Skip to content

bump submodule and fix issues - #723

Merged
evertlammerts merged 29 commits into
duckdb:mainfrom
evertlammerts:main-engine-follow
Oct 29, 2026
Merged

evertlammerts merged 29 commits into
duckdb:mainfrom
evertlammerts:main-engine-follow

Conversation

@evertlammerts

Copy link
Copy Markdown
Member

No description provided.

@evertlammerts evertlammerts added the submodule-bump Intentional DuckDB submodule pin move label Oct 28, 2026
evertlammerts and others added 8 commits October 28, 2026 19:00
…b#728)

Every CI test job exported the test group to a pylock.toml and installed
it. uv.lock is untracked, so each job resolves afresh, and the pylock
exporter downloads every wheel the index lists without a hash to compute
one, for every platform of the universal resolution. The pytorch-cpu
index lists torch 2.15.0's 36 wheels without hashes since this week; on
2026-10-28 four macOS Test abi3 jobs on duckdb#710, duckdb#711 and duckdb#723 died at the
export, timing out on a Linux cp315 torch wheel, while every Linux and
Windows job passed.
… in the pandas and numpy scans (duckdb#711)

A pandas 2 categorical read through a strided view (`df[::2]`,
`iloc[1::2]`) returned the rows of the full column: the category scan
stepped by element while the codes view had a stride (299 of 300 rows
wrong with int16 codes). A tuple of arrays, a pandas 2 Series, a range
or a set, and any reversed view (`df[::-1]`) raised a bare
`std::bad_cast`, where 1.5.6 read the tuple and refused the rest with a
clear error.
…b#710)

Fetching a TIMESTAMPTZ as a Python value needed pytz, which no
dependency group declares, so a clean install failed with "Required
module 'pytz' failed to import" (duckdb#406). pytz's tables end at 2037, so
every DST instant from 2038 on came back one hour off (duckdb#579).

Change: `PythonObject::FromValue` builds the value from
`datetime.timezone.utc` and `zoneinfo.ZoneInfo(TimeZone)`. The pytz
import cache item and every pytz pin in pyproject, the benchmarks and
the tests are gone.

We throw an error instead of relying on a `pytz` dependency when
zoneinfo has no database (Windows without the tzdata package, or an ICU
name IANA lacks such as ACT). Only this fetch path needs it, and a
dependency would reach every install.
The engine's v2.0 branch moved to fbd180f3a40 with a backport of 181
fixes (duckdb/duckdb#27965), among them the fix for Arrow scans of empty
list dictionaries (duckdb/duckdb#27626). The nightly of 2026-10-29 built
against it and failed 4 of 16729 tests: the four strict xfails of
`test_a_list_of_dictionaries_after_a_vector_of_empty_lists_reads_its_lists`
now pass. Every nightly stays red until the markers move with the pin,
which also keeps the bot from opening the bump itself.
@evertlammerts
evertlammerts added this pull request to the merge queue Oct 29, 2026
@evertlammerts
evertlammerts removed this pull request from the merge queue due to a manual request Oct 29, 2026
@evertlammerts
evertlammerts merged commit c5f8eba into duckdb:main Oct 29, 2026
30 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

submodule-bump Intentional DuckDB submodule pin move

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants