StringLiteral to IntegerType/LongType silently truncates fractional strings, changing predicate meaning #4382

Apache Iceberg version

0.12.0 (latest release)

Please describe the bug 🐞

literal("2.6").to(IntegerType()) returns LongLiteral(2) instead of raising, while DecimalLiteral(Decimal("2.6")).to(IntegerType()) raises ValueError. A bound predicate such as x < "2.6" on an int column therefore becomes x < 2 and wrongly drops rows with x = 2 (verified with expression_evaluator); the same bound literal feeds partition pruning, so files can be skipped too.

The sibling DecimalLiteral.to(IntegerType/LongType) already rejects fractional values via the existing _to_integral helper, with a test spelling out why: rounding would change the predicate (e.g. x > 2.6 becoming x > 3 drops x = 3). The StringLiteral path was missed. The reference Java implementation has no INTEGER/LONG case in StringLiteral.to() and refuses the conversion.

Suggested fix: use the existing _to_integral helper in both StringLiteral.to(IntegerType) and StringLiteral.to(LongType), and update test_string_literal to expect ValueError for fractional strings.

Reproduction (pyiceberg 0.12.0, main @ fce8185):

• literal("2.6").to(IntegerType()) returns LongLiteral(2); same for LongType
• literal("-2.5").to(IntegerType()) returns LongLiteral(-2); literal("0.1").to(LongType()) returns LongLiteral(0)
• LessThan("x", "2.6") bound to an int column becomes x < 2, and expression_evaluator returns False for a row with x = 2, although x < 2.6 is True
• Integral strings still convert exactly: literal("2.0").to(IntegerType()) returns 2, literal("1e3").to(IntegerType()) returns 1000

Willingness to contribute

• I can contribute a fix for this bug independently
• I would be willing to contribute a fix for this bug with guidance from the Iceberg community
• I cannot contribute a fix for this bug at this time