Fix DataFrame accessor mutating original on assignment

[mborodii-prog]

Fix: DataFrame accessor no longer mutates the original on assignment

Problem

When using the .wrangles accessor and assigning the result to a new variable, the original DataFrame was also mutated. Both variables ended up pointing to the same in-memory object.

df = DataFrame({'col': range(20)})
df_sample = df.wrangles.select.sample(rows=5)

print(len(df))        # 5  ← should still be 20
print(len(df_sample)) # 5
print(df is df_sample) # True ← same object

The same problem affected any wrangle that modifies columns:

df = DataFrame({'name': ['alice', 'bob', 'carol']})
df_upper = df.wrangles.convert.case(input='name', case='upper')

print(df['name'].tolist())       # ['ALICE', 'BOB', 'CAROL'] ← should be unchanged
print(df_upper['name'].tolist()) # ['ALICE', 'BOB', 'CAROL']
print(df is df_upper)            # True

Root cause

Every make_method closure in dataframe.py used this pattern:

def method(self, *args, **kwargs):
    self._df.__init__(target_func(self._df, *args, **kwargs))  # reinitialises df in-place
    return self._df                                             # returns the same object as df

Two things go wrong:

1. self._df.__init__(result) calls __init__ on the existing object, which reinitialises its data in-place without changing its identity (id()). So df is mutated.
2. return self._df hands back the same reference as df. Any assignment (df_sample = ...) just creates a second name for the same object.

Many wrangle functions also assign columns directly (df[col] = ...), so even without the __init__ trick the original would be modified.

The pattern appeared in four places: _wrangles_accessor.make_method, _wrangles.make_method, _read.make_method, and _read.file.

Fix

Pass a copy of the DataFrame into target_func and return its result directly — no __init__, no returning self._df:

# Before
def method(self, *args, **kwargs):
    self._df.__init__(target_func(self._df, *args, **kwargs))
    return self._df

# After
def method(self, *args, **kwargs):
    return target_func(self._df.copy(), *args, **kwargs)

Using .copy() means wrangle functions that assign columns in-place operate on a throwaway copy, so the caller's original DataFrame is never touched. The returned value is always an independent object.

Behaviour after fix

Assigning to a new variable — original unchanged:

df = DataFrame({'col': range(20)})
df_sample = df.wrangles.select.sample(rows=5)

print(len(df))         # 20 ✓
print(len(df_sample))  # 5  ✓
print(df is df_sample) # False ✓

df = DataFrame({'name': ['alice', 'bob', 'carol']})
df_upper = df.wrangles.convert.case(input='name', case='upper')

print(df['name'].tolist())       # ['alice', 'bob', 'carol'] ✓
print(df_upper['name'].tolist()) # ['ALICE', 'BOB', 'CAROL'] ✓
print(df is df_upper)            # False ✓

Reassignment still works as before:

df = DataFrame({'col': ['hello', 'world']})
df = df.wrangles.convert.case(input='col', case='upper')

print(df['col'].tolist())  # ['HELLO', 'WORLD'] ✓

Multiple calls on the same DataFrame — each independent:

df = DataFrame({'col': ['hello', 'world', 'foo', 'bar', 'baz']})
_ = df.wrangles.convert.case(input='col', case='upper')
_ = df.wrangles.select.sample(rows=2)

print(df['col'].tolist())  # ['hello', 'world', 'foo', 'bar', 'baz'] ✓
print(len(df))             # 5 ✓

[ebhills]

Queue triage (2026-07-27)

• Disposition: Ready for review
• Delivery owner: @mborodii-prog
• Primary reviewer: @thomasstvr
• Next action: Review the DataFrame copy/identity behavior and regression tests.

Please keep the branch current and put the decision in GitHub. This is one of the five active review slots.