Recover ANY-typed column schema on the analytics route (fixes eval max/min)

[ahkcs]

Description

Brings CalcitePPLEvalMaxMinFunctionIT to parity on the analytics-engine route (-Dtests.analytics.parquet_indices=true) by fixing a real schema-type defect on the analytics path, then skipping the two genuinely-unsupported residual cases. The IT passes on the v2/Calcite route; the engine fix is shared, the skips are analytics-route-only.

The defect (commit 1)

PPL eval max(...) / min(...) lower to the SCALAR_MAX / SCALAR_MIN UDFs, which declare their return type as ANY (they accept mixed numeric/string operands, with OpenSearch's "strings sort above numbers" semantics). ANY maps to UNDEFINED → the schema reports the result column as "undefined".

The v2/Calcite path already handles this: OpenSearchExecutionEngine.buildResultSet sees ANY and recovers the concrete type from the first row's runtime ExprValue. The analytics path (AnalyticsExecutionEngine.buildSchema) built the schema purely from the planned RelDataType, with no such recovery — so the column came back undefined even though the values were correct.

Fix: mirror the Calcite-path behavior in AnalyticsExecutionEngine.buildSchema — when a column's converted type is UNDEFINED and a result row exists, take the concrete type from the first row's value. The rows are already converted before the schema is built, so there's no extra work. Behavior on the v2 path is unchanged. This is generic — it fixes any ANY-typed UDF column on the analytics route, not just max/min. Covered by two new AnalyticsExecutionEngineTest cases (recovery from first row; stays UNDEFINED when empty, matching the Calcite fallback).

Residual divergences (commit 2)

With the fix, 5 of 8 tests pass on the analytics route. The remaining 3 are genuine route behaviors, skipped via assumeNotAnalytics(...) + excludeTestsMatching:

• EVAL_MAX_MIN_MIXED_TYPES (2 tests) — mixed numeric+string operands (max(14, age, 'Fred', holdersName)) throw SqlValidatorException: Cannot infer return type for GREATEST. The UDF is converted to a DataFusion GREATEST/LEAST during Substrait conversion (in OpenSearch core), which can't unify heterogeneous operand types. Out of scope for this repo.
• EVAL_MAX_MIN_INT_WIDENING (1 test) — max(1, 3, age) reports bigint on the route (DataFusion widens integer results to Int64) where v2/Calcite reports int. Values match; only the integer width differs.

Results

Run via :integ-test:integTestRemote against an analytics :run cluster.

Test class	Analytics route — before	Analytics route — after	v2/Calcite route
CalcitePPLEvalMaxMinFunctionIT	0/8 pass, 8 fail	5/8 pass, 3 excluded, 0 fail	8/8 pass

The 8 failures before split into 6 undefined-schema (now fixed by the engine change) and 2 mixed-type 500s. After the fix, 5 pass; the 3 skips are 2 mixed-type + 1 int-widening. The v2 route runs all 8 (the engine fix doesn't change v2 types; assumeNotAnalytics is a no-op off-route; gradle excludes apply only to integTestRemote).

Check List

• New functionality includes testing.
• New functionality has been documented.
• Commits are signed per the DCO using --signoff.

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.
For more information on following Developer Certificate of Origin and signing off your commits, please check here.