Eventual-Inc/Daft - CodSpeed

Daft

Blog Docs Changelog

Performance History

Latest Results

Merge branch 'Eventual-Inc:main' into gvfs_local_issue

qingfeng-occ:gvfs_local_issue

53 minutes ago

ci(deps): fix CI failures from dependabot bump #6570 (#6596) Addresses the widespread CI failures in #6570 by selectively reverting deps that either break builds or violate the `exclude-newer = "7 days"` policy. Safe updates are kept: upper bound bumps for optional extras (clickhouse, deltalake, huggingface, iceberg, openai, sqlglot, turbopuffer, transformers, video, av), dev tooling (ruff, black, isort, maturin, pytest-cov, memray), data/docs deps (duckdb, datasets, mkdocs-jupyter, mkdocs-material, mkdocstrings-python, mkdocs-gen-files, mkdocs-literate-nav), and library bumps (huggingface-hub, openai, sqlalchemy, trino, databricks-sdk, pyiceberg, imagehash). Reverted for CI compat: - torch/torchvision (GPU build issues) - opencv-python (binary compat) - deltalake dev pin (major version on Windows) - s3fs/boto3/adlfs/gcsfs (cloud SDK compat) - vllm (API changes) - fsspec (upper bound tightened) - clickhouse-connect dev pin Reverted for `exclude-newer` compliance (published after the 7-day cutoff): - aiohttp, orjson, Pillow, pylance, google-genai, pymdown-extensions, mkdocs-redirects, griffe, mypy See #6570 --------- Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

main

2 hours ago

fix(io): support writing to local fs via Gravitino

qingfeng-occ:gvfs_local_issue

2 hours ago

fix(ai): handle transformers 5.2+ API changes in image embedder and classifier `get_image_features()` now returns `BaseModelOutputWithPooling` instead of a tensor in newer transformers. `Pipeline.framework` attribute may not be set on subclasses - use `getattr` with `"pt"` default. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

codex/fix-pr-6570-ci

3 hours ago

feat(distributed): enable two-stage pre-aggregation for approx_count_distinct The distributed aggregation pipeline special-cased ApproxCountDistinct to skip two-stage pre-aggregation because MergeSketch(HyperLogLog) collapsed sketches directly to a UInt64 count, making the intermediate sketch type unusable across stages. To fix this, we align with the Spark/Trino/DuckDB convention: - MergeSketch(HLL) now returns a merged FixedSizeBinary sketch (not a final count) - A new hll_cardinality scalar function extracts the UInt64 count from a sketch, used as the final projection in the ApproxCountDistinct decomposition - The ApproxCountDistinct bypass in the distributed AggregateNode is removed, letting the standard two-stage path handle it Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

codex/approx-count-distinct-hll-schema

3 hours ago

feat(distributed): enable two-stage pre-aggregation for approx_count_distinct Remove the special case that forced single-stage execution for ApproxCountDistinct. The two-stage split logic already decomposes this correctly into ApproxSketch (first stage) and MergeSketch (second stage), and HLL sketch merging works - the bypass was based on a stale assumption. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

codex/approx-count-distinct-hll-schema

4 hours ago

fix(deps): align unity extra deltalake pin with deltalake extra The `deltalake<1.3.0` constraint in the unity extra was copied from the main deltalake pin in #5785 and never updated when the main pin was bumped. The unity integration only passes URIs and IOConfig to Daft's standard deltalake reader/writer, so it's compatible with whatever version the deltalake extra supports. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

codex/fix-pr-6570-ci

5 hours ago

feat: support get function from catalog (#6524) ## Changes Made To enable the retrieval of function definitions from the metastore service, I propose the following first step: add a get_function interface to the Catalog, and introduce a call to Catalog.get_function within the function planning phase (plan_function) to resolve functions registered in the metastore. This design ensures that only the specific Catalog implementation needs to provide the underlying capability. Accordingly, I plan to implement this functionality in the PaimonCatalog in a future update. ## Related Issues #6523 --------- Co-authored-by: 夷羿 <yiyi.zt@alibaba-inc.com> Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com> Co-authored-by: R. Conner Howell <5731503+rchowell@users.noreply.github.com>

main

5 hours ago

Latest Branches

+1%

fix(io): support writing to local fs via GravitinoGvfs local issue#6579

1 hour ago

e6f287e

qingfeng-occ:gvfs_local_issue

0%

ci(deps): fix CI failures from dependabot bump #6570#6596

3 hours ago

47178f4

codex/fix-pr-6570-ci

-1%

feat(distributed): enable two-stage HLL aggregation for approx_count_distinct#6597

4 hours ago

00afcda

codex/approx-count-distinct-hll-schema

© 2026 CodSpeed Technology

Home Terms Privacy Docs