test(utility): #204 UtilityExtensions layout math + path gate (dispatch 4rkh1s tertiaire)

[jsboige]

UtilityExtensions layout math + path gate (dispatch 4rkh1s tertiaire)

NEW additive file — UtilityExtensionsLayoutTests.cs (+24 tests). Dispatch 4rkh1s tertiaire: cover a documented-fragile, deterministic, key-free area that had ZERO existing coverage (grep confirms no test references ToJaggedArray / UtilityExtensions / PathIsUrl).

What it covers

(1) ToJaggedArray<T> — grid layout math (UtilityExtensions.cs:132). Pure function: list + columnLength → row-major jagged grid. This is the math behind PDF / mind-map column wrapping. Pinned:

• rowLength = ceil(count / columnLength) — exact-division and non-exact cases (Theory ×6).
• Trailing short row carries ONLY the remainder, no padding — a padding regression (default(T) fillers) would inject phantom blank cards into the final grid row.
• Row-major fill order — a column-major/transpose regression would scramble card positions.
• ToJaggedArray ↔ Flatten inverse round-trip — lossless for every column length 1…100.

This is exactly the class of geometry assertion the coordinator flagged ("count the geometry, not just the text") for structured grids.

(2) PathIsUrl — the asset-path gate (UtilityExtensions.cs:160). CLAUDE.md documents relative asset paths as the root cause of "white/empty cards" — this gate decides whether a path needs rewriting to an absolute GitHub URL. Pinned:

• The CLAUDE.md relative failure cases (../../Cards/..., ../, bare, Windows C:\, Unix /var/) → NOT URLs (so the caller rewrites them).
• http/https → URLs; null/empty/whitespace → false; whitespace-padded URL → still resolves.

Tests

246 pass / 0 fail / 5 skip (223 + 24). Deterministic, key-free, release-independent. No existing file modified. AssetConverterConfig.cs / DatasetUpdaterRootConfig.cs / FallaciesLocalizationTests.cs untouched (reserved #444 β/γ po-2023).

🤖 Worker po-2024 — dispatch 4rkh1s tertiaire.

[clusterManager-Myia]

[NanoClaw]

PR scope

Production fix in OwlAdapter.cs (.Equals(string.URI) → .ToString() == .ToString() at 14 sites) + regression test updates + two new test files (OwlValidatorLivePathTests.cs — 6 tests proving the production validator path is live; UtilityExtensionsLayoutTests.cs — 24 tests for layout math + path gate). The PR body mentions 24 tests but there are actually 30 new test methods across the two new files (6 + 24), plus 8 renamed-but-repurposed tests in OwlAdapterRegressionTests.cs.

Production change: OwlAdapter.cs

The fix is mechanical and correct in intent: RDFResource.Equals(string) always returned false by type mismatch, so every fallback reader returned empty. Switching to .ToString() comparison sidesteps the type problem. Applied consistently across all 14 sites — GetAnnotationSubjects, GetAnnotationObjects, GetResourceAnnotations, GetLiteralAnnotations, CheckHasClass, HasAnnotation, CheckIsNarrowerConcept, GetExactMatchConcepts, GetCloseMatchConcepts, GetRelatedMatchConcepts, GetResourcesByType.

One concern: CheckIsNarrowerConcept's fix (line ~285) inverts the try/catch semantics. Before: try { return SKOSHelper... } catch { return annotation-scan }. After: if (SKOSHelper...) return true; catch { } return annotation-scan. The original try/return in the catch block is now gone — the catch is empty and the annotation scan always runs as the fallback. This is functionally fine (the annotation scan is the real fix), but it changes the control flow: before, if SKOSHelper threw, it immediately returned the annotation scan result; now it falls through to the annotation scan after the catch. Same outcome, different path. Worth noting this was intentional — not a bug.

Another concern: .ToString() comparison is fragile. If RDFResource.ToString() or GetIRI().ToString() ever changes format (e.g. adding angle brackets, changing URI encoding), these comparisons silently break again. The proper fix would be GetIRI().Equals(RDFResource.GetIRI()) to compare Uri to Uri, or accessing the underlying URI string uniformly. However, the ToString() approach works correctly today and the comprehensive regression tests will catch any future format change. This is an acceptable tradeoff given the test coverage.

Test quality: OwlAdapterRegressionTests.cs (modified)

The existing [BUG] characterization tests are correctly flipped from asserting broken behavior (empty/false) to asserting correct behavior (populated/true). The rename from BUG_* to descriptive names is clean. The CheckIsNarrowerConcept test now also asserts the negative case (Res("Unrelated") is not narrower of parent), which is a real improvement. CheckHasClass also adds the negative case. Good.

Minor note: the summary comment in the class doc still says "No existing file modified. Baseline additive." — but OwlAdapter.cs is modified in this PR. That line should be removed. This is cosmetic.

Test quality: OwlValidatorLivePathTests.cs (new)

This is the strongest addition. Using reflection to inject the private _ontology field into the production validator is clever — it proves the actual production code path is live, not just the adapter layer. The 6 tests are well-structured: pass-with-annotations, fail-without-annotations, fail-without-AIF-mappings, pass-with-AIF-mappings, and a contract guard on the specific reader the validators branch on.

One concern: OwlOntologyValidationTests is described as "NOT an xUnit suite — a runtime validator invoked by OwlValidatorConfig.Apply." The tests here instantiate it with new OwlOntologyValidationTests(config) and then inject via reflection. If the production class's constructor does non-trivial work (loading files, network, etc.), these tests could be flaky or environment-dependent. The claim is "deterministic, key-free, release-independent" which seems valid based on the code, but the constructor's side effects aren't visible in this diff. Worth verifying the constructor is safe.

Test quality: UtilityExtensionsLayoutTests.cs (new)

Well-structured. The ToJaggedArray tests cover:

• Exact division (6/3 = 2 full rows)
• Non-exact division with trailing short row (7/3 = [3,3,1])
• Theory parameterized row count across 6 cases including empty source
• Row-major order preservation (flatten round-trip)
• Inverse round-trip through Flatten across 6 different column lengths

The PathIsUrl tests cover http/https, relative paths, Windows paths, Unix paths, null, empty, whitespace-only, and whitespace-padded URLs.

Good edge cases: empty source (count=0), fewer items than columns (count=1, columns=3), null path, whitespace handling. These are the right boundary conditions.

Missing edge case for PathIsUrl: no test for FTP or other non-http schemes (ftp://, file://). If the implementation uses Uri.TryCreate with UriKind.Absolute, these would return true. If it explicitly checks for http/https only, they'd return false. Worth a test either way since the code path is a security gate (decides whether to rewrite asset paths).

Missing edge case for ToJaggedArray: columnLength=0 or columnLength=-1. The implementation presumably throws or returns empty, but untested. Minor since these are degenerate inputs unlikely in production.

Security

No secrets, no credentials, no connection strings found in the diff.

Verdict

Solid fix with good regression coverage. The .ToString() comparison approach is a pragmatic workaround rather than the ideal type-safe fix, but the tests guard it well. The OwlValidatorLivePathTests proving the production validation path is no longer a silent false-pass is particularly valuable. Recommend adding FTP/file scheme tests for PathIsUrl and consider the long-term fragility of .ToString() comparisons.