Next skip

[samuelcolvin]

Support skipping vales with Jiter, this is specifically for https://github.com/apache/datafusion/issues/7845.

The performance seems good, benchmarks show skipping values is significantly faster than not.

Running the query from https://github.com/apache/datafusion/issues/7845#issuecomment-2061121332, gives:

-- datafusion with next_value
SELECT count(*) FROM records where json_contains(attributes, 'size');
  -- 6165586 in 17.2s

-- datafusion with next_skip
SELECT count(*) FROM records where json_contains(attributes, 'size');
  -- 6165586 in 14.3s

-- datafusion using serde-json allocating keys to a Vec<String> then checking for the key
SELECT count(*) FROM records where json_contains(attributes, 'size');
  -- 6165586 in 21.7s

-- datafusion return false - e.g. fastest case for a UDF
SELECT count(*) FROM records where json_contains(attributes, 'size');
  -- 6165586 in 11.8s

-- datafusion like
SELECT count(*) FROM records where attributes like '%"size":%';
  -- 6165586 in 14.5s

-- duckdb
SELECT count(*) FROM read_parquet('file.parquet') where list_contains(json_keys(attributes), 'size')
  -- 6165586 in 14.2s

[codspeed-hq[bot]]

CodSpeed Performance Report

Merging #84 will not alter performance

_{Comparing next_skip (bf9c6d1) with main (0ab4dd4)}

Summary

✅ 59 untouched benchmarks

🆕 14 new benchmarks

Benchmarks breakdown

	Benchmark	main	next_skip	Change
🆕	big_jiter_skip	N/A	107.8 ms	N/A
🆕	bigints_array_jiter_skip	N/A	500.4 µs	N/A
🆕	floats_array_jiter_skip	N/A	573.2 µs	N/A
🆕	massive_ints_array_jiter_skip	N/A	1.2 ms	N/A
🆕	medium_response_jiter_skip	N/A	73.4 µs	N/A
🆕	pass1_jiter_skip	N/A	53.4 µs	N/A
🆕	pass2_jiter_skip	N/A	5.4 µs	N/A
🆕	sentence_jiter_skip	N/A	6.5 µs	N/A
🆕	short_numbers_jiter_skip	N/A	332.7 µs	N/A
🆕	string_array_jiter_skip	N/A	37.1 µs	N/A
🆕	true_array_jiter_skip	N/A	22.8 µs	N/A
🆕	true_object_jiter_skip	N/A	56.2 µs	N/A
🆕	unicode_jiter_skip	N/A	6.7 µs	N/A
🆕	x100_jiter_skip	N/A	2.4 µs	N/A