Segfault when trying to serialize a DataFrame with Null-only Categorical Column

[asfimport]

import pandas as pd
import feather
print("Pandas: %s" % pd.__version__)
print("Feather: %s" % feather.version.version)
df = pd.DataFrame({"A": (1, 2, 3), "B": ("a", "b", "c")})   # if we use this line everything works fine
df = pd.DataFrame({"A": (1, 2, 3), "B": (None, None, None)})
df = df.assign(B=df.B.astype("category"))
#feather.write_dataframe(df, "/tmp/foo.feather")   # this segfaults!

Pandas: 0.19.2
Feather: 0.4.0

Reporter: Panagiotis Mavrogiorgos
Assignee: Wes McKinney / @wesm

_{Note: This issue was originally created as ARROW-1188. Please see the migration documentation for further details.}

[asfimport]

Wes McKinney / @wesm:
Marked for Arrow 0.5.0; I think this may be fixed already

[asfimport]

Wes McKinney / @wesm:
Confirming this bug still exists (it hits a DCHECK in debug builds). Going to take a look

[asfimport]

Wes McKinney / @wesm:
PR: #863

[asfimport]

Wes McKinney / @wesm:
Issue resolved by pull request 863
#863