test (test074, NegativeEvalTest) expects exactly that: refusing under a resource budget IS conforming. Verification: contract tests (reload-equality across relabelling; conclusive difference) plus a ten-case W3C rdf-canon conformance subset asserting our digest equals sha256 of the suite's OWN expected canonical documents - byte-for-byte spec exactness, first run. Vendored fixtures under tests/fixtures/rdfc10/ with W3C dual-license attribution. Closes #117
Canonical graph identity lands: pgrdf.graph_digest, RDFC-1.0, proven byte-for-byte against the W3C suite's own expected documents on first run, poison refusing under budget exactly as the spec demands. Two silent failure modes become refusals: VALUES-on-graph-variable (pgRDF#109) and staged-in-transaction-block (pgRDF#123). Release plumbing per the standing pattern: version 0.6.32 across Cargo.toml/lock, pgrdf.control, META.json (both fields), compose mount, CHANGELOG entry; upgrade script carries the generated graph_digest DDL; rolling bridge renamed with the delta appended; 00-smoke expected version strings advance. Full suite 381/381. Release modeled before code as ticket-1786916192416285000 (proposed); the recheck inverts on the bench after the chain. Closes #109
styk-tv
marked this pull request as ready for review
August 16, 2026 22:16
This was referenced Aug 17, 2026
Closed
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.
v0.6.32 — the truth cut
Closes #109. (#117 closed by the feat commit; #123 closed by the guards commit.)
Three deliverables, every one measured before it was coded:
pgrdf.graph_digest(graph_id) -> text(#117)RDFC-1.0 canonical graph digest — identity of meaning that survives reload. In-engine
over our own quad rows (no second RDF term model): first-degree hashing, related-hash,
full hash-n-degree gossip-path tie-breaking, canonical issuance, sorted canonical
N-Triples, sha256. Algorithm label
rdfc-1.0-sha256. Asserted-only (I13).Conformance proven, not claimed: ten vendored W3C rdf-canon cases pass with our
digest equal to sha256 of the suite's own expected canonical documents — byte-for-byte,
first run. The poison graph (test074, NegativeEval) hits the budgets and raises
pgRDF#117: refusing under budget is the conforming behaviour, so the fail-closeddoctrine and the spec turned out to be the same sentence. Fixtures vendored with W3C
dual-license attribution.
VALUESon a graph variable refuses (#109)Re-confirmed live on 0.6.31 before coding: three nonexistent graphs in a
VALUES, andthe query answered over every graph in the store. Now: stable
pgRDF#109refusal onboth assembly paths, unconditionally (the repro has no UNION),
filter_clauses_droppedincremented, plain-variable
VALUESregression-pinned as still working.Staged loader refuses in transaction blocks (#123)
The 31-minute silent hang (workers commit per phase; a caller transaction can never
allow it) becomes a
pgRDF#123refusal-with-rewrite atload_turtle_staged_runentry,and
load_turtle's N-Triples auto-dispatch falls back to the standard parser in thatcontext — auto-selection never picks a path that cannot work. The deeper discovery
(the coordinator's wait observes no interrupts in any context —
statement_timeoutnever fired across 600 s) is filed as #125 with the measurement.
Verification
guards on absent errors; error strings are exact-match contracts per the lock-module
precedent — pgrx compares
==, verified in its source)-D warningsclean · fmt cleanproposed); the recheck inverts onthe bench after the chain
Merge is not release: the tag follows the standing cadence on the merge word + tag
confirm.