Fix quirks in pandas test behaviour

For the current state of the code, this change is not required.
It will, however, save someone fixing this in future, when actually
wanting to use a newer OpenSAFELY Python image.

However, the tests currently rely on a quirky behaviour of `.all()` in
pandas that has been fixed.

In the version of pandas inside the `python:latest` image (v1.0.1), the
behaviour is that `.all()` can return an `dtype=object` instead of a Boolean,
and our tests as written pass.

In later versions of pandas, `.all()` returns a Boolean instead, and our
tests fail.

We can rewrite the tests to work both prior to and post this pandas fix.

This was checked by running the tests with the `Dockerfile` as provided,
and editing the `Dockerfile` to use the `v2` OpenSAFELY Python image.

See pandas-dev/pandas#12863.

Author: StevenMaude

interactive_templates/templates/v2/tests/test_measures.py

@@ -100,8 +100,8 @@ def test_calculate_total_counts(df):
 
     obs = measures.calculate_total_counts(df, date, group="total", group_value="total")
 
-    assert obs["group"].all() == "total"
-    assert obs["group_value"].all() == "total"
+    assert obs["group"].eq("total").all()
+    assert obs["group_value"].eq("total").all()
 
     assert len(obs) == 1
     assert obs.columns.tolist() == [
@@ -113,7 +113,7 @@ def test_calculate_total_counts(df):
     ]
 
     # assert date is correct
-    assert obs["date"].all() == date
+    assert obs["date"].eq(date).all()
 
     # assert that the sum of the event_measure and population columns is correct
     assert obs["event_measure"].iloc[0] == df["event_measure"].sum()
@@ -139,7 +139,7 @@ def test_calculate_group_counts(df):
     ]
 
     # assert date is correct
-    assert obs["date"].all() == date
+    assert obs["date"].eq(date).all()
 
     # assert that the sum of the event_measure and population columns is correct
     assert (