Eventual-Inc/Daft - CodSpeed

Daft

Blog Docs Changelog

Performance History

Latest Results

document Common Crawl on HF Buckets in dedicated docs page

lhoestq:document-s3-alternative-for-cc

11 hours ago

Merge branch 'main' into fix/plan-cache-fingerprint-collision

TechyMT:fix/plan-cache-fingerprint-collision

11 hours ago

fix: include null placement in range clustering spec for sort shuffle elision A sort can only skip its shuffle when nulls already live in the partition its nulls_first expects. RangeClusteringConfig now carries nulls_first, sorts record and compare it, and ClusteringKeys.range() lets sources declare null placement (defaulting to sort semantics: nulls last ascending, first descending). Output assertions for non-default nulls_first through an actual shuffle are xfailed: the range shuffle routes rows via search_sorted, which has no nulls_first support, so a distributed sort with non-default nulls_first already places nulls incorrectly today (pre-existing, independent of this PR). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

euan/range-repartition-clustering-spec-1

12 hours ago

test: make aligned asof join PR test-only Remove the copied AsofJoinAlignedNode; _assume_sorted_and_aligned now routes to the regular shuffle-based AsofJoinNode (correct for any layout) with the scan-task-split guard kept in the translator. Rewrite the test suite around a parameterized scenario table (per-strategy expectations on shared data), dedupe mirrored tests, and add the coverage a shuffle-skipping implementation must pass: >=3-partition layouts with empty / group-sparse middle partitions, and a differential test against a pure-Python reference. The nearest differential cases are xfailed on a pre-existing local kernel bug (search_nearest one-sided offers, see test comment). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

euan/asof-join-aligned

12 hours ago

file-path identity checkpoint mode

chenghuichen:path-checkpoint

13 hours ago

fix(tests): skip regression tests on Ray runner, fix unused pytest import The interleaved-iteration tests call .to_arrow() on MicroPartition objects, which is NativeRunner-specific — on Ray runner iter_partitions() returns ray.ObjectRef and to_arrow_iter() would need ray.get(). The bug itself only exists in NativeRunner's plan cache, so skip both tests on Ray. Also fixes the F401 ruff error (pytest was imported but unused before the skipif marks were added).

TechyMT:fix/plan-cache-fingerprint-collision

15 hours ago

test: add regression tests for #7087 (tests only, no fix)

TechyMT:test-only/plan-cache-regression

15 hours ago

file-path identity checkpoint mode

chenghuichen:path-checkpoint

15 hours ago

Latest Branches

+11%

Document HF Buckets as s3 alternative for Common Crawl#7103

11 hours ago

4ae3600

lhoestq:document-s3-alternative-for-cc

0%

fix(native-runner): assign unique plan-cache fingerprint when caller omits one#7088

12 hours ago

662e3ab

TechyMT:fix/plan-cache-fingerprint-collision

0%

feat: add support for range partition hints in clustering spec #7050

12 hours ago

0ff71b0

euan/range-repartition-clustering-spec-1

© 2026 CodSpeed Technology

Home Terms Privacy Docs