Circumvent Pandas bugs for years > 2262

equinor · Jan 20, 2023 · bfd3574 · bfd3574
1 parent 338df6d
commit bfd3574
Show file tree

Hide file tree

Showing 2 changed files with 33 additions and 4 deletions.
diff --git a/python/ecl/summary/ecl_sum.py b/python/ecl/summary/ecl_sum.py
@@ -581,7 +581,7 @@ def pandas_frame(self, time_index=None, column_keys=None):
             raise ValueError("No valid key")
 
         if time_index is None:
-            time_index = self.numpy_dates
+            time_index = self.dates
             data = numpy.zeros([len(time_index), len(keywords)])
             EclSum._init_pandas_frame(
                 self, keywords, data.ctypes.data_as(ctypes.POINTER(ctypes.c_double))
@@ -635,16 +635,21 @@ def _compile_headers_list(headers, dims):
     @classmethod
     def from_pandas(cls, case, frame, dims=None, headers=None):
         start_time = frame.index[0]
+
+        # Avoid Pandas or numpy timestamps, to avoid Pandas attempting to create
+        # timestamp64[ns] indices (which can't go beyond year 2262)
+        # https://github.com/pandas-dev/pandas/issues/39727
+        if isinstance(start_time, pandas.Timestamp):
+            start_time = start_time.to_pydatetime()
+
         var_list = []
         if headers is None:
             header_list = EclSum._compile_headers_list(frame.columns.values, dims)
         else:
             header_list = EclSum._compile_headers_list(headers, dims)
         if dims is None:
             dims = [1, 1, 1]
-        ecl_sum = EclSum.writer(
-            case, start_time.to_pydatetime(), dims[0], dims[1], dims[2]
-        )
+        ecl_sum = EclSum.writer(case, start_time, dims[0], dims[1], dims[2])
         for kw, wgname, num, unit in header_list:
             var_list.append(
                 ecl_sum.addVariable(kw, wgname=wgname, num=num, unit=unit).getKey1()

diff --git a/python/tests/ecl_tests/test_ecl_sum.py b/python/tests/ecl_tests/test_ecl_sum.py
@@ -1,6 +1,8 @@
 #!/usr/bin/env python
 import datetime
 import os.path
+import pandas as pd
+import numpy as np
 
 from cwrap import CFILE
 from cwrap import Prototype, load, open as copen
@@ -19,6 +21,28 @@ def test_write_repr():
     )
 
 
+def test_that_year2263_through_pandas_works():
+    """Date 2262-04-11 is the upper limit for nanosecond 64-bit
+    timestamps, which is what pandas<2 prefers.
+    """
+    eclsum = EclSum.from_pandas(
+        "TESTCASE",
+        pd.DataFrame(
+            [
+                {"DATE": datetime.date(2000, 1, 1), "FPR": 200},
+                {"DATE": datetime.date(2263, 1, 1), "FPR": 1},
+            ]
+        ).set_index("DATE"),
+    )
+    assert eclsum.numpy_dates[0] == np.datetime64("2000-01-01T00:00:00.000")
+
+    # (This is hit by a round-off error due to 32-bit floats):
+    assert eclsum.numpy_dates[1] == np.datetime64("2262-12-31T23:57:52.000")
+
+    # Roundtrip:
+    assert eclsum.pandas_frame().index[1] == datetime.datetime(2262, 12, 31, 23, 57, 52)
+
+
 @equinor_test()
 class EclSumTest(EclTest):
     def setUp(self):