Skip to content

Plan, apply and roll back changes to a production ontology, with a blast radius report and a proof an auditor can re-check.

Withdraw the result deduplication: it broke agreement with the oracle

FABIOTESSc271fbe768ad
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 696eb94d..993c50af 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -57,9 +57,6 @@ All notable changes to Open Ontologies are documented here.
indistinguishable, which is the case #131 was filed about. pyshacl returns
the property shape and this now agrees. The node shape is still reported,
under `node_shape`.
-- **Duplicate results.** One shape carrying two target declarations that select
- the same node reported every violation on that node twice. A validation
- report is a set.
- **The reasoner was not a fixpoint of its own rule set.** Only the `rdf:type`,
`rdfs:subClassOf` and `rdfs:subPropertyOf` indices were rebuilt each
iteration; every schema index was filtered once out of the run-start
diff --git a/docs/lean-certificates.md b/docs/lean-certificates.md
index cd06b4fa..c98c3879 100644
--- a/docs/lean-certificates.md
+++ b/docs/lean-certificates.md
@@ -154,5 +154,8 @@ Stated rather than discovered later.
but if you materialise into a store that already held inferences they appear in `asserted.tsv` as
assumptions with nothing marking them derived. Use `inference_graph: true` (decision 0001) when
that distinction matters.
-- **`focus_nodes` in a SHACL report is a sum over target declarations**, so a node selected by two
- declarations of one shape is counted twice. The violations themselves are deduplicated.
+- **A SHACL report double-counts a node selected by two target declarations of one shape.** Both
+ `focus_nodes` and `violation_count` are affected. Collapsing identical results is not the fix:
+ SHACL emits one result per SPARQL solution, pyshacl does too, and deduplicating breaks an exact
+ agreement with it. The fix is to union a shape's focus nodes across its declarations, which is a
+ change to the evaluation loop and is not done.
diff --git a/src/shacl.rs b/src/shacl.rs
index 7aab8cb4..ac3b93c8 100644
--- a/src/shacl.rs
+++ b/src/shacl.rs
@@ -1721,21 +1721,24 @@ impl ShaclValidator {
let nothing_matched = (!targets.is_empty() && focus_nodes_total == 0)
|| (targets.is_empty() && declared_any_shape);
- // A validation report is a SET of results. One shape may carry two
- // target declarations that select the same node (a `sh:targetClass` and
- // a `sh:targetNode` for one of its instances), and the loop above walks
- // targets, so that node was checked twice and every violation on it was
- // reported twice. Two results identical in every field are one result.
+ // Deliberately NOT deduplicated, and this was measured before it was
+ // decided. One shape carrying two target declarations that select the
+ // same node does get checked twice, which inflates `violation_count`
+ // and `focus_nodes`. Collapsing identical results looked like the fix
+ // and is not: SHACL 5.3.2 emits one result per SPARQL solution, so a
+ // constraint returning several solutions for one focus node produces
+ // several results that are identical wherever the extra solution binds
+ // nothing this report carries. pyshacl emits those too. On the
+ // 39-shape corpus from issue #132 both engines return 249 results over
+ // 245 distinct (focus node, shape) pairs, and deduplicating took this
+ // engine to 245 and broke an exact agreement.
//
- // This does not touch `focus_nodes`, which is still a sum over targets
- // and so still counts such a node twice. Making that a distinct count
- // means enumerating focus nodes rather than counting them in SPARQL,
- // which is a memory trade on large graphs and a separate decision.
- {
- let mut seen: HashSet<String> = HashSet::new();
- violations.retain(|v| seen.insert(v.to_string()));
- }
-
+ // The real fix for the double-checked node is to take the union of a
+ // shape's focus nodes across its target declarations rather than to
+ // walk targets, which is a change to the evaluation loop and not to
+ // the report. Until then the inflation is a documented limitation,
+ // because a wrong count is cheaper than a validator that drops results
+ // the specification says to emit.
let mut report = serde_json::json!({
"violation_count": violations.len(),
"violations": violations,