fix(expr): reject non-string StartsWith operands instead of pruning the file

[jackylee-ch]

Which issue does this PR close?

None — filed directly.

What changes are included in this PR?

ExpressionEvaluator::starts_with matched only (String, String) and fell through to
Ok(false). In this visitor false means "this data file cannot match", so a StartsWith
predicate on a non-string partition column silently pruned files that do contain matching rows.

It is reachable: Reference::new("a").starts_with(Datum::binary(b"ab")) binds without error, and
with an identity-partitioned binary column a file whose partition value is X'616263' was
dropped from the scan.

Now it returns the same error the other four visitors already return for this input —
inclusive_metrics_evaluator.rs:308, row_group_metrics_evaluator.rs:347,
page_index_evaluator.rs:586, manifest_evaluator.rs:331.

Are these changes tested?

Yes — test_expr_starts_with_non_string_errors covers StartsWith and NotStartsWith on a binary
partition; it returns false on the parent commit. cargo test --release -p iceberg --lib expr::
→ 287 passed.

AI Disclosure

Written with AI assistance (Claude Code); I reviewed the change and ran the tests above.
Worth a reviewer's attention: this turns NOT STARTS WITH on a non-string column from "no pruning"
into an error. Java rejects the predicate earlier, when binding, so validating in
BinaryExpression::bind would be a reasonable follow-up.

[Copilot]

🟢 Approval recommended

The change corrects a pruning bug with consistent error behavior and includes a focused regression test covering the reported failure mode.

Pull request overview

Fixes a correctness issue in partition pruning where StartsWith on a non-string partition column could silently evaluate to false, causing data files that do contain matching rows to be incorrectly pruned during scans.

Changes:

• Update ExpressionEvaluatorVisitor::starts_with to return an ErrorKind::Unexpected error when either operand is non-string, instead of falling back to Ok(false).
• Add a regression test covering both StartsWith and NotStartsWith against an identity-partitioned binary column.

File summaries

File	Description
crates/iceberg/src/expr/visitors/expression_evaluator.rs	Rejects non-string StartsWith operands with an error and adds regression coverage to prevent silent mis-pruning.

Review details

• Files reviewed: 1/1 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.