[sdairs]

Part of #1175.

Summary

Applies the one-layer rule (#1187) to the Cloud services tests in tests/cli_request_shape_test.rs. That covers service CRUD and lifecycle, settings, tags, private endpoints, backups, scaling, upgrade windows, service query, Query API endpoints and saved queries. The binary goes from 527 to 472 tests, about −2,400 lines.

• Validation (25 deleted, 4 trimmed). These subprocess tests only showed clap, post-parse validation or a builder rejecting bad input. The largest was the 30-run settings inexact-number loop. Before deleting, new unit tests were added:

  • settings number boundaries and inexact numbers
  • malformed settings documents and assignments
  • unreadable settings and config files
  • saved-query file I/O

  One subprocess run is kept where only the binary shows the behaviour: settings are parsed before organization discovery, and the selector forms.

• Body shape (14 deleted, 2 cut to one run). Each deleted field is set by a builder or passed through a destructured dispatcher binding.

• Shared behaviour (17 deleted). These were per-command repeats of convert_error, the 401/403 → exit 4 mapping, by-id not-found and the OAuth write check. The last end-to-end copy of each still exists, for example dispatched_cloud_401_*, deleted_resource_404_* and udf_every_write_rejects_oauth_before_api.

• Output owned by unit tests (6 deleted).

• Repeated runs and never-fail assertions:

  • The --file alias loops and single-mode halves are cut.
  • The reset-password test now runs in an isolated environment. Agent detection could previously force JSON in both of its "modes". It now asserts exit 1 rather than !success().
  • A contains("-") table check now checks the parsed rows instead.

Verifier. An independent verifier mutation-tested 35 handler and dispatcher sites. It restored or added coverage where a deletion, or a gap that predates this PR, let a mutation survive:

• --replace-open-api-keys with an existing endpoint, a read-modify-write. The deleted test is restored.
• Reset-password --new-password-hash and --new-double-sha1-hash reaching the wire.
• service settings get in human mode.
• A new service scale wiring test. The family had none before.
• service create --tag and --enable-endpoint, and private-endpoint create --description, now folded into existing exact-body wiring tests.

Known gaps, not addressed here:

• list_services_filtered losing its org scope on errors is not tested; only the unfiltered list has an error test.
• Scaling-schedule get in human mode is not tested.
• service private-endpoint get-config has no wiring test.
• The stored-key a_* and first_* tests each take about 1.1s and look like they hit a fixed wait. That would be a follow-up for Cut local test run time: six tests wait out real timeouts (~70s of ~2m) #1174.

Test plan

• cargo fmt --all --check, both clippy configurations, cargo check -p clickhousectl --no-default-features
• cargo test -p clickhousectl
• Author and verifier mutation checks: 51 mutations in total. Each surviving mutation was either fixed or is listed above.

🤖 Generated with Claude Code

---

Note

Low Risk
Test-only refactor with no production code changes; risk is reduced regression coverage in edge cases called out in the PR description.

Overview
Refactors clickhousectl Cloud services coverage by moving parse/validation and builder logic out of tests/cli_request_shape_test.rs into unit tests, and trimming the wiremock suite to HTTP wiring, auth routing, and behaviors only the full binary can show (~527 → 472 tests, ~−2,400 lines).

New unit tests cover shared input paths: read_config_value (missing/malformed JSON), read_sql (unreadable file → I/O failure), and services::settings (integer boundaries, inexact numbers, malformed documents/assignments, unreadable settings files).

Integration tests removed or shortened where clap, builders, or shared error mapping already had coverage—backup bucket validation loops, backup-config edge cases, duplicate --file alias runs, query-endpoint permutations, OAuth-before-HTTP repeats, API error classification duplicates, and large settings subprocess loops. Comments point reviewers to the unit tests or clap contracts instead.

Restored or added wiremock tests where mutation review found gaps: service scale PATCH to replicaScaling, service settings get human output, reset-password hash fields on the wire, query-endpoint --replace-open-api-keys**, and richer **BYOC create** / **private-endpoint** bodies. A few subprocess checks remain (e.g. settings fail **before org discovery**, selector forms for **service query`**).

^{Reviewed by Cursor Bugbot for commit 2e01803. Bugbot is set up for automated code reviews on this repo. Configure here.}