Eventual-Inc/Daft - CodSpeed

Daft

Blog Docs Changelog

Performance History

Latest Results

fix(sql): report read_iceberg argument errors instead of panicking `read_iceberg()` with no path used `.expect()`, so a Rust panic crossed the FFI boundary as PanicException instead of the InvalidSQLException the other readers raise. It also rejected `path => '...'`, which read_parquet, read_csv, read_json and read_deltalake all accept.

jackylee-ch:sql-table-function-arg-validation

20 minutes ago

chore: rerun CI after upstream refresh

GENG-CHOVYYYY:perf/dense-integer-argsort

58 minutes ago

Merge branch 'main' into perf/dense-integer-argsort

GENG-CHOVYYYY:perf/dense-integer-argsort

2 hours ago

refactor(sql): reuse daft-core's integer conversion for usize options `impl SQLLiteral for usize` matched only `Literal::Int64` through `as_i64`, so a `UInt64` literal was rejected as "not an integer". `try_as_usize` in daft-core already handles every integer width and is used elsewhere; delegate to it instead of keeping a second policy in daft-sql. The user-facing message stays ours, since the core error mentions `usize`.

jackylee-ch:sql-table-function-arg-validation

10 hours ago

fix(sql): normalize function names when registering, not per call site Lowercasing the five config constructors at their registration sites left the same trap for the next contributor. `add_fn` now lowercases the key, so a mixed-case registration can no longer become unreachable. `list_sql_functions` reports lowercase names as a result; every other function was already registered lowercase, so only the five config constructors change. Also drops the `Option` from `resolve_io_config`, which could never return `Ok(None)`, and records why it lives in daft-sql rather than in the scan builders.

jackylee-ch:sql-reader-default-io-config

10 hours ago

refactor(iceberg): delegate partition value conversion to pyiceberg `to_partition_representation` reimplemented pyiceberg's epoch arithmetic. Call `datetime_to_micros`, `date_to_days` and `time_to_micros` instead: those already pick an aware or naive epoch to match the value, which is what the timezone bug was about, and the conversion can no longer drift from what the manifest readers expect. The parity test now compares against those functions rather than against hardcoded integers, so a pyiceberg semantics change surfaces here.

jackylee-ch:iceberg-partition-repr-parity

10 hours ago

refactor(parquet): derive visible row counts in one place Row group pruning and the zero-column count shortcut both need "physical rows minus positional deletes". Extract `visible_rows_in_row_groups` so the two agree by construction, and normalize the delete positions once per call instead of once per row group. I could not construct a query that reaches `count_only_stream` with delete files attached — the metadata count-pushdown path declines when deletes are present, and the normal read path keeps at least one data column — so this is consistency, not a reproducible fix. Adds unit tests for the shared helper and for `start_offset` composed with deletes.

jackylee-ch:parquet-limit-deleted-rows

10 hours ago

ci: rerun PR checks integration-test-catalogs failed while building the Iceberg docker-compose image: deb.debian.org served an index referencing openjdk-11-jdk 11.0.32.1+1-1~deb11u1 and sudo 1.9.5p2-3+deb11u4 before those .debs reached the pool, so apt 404'd. Both URLs return 200 now. The same job failed on every open PR during that window.

jackylee-ch:iceberg-partition-repr-parity

11 hours ago

Latest Branches

0%

fix(sql): report read_iceberg argument errors instead of panicking#7354

45 minutes ago

834dc8a

jackylee-ch:sql-table-function-arg-validation

0%

perf: add guarded dense integer argsort fast path#7430

1 hour ago

14b2194

GENG-CHOVYYYY:perf/dense-integer-argsort

+1%

fix(sql): make the io_config argument work in SQL readers#7449

10 hours ago

0fe1e5e

jackylee-ch:sql-reader-default-io-config

© 2026 CodSpeed Technology

Home Terms Privacy Docs