Surfaced by turning Newman on (#156). Not a regression — the divergence has always been there; nothing measured it until the collection's earlier failures were cleared away.
The spec
openspec/specs/openbuild-version-snapshots/spec.md (status: done), REQ-OBV-005:
The diff endpoint changes shape under the versioned model: diffing two ApplicationVersion rows is the canonical case; comparing two historical states of one ApplicationVersion (time-travel diff on a single row) is the second supported case.
where {fromRef} and {toRef} are either:
- An ApplicationVersion
slug (e.g. staging) — diff is against the current saved state of that version's manifest.
- The literal
current:<versionSlug> — reserved syntax for forward compatibility.
- A version-history reference
history:<versionSlug>:<revisionId>.
The controller
ApplicationsController::resolveVersionBlob() accepts the literal draft, or an ApplicationVersion UUID, and nothing else. None of slug, current:<slug> or history:<slug>:<rev> resolves.
So the second supported case — a time-travel diff on a single row, which is the whole point of the model ADR-002 moved to — is not reachable through the endpoint at all.
Why it went unnoticed
tests/integration/openbuild-versioning.postman_collection.json drove the endpoint with two sibling snapshot UUIDs, which is the append-only model REQ-OBV-002 retired. Those rows are never created, so to interpolated empty and the request 500'd — and the 500 was itself a second defect masking this one.
That 500 is fixed in #156: resolveVersionBlob() documented "Returns null on miss so the caller can surface 404" while ObjectService::find() throws, so both of diffVersions()'s 404 branches were dead code and the throw fell into the outer catch (Throwable) as internal_error. It now answers 404 — the eighth instance of the family #159 fixed, missed there because gate-49 only flags an untranslated lookup outside a try/catch.
With the 500 gone, the endpoint answers 404 and the divergence is what is left.
Deliberately not papered over
The three REQ-OBV-005 assertions are left red in #156, with the reason written into the collection's folder description. Rewriting them to match today's UUID-only behaviour would lock in the divergence and delete the only evidence of it.
What a fix needs
resolveVersionBlob() to accept a version slug (scoped to the parent Application), current:<slug>, and history:<slug>:<revisionId> via OR's object-history API.- The Newman folder re-pointed at the slug grammar, plus a case that diffs two historical states of one row — the case that has never been testable.
Surfaced by turning Newman on (#156). Not a regression — the divergence has always been there; nothing measured it until the collection's earlier failures were cleared away.
The spec
openspec/specs/openbuild-version-snapshots/spec.md(status: done), REQ-OBV-005:The controller
ApplicationsController::resolveVersionBlob()accepts the literaldraft, or an ApplicationVersion UUID, and nothing else. None ofslug,current:<slug>orhistory:<slug>:<rev>resolves.So the second supported case — a time-travel diff on a single row, which is the whole point of the model ADR-002 moved to — is not reachable through the endpoint at all.
Why it went unnoticed
tests/integration/openbuild-versioning.postman_collection.jsondrove the endpoint with two sibling snapshot UUIDs, which is the append-only model REQ-OBV-002 retired. Those rows are never created, sotointerpolated empty and the request 500'd — and the 500 was itself a second defect masking this one.That 500 is fixed in #156:
resolveVersionBlob()documented "Returns null on miss so the caller can surface 404" whileObjectService::find()throws, so both ofdiffVersions()'s 404 branches were dead code and the throw fell into the outercatch (Throwable)asinternal_error. It now answers 404 — the eighth instance of the family #159 fixed, missed there because gate-49 only flags an untranslated lookup outside a try/catch.With the 500 gone, the endpoint answers 404 and the divergence is what is left.
Deliberately not papered over
The three REQ-OBV-005 assertions are left red in #156, with the reason written into the collection's folder description. Rewriting them to match today's UUID-only behaviour would lock in the divergence and delete the only evidence of it.
What a fix needs
resolveVersionBlob()to accept a version slug (scoped to the parent Application),current:<slug>, andhistory:<slug>:<revisionId>via OR's object-history API.