[paveltiunov]

Check List

• Tests have been run in packages where changes have been made if available
• Linter has been run for changed code
• Tests for the changes have been added if not covered yet
• Docs have been added / updated if required

Issue Reference this PR resolves

Refs #5652, #6158, #8561, #10181, #10962, #11240. These tests only reproduce the bugs and don't fix them. Each one fails on current master. I also reproduced every bug end to end on the v1.8.16 release, running Cube and Cube Store against Postgres 16.

• Generated SQL for subquery dimensions does not respect sql_alias #8561 packages/cubejs-schema-compiler/test/integration/postgres/issue-8561.test.ts: the join alias of a subquery dimension is built from the cube name and ignores sql_alias. With two subquery dimensions on a cube whose name is longer than 63 characters, Postgres truncates both aliases to the same name and fails with table name "..." specified more than once. Fails on both planners.
• API error "One or more Primary key is required for '${cubeName}' cube" even when allowUngroupedWithoutPrimaryKey set to true #10181 packages/cubejs-schema-compiler/test/integration/postgres/issue-10181.test.ts: an ungrouped query on a cube without a primary key, with allowUngroupedWithoutPrimaryKey: true. The SQL builds, but PreAggregations.transformQueryToCanUseForm calls primaryKeyNames() and throws One or more Primary key is required for 'no_pk' cube. The server runs that path on every /v1/load request. Fails on both planners.
• Zero-option "case" dimensions compile but don't work #6158 packages/cubejs-schema-compiler/test/integration/postgres/issue-6158.test.ts: a case dimension with when: [] compiles but renders CASE ELSE 'Unknown' END, which Postgres rejects. Fails on both planners.
• Multi-stage time_shift measures produce empty results when combined with dimensions from joined cubes #11240 packages/cubejs-schema-compiler/test/integration/postgres/issue-11240.test.ts: a time_shift measure queried with a dimension from a joined cube. The shifted CTE joins dim_date on the unshifted date, and the final join aligns rows on day_name too, so each day comes back as two rows: one with total and one with total_prior_day. Tesseract only; skipped on the legacy planner, which doesn't support multi-stage measures.
• Min measure for a Date field throws "Float cannot represent non numeric value" #5652 packages/cubejs-api-gateway/test/graphql.test.ts (time measures): a time measure is typed as TimeDimension in the GraphQL schema. The result mapping only adds the value path for time dimensions, so a time measure always resolves to null. Because of this, the workaround for the original report (a type: time measure with MIN(...) instead of type: min on a date) doesn't work over GraphQL either.
• bug(cubesql): Bind message still processed after Parse failure causes protocol desync with npgsql/.NET clients #10962 rust/cubesql/cubesql/src/compile/test/test_extended_query_errors.rs: sends Parse (a failing query), Bind, Describe, Execute and Sync over the wire. Cube answers BindComplete, ErrorResponse, ReadyForQuery; Postgres answers only ErrorResponse, ReadyForQuery. The Bind handler in shim.rs still calls self.bind() when tracked_error is set.

Test runs:

• TEST_LOCAL=1 jest dist/test/integration/postgres/issue-* with CUBEJS_TESSERACT_SQL_PLANNER=true and false: all fail as described above.
• jest dist/test/graphql.test.js: only the new time measures test fails.
• cargo test -p cubesql --lib test_bind_after_failed_parse_is_skipped: fails with 2EZ instead of EZ.

---

Generated by Claude Code

[claude]

Claude finished @paveltiunov's task in 1m 35s —— View job

---

Changes requested: 1 high, 1 medium, 2 low. All six new tests fail on master and run in CI, so merging this as-is turns CI red.

Review details

• Read diff and changed files
• Check tests against referenced issues and CI wiring
• Review threads
• Post findings

Findings

Severity	Where	Issue
High	packages/cubejs-api-gateway/test/graphql.test.ts:478 (also issue-6158, issue-8561, issue-10181, issue-11240)	The tests fail on master and CI runs them: the api-gateway unit step, and integration:postgres on both planners. Use test.failing (the repo already does this in cubejs-testing/test/smoke-view-join.test.ts:75).
Medium	rust/cubesql/cubesql/src/compile/test/test_extended_query_errors.rs:109	cargo insta test --workspace (rust-cubesql.yml:113) runs this test and it fails. Add #[ignore = "…/issues/10962"]. #[should_panic] is a worse fit because it would also pass on a timeout or a handshake failure.
Low	issue-11240.test.ts:1-5 and the other file headers	Header comments run past 3 lines and re-narrate the issue. Keep the load-bearing sentence.
Low	issue-10181.test.ts:34	The as any is unnecessary, since allowUngroupedWithoutPrimaryKey is in BaseQueryOptions.

Notes

• The test logic matches each linked issue. The expected values check out: in Multi-stage time_shift measures produce empty results when combined with dimensions from joined cubes #11240, the prior-day totals 9 and 10 come from extract(day). In Min measure for a Date field throws "Float cannot represent non numeric value" #5652, parseDates appends Z through moment.tz(...).toISOString(). Skipping Multi-stage time_shift measures produce empty results when combined with dimensions from joined cubes #11240 on the legacy planner via getEnv('nativeSqlPlanner') follows the existing precedent.
• With test.failing, any throw counts as a pass, so a typo in a fixture would also look like "still broken". Pinning the expected error message would prevent that.
• I didn't run the tests. I didn't set up Postgres or build in this run, so the failure modes are as reported in the PR description.
• Docs: none needed, since the PR only adds tests.
• Review threads: the listing returned 0 of my own threads (no prior threads), so none were resolved or re-affirmed.

[claude]

High: merging this turns CI red. Every new test fails on master, and CI runs all of them:

• this one: the api-gateway unit step (push.yml:139, lerna run unit)
• issue-{6158,8561,10181,11240}.test.ts: the integration job, postgres matrix entry (.github/actions/integration/postgres.sh → integration:postgres), on both use_tesseract_sql_planner values
• test_bind_after_failed_parse_is_skipped: cargo insta test --workspace in rust-cubesql.yml:113

The repo already has a pattern for this: packages/cubejs-testing/test/smoke-view-join.test.ts:75 uses test.failing(...). Jest 29 supports it here. It also signals when a fix lands, because the test then "unexpectedly passes" and has to be flipped back to test. In Rust, see the comment on test_extended_query_errors.rs:109.

Suggested change

	test('should return the value of a time measure', async () => {
	test.failing('should return the value of a time measure', async () => {

The same change applies to the it(...) in issue-6158, issue-8561 and issue-10181, and to testIfNative in issue-11240 (getEnv('nativeSqlPlanner') ? it.failing : it.skip). One caveat: test.failing passes on any throw, so a broken DB connection or a typo in the fixture also counts as "still failing". Pinning the expected error, as smoke-view-join.test.ts:119 does, keeps the test honest.

[claude]

This test fails on master (2EZ vs EZ), and rust-cubesql.yml:113 runs cargo insta test --workspace, so the cubesql job goes red once this merges. Mark it ignored until #10962 is fixed. #[should_panic] is a worse fit: it would also pass on the 10s read timeout or on any expect in the handshake.

Suggested change

	#[tokio::test]
	#[tokio::test]
	#[ignore = "https://github.com/cube-js/cube/issues/10962"]

[claude]

Low: this 5-line header re-narrates the issue. The same goes for the headers in issue-8561.test.ts:1-5, issue-10181.test.ts:1-5, test_extended_query_errors.rs:1-5 and the 3–4 line blocks in issue-6158.test.ts and graphql.test.ts:422. The issue link and the test name already carry the symptom. Keep the load-bearing sentence, i.e. what a later reader would get wrong without it. Here that is:

Suggested change

	// Reproduces https://github.com/cube-js/cube/issues/11240: a `time_shift` measure
	// queried with a dimension of a joined cube. The shifted CTE joins that cube on the
	// unshifted fact date, so its `day_name` belongs to the previous day, and the final
	// join uses `day_name` to align the shifted and current rows. Each day comes back as
	// two rows, one with `total` and one with `total_prior_day`.
	// https://github.com/cube-js/cube/issues/11240: the shifted CTE joins dim_date on the
	// unshifted date, so aligning on day_name splits each day into two rows.

[claude]

Low: the as any isn't needed. allowUngroupedWithoutPrimaryKey is declared in BaseQueryOptions (src/adapter/BaseQuery.ts:128), and order is any[]. The cast only hides type errors if the option is renamed.

Suggested change

	} as any);
	});