[fast-ut] [reduced-it] [SkipRecovery] Narrow CSV timestamp inference exemptions [databricks]

[wjxiz1992]

JaCoCo sql-plugin line coverage: +0 lines (26.08 b14 shim-350 nightly; 56012/69176 -> 56012/69176)

Summary

• Recover CSV V2 timestamp inference coverage on Apache Spark by replacing the unconditional skip with a Databricks-only guard.
• Narrow the V1/V2 LTZ xfail from Spark 3.5+ to the four affected format/time tuples on exact Spark 3.5.0.
• Keep expected values, generated data, and CPU/GPU parity assertions unchanged.
• Retain the Databricks V2 guard because no authorized DBR runtime evidence was available.

Contributes to #9325

Validation

• Spark 3.4.1 V2 matrix: 160 passed, 966 deselected, 4 warnings in 81.05s.
• Spark 3.5.0 negative boundary with --runxfail: expected PARSE_DATETIME_BY_NEW_PARSER failure, 1 failed in 5.71s.
• Spark 3.5.0 final V1/V2 matrix: 312 passed, 806 deselected, 8 xfailed, 2 warnings in 159.55s.
• Exact nightly-anchor JaCoCo run: 312 passed, 39444 deselected, 8 xfailed, 14 warnings in 168.50s.
• Spark 3.5.5 V1/V2 matrix: 320 passed, 806 deselected, 2 warnings in 159.19s.
• Spark 3.5.5 complete csv_test.py: 1078 passed, 40 xfailed, 8 xpassed, 698 warnings in 281.07s.
• Spark 4.1.1 V1/V2 matrix: 320 passed, 806 deselected, 4 warnings in 156.34s.
• Spark 3.5.5 America/Los_Angeles exact spot check: 4 passed, 2 warnings in 8.69s.
• Final rebased-head Spark 3.5.5 exact GPU gate: 4 passed, 2 warnings in 9.05s.
• Python syntax and git diff checks passed.
• NT local review: 0 must-fix, 0 should-fix; one caching suggestion applied before final validation.

Documentation

• Updated for new or modified user-facing features or behaviors
• No user-facing change

Testing

• Added or modified tests to cover new code paths
• Covered by existing tests
  (Please provide the names of the existing tests in the PR description.)
• Not required

Performance

• Tests ran and results are added in the PR description
• Issue filed with a link in the PR description
• Not required

[greptile-apps]

Greptile Summary

This PR restores Apache Spark coverage for CSV V2 timestamp inference while retaining the Databricks exemption and narrowing Spark 3.5.0 expected failures to four LTZ format combinations.

• Replaces the unconditional V2 skip with a Databricks-only guard.
• Uses one shared V1/V2 parameter matrix with tuple-specific Spark 3.5.0 xfail marks.
• Preserves CPU/GPU equality, fallback, and plan-capture assertions.

Confidence Score: 5/5

The PR appears safe to merge with no actionable defects identified in the narrowed CSV timestamp-inference test exemptions.

The new parameter matrix consistently covers both V1 and V2 paths, limits expected failures to the validated Spark 3.5.0 LTZ tuples, and retains the intended Databricks-only V2 skip.

Important Files Changed

Filename	Overview
integration_tests/src/main/python/csv_test.py	Narrows timestamp-inference skips and xfails while preserving the existing generated data and GPU-versus-CPU verification paths.

_{Reviews (1): Last reviewed commit: "test: narrow CSV timestamp inference exe..." | Re-trigger Greptile}

[Copilot]

Pull request overview

Adjusts Python CSV integration-test markers to better match known Spark 3.5.0 timestamp-inference failures while restoring Apache Spark coverage for CSV V2 (and keeping Databricks guarded).

Changes:

• Replace the unconditional skip on the CSV V2 timestamp inference test with a Databricks-only skipif.
• Narrow the TIMESTAMP_LTZ xfail from “Spark 3.5+” to only Spark 3.5.0 and only the four known (date_format, ts_part) combinations tied to #9325.
• Consolidate the V1/V2 parametrization into a shared precomputed parameter list with conditional xfail marks.