Field metadata is lost on serialization round-trip

[asfimport]

It seems only schema metadata roundtrips, while field metadata is lost:

import pandas as pd
import pyarrow as pa

fnm = "/path/to/file.arr"

df = pd.DataFrame({"x": [0,1,2,3]})
tbl = pa.Table.from_pandas(df)

metadata = {"custom": "test"}

# Update field metadata, and schema metadata
fields = [col.field.add_metadata(metadata) for col in tbl.itercolumns()]
schema_metadata = {**tbl.schema.metadata, **metadata}
schema = pa.schema(fields, metadata=schema_metadata)
tbl = pa.Table.from_batches(tbl.to_batches(), schema=schema)
print(tbl.column(0).field.metadata)              # correct :)
print(tbl.schema.field_by_name("x").metadata)    # correct :)
print(tbl.schema)                                # correct :) 

# Roundtrip
writer = pa.RecordBatchStreamWriter(fnm, tbl.schema)
writer.write_table(tbl)
writer.close()
reader = pa.RecordBatchStreamReader(fnm)
tbl = reader.read_all()

# Check
print(tbl.column(0).field.metadata)               # None :(
print(tbl.schema.field_by_name("x").metadata)     # None :(
print(tbl.schema)                                 # Metadata good :)

Reporter: Thomas Buhrmann / @buhrmann

PRs and other links:

• GitHub Pull Request #2053

_{Note: This issue was originally created as ARROW-2573. Please see the migration documentation for further details.}

[asfimport]

Wes McKinney / @wesm:
Duplicate of ARROW-906