test(e2e): cover the oci catalog's setup group

Suleimi Ahmedrequested to merge
sahmed/e2e-oci-setup-scenarios into sahmed/e2e-caproni-authz

What this adds

The two automatable rows of the OCI catalog's Setup and configuration group, authored against the pattern registry and flipped to implemented:

Scenario Priority Covered by
e2e.oci.setup.create-repository critical TestRepositoryIsCreatedAndUsable
e2e.oci.setup.delete-repository high TestRepositoryDeleteHonorsDestructiveIntent

They live in a new client surface, e2e/bootstrap/oci/, because the rows assert no authorization and the bootstrap profile is the cheaper rig. The profile's harness surface keeps the route contracts it already had; these two files hold nothing but the scenarios, per e2e/README.md's rule.

create-repository reads the 201 off CreateRepositoryStatus rather than the mapping helper, which replays a conflict to a GET and so cannot tell a create from a repository that was already there. Then both halves of the format filter, then an oras push — which is what makes "immediately usable as a push target" an assertion rather than a claim about the status code.

delete-repository covers the destructive guard's 400 across six spellings including an omitted parameter, the 204 under either value with the name reusable at once, the 409 with the artifact pulled back to prove the images are intact, and the 202 with what it promises: pushes and pulls fail and the repository leaves the listing.

registry.Client gains DeleteRepository and DeleteRepositoryRawQuery. The raw-query variant is what lets a test spell the parameter the ways the guard rejects, an omitted one included, which a typed bool cannot express.

What is deliberately not asserted

Two arms of the delete row, both named here so a reader does not take the implemented status for full coverage of that cell:

  • The #371 retry. That emptying a hosted container repository through the /v2/ deletes leaves the 409 standing needs manifest and blob deletes, which no client in this tree has. That capability belongs with the lifecycle rows that are about deletion, not built inside a setup-group test.
  • The name reservation, caught at an instant. The row says a create at the name returns 409 until the removal finishes. The purge worker runs in the service's own binary and lifecycle.retention_window defaults to 0s, so the window is milliseconds and a hard assertion would flake. The test polls for the release instead and logs whether a 409 was observed. On the verification run below it was, on attempt 2.

The group's other two rows stay not started: create-repository-ui has no UI driver in this tree, and update-visibility defers itself to GA, since closed beta is private-only and there is no visibility to flip to.

Verification

Against a caproni rig deployed from ea6c9379, this branch's parent:

DONE 18 tests in 6.124s          exit=0
PASS e2e/bootstrap/oci.TestRepositoryIsCreatedAndUsable/e2e.oci.setup.create-repository (2.14s)
PASS e2e/bootstrap/oci.TestRepositoryDeleteHonorsDestructiveIntent/e2e.oci.setup.delete-repository (5.47s)

Every pre-existing bootstrap test still passes, so the new surface costs the profile nothing but its own binary.

The rig run earned its keep: it caught envelopeMessage decoding a top-level message when the envelope nests it under error, so the 409 message assertion read an empty string against a well-formed body.

Also run: e2e-catalog check (0 disagreements, covered 3 to 5), e2e-catalog generate (no id change, the rows' Status alone moved), go build -tags=e2e ./e2e/..., go test ./e2e/..., golangci-lint run --build-tags=e2e --max-same-issues=0 --max-issues-per-linter=0 --uniq-by-line=false ./e2e/... (0 issues), and scripts/ci/check-comment-caps.sh.

Notes for the reviewer

  • 371 insertions, of which 289 are the two new files. Under the reviewable-LOC line, so no split justification.
  • Stacked on !2052, which owns the catalogs and the pattern registry this builds on. Target it, not main.
  • pre-commit is not installed in the checkout this was authored in, so the hook chain was run by hand rather than on commit. It reports one failure, TestWireEncryption_NilInfraConfigResolvesLabKitDefault in cmd/artifact-registry, which reproduces at ea6c9379 with this commit absent and touches nothing this branch changes.

Merge request reports