Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

ENH(string dtype): fallback for HDF5 with UTF-8 surrogates #60993

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
+96 −39
Open

ENH(string dtype): fallback for HDF5 with UTF-8 surrogates #60993

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
d621cd0
ENH(string dtype): Add object fallback for HDF5 read/write with UTF-8…
rhshadrach Feb 23, 2025
e3e64f4
Fallback to StringDtype(python) instead
rhshadrach Feb 23, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
114 changes: 85 additions & 29 deletions pandas/io/pytables.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,6 +39,7 @@
)
from pandas._libs.lib import is_string_array
from pandas._libs.tslibs import timezones
from pandas.compat import HAS_PYARROW
from pandas.compat._optional import import_optional_dependency
from pandas.compat.pickle_compat import patch_pickle
from pandas.errors import (
Expand Down Expand Up @@ -376,6 +377,13 @@ def read_hdf(
object
The selected object. Return type depends on the object stored.

Notes
-----
When ``errors="surrogatepass"``, ``pd.options.future.infer_string`` is true,
and PyArrow is installed, if a UTF-16 surrogate is encountered when decoding
to UTF-8, the resulting dtype will be
``pd.StringDtype(storage="python", na_value=np.nan)``.

See Also
--------
DataFrame.to_hdf : Write a HDF file from a DataFrame.
Expand Down Expand Up @@ -2257,6 +2265,20 @@ def convert(
# making an Index instance could throw a number of different errors
try:
new_pd_index = factory(values, **kwargs)
except UnicodeEncodeError as err:
if (
errors == "surrogatepass"
and get_option("future.infer_string")
and str(err).endswith("surrogates not allowed")
and HAS_PYARROW
):
new_pd_index = factory(
values,
dtype=StringDtype(storage="python", na_value=np.nan),
**kwargs,
)
else:
raise
except ValueError:
# if the output freq is different that what we recorded,
# it should be None (see also 'doc example part 2')
Expand Down Expand Up @@ -3170,12 +3192,29 @@ def read_index_node(
**kwargs,
)
else:
index = factory(
_unconvert_index(
data, kind, encoding=self.encoding, errors=self.errors
),
**kwargs,
)
try:
index = factory(
_unconvert_index(
data, kind, encoding=self.encoding, errors=self.errors
),
**kwargs,
)
except UnicodeEncodeError as err:
if (
self.errors == "surrogatepass"
and get_option("future.infer_string")
and str(err).endswith("surrogates not allowed")
and HAS_PYARROW
):
index = factory(
_unconvert_index(
data, kind, encoding=self.encoding, errors=self.errors
),
dtype=StringDtype(storage="python", na_value=np.nan),
**kwargs,
)
else:
raise

index.name = name

Expand Down Expand Up @@ -3311,13 +3350,24 @@ def read(
self.validate_read(columns, where)
index = self.read_index("index", start=start, stop=stop)
values = self.read_array("values", start=start, stop=stop)
result = Series(values, index=index, name=self.name, copy=False)
if (
using_string_dtype()
and isinstance(values, np.ndarray)
and is_string_array(values, skipna=True)
):
result = result.astype(StringDtype(na_value=np.nan))
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This was added back in #54431. I do not believe it is no longer necessary - we will infer string in the Series constructor.

try:
result = Series(values, index=index, name=self.name, copy=False)
except UnicodeEncodeError as err:
if (
self.errors == "surrogatepass"
and get_option("future.infer_string")
and str(err).endswith("surrogates not allowed")
and HAS_PYARROW
):
result = Series(
values,
index=index,
name=self.name,
copy=False,
dtype=StringDtype(storage="python", na_value=np.nan),
)
else:
raise
return result

def write(self, obj, **kwargs) -> None:
Expand Down Expand Up @@ -4764,7 +4814,24 @@ def read(
values = values.reshape((1, values.shape[0]))

if isinstance(values, (np.ndarray, DatetimeArray)):
df = DataFrame(values.T, columns=cols_, index=index_, copy=False)
try:
df = DataFrame(values.T, columns=cols_, index=index_, copy=False)
except UnicodeEncodeError as err:
if (
self.errors == "surrogatepass"
and get_option("future.infer_string")
and str(err).endswith("surrogates not allowed")
and HAS_PYARROW
):
df = DataFrame(
values.T,
columns=cols_,
index=index_,
copy=False,
dtype=StringDtype(storage="python", na_value=np.nan),
)
else:
raise
elif isinstance(values, Index):
df = DataFrame(values, columns=cols_, index=index_)
else:
Expand All @@ -4774,23 +4841,10 @@ def read(
assert (df.dtypes == values.dtype).all(), (df.dtypes, values.dtype)

# If str / string dtype is stored in meta, use that.
converted = False
for column in cols_:
dtype = getattr(self.table.attrs, f"{column}_meta", None)
if dtype in ["str", "string"]:
df[column] = df[column].astype(dtype)
converted = True
# Otherwise try inference.
if (
not converted
and using_string_dtype()
and isinstance(values, np.ndarray)
and is_string_array(
values,
skipna=True,
)
):
df = df.astype(StringDtype(na_value=np.nan))
frames.append(df)

if len(frames) == 1:
Expand Down Expand Up @@ -5224,7 +5278,7 @@ def _convert_string_array(data: np.ndarray, encoding: str, errors: str) -> np.nd
# encode if needed
if len(data):
data = (
Series(data.ravel(), copy=False)
Series(data.ravel(), copy=False, dtype="object")
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We go immediately back to NumPy, so no reason to use string dtypes here.

.str.encode(encoding, errors)
._values.reshape(data.shape)
)
Expand Down Expand Up @@ -5264,7 +5318,9 @@ def _unconvert_string_array(
dtype = f"U{itemsize}"

if isinstance(data[0], bytes):
ser = Series(data, copy=False).str.decode(encoding, errors=errors)
ser = Series(data, copy=False).str.decode(
encoding, errors=errors, dtype="object"
)
Comment on lines +5321 to +5323
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We go immediately back to NumPy, so no reason to use string dtypes here.

data = ser.to_numpy()
data.flags.writeable = True
else:
Expand Down
21 changes: 11 additions & 10 deletions pandas/tests/io/pytables/test_store.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,8 +7,6 @@
import numpy as np
import pytest

from pandas._config import using_string_dtype

from pandas.compat import PY312

import pandas as pd
Expand All @@ -25,7 +23,6 @@
timedelta_range,
)
import pandas._testing as tm
from pandas.conftest import has_pyarrow
from pandas.tests.io.pytables.common import (
_maybe_remove,
ensure_clean_store,
Expand Down Expand Up @@ -385,20 +382,24 @@ def test_to_hdf_with_min_itemsize(tmp_path, setup_path):
tm.assert_series_equal(read_hdf(path, "ss4"), concat([df["B"], df2["B"]]))


@pytest.mark.xfail(
using_string_dtype() and has_pyarrow,
reason="TODO(infer_string): can't encode '\ud800': surrogates not allowed",
)
@pytest.mark.parametrize("format", ["fixed", "table"])
def test_to_hdf_errors(tmp_path, format, setup_path):
def test_to_hdf_errors(tmp_path, format, setup_path, using_infer_string):
data = ["\ud800foo"]
ser = Series(data, index=Index(data))
ser = Series(data, index=Index(data, dtype="object"), dtype="object")
path = tmp_path / setup_path
# GH 20835
ser.to_hdf(path, key="table", format=format, errors="surrogatepass")

result = read_hdf(path, "table", errors="surrogatepass")
tm.assert_series_equal(result, ser)

if using_infer_string:
# https://github.com/pandas-dev/pandas/pull/60993
# Surrogates fallback to python storage.
dtype = pd.StringDtype(storage="python", na_value=np.nan)
else:
dtype = "object"
expected = Series(data, index=Index(data, dtype=dtype), dtype=dtype)
tm.assert_series_equal(result, expected)


def test_create_table_index(setup_path):
Expand Down
Loading