Speed up scans with a roaring row selection

[joseph-isaacs]

Summary

Scans that take their rows from an external index pass those rows as a Selection. With IncludeRoaring, every split did the following:

• built a one-range treemap,
• intersected it with a clone of the whole selection,
• collected the result into a Vec<usize>,
• and only then built the mask.

Sparse roaring selections also never got the exact split ranges that IncludeByIndex selections get, so they were planned over natural splits.

Changes

• Selection::row_mask for IncludeRoaring/ExcludeRoaring counts the selected rows in the split with range_cardinality. It returns an all-false or all-true mask directly when it can. Otherwise it seeks the treemap iterator to the split and sets the bits in place.
• attempt_split_ranges plans exact sparse ranges for IncludeRoaring the same way it does for IncludeByIndex. The range-building loop is shared between the two.

New tests check that roaring and index selections plan the same split ranges, and that dense roaring selections fall back to natural splits.

On a 14M-row file with 840K rows selected, a count(*) over the selection drops from 29.5 ms to 20.6 ms of CPU. With 58K rows selected it drops from 7.0 ms to 5.9 ms.

🤖 Generated with Claude Code

https://claude.ai/code/session_01JRTKSkM9xUFT8WHRyS97aR

---

Generated by Claude Code

[codspeed]

Merging this PR will regress 2 benchmarks

⚠️ 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.

⚠️ 19 benchmarks spent significant time in system calls

System calls cannot be consistently instrumented, so they are not included in the measure, which understates the real cost. Please switch to the Walltime instrument to accurately measure system calls.

Measurement and system calls

• density_sweep_dense_runs[0.001]
• chunked_into_canonical[(20, 100, 16, 4)]
• chunked_dict_primitive_canonical_into[f64, (1000, 1000, 10)]
• pcodec_compress_f64
• density_sweep_dense_runs[0.9]
• bench_dict_mask[(0.9, 0.9)]
• filter_run_end[len=4096_run=16_density=0.1]
• multipolygon_mixed_validity
• random_i128[0.01]
• take_varbin
• …and 9 more

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 1 improved benchmark
❌ 2 regressed benchmarks
✅ 2180 untouched benchmarks
⏩ 359 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	WallTime	bitpack_blocked_compress_avx2	6.7 µs	7.6 µs	-11.69%
❌	WallTime	scalar_subtract_neon	10.9 µs	12.1 µs	-10.29%
⚡ ⚠️	Simulation	density_sweep_dense_runs[0.001]	47.8 µs	29.7 µs	+60.73%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing ji/roaring-selection-scan (1939f12) with develop (731a231)}

Footnotes

1. 359 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. ↩