DataFrame.append removal in pandas 2.0 and concat replacement

Direct answer

In pandas 2.0+, DataFrame.append and Series.append are removed; calls raise AttributeError. The supported row-wise replacement is pd.concat(frames, axis=0).

ignore_index semantics are effectively unchanged from the old append default. Both default to False, which preserves original index labels and permits duplicates. Setting ignore_index=True in either API replaces the index with RangeIndex(0, n) and discards the old labels.

For column alignment, concat(axis=0) performs an outer union of columns by name, as append did. Rows missing a column are filled with NaN for object or numeric frames, or pd.NA for nullable extension dtypes. There is no behavioral difference in the basic alignment rule.

Dtype upcasting is the main migration risk. Concatenating integer columns with missing values typically widens the column to float unless nullable dtypes such as Int64 are used. Mixing int and float yields float; object, categorical, timezone-aware datetime, and mixed extension columns may unify to a broader dtype or object. append tolerated similar coercion, but concat results must be explicitly checked and cast.

Confirmed facts versus likely explanation

Migration steps

1. Replace df.append(other) with pd.concat([df, other], axis=0, ignore_index=False).
2. Use ignore_index=True only if the legacy code reset the index.
3. Accumulate frames in a list and call pd.concat once; do not concat repeatedly inside a loop.
4. After concatenation, verify index.duplicated().any(), column set, isnull().sum(), and dtypes against the target schema.
5. If the legacy call used verify_integrity=True, raise or drop duplicates when index.duplicated().any() is true.

Verification

import pandas as pd

print(pd.__version__)  # expect 2.0+

frames = [df_left, df_right]
result = pd.concat(frames, axis=0, ignore_index=False)

print(result.index.duplicated().any())
print(result.dtypes)
print(result.isnull().sum())

One missing diagnostic detail changes the recommendation: did the legacy code pass verify_integrity=True to append? If yes, add the duplicate-index check shown above; otherwise the basic concat replacement is sufficient.