Add support ANSI cast float/double to timestamp

[Harthi7]

Implemented ANSI support for cast float/double to timestamp.

Changes made:

• Enabled ANSI routing for REAL/DOUBLE -> TIMESTAMP in SparkCastExpr.
• Updated castDoubleToTimestamp so ANSI CAST throws on non-finite and overflow inputs, while non-ANSI behavior remains unchanged.
• Added and split tests for ANSI ON and ANSI OFF behavior, including TRY_CAST coverage.

Validation:

• Targeted float/double to timestamp tests passed under ANSI ON and ANSI OFF.
• Full SparkCastExpr suite passed locally: 46/46 tests.

[netlify]

✅ Deploy Preview for meta-velox canceled.

Name	Link
🔨 Latest commit	9b97bc3
🔍 Latest deploy log	https://app.netlify.com/projects/meta-velox/deploys/69e22df3ebc3d70008bf4fd8

[github-actions]

Build Impact Analysis

Selective Build Targets (building these covers all 21 affected)

cmake --build _build/release --target spark_expression_fuzzer_test velox_expression_runner_test velox_expression_runner_unit_test velox_format_datetime_benchmark velox_functions_spark_aggregates_test velox_functions_spark_test velox_spark_function_registry_test velox_spark_query_runner_test velox_sparksql_benchmarks_cast velox_sparksql_benchmarks_compare velox_sparksql_benchmarks_from_json velox_sparksql_benchmarks_get_funcs velox_sparksql_benchmarks_hash velox_sparksql_benchmarks_in velox_sparksql_benchmarks_simd_compare velox_sparksql_benchmarks_split velox_sparksql_coverage

Total affected: 21/565 targets

Affected targets (21)

Directly changed (2)

Target	Changed Files
velox_functions_spark_specialforms	SparkCastExpr.cpp, SparkCastHooks.cpp
velox_functions_spark_test	SparkCastExprTest.cpp

Transitively affected (19)

• spark_expression_fuzzer_test
• velox_expression_runner
• velox_expression_runner_test
• velox_expression_runner_unit_test
• velox_format_datetime_benchmark
• velox_functions_spark
• velox_functions_spark_aggregates_test
• velox_functions_spark_impl
• velox_spark_function_registry_test
• velox_spark_query_runner_test
• velox_sparksql_benchmarks_cast
• velox_sparksql_benchmarks_compare
• velox_sparksql_benchmarks_from_json
• velox_sparksql_benchmarks_get_funcs
• velox_sparksql_benchmarks_hash
• velox_sparksql_benchmarks_in
• velox_sparksql_benchmarks_simd_compare
• velox_sparksql_benchmarks_split
• velox_sparksql_coverage

---

Fast path • Graph from main@3a86cb261b0b0dfdfef527e679b93a7f1d87999d

[philo-he]

Looks good overall. Some comments. Please check if they make sense. Thanks.

[philo-he]

Suggest creating a static constexpr variable to hold static_cast(std::numeric_limits<int64_t>::max())

[Harthi7]

Thanks. The first two suggestions make sense. I’ll factor out the int64 bounds into local constexpr variables to simplify the overflow check.

[philo-he]

ditto

[philo-he]

Could you confirm Spark's error message? It would be better to make them consistent.

[Harthi7]

Confirmed. Spark’s ANSI float/double -> timestamp path uses doubleToTimestampAnsi(...), which throws invalidInputInCastToDatetimeError(...) for NaN/Infinity, and finite overflow goes through exact numeric conversion. So Spark’s behavior maps to the generic CAST_INVALID_INPUT and CAST_OVERFLOW error styles rather than a timestamp specific custom message.

• Cast routing:
  https://github.com/apache/spark/blob/e221b56be7b6d9e48e107fc4d1cf0c15f02700f8/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Cast.scala

• ANSI implementation:
  https://apache.googlesource.com/spark/%2B/master/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/util/DateTimeUtils.scala?autodive=0%2F%2F%2F

In this Velox hook we only have the converted double value, not the full source expression/type context needed to reproduce Spark’s fully rendered message exactly. So, I’m keeping the logic as is and updating the wording to be closer to Spark’s generic error style.

[rui-mo]

Thanks!

[rui-mo]

Since castNumberToTimestamp is also used by castIntToTimestamp, could we extend it with a field (e.g., allowOverflow) so that a single implementation supports both floating-point and integral types on ANSI ON & OFF?