fix(backends): preserve order_by position in window function when s…

…ubsequent expressions are duplicated (#7943)

## Description of changes

This PR fixes an issue where duplicate window expressions were given the
position of the new (duplicated value)
of the expression, rather than preserving the original order.

This was because we were iterating over the new keys first and putting
those expressions into a `dict`.

BY iterating over the old keys first, we preserve the original order.

## Issues closed

Closes #7940.

ibis/backends/tests/test_window.py

@@ -1251,3 +1251,36 @@ def test_rank_followed_by_over_call_merge_frames(backend, alltypes, df):
     )
 
     backend.assert_series_equal(result, expected)
+
+
+@pytest.mark.notyet(
+    ["pandas", "dask"],
+    reason="multiple ordering keys in a window function not supported for ranking",
+    raises=ValueError,
+)
+@pytest.mark.notyet(
+    ["mssql"],
+    reason="IS NULL not valid syntax for mssql",
+    raises=sa.exc.ProgrammingError,
+)
+@pytest.mark.notimpl(["polars"], raises=com.OperationNotDefinedError)
+@pytest.mark.notyet(["flink"], raises=com.UnsupportedOperationError)
+@pytest.mark.broken(
+    ["pyspark"], reason="pyspark requires CURRENT ROW", raises=PySparkAnalysisException
+)
+def test_ordering_order(con):
+    table = ibis.memtable({"bool_col": [True, False, False, None, True]})
+    window = ibis.window(
+        order_by=[
+            ibis.asc(table["bool_col"].isnull()),
+            ibis.asc(table["bool_col"]),
+        ],
+    )
+    expr = table.select(
+        rank=table["bool_col"].rank().over(window),
+        bool_col=table["bool_col"],
+    )
+
+    result = con.execute(expr)
+    value = result.bool_col.loc[result["rank"] == 4].item()
+    assert pd.isna(value)

ibis/expr/analysis.py

@@ -182,7 +182,14 @@ def merge_windows(_, default_frame):
     group_by = tuple(toolz.unique(_.frame.group_by + default_frame.group_by))
 
     order_by = {}
-    for sort_key in _.frame.order_by + default_frame.order_by:
+    # iterate in the order of the existing keys followed by the new keys
+    #
+    # this allows duplicates to be overridden with no effect on the original
+    # position
+    #
+    # see https://github.com/ibis-project/ibis/issues/7940 for how this
+    # originally manifested
+    for sort_key in default_frame.order_by + _.frame.order_by:
         order_by[sort_key.expr] = sort_key.ascending
     order_by = tuple(ops.SortKey(k, v) for k, v in order_by.items())
 

ibis/expr/types/generic.py

@@ -765,7 +765,6 @@ def bind(table):
                 )
             return expr
 
-        op = self.op()
         if isinstance(window, bl.WindowBuilder):
             if table := an.find_first_base_table(self.op()):
                 return bind(table)