Add initial pandas 3 support

.github/workflows/subscript.yml

@@ -26,9 +26,17 @@ jobs:
     strategy:
       matrix:
         python-version: ["3.11", "3.12", "3.13"]
+        pandas-version: [""]
+        include:
+          - python-version: "3.13"
+            pandas-version: ">=2,<3"
+          - python-version: "3.13"
+            pandas-version: ">=3.0.0"
+        exclude:
+          - python-version: "3.13"
+            pandas-version: ""
 
     steps:
-
       - name: Install Ubuntu dependencies
         run: |
           sudo apt-get update
@@ -63,6 +71,10 @@ jobs:
         run: |
           uv pip install ".[tests, docs]"
 
+      - name: Pin pandas version
+        if: matrix.pandas-version
+        run: uv pip install "pandas${{ matrix.pandas-version }}"
+
       - name: Log OPM-Flow version
         run: |
           flow --version

pyproject.toml

@@ -43,7 +43,7 @@ dependencies = [
     "matplotlib",
     "numpy",
     "opm>=2023.04",
-    "pandas",
+    "pandas >= 2",
     "pydantic",
     "pyscal",
     "pyyaml",

src/subscript/fmuobs/parsers.py

@@ -5,7 +5,6 @@
 import re
 from pathlib import Path
 
-import numpy as np
 import pandas as pd
 
 from subscript import getLogger
@@ -431,20 +430,20 @@ def compute_date_from_days(
             to this starttime, and converted to DATE.
 
     Returns:
-        pd.DataFrame. DATE column is always of type datetime64
+        pd.DataFrame. DATE column is always datetime-like
+        (datetime64 unit depends on pandas)
     """
     assert isinstance(dframe, pd.DataFrame)
-    if starttime and "DAYS" in dframe:
-        if "DATE" not in dframe:
-            dframe["DATE"] = np.nan
-        start = pd.to_datetime(starttime)
-        date_needed_rows = ~dframe["DAYS"].isna() & dframe["DATE"].isna()
-        dframe["DATE"] = pd.to_datetime(dframe["DATE"])
-        dframe.loc[date_needed_rows, "DATE"] = start + pd.to_timedelta(
-            dframe.loc[date_needed_rows, "DAYS"], "d"
-        )
+
     if "DATE" in dframe:
         dframe["DATE"] = pd.to_datetime(dframe["DATE"])
+
+    if starttime and "DAYS" in dframe:
+        computed = pd.to_datetime(starttime) + pd.to_timedelta(dframe["DAYS"], unit="D")
+        dframe["DATE"] = (
+            dframe["DATE"].fillna(computed) if "DATE" in dframe else computed
+        )
+
     return dframe
 
 

src/subscript/fmuobs/writers.py

@@ -292,14 +292,10 @@ def convert_dframe_date_to_str(dframe: pd.DataFrame) -> pd.DataFrame:
         pd.DataFrame: DATE as a string type
     """
     if "DATE" in dframe:
-        with pd.option_context("future.no_silent_downcasting", True):
-            dframe = dframe.copy()
-            dframe["DATE"] = (
-                dframe["DATE"]
-                .astype(str)
-                .replace(["NaT", "NaN", "nan"], np.nan)
-                .infer_objects(copy=False)
-            )
+        dframe = dframe.copy()
+        dframe["DATE"] = (
+            dframe["DATE"].astype(str).replace(["NaT", "NaN", "nan"], np.nan)
+        )
 
     return dframe
 

src/subscript/summaryplot/summaryplot.py

@@ -34,9 +34,9 @@
 
 # Get rid of FutureWarning from pandas/plotting.py
 from pandas.plotting import register_matplotlib_converters
-from resdata.grid import Grid  # type: ignore
-from resdata.resfile import ResdataFile  # type: ignore
-from resdata.summary import Summary  # type: ignore
+from resdata.grid import Grid
+from resdata.resfile import ResdataFile
+from resdata.summary import Summary
 
 import subscript
 

tests/test_check_swatinit.py

@@ -558,7 +558,9 @@ def test_eqlnum2(tmp_path, mocker):
 def test_reorder_dframe_for_nonnans(inputrows, expected):
     """Test that rows with less NaNs will be prioritized through the reorder function"""
     pd.testing.assert_frame_equal(
-        reorder_dframe_for_nonnans(pd.DataFrame(inputrows)), pd.DataFrame(expected)
+        reorder_dframe_for_nonnans(pd.DataFrame(inputrows)),
+        pd.DataFrame(expected),
+        check_column_type=False,
     )
 
 

tests/test_csv2ofmvol.py

@@ -279,7 +279,9 @@ def test_df2vol(dframe, expected_lines):
     else:
         # (bogus columns in dframe must be ignored)
         pd.testing.assert_frame_equal(
-            dframe[backagain_df.columns].fillna(value=0.0), backagain_df
+            dframe[backagain_df.columns].fillna(value=0.0),
+            backagain_df,
+            check_index_type=False,
         )
 
 

tests/test_fmuobs.py

@@ -198,6 +198,7 @@ def test_roundtrip_yaml(filename, readonly_testdata_dir):
         yaml_roundtrip_dframe.sort_index(axis="columns").sort_values("LABEL"),
         dframe.sort_index(axis="columns").sort_values("LABEL"),
         check_like=True,
+        check_dtype=False,
     )
 
 

tests/test_fmuobs_parsers.py

@@ -508,11 +508,13 @@ def test_ertobs2df_starttime(string, expected):
     pd.testing.assert_frame_equal(
         ertobs2df(string, starttime="2020-01-01").sort_index(axis=1),
         expected.sort_index(axis=1),
+        check_dtype=False,
     )
     # Test again with datetime object passed, not string:
     pd.testing.assert_frame_equal(
         ertobs2df(string, starttime=datetime.date(2020, 1, 1)).sort_index(axis=1),
         expected.sort_index(axis=1),
+        check_dtype=False,
     )
 
 

tests/test_fmuobs_writers.py

@@ -533,6 +533,7 @@ def test_convert_dframe_date_to_str(dframe, expected_dframe):
     pd.testing.assert_frame_equal(
         convert_dframe_date_to_str(dframe),
         expected_dframe,
+        check_dtype=False,
     )
 
 

tests/test_ofmvol2csv.py

@@ -270,7 +270,7 @@ def test_parse_well(inputlines, expected):
     inputlines = ofmvol2csv.cleanse_ofm_lines(inputlines)
     colnames = ofmvol2csv.extract_columnnames(inputlines)
     dframe = ofmvol2csv.parse_well(inputlines[1:], colnames)
-    pd.testing.assert_frame_equal(dframe, expected)
+    pd.testing.assert_frame_equal(dframe, expected, check_index_type=False)
 
 
 @pytest.mark.parametrize(
@@ -362,7 +362,11 @@ def test_process_volstr(inputlines, expected):
         expected["DATE"] = pd.to_datetime(expected["DATE"])
         expected = expected.set_index(["WELL", "DATE"])
     dframe = ofmvol2csv.process_volstr("\n".join(inputlines))
-    pd.testing.assert_frame_equal(dframe, expected)
+    pd.testing.assert_frame_equal(
+        dframe,
+        expected,
+        check_index_type=False,
+    )
 
 
 @pytest.mark.parametrize(