fix(layout): apply struct nulls before evaluating field expressions

[robert3005]

Summary

A nullable struct's fields store values even for rows where the struct itself is null. When reading a file, StructReader evaluated field expressions without taking that into account:

• Filters were pushed straight into the field readers, which know nothing about the parent's validity. So x.value >= 0 matched rows where x is null. This is the bug behind the failing struct tests in fix(python): translate Polars struct field access #10194: a file scan returned 2 rows where Polars, and Vortex's own in-memory evaluation, return 1.
• Projections applied the struct's nulls to the result, not the input. That is right for expressions that turn a null input into a null output, but wrong for others: is_null(x.value) gave null instead of true on rows where x is null.
• Pruning could use field statistics, which include the hidden values, to rule out rows that should match.

Changes

All changes are in vortex-layout/src/layouts/struct_/reader.rs, and only affect structs that are nullable (have a validity child):

• Filter and projection now read each field, apply the struct's validity, and only then evaluate the field-scoped expression in memory. This handles single-field and multi-field expressions alike.
• A struct-valued (pack/merge) projection still masks each output field with the struct's validity, as before. That is how a nested nullable struct stays nullable when projected through its parent; vortex-file's write_nullable_nested_struct covers it.
• Statistics pruning is skipped for nullable structs.

Non-nullable structs, including every file's root struct, keep the existing pushdown path unchanged. The cost is that predicates on fields of nullable nested structs are no longer pushed into the field readers and their statistics are no longer used for pruning. They are evaluated in memory after reading the field. Pushing down predicates that are provably false on null input would be a possible follow-up.

Tests

Added rstest cases to struct_::reader::tests. The fixture is a nullable struct whose null row 0 still stores a = 7:

• Filters a > 1, is_null(a), and the two-field a > 5 OR b > 4 respect the parent's nulls.
• The projection is_null(a) returns [true, false, false].

All four fail on develop and pass with this change.

Checks run

• cargo clippy -p vortex-layout --all-targets --all-features -- -D warnings: clean.
• cargo +nightly-2026-09-10 fmt -p vortex-layout.
• cargo nextest run -p vortex-layout -p vortex-file -p vortex-datafusion --all-features: 813 passed. -p vortex-scan: passed too.
• Python suite with fix(python): translate Polars struct field access #10194 merged on top and the extension rebuilt: 416 passed, including fix(python): translate Polars struct field access #10194's 5 struct null-parent scan tests that fail on develop.

🤖 Generated with Claude Code

https://claude.ai/code/session_01E92wUoS3LaqWVJmmS7vKVq

---

Generated by Claude Code

[codspeed]

Merging this PR will improve performance by 12.51%

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

⚡ 1 improved benchmark
✅ 2174 untouched benchmarks
⏩ 534 skipped benchmarks¹

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	WallTime	bitpack_blocked_compress_avx2	7.6 µs	6.8 µs	+12.51%

Tip

Curious why performance improved? Comment @codspeedbot explain why performance improved on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing rk/struct-filter-parent-validity (2133e00) with develop (c93ffa3)}

Footnotes

1. 534 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[robert3005]

will wait for @myrrc pr