Exemple

Port pandas DataFrame code to polars on Alpine, without the wheel surprise or the silently wrong answers

sha256:a238e626062c628c548646e46144b6aac21a07492cf10e1617eb4286c853ac05

Cas

Objectif

Port pandas DataFrame code to polars on Alpine, without the wheel surprise or the silently wrong answers MIGRATION

Paquets

polars 1.43.2 polars-runtime-32 1.43.2

Environnement

python 3.12

Créé

2026-08-14T12:56:07Z

Contrat

• assert pip took the musllinux wheel for the engine, polars_runtime_32-1.43.2-cp310-abi3-musllinux_1_2_x86_64.whl, reconstructed from the recorded dist-info tag
• assert polars itself is a pure-Python py3-none-any distribution carrying no compiled file, which depends on polars-runtime-32==1.43.2 for the engine
• assert the engine's shared object is named .abi3.so rather than this interpreter's EXT_SUFFIX, so one cp310-abi3 build serves 3.12
• assert the interpreter reports musl and polars.__version__ agrees with the distribution metadata, which is the check a half-installed polars fails
• assert a DataFrame has none of index, loc, iloc, at, iat, values, reset_index, sort_values, assign, query or apply
• assert filtering renumbers rows from zero, and with_row_index before the filter is what carries the original positions through, as UInt32 data
• assert brackets take a column by name and a row by position, and an unknown name raises ColumnNotFoundError
• assert a boolean mask in brackets selects columns rather than rows, raising ValueError when its length does not match the column count
• assert the same mask on a frame with as many columns as rows returns the wrong columns with no error at all
• assert filter with a pl.col boolean expression returns the intended rows and select takes strings, pl.col and an anchored regex
• assert an Expr has no truth value, so `and` and `or` both raise TypeError, and dropping the parentheses around & raises that identical message
• assert separate positional predicates mean AND, and the keyword form is equality against a literal
• assert with_columns returns a new frame, leaves the input's columns and values untouched, and has no inplace parameter
• assert item assignment on a DataFrame raises TypeError naming DataFrame.with_columns as the replacement
• assert the new frame shares the untouched column's Arrow allocation with the old one, so returning a frame copies the column list rather than the data
• assert polars does keep a mutating escape hatch: hstack and vstack take in_place= and hand the same object back, and extend, insert_column and drop_in_place mutate the receiver
• assert an unaliased expression is named after the first column it mentions, so with_columns replaces that column instead of adding one
• assert explain returns the plan as a string while the per-row Python hook has still not been called once
• assert the optimizer moves the filter below the projection, the reverse of both the unoptimized plan and the order written
• assert collect runs the hook on only the two surviving rows, where the identical eager pipeline runs it on all six
• assert a LazyFrame naming a missing column is built without complaint and raises ColumnNotFoundError only when the plan is resolved, which explain also does
• assert a LazyFrame has no shape, and asking it for columns emits a PerformanceWarning because that resolves the schema
• assert is_null and is_nan disagree on every hole, and is_nan over a null is null rather than False
• assert null_count reports one null while the mean is nan, count excludes the null and pl.len counts every row
• assert a column holding a NaN and no null reports null_count zero with a nan mean, so a missing-data check written as null_count() == 0 passes on it
• assert min and max ignore the NaN that sum and mean propagate
• assert comparison against a null yields null while eq_missing yields False
• assert polars orders NaN above every number and compares it equal to itself, the opposite of Python, so a > 0 filter keeps the NaN and drops the null
• assert nulls sort first in both directions unless nulls_last is set
• assert drop_nulls keeps NaN and drop_nans keeps null, neither fill touches the other's hole, and fill_nan(None) collapses the two into one
• assert an integer column with a missing value stays Int64 and its aggregates skip the null on both sides of the division
• assert group_by returns the group keys first, then one column per agg expression in the order given
• assert maintain_order gives first-appearance order, which is not sorted order, and without it only the set of rows is defined
• assert two aggregates over one column collide as DuplicateError, worded one way inside agg and another inside select

Fichiers

• csx.json
• requirements.txt
• src/__init__.py
• src/frames.py
• src/install.py
• test/contract.py

Seeder d'origine

anonymous

Reçus de vérification

• python 3.12 · CONTAINER_RUN · compile:SKIPPED · contract:PASS · load:PASS · resolve:PASS · pypi@1 · 2026-08-14 · ed25519:a2ec939a4c60e243
• · CONTAINER_RUN · compile:SKIPPED · contract:FAIL · load:SKIPPED · resolve:PASS · pypi@1 · 2026-08-14 · ed25519:7da3f610facce8bf