CLI tests red on main: stale init/help expectations (supabase default, expanded description, non-TTY iceberg)

[ihistand]

Found while doing #18 (typed-yargs refactor). These failures reproduce on pristine main (verified by git stash + run), so they're pre-existing test rot, not regressions. Three //cli:* targets are red:

1. //cli:index_init_test — "workflow_settings.yaml generated from init"

Expects a BigQuery project but init now defaults to supabase:

expected 'sqlanvilCoreVersion: 1.3.0\nwarehouse: supabase\ndefaultDataset: sqlanvil\n…'
to equal  'sqlanvilCoreVersion: 1.3.0\ndefaultProject: sqlanvil-database\ndefaultLocation: us-central1\ndefaultDataset: sqlanvil\n…'

warehouseOption default was changed to "supabase"; the test was never updated. Fix: pass --warehouse bigquery in the test (or update the expected YAML to the supabase shape, and/or add a supabase-default case).

2. //cli:index_init_test — iceberg init cases (5 failures)

AssertionError: expected null to deeply equal { bucketName, tableFolderRoot, … }. promptForIcebergConfig() returns null under the non-TTY test runner, so the iceberg config never materializes. Fix: stub/inject the iceberg prompt, or gate these on an interactive harness.

3. //cli:index_help_test — "shows help for 'init' command"

Stale expected description:

expected help text to include 'Create a new sqlanvil project.'
actual description: 'Create a new sqlanvil project (BigQuery, Postgres, or Supabase).'

Fix: update the expected substring.

4. //cli:index_compile_test — "disable-assertions flag (compilation)" (needs triage)

Failed in a --jobs=2 run with a Bazel-sandbox tmp-dir error (tmp_dir_N does not appear to be a sqlanvil directory, unexpected; remove it) — smells like a sandbox/npm install race under parallelism rather than a logic failure. Confirm whether it's flaky (re-run with --jobs=1/--runs_per_test) or a real stale expectation before fixing.

Why it matters

//cli:... is partly red, which masks real regressions (it nearly masked the verification for #18 — had to diff before/after to prove no new breakage). Worth getting the CLI suite green so it's a trustworthy signal again.

[ihistand]

Fixed and merged to main (21923d45, fast-forward).

One root cause for all of items 1–4: init now defaults to the supabase warehouse (commit d497e441), but these tests assert BigQuery behavior without passing --warehouse=bigquery.

• refactor: rename dataform → sqlanvil #1 index_init_test "workflow_settings.yaml generated from init": added --warehouse=bigquery → produces the expected defaultProject/defaultLocation settings.
• fix(deps): refresh Node toolchain pin to v20.20.2 LTS #2 index_init_test iceberg (5 cases): turned out not to be a prompt/env bug — the test inputs were read fine. Iceberg config is persisted only for BigQuery (init.ts: isBigQuery && projectConfig.defaultIcebergConfig), which is correct (iceberg is a BQ-only feature), so the supabase default silently dropped it. Added --warehouse=bigquery.
• fix(postgres-fixture): drop deprecated rules_docker dependency #3 index_help_test "shows help for 'init'": description was expanded to "… (BigQuery, Postgres, or Supabase)."; match on the substring without the trailing period.
• feat(deps): add pg, pg-query-stream, uuid for Postgres adapter (Phase 1) #4 index_compile_test disable-assertions: the "sandbox race" in the issue was a red herring — the real failure is that expectedCompileResult is BigQuery output (backtick-qualified SQL, defaultDatabase/defaultLocation) while the project was inited as supabase. Added --warehouse=bigquery to its setupTestProject; now deterministically green.

No product code changed.

Verification: //cli:{index_init,index_help,index_compile,index_project,index_connections,util}_test → 6/6 PASS. index_run_e2e_test still requires GCP creds (unchanged, out of scope here).

The offline CLI suite is green again, so it's a trustworthy signal once more. Closing as completed.