timestamps are interpreted as having a timezone when reading a file written by deltalake

[cpcloud]

Here's a reproducible example showing that writing timestamp without a timezone comes back as timestamp with a timezone when using delta_scan.

Not sure whether this is expected behavior!

import datetime

import duckdb

import deltalake as dl
import pyarrow as pa


t = pa.Table.from_pydict({"ts": [datetime.datetime(2024, 6, 27, 7, 13, 13)]})
# timezone should be None
assert t["ts"].type.tz is None

# write the deltalake file
dl.write_deltalake("test.delta", t)
ddb = duckdb.connect()

ddb.install_extension("delta")
ddb.load_extension("delta")

res = ddb.sql("DESCRIBE FROM delta_scan('test.delta')")
[(coltype,)] = res.column_type.fetchall()

# should be timestamp, but comes back as TIMESTAMP WITH TIME ZONE
assert coltype == "TIMESTAMP"

[cpcloud]

@samansmink Wanted to make sure this was on your radar! Any thoughts on whether this is a bug or not?

[samansmink]

@cpcloud thanks for reporting! That looks like a bug indeed. Roundtripping through parquet does result in a timestamp