[Juliano1612]

Description

Fixes several issues found while verifying and deriving OpenBadge v3 credentials issued by other Data Integrity implementations.

• Canonicalization now serializes claims through json-ld to_rdf. The linked_data path rewrote context-coerced datatypes (for example https://www.w3.org/2001/XMLSchema#boolean) to the http form, so signatures over such quads never verified. Applies to the standard suites, ecdsa-sd-2023 and bbs-2023.
• Expansion stays strict through a shared policy in ssi-json-ld. Undefined terms raise an error as required by Data Integrity 2.4.3, never silently dropped.
• Label replacement canonicalization sorts by N-Quad string form, matching RDFC-1.0 and the existing verifier. The structural sort diverged when IRI and blank node subjects were mixed and broke mandatory indexes.
• URDNA2015 hash n-degree quads step 5.4.4.1 appends an already-canonical identifier to the path instead of recursing. Verified against the W3C rdf-canon suite.
• DataIntegrity::select picks the single selective proof from a proof set and ignores full-document signatures. New SelectiveCryptographicSuite::is_selective has a default of true, so existing implementors are unaffected.

Compatibility

• No public signatures change. Select returns MissingProof instead of NonSelectiveSuite or AmbiguousProof when no selective proof is present.
• The RDFC transformation impl now requires T: Serialize, which all in-tree claim types already satisfy.

Optional section

• Reviewers, please pay special attention to the strict expansion policy in ssi-json-ld and its use in canonicalization.rs, ecdsa_sd_2023/transformation/derive.rs and bbs_2023/transformation/derived.rs.
• Also worth a close look: the sort change in sd-primitives/canonicalize.rs and the proof selection change in core/src/lib.rs, since both affect which quads the mandatory indexes point at.
• This PR exceeds 500 lines because of test fixtures, not code. Around 2,100 of the 2,700 added lines are JSON-LD contexts bundled for offline tests, two OpenBadge credentials, a multi-proof credential and a 672-quad canonical N-Quads file. Source changes are under 200 lines across nine files.

Tested

• Undefined term rejection for signing, verification and derived SD verification.
• OpenBadge credentials with coerced datatypes, full-document and SD proofs, verified offline.
• Multi-proof credential derivation and verification.
• Canonicalization idempotency on symmetric blank node lists, cross-checked with jsonld.js canonize.

[timothee-haudebourg]

Approving, with just a request to select the selective proof between all the proofs, not just the first two.

[timothee-haudebourg]

Hmm this is weird. Why do you only look at the first two proofs? You should iterate over all the proofs:

let mut selected = None;

for proof in &self.proofs {
  if proof.suite().is_selective() && selected.replace(proof).is_some() {
    return Err(AmbiguousProof)
  }
}

select.ok_or(MissingProof).select(...).await

[timothee-haudebourg]

Sounds like this is an issue with rdf-types more than anything. Let's keep it as a work around for now, but can you flag that?

[timothee-haudebourg]

sounds like an AI comment. You don't need to justify why you don't do something.

[timothee-haudebourg]

I double checked the spec and indeed it should be added to path. I see no version of the spec that says to add to recursion_list, so it's a genuine and probably old bug that we have in there. I don't even know how we missed it with all the tests that we have, we must be lucky.