Reproduction of test_map_meta failure with pandas 3.0

I was able to reproduce the failure locally in test_map_meta when testing against pandas 3.0.

Environment
Python 3.12.3
pytest 9.0.2
pandas 3.0
local checkout of main

Command used

PYTEST_DISABLE_PLUGIN_AUTOLOAD=1 \
pytest -o addopts= \
dask/dataframe/dask_expr/tests/test_collection.py::test_map_meta -s

Debugging details
To understand the failure, I added debug prints just before the first assertion in test_map_meta:

print(f"DEBUG - RESULT META INDEX: {result._meta.index}")
print(f"DEBUG - EXPECTED INDEX: {expected.index}")

Observed output

DEBUG - RESULT META INDEX: RangeIndex(start=0, stop=0, step=1)
DEBUG - EXPECTED INDEX: RangeIndex(start=0, stop=100, step=1)

The test then fails at:
assert result._meta.index.equals(expected.index)

Interpretation
This indicates that when calling:
df.x.map(lambda x: x + 1, meta=expected.iloc[:0])

the resulting _meta does not preserve the index information from the provided meta. Instead, _meta.index is inferred as an empty RangeIndex(0, 0), even though the supplied meta carries the full index type.

This reproduces the CI failure and suggests that index handling during _meta inference for Series.map has changed (or regressed) with pandas 3.0.

Uh oh!

Pandas 3 test failures #12225

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions