add microseconds conversion test (#2632)

pola-rs · Feb 13, 2022 · 6d791bd · 6d791bd
1 parent 44e1fdf
commit 6d791bd
Show file tree

Hide file tree

Showing 4 changed files with 26 additions and 6 deletions.
diff --git a/py-polars/null.parquet b/py-polars/null.parquet
diff --git a/py-polars/polars/utils.py b/py-polars/polars/utils.py
@@ -164,10 +164,12 @@ def _prepare_row_count_args(
         return None
 
 
+EPOCH = datetime(1970, 1, 1).replace(tzinfo=None)
+
+
 def _to_python_datetime(
     value: Union[int, float], dtype: Type[DataType], tu: Optional[str] = "ns"
 ) -> Union[date, datetime]:
-    print(tu, value)
     if dtype == Date:
         # days to seconds
         # important to create from utc. Not doing this leads
@@ -176,10 +178,9 @@ def _to_python_datetime(
     elif dtype == Datetime:
         if tu == "ns":
             # nanoseconds to seconds
-            return datetime.utcfromtimestamp(value / 1_000_000_000)
+            return EPOCH + timedelta(microseconds=value / 1000)
         if tu == "us":
-            # microseconds to seconds
-            return datetime.utcfromtimestamp(value / 1_000_000)
+            return EPOCH + timedelta(microseconds=value)
         elif tu == "ms":
             # milliseconds to seconds
             return datetime.utcfromtimestamp(value / 1_000)

diff --git a/py-polars/tests/lazy_io/test_parquet.py b/py-polars/tests/lazy_io/test_parquet.py
@@ -36,7 +36,8 @@ def test_categorical_parquet_statistics() -> None:
 
 
 def test_null_parquet() -> None:
+    file = path.join(path.dirname(__file__), "null.parquet")
     df = pl.DataFrame([pl.Series("foo", [], dtype=pl.Int8)])
-    df.to_parquet("null.parquet")
-    out = pl.read_parquet("null.parquet")
+    df.to_parquet(file)
+    out = pl.read_parquet(file)
     assert out.frame_equal(df)
diff --git a/py-polars/tests/test_datelike.py b/py-polars/tests/test_datelike.py
@@ -479,3 +479,21 @@ def test_upsample() -> None:
     )
 
     assert up.frame_equal(expected)
+
+
+def test_microseconds_accuracy() -> None:
+    timestamps = [
+        datetime(2600, 1, 1, 0, 0, 0, 123456),
+        datetime(2800, 1, 1, 0, 0, 0, 456789),
+    ]
+    a = pa.Table.from_arrays(
+        arrays=[timestamps, [128, 256]],
+        schema=pa.schema(
+            [
+                ("timestamp", pa.timestamp("us")),
+                ("value", pa.int16()),
+            ]
+        ),
+    )
+
+    assert pl.from_arrow(a)["timestamp"].to_list() == timestamps  # type: ignore