[ykisana]

What changes were proposed in this pull request?

IntervalUtils.parseSecondNano took the sign of the fractional part from the parsed whole seconds:

val seconds = parseSeconds(secondsStr)
Math.addExact(seconds, parseNanos(nanosStr, seconds < 0))

When casting a string to INTERVAL SECOND, a negative input such as -0.5 reaches this method as "-0.5". Its whole seconds parse to 0, so seconds < 0 is false and the fraction is added as a positive value. This PR takes the sign from the seconds string instead (secondsStr.startsWith("-")). For any input whose whole seconds are non-zero, the result is unchanged.

The other callers of parseSecondNano (toDTInterval for ... TO SECOND types) pass unsigned seconds and apply the sign outside, so they are not affected.

Why are the changes needed?

Every value between -1 and 0 seconds loses its sign when cast from a string:

SELECT CAST('-0.5' AS INTERVAL SECOND);
-- before: INTERVAL '00.5' SECOND
-- after:  INTERVAL '-00.5' SECOND

Spark formats such an interval as INTERVAL '-00.5' SECOND, and the JSON and CSV readers parse INTERVAL SECOND with the same code. As a result, writing -0.5 seconds to JSON or CSV and reading it back returns +0.5 seconds.

Does this PR introduce any user-facing change?

Yes, it fixes a bug. Casting a string to INTERVAL SECOND now keeps the sign of values between -1 and 0 seconds, as the INTERVAL '-0.5' SECOND literal already does. JSON and CSV round trips of such values are fixed as well. The bug reproduces on 3.5.8, 4.0.0, 4.1.1 and 4.2.0.

How was this patch tested?

Added SPARK-60484: Cast string to INTERVAL SECOND keeps the sign of values in (-1, 0) to CastSuiteBase. It covers:

• plain strings (-0.5, -00.5, -0.000001, -0.999999999, and 0.5 as a positive control);
• the literal forms (INTERVAL '-0.5' SECOND, INTERVAL -'0.5' SECOND, INTERVAL -'-0.5' SECOND);
• an interval -> string -> interval round trip.

Without the fix, the new test fails in all three suites with cast(-0.5 as interval second), actual: 500000, expected: -500000.

build/sbt 'catalyst/testOnly *CastWithAnsiOffSuite *CastWithAnsiOnSuite *TryCastSuite *IntervalUtilsSuite'

All 458 tests in these four suites pass with the fix. Scalastyle (build/sbt catalyst/scalastyle catalyst/Test/scalastyle) reports 0 errors.

Was this patch authored or co-authored using generative AI tooling?

Generated-by: Claude Code (Claude Opus 5.5)
Fix, tests, PR description, and comments by Claude.
All code manually reviewed by @ykisana.