feat(cubesql): Support FULL and RIGHT joins with non-push-to-Cube SQL push down

[MazterQyou]

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

Description of Changes Made

This PR fixes a regression introduced by grouped joins with SQL push down (#10234): a FULL OUTER JOIN of two grouped subqueries (e.g. period-over-period cost comparison) now fails with Unsupported join type for join subquery: Full. The rewrite already produced the correct plan; only SQL generation and the join-type templates were limited to Inner/Left.

This adds FULL and RIGHT support for the non-push-to-Cube path (both sides as standalone subqueries, where outer-join semantics map cleanly to SQL). The push-to-Cube path — where the join is folded inside the Cube query with
its own grouping/measures — explicitly keeps refusing FULL/RIGHT, since NULL-extended outer rows there are not validated.

Push down is template-gated: FULL/RIGHT only push down when the target data source declares the join_types/* template, otherwise the join falls back to DataFusion.

The PR also makes Tesseract use separate templates for joins.

[claude]

Claude finished @MazterQyou's task in 2m 16s —— View job

---

Code Review

Verdict: Looks good overall — clean, well-scoped fix. 0 high, 0 medium, 3 low (all suggestions / nits posted inline).

Full review (expand)

Summary

The fix is well-scoped: the rewrite already produced a correct plan for FULL/RIGHT joins of two grouped subqueries; the PR just wires SQL generation and join-type templates through. Separation between the push-to-Cube path (refuses FULL/RIGHT) and the non-push-to-Cube path (now supports them) is clearly commented and consistently enforced.

What's good

• Template-gating via is_subquery_join_type_supported (join.rs:996) is the right abstraction — it composes the join-type rule with can_rewrite_template, so non-supporting drivers fall back to DataFusion rather than emitting unsupported SQL.
• push_to_cube parameter correctly differentiates the two paths: hardcoded true for transform_ungrouped_join_grouped (always refuses FULL/RIGHT), left_push_to_cube.0 for transform_grouped_join_grouped (refuses FULL/RIGHT only when folding into Cube query).
• continue vs return false in the grouped-grouped variant (join.rs:1320) is the right choice — lets the non-push variant of the eclass still match.
• Driver-specific opt-outs (CubeStoreQuery.ts, MysqlQuery.ts) are well-commented with the underlying reason.
• Comments at wrapper.rs:3462 and :3600 preserve the invariant for future maintainers.

Findings (all low)

1. BaseQuery.js — making right/full the default exposes all non-overriding drivers; verify each one. Snowflake's explicit templates.join_types.full = 'FULL' (SnowflakeQuery.ts:122) is now redundant. Inline comment posted.
2. Test coverage — only FULL is exercised; RIGHT is not, and there's no negative test for the template-gating fallback or for the push-to-Cube path still rejecting FULL/RIGHT. Inline comment posted.
3. Minor redundancy in transform_ungrouped_join_grouped — the data-source lookup runs even though push_to_cube=true makes Right/Full unreachable. Functionally correct but slightly misleading. Inline comment posted.

Security / performance

• No new user-input surfaces; the templates are static strings. No injection vector introduced.
• No allocation/hot-path concerns: the rewrite-rule helper does one extra MetaContext clone per rule construction (cheap, one-time at setup) and one template lookup per match attempt (already done for other gates).

Checklist

• Reviewed Rust changes (join.rs, wrapper.rs, service.rs, tests)
• Reviewed JS/TS adapter changes (BaseQuery.js, CubeStoreQuery.ts, MysqlQuery.ts)
• Posted 3 inline findings
• Verified push-to-Cube path still rejects FULL/RIGHT (both code paths in generate_sql_for_push_to_cube)

• [Branch](https://github.com/cube-js/cube/tree/cubesql/grouped-join-grouped-right-full)

[codecov]

Codecov Report

❌ Patch coverage is 94.41624% with 11 lines in your changes missing coverage. Please review.
✅ Project coverage is 79.10%. Comparing base (c70e7ec) to head (74331ab).
⚠️ Report is 1 commits behind head on master.

Files with missing lines	Patch %	Lines
.../cubesql/src/compile/rewrite/rules/wrapper/join.rs	90.90%	4 Missing ⚠️
...t/cubesql/cubesql/src/compile/test/test_wrapper.rs	97.79%	3 Missing ⚠️
...bejs-schema-compiler/src/adapter/CubeStoreQuery.ts	0.00%	2 Missing ⚠️
...s/cubejs-schema-compiler/src/adapter/MysqlQuery.ts	0.00%	1 Missing ⚠️
...bejs-schema-compiler/src/adapter/SnowflakeQuery.ts	0.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master   #11008      +/-   ##
==========================================
+ Coverage   79.06%   79.10%   +0.04%     
==========================================
  Files         472      472              
  Lines       95983    96172     +189     
  Branches     3524     3524              
==========================================
+ Hits        75888    76077     +189     
  Misses      19580    19580              
  Partials      515      515              

Flag	Coverage Δ
cube-backend	58.50% <0.00%> (-0.02%)	⬇️
cubesql	83.61% <96.37%> (+0.04%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[claude]

Claude finished @MazterQyou's task in 1m 6s —— View job

---

Code Review

Re-checked after PR synchronize: no new commits since the prior review (still 37109d0). All 3 prior inline findings remain unresolved and still applicable to the current diff — re-affirming them rather than re-posting.

Verdict: Looks good overall — clean, well-scoped fix. 0 high, 0 medium, 3 low (suggestions only; all already inline on the PR).

Prior findings (still apply)

1. rust/cubesql/cubesql/src/compile/rewrite/rules/wrapper/join.rs:1068 — push_to_cube=true in transform_ungrouped_join_grouped makes the template lookup unreachable for Right/Full; consider a clarifying comment or inline match guard.
2. rust/cubesql/cubesql/src/compile/test/test_wrapper.rs:1249 — coverage gap: RIGHT JOIN isn't asserted, no negative test for template-gating fallback, no test that push-to-Cube path still rejects FULL/RIGHT.
3. packages/cubejs-schema-compiler/src/adapter/BaseQuery.js:4608 — making FULL/RIGHT base-default exposes every non-overriding driver; double-check exotic adapters (Druid, Pinot, Elasticsearch SQL, Materialize, MongoDB). Snowflake's templates.join_types.full = 'FULL' is now redundant.

No new high/medium issues introduced. Security and performance unchanged from prior pass.

• [Branch](https://github.com/cube-js/cube/tree/cubesql/grouped-join-grouped-right-full)

[claude]

Claude finished @MazterQyou's task in 1m 26s —— View job

---

Code Review

Re-reviewed after 74331ab fix(tesseract): Decouple join types from SQL push down.

Verdict: Looks good. The new commit cleanly addresses my prior Snowflake redundancy concern by splitting the namespaces. 0 high, 0 medium, 0 new low. All 3 prior threads are already resolved.

Details

What changed in 74331ab

Tesseract now uses a distinct template namespace (tesseract.join_types_inner/left/full) instead of sharing join_types/* with the SQL API push down path. Concretely:

• BaseQuery.js:4634-4637 — adds tesseract.join_types_inner = 'INNER' and tesseract.join_types_left = 'LEFT' as base defaults; FULL is intentionally not a base default (opt-in per dialect).
• SnowflakeQuery.ts:122 — what was templates.join_types.full = 'FULL' is now templates.tesseract.join_types_full = 'FULL', restoring Snowflake's Tesseract FULL support under the new namespace.
• plan.rs:447-452, 500-502 — PlanSqlTemplates::render_join and supports_full_join now read from tesseract/join_types_* instead of join_types/*.
• mock_sql_templates_render.rs:464-470 — test fixture mirrors the new namespace.

Why this is the right call

The two paths had different semantics piled into one set of templates:

• SQL API push down (join_types/*) — driven by the BaseQuery rewrite rules added in this PR, gated by is_subquery_join_type_supported. Base defaults all 4 (inner/left/right/full); MySQL/CubeStore opt out.
• Tesseract (tesseract.join_types_*) — used by the native planner via PlanSqlTemplates::render_join. Base defaults inner/left only; FULL is opt-in (currently Snowflake only).

Before this commit, Snowflake's join_types.full = 'FULL' was simultaneously (a) the SQL API push-down opt-in and (b) the Tesseract FULL gate, which is what made my earlier "now redundant" call slightly wrong — it wasn't actually redundant for the Tesseract path. The split makes each path's opt-in surface explicit.

Nits (non-blocking)

• The fixture comment at mock_sql_templates_render.rs:461 still says "based on BaseQuery.js:4424-4427" but the referenced lines have shifted to around 4665 in current BaseQuery.js. Pre-existing, not introduced here.
• supports_full_join is now the single source of truth for Tesseract FULL support; worth a quick mental note that a dialect could in principle opt into FULL on one path but not the other (e.g. set tesseract.join_types_full without join_types.full). Not a real concern today, just an architectural consequence of the split.

Prior findings — re-checked

#	Path:line	Status
1	join.rs:1068 (template lookup unreachable on push-to-Cube path)	thread resolved — non-blocking nit
2	test_wrapper.rs:1249 (RIGHT/negative/push-to-Cube test gaps)	thread resolved
3	BaseQuery.js:4669 (base-default FULL/RIGHT exposure)	thread resolved; partially superseded by namespace split for the Tesseract path

• [Branch](https://github.com/cube-js/cube/tree/cubesql/grouped-join-grouped-right-full)