BUG: interpolate(method="linear", limit_direction="forward") leaves NAs unchanged for pyarrow dtype

[jdamp]

Pandas version checks

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pandas.

• I have confirmed this bug exists on the main branch of pandas.

Reproducible Example

import pandas as pd

data = [1, 2, 3, None, None, 6, 7]
s_arrow = pd.Series(pd.array(data, dtype="int64[pyarrow]"))
r_nullable = s_arrow.interpolate(method="linear", limit_direction="forward")
print(r_nullable.tolist())

Issue Description

When interpolating a Series backed by a pyarrow dtype such as int64[pyarrow], method="linear" with limit_direction="forward" leaves missing values unchanged, while the equivalent series with a numpy backend interpolates as expected. The same pyarrow-backed series also appears to work with limit_direction="backward".

Expected Behavior

The pyarrow-backed series should interpolate consistently with the numpy backend, expected result is a of the form [1, 2, 3, 4, 5, 6, 7].

The expected behavior can be achieved in three ways:

1. Running the same code with pandas 2.3.3
2. Using limit_direction=backward:

import pandas as pd

data = [1, 2, 3, None, None, 6, 7]
s_arrow = pd.Series(pd.array(data, dtype="int64[pyarrow]"))
r_backwards = s_arrow.interpolate(method="linear", limit_direction="backward")
print(r_backwards.to_list())

3. Using a numpy data type instead of the pyarrow type:

import pandas as pd

data = [1, 2, 3, None, None, 6, 7]
s= pd.Series(pd.array(data, dtype="Int64"))
p= s.interpolate(method="linear", limit_direction="forward")
print(p.tolist())

Installed Versions

Details

INSTALLED VERSIONS

commit : ab90747
python : 3.11.10
python-bits : 64
OS : Linux
OS-release : 6.6.87.2-microsoft-standard-WSL2
Version : #1 SMP PREEMPT_DYNAMIC Thu Jun 5 18:30:46 UTC 2025
machine : x86_64
processor : x86_64
byteorder : little
LC_ALL : None
LANG : C.UTF-8
LOCALE : en_US.UTF-8

pandas : 3.0.2
numpy : 2.4.4
dateutil : 2.9.0.post0
pip : None
Cython : None
sphinx : None
IPython : None
adbc-driver-postgresql: None
adbc-driver-sqlite : None
bs4 : None
bottleneck : None
fastparquet : None
fsspec : None
html5lib : None
hypothesis : None
gcsfs : None
jinja2 : None
lxml.etree : None
matplotlib : None
numba : None
numexpr : None
odfpy : None
openpyxl : None
psycopg2 : None
pymysql : None
pyarrow : 24.0.0
pyiceberg : None
pyreadstat : None
pytest : None
python-calamine : None
pytz : 2026.1.post1
pyxlsb : None
s3fs : None
scipy : None
sqlalchemy : None
tables : None
tabulate : None
xarray : None
xlrd : None
xlsxwriter : None
zstandard : None
qtpy : None
pyqt5 : None

[rhshadrach]

Thanks for the report. I haven't confirmed this, but I believe it to be a regression.

[jorisvandenbossche]

This was introduced by #56347 (cc @mroeschke). It seems the pyarrow-native method only works for filling in single NA values and for cases where there are multiple consecutive NAs? (while it does not check for that condition when deciding to take that code path)

pandas/pandas/core/arrays/arrow/array.py

Lines 2745 to 2756 in 501e319

	if (
	method == "linear"
	and limit_area is None
	and limit is None
	and limit_direction == "forward"
	):
	values = self._pa_array.combine_chunks()
	na_value = pa.array([None], type=values.type)
	y_diff_2 = pc.fill_null_backward(pc.pairwise_diff_checked(values, period=2))
	prev_values = pa.concat_arrays([na_value, values[:-2], na_value])
	interps = pc.add_checked(prev_values, pc.divide_checked(y_diff_2, 2))
	return self._from_pyarrow_array(pc.coalesce(self._pa_array, interps))