feat(api): add nulls_first=False argument to order_by (#9385)

Co-authored-by: Phillip Cloud <[email protected]>

ibis/backends/dask/executor.py

@@ -364,10 +364,25 @@ def visit(cls, op: ops.Sort, parent, keys):
         # 2. sort the dataframe using those columns
         # 3. drop the sort key columns
         ascending = [key.ascending for key in op.keys]
+        nulls_first = [key.nulls_first for key in op.keys]
+
+        if all(nulls_first):
+            na_position = "first"
+        elif not any(nulls_first):
+            na_position = "last"
+        else:
+            raise ValueError(
+                "dask does not support specifying null ordering for individual columns"
+            )
+
         newcols = {gen_name("sort_key"): col for col in keys}
         names = list(newcols.keys())
         df = parent.assign(**newcols)
-        df = df.sort_values(by=names, ascending=ascending)
+        df = df.sort_values(
+            by=names,
+            ascending=ascending,
+            na_position=na_position,
+        )
         return df.drop(names, axis=1)
 
     @classmethod

ibis/backends/impala/tests/snapshots/test_analytic_functions/test_analytic_exprs/first/out.sql

@@ -1,3 +1,3 @@
 SELECT
-  FIRST_VALUE(`t0`.`double_col`) OVER (ORDER BY `t0`.`id` ASC NULLS LAST) AS `First(double_col)`
+  FIRST_VALUE(`t0`.`double_col`) OVER (ORDER BY `t0`.`id` ASC) AS `First(double_col)`
 FROM `functional_alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_analytic_functions/test_analytic_exprs/lag_arg/out.sql

@@ -1,3 +1,3 @@
 SELECT
-  LAG(`t0`.`string_col`, 2) OVER (ORDER BY NULL ASC NULLS LAST) AS `Lag(string_col, 2)`
+  LAG(`t0`.`string_col`, 2) OVER (ORDER BY NULL ASC) AS `Lag(string_col, 2)`
 FROM `functional_alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_analytic_functions/test_analytic_exprs/lag_default/out.sql

@@ -1,3 +1,3 @@
 SELECT
-  LAG(`t0`.`string_col`) OVER (ORDER BY NULL ASC NULLS LAST) AS `Lag(string_col)`
+  LAG(`t0`.`string_col`) OVER (ORDER BY NULL ASC) AS `Lag(string_col)`
 FROM `functional_alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_analytic_functions/test_analytic_exprs/lag_explicit_default/out.sql

@@ -1,3 +1,3 @@
 SELECT
-  LAG(`t0`.`string_col`, 1, 0) OVER (ORDER BY NULL ASC NULLS LAST) AS `Lag(string_col, 0)`
+  LAG(`t0`.`string_col`, 1, 0) OVER (ORDER BY NULL ASC) AS `Lag(string_col, 0)`
 FROM `functional_alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_analytic_functions/test_analytic_exprs/last/out.sql

@@ -1,3 +1,3 @@
 SELECT
-  LAST_VALUE(`t0`.`double_col`) OVER (ORDER BY `t0`.`id` ASC NULLS LAST) AS `Last(double_col)`
+  LAST_VALUE(`t0`.`double_col`) OVER (ORDER BY `t0`.`id` ASC) AS `Last(double_col)`
 FROM `functional_alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_analytic_functions/test_analytic_exprs/lead_arg/out.sql

@@ -1,3 +1,3 @@
 SELECT
-  LEAD(`t0`.`string_col`, 2) OVER (ORDER BY NULL ASC NULLS LAST) AS `Lead(string_col, 2)`
+  LEAD(`t0`.`string_col`, 2) OVER (ORDER BY NULL ASC) AS `Lead(string_col, 2)`
 FROM `functional_alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_analytic_functions/test_analytic_exprs/lead_default/out.sql

@@ -1,3 +1,3 @@
 SELECT
-  LEAD(`t0`.`string_col`) OVER (ORDER BY NULL ASC NULLS LAST) AS `Lead(string_col)`
+  LEAD(`t0`.`string_col`) OVER (ORDER BY NULL ASC) AS `Lead(string_col)`
 FROM `functional_alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_analytic_functions/test_analytic_exprs/lead_explicit_default/out.sql

@@ -1,3 +1,3 @@
 SELECT
-  LEAD(`t0`.`string_col`, 1, 0) OVER (ORDER BY NULL ASC NULLS LAST) AS `Lead(string_col, 0)`
+  LEAD(`t0`.`string_col`, 1, 0) OVER (ORDER BY NULL ASC) AS `Lead(string_col, 0)`
 FROM `functional_alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_analytic_functions/test_analytic_exprs/ntile/out.sql

@@ -1,3 +1,3 @@
 SELECT
-  NTILE(3) OVER (ORDER BY `t0`.`double_col` ASC NULLS LAST) - 1 AS `NTile(3)`
+  NTILE(3) OVER (ORDER BY `t0`.`double_col` ASC) - 1 AS `NTile(3)`
 FROM `functional_alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_analytic_functions/test_analytic_exprs/percent_rank/out.sql

@@ -1,3 +1,3 @@
 SELECT
-  PERCENT_RANK() OVER (ORDER BY `t0`.`double_col` ASC NULLS LAST) AS `PercentRank()`
+  PERCENT_RANK() OVER (ORDER BY `t0`.`double_col` ASC) AS `PercentRank()`
 FROM `functional_alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_exprs/test_filter_with_analytic/out.sql

@@ -1 +1 @@
-SELECT * FROM (SELECT `t1`.`col`, COUNT(*) OVER (ORDER BY NULL ASC NULLS LAST) AS `analytic` FROM (SELECT `t0`.`col`, NULL AS `filter` FROM `x` AS `t0` WHERE NULL IS NULL) AS `t1`) AS `t2`
+SELECT * FROM (SELECT `t1`.`col`, COUNT(*) OVER (ORDER BY NULL ASC) AS `analytic` FROM (SELECT `t0`.`col`, NULL AS `filter` FROM `x` AS `t0` WHERE NULL IS NULL) AS `t1`) AS `t2`

ibis/backends/impala/tests/snapshots/test_sql/test_group_by_with_window_preserves_range/out.sql

@@ -1 +1 @@
-SELECT `t0`.`one`, `t0`.`two`, `t0`.`three`, SUM(`t0`.`two`) OVER (PARTITION BY `t0`.`three` ORDER BY `t0`.`one` ASC NULLS LAST) AS `four` FROM `my_data` AS `t0`
+SELECT `t0`.`one`, `t0`.`two`, `t0`.`three`, SUM(`t0`.`two`) OVER (PARTITION BY `t0`.`three` ORDER BY `t0`.`one` ASC) AS `four` FROM `my_data` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_add_default_order_by/out.sql

@@ -10,9 +10,9 @@ SELECT
   `t0`.`i`,
   `t0`.`j`,
   `t0`.`k`,
-  LAG(`t0`.`f`) OVER (PARTITION BY `t0`.`g` ORDER BY NULL ASC NULLS LAST) AS `lag`,
-  LEAD(`t0`.`f`) OVER (PARTITION BY `t0`.`g` ORDER BY NULL ASC NULLS LAST) - `t0`.`f` AS `fwd_diff`,
-  FIRST_VALUE(`t0`.`f`) OVER (PARTITION BY `t0`.`g` ORDER BY NULL ASC NULLS LAST) AS `first`,
-  LAST_VALUE(`t0`.`f`) OVER (PARTITION BY `t0`.`g` ORDER BY NULL ASC NULLS LAST) AS `last`,
-  LAG(`t0`.`f`) OVER (PARTITION BY `t0`.`g` ORDER BY `t0`.`d` ASC NULLS LAST) AS `lag2`
+  LAG(`t0`.`f`) OVER (PARTITION BY `t0`.`g` ORDER BY NULL ASC) AS `lag`,
+  LEAD(`t0`.`f`) OVER (PARTITION BY `t0`.`g` ORDER BY NULL ASC) - `t0`.`f` AS `fwd_diff`,
+  FIRST_VALUE(`t0`.`f`) OVER (PARTITION BY `t0`.`g` ORDER BY NULL ASC) AS `first`,
+  LAST_VALUE(`t0`.`f`) OVER (PARTITION BY `t0`.`g` ORDER BY NULL ASC) AS `last`,
+  LAG(`t0`.`f`) OVER (PARTITION BY `t0`.`g` ORDER BY `t0`.`d` ASC) AS `lag2`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_aggregate_in_projection/out.sql

@@ -10,5 +10,5 @@ SELECT
   `t0`.`i`,
   `t0`.`j`,
   `t0`.`k`,
-  `t0`.`f` / SUM(`t0`.`f`) OVER (ORDER BY NULL ASC NULLS LAST) AS `normed_f`
+  `t0`.`f` / SUM(`t0`.`f`) OVER (ORDER BY NULL ASC) AS `normed_f`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_cumulative_functions/max/out1.sql

@@ -1,3 +1,3 @@
 SELECT
-  MAX(`t0`.`f`) OVER (ORDER BY `t0`.`d` ASC NULLS LAST) AS `foo`
+  MAX(`t0`.`f`) OVER (ORDER BY `t0`.`d` ASC) AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_cumulative_functions/max/out2.sql

@@ -1,3 +1,3 @@
 SELECT
-  MAX(`t0`.`f`) OVER (ORDER BY `t0`.`d` ASC NULLS LAST) AS `foo`
+  MAX(`t0`.`f`) OVER (ORDER BY `t0`.`d` ASC) AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_cumulative_functions/mean/out1.sql

@@ -1,3 +1,3 @@
 SELECT
-  AVG(`t0`.`f`) OVER (ORDER BY `t0`.`d` ASC NULLS LAST) AS `foo`
+  AVG(`t0`.`f`) OVER (ORDER BY `t0`.`d` ASC) AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_cumulative_functions/mean/out2.sql

@@ -1,3 +1,3 @@
 SELECT
-  AVG(`t0`.`f`) OVER (ORDER BY `t0`.`d` ASC NULLS LAST) AS `foo`
+  AVG(`t0`.`f`) OVER (ORDER BY `t0`.`d` ASC) AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_cumulative_functions/min/out1.sql

@@ -1,3 +1,3 @@
 SELECT
-  MIN(`t0`.`f`) OVER (ORDER BY `t0`.`d` ASC NULLS LAST) AS `foo`
+  MIN(`t0`.`f`) OVER (ORDER BY `t0`.`d` ASC) AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_cumulative_functions/min/out2.sql

@@ -1,3 +1,3 @@
 SELECT
-  MIN(`t0`.`f`) OVER (ORDER BY `t0`.`d` ASC NULLS LAST) AS `foo`
+  MIN(`t0`.`f`) OVER (ORDER BY `t0`.`d` ASC) AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_cumulative_functions/sum/out1.sql

@@ -1,3 +1,3 @@
 SELECT
-  SUM(`t0`.`f`) OVER (ORDER BY `t0`.`d` ASC NULLS LAST) AS `foo`
+  SUM(`t0`.`f`) OVER (ORDER BY `t0`.`d` ASC) AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_cumulative_functions/sum/out2.sql

@@ -1,3 +1,3 @@
 SELECT
-  SUM(`t0`.`f`) OVER (ORDER BY `t0`.`d` ASC NULLS LAST) AS `foo`
+  SUM(`t0`.`f`) OVER (ORDER BY `t0`.`d` ASC) AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_multiple_windows/out.sql

@@ -1,4 +1,4 @@
 SELECT
   `t0`.`g`,
-  SUM(`t0`.`f`) OVER (PARTITION BY `t0`.`g` ORDER BY NULL ASC NULLS LAST) - SUM(`t0`.`f`) OVER (ORDER BY NULL ASC NULLS LAST) AS `result`
+  SUM(`t0`.`f`) OVER (PARTITION BY `t0`.`g` ORDER BY NULL ASC) - SUM(`t0`.`f`) OVER (ORDER BY NULL ASC) AS `result`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_nested_analytic_function/out.sql

@@ -1,3 +1,3 @@
 SELECT
-  LAG(`t0`.`f` - LAG(`t0`.`f`) OVER (ORDER BY `t0`.`f` ASC NULLS LAST)) OVER (ORDER BY `t0`.`f` ASC NULLS LAST) AS `foo`
+  LAG(`t0`.`f` - LAG(`t0`.`f`) OVER (ORDER BY `t0`.`f` ASC)) OVER (ORDER BY `t0`.`f` ASC) AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_order_by_desc/out1.sql

@@ -1,4 +1,4 @@
 SELECT
   `t0`.`f`,
-  ROW_NUMBER() OVER (ORDER BY `t0`.`f` DESC) - 1 AS `revrank`
+  ROW_NUMBER() OVER (ORDER BY `t0`.`f` DESC NULLS LAST) - 1 AS `revrank`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_order_by_desc/out2.sql

@@ -1,4 +1,4 @@
 SELECT
-  LAG(`t0`.`d`) OVER (PARTITION BY `t0`.`g` ORDER BY `t0`.`f` DESC) AS `foo`,
-  MAX(`t0`.`a`) OVER (PARTITION BY `t0`.`g` ORDER BY `t0`.`f` DESC) AS `Max(a)`
+  LAG(`t0`.`d`) OVER (PARTITION BY `t0`.`g` ORDER BY `t0`.`f` DESC NULLS LAST) AS `foo`,
+  MAX(`t0`.`a`) OVER (PARTITION BY `t0`.`g` ORDER BY `t0`.`f` DESC NULLS LAST) AS `Max(a)`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_propagate_nested_windows/out.sql

@@ -1,5 +1,3 @@
 SELECT
-  LAG(
-    `t0`.`f` - LAG(`t0`.`f`) OVER (PARTITION BY `t0`.`g` ORDER BY `t0`.`f` ASC NULLS LAST)
-  ) OVER (PARTITION BY `t0`.`g` ORDER BY `t0`.`f` ASC NULLS LAST) AS `foo`
+  LAG(`t0`.`f` - LAG(`t0`.`f`) OVER (PARTITION BY `t0`.`g` ORDER BY `t0`.`f` ASC)) OVER (PARTITION BY `t0`.`g` ORDER BY `t0`.`f` ASC) AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_rank_functions/out.sql

@@ -1,5 +1,5 @@
 SELECT
   `t0`.`g`,
-  RANK() OVER (ORDER BY `t0`.`f` ASC NULLS LAST) - 1 AS `minr`,
-  DENSE_RANK() OVER (ORDER BY `t0`.`f` ASC NULLS LAST) - 1 AS `denser`
+  RANK() OVER (ORDER BY `t0`.`f` ASC) - 1 AS `minr`,
+  DENSE_RANK() OVER (ORDER BY `t0`.`f` ASC) - 1 AS `denser`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_row_number_does_not_require_order_by/out1.sql

@@ -10,5 +10,5 @@ SELECT
   `t0`.`i`,
   `t0`.`j`,
   `t0`.`k`,
-  ROW_NUMBER() OVER (PARTITION BY `t0`.`g` ORDER BY NULL ASC NULLS LAST) AS `foo`
+  ROW_NUMBER() OVER (PARTITION BY `t0`.`g` ORDER BY NULL ASC) AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_row_number_does_not_require_order_by/out2.sql

@@ -10,5 +10,5 @@ SELECT
   `t0`.`i`,
   `t0`.`j`,
   `t0`.`k`,
-  ROW_NUMBER() OVER (PARTITION BY `t0`.`g` ORDER BY `t0`.`f` ASC NULLS LAST) - 1 AS `foo`
+  ROW_NUMBER() OVER (PARTITION BY `t0`.`g` ORDER BY `t0`.`f` ASC) - 1 AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_row_number_properly_composes_with_arithmetic/out.sql

@@ -11,6 +11,6 @@ SELECT
   `t0`.`j`,
   `t0`.`k`,
   (
-    ROW_NUMBER() OVER (ORDER BY `t0`.`f` ASC NULLS LAST) - 1
+    ROW_NUMBER() OVER (ORDER BY `t0`.`f` ASC) - 1
   ) / 2 AS `new`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_window_frame_specs/cumulative/out.sql

@@ -1,3 +1,3 @@
 SELECT
-  SUM(`t0`.`d`) OVER (ORDER BY `t0`.`f` ASC NULLS LAST) AS `foo`
+  SUM(`t0`.`d`) OVER (ORDER BY `t0`.`f` ASC) AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_window_frame_specs/foll_0/out.sql

@@ -1,3 +1,3 @@
 SELECT
-  SUM(`t0`.`d`) OVER (ORDER BY `t0`.`f` ASC NULLS LAST) AS `foo`
+  SUM(`t0`.`d`) OVER (ORDER BY `t0`.`f` ASC) AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_window_frame_specs/foll_10_5/out.sql

@@ -1,3 +1,3 @@
 SELECT
-  SUM(`t0`.`d`) OVER (ORDER BY `t0`.`f` ASC NULLS LAST ROWS BETWEEN 10 preceding AND 5 preceding) AS `foo`
+  SUM(`t0`.`d`) OVER (ORDER BY `t0`.`f` ASC ROWS BETWEEN 10 preceding AND 5 preceding) AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_window_frame_specs/foll_2/out.sql

@@ -1,3 +1,3 @@
 SELECT
-  SUM(`t0`.`d`) OVER (ORDER BY `t0`.`f` ASC NULLS LAST ROWS BETWEEN UNBOUNDED PRECEDING AND 2 following) AS `foo`
+  SUM(`t0`.`d`) OVER (ORDER BY `t0`.`f` ASC ROWS BETWEEN UNBOUNDED PRECEDING AND 2 following) AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_window_frame_specs/foll_2_prec_0/out.sql

@@ -1,3 +1,3 @@
 SELECT
-  SUM(`t0`.`d`) OVER (ORDER BY `t0`.`f` ASC NULLS LAST ROWS BETWEEN CURRENT ROW AND 2 following) AS `foo`
+  SUM(`t0`.`d`) OVER (ORDER BY `t0`.`f` ASC ROWS BETWEEN CURRENT ROW AND 2 following) AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_window_frame_specs/foll_5_10/out.sql

@@ -1,3 +1,3 @@
 SELECT
-  SUM(`t0`.`d`) OVER (ORDER BY `t0`.`f` ASC NULLS LAST ROWS BETWEEN 5 following AND 10 following) AS `foo`
+  SUM(`t0`.`d`) OVER (ORDER BY `t0`.`f` ASC ROWS BETWEEN 5 following AND 10 following) AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_window_frame_specs/prec_0/out.sql

@@ -1,3 +1,3 @@
 SELECT
-  SUM(`t0`.`d`) OVER (ORDER BY `t0`.`f` ASC NULLS LAST ROWS BETWEEN CURRENT ROW AND UNBOUNDED FOLLOWING) AS `foo`
+  SUM(`t0`.`d`) OVER (ORDER BY `t0`.`f` ASC ROWS BETWEEN CURRENT ROW AND UNBOUNDED FOLLOWING) AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_window_frame_specs/prec_5/out.sql

@@ -1,3 +1,3 @@
 SELECT
-  SUM(`t0`.`d`) OVER (ORDER BY `t0`.`f` ASC NULLS LAST ROWS BETWEEN 5 preceding AND UNBOUNDED FOLLOWING) AS `foo`
+  SUM(`t0`.`d`) OVER (ORDER BY `t0`.`f` ASC ROWS BETWEEN 5 preceding AND UNBOUNDED FOLLOWING) AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_window_frame_specs/prec_5_foll_0/out.sql

@@ -1,3 +1,3 @@
 SELECT
-  SUM(`t0`.`d`) OVER (ORDER BY `t0`.`f` ASC NULLS LAST ROWS BETWEEN 5 preceding AND CURRENT ROW) AS `foo`
+  SUM(`t0`.`d`) OVER (ORDER BY `t0`.`f` ASC ROWS BETWEEN 5 preceding AND CURRENT ROW) AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_window_frame_specs/prec_5_foll_2/out.sql

@@ -1,3 +1,3 @@
 SELECT
-  SUM(`t0`.`d`) OVER (ORDER BY `t0`.`f` ASC NULLS LAST ROWS BETWEEN 5 preceding AND 2 following) AS `foo`
+  SUM(`t0`.`d`) OVER (ORDER BY `t0`.`f` ASC ROWS BETWEEN 5 preceding AND 2 following) AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/impala/tests/snapshots/test_window/test_window_frame_specs/trailing_10/out.sql

@@ -1,3 +1,3 @@
 SELECT
-  SUM(`t0`.`d`) OVER (ORDER BY `t0`.`f` ASC NULLS LAST ROWS BETWEEN 10 preceding AND CURRENT ROW) AS `foo`
+  SUM(`t0`.`d`) OVER (ORDER BY `t0`.`f` ASC ROWS BETWEEN 10 preceding AND CURRENT ROW) AS `foo`
 FROM `alltypes` AS `t0`

ibis/backends/pandas/executor.py

@@ -69,7 +69,7 @@ def visit(cls, op: ops.Alias, arg, name):
             return arg
 
     @classmethod
-    def visit(cls, op: ops.SortKey, expr, ascending):
+    def visit(cls, op: ops.SortKey, expr, ascending, nulls_first):
         return expr
 
     @classmethod
@@ -593,11 +593,26 @@ def visit(cls, op: ops.Sort, parent, keys):
         # 2. sort the dataframe using those columns
         # 3. drop the sort key columns
         ascending = [key.ascending for key in op.keys]
+        nulls_first = [key.nulls_first for key in op.keys]
+
+        if all(nulls_first):
+            na_position = "first"
+        elif not any(nulls_first):
+            na_position = "last"
+        else:
+            raise ValueError(
+                "pandas does not support specifying null ordering for individual columns"
+            )
+
         newcols = {gen_name("sort_key"): col for col in keys}
         names = list(newcols.keys())
         df = parent.assign(**newcols)
         df = df.sort_values(
-            by=names, ascending=ascending, ignore_index=True, kind="mergesort"
+            by=names,
+            ascending=ascending,
+            na_position=na_position,
+            ignore_index=True,
+            kind="mergesort",
         )
         return df.drop(columns=names)
 

ibis/backends/polars/compiler.py

@@ -227,10 +227,9 @@ def sort(op, **kw):
 
     by = list(newcols.keys())
     descending = [key.descending for key in op.keys]
-    try:
-        lf = lf.sort(by, descending=descending, nulls_last=True)
-    except TypeError:  # pragma: no cover
-        lf = lf.sort(by, reverse=descending, nulls_last=True)  # pragma: no cover
+    nulls_last = [not key.nulls_first for key in op.keys]
+
+    lf = lf.sort(by, descending=descending, nulls_last=nulls_last)
 
     return lf.drop(*by)
 

ibis/backends/sql/compiler.py

@@ -1044,8 +1044,8 @@ def visit_Coalesce(self, op, *, arg):
 
     ### Ordering and window functions
 
-    def visit_SortKey(self, op, *, expr, ascending: bool):
-        return sge.Ordered(this=expr, desc=not ascending)
+    def visit_SortKey(self, op, *, expr, ascending: bool, nulls_first: bool):
+        return sge.Ordered(this=expr, desc=not ascending, nulls_first=nulls_first)
 
     def visit_ApproxMedian(self, op, *, arg, where):
         return self.agg.approx_quantile(arg, 0.5, where=where)

ibis/backends/sql/dialects.py

@@ -273,6 +273,8 @@ class Tokenizer(Hive.Tokenizer):
 
 
 class Impala(Hive):
+    NULL_ORDERING = "nulls_are_large"
+
     class Generator(Hive.Generator):
         TRANSFORMS = Hive.Generator.TRANSFORMS.copy() | {
             sge.ApproxDistinct: rename_func("ndv"),
@@ -336,6 +338,8 @@ def _create_sql(self, expression: sge.Create) -> str:
     return self.create_sql(expression)
 
 
+# hack around https://github.com/tobymao/sqlglot/issues/3684
+Oracle.NULL_ORDERING = "nulls_are_large"
 Oracle.Generator.TRANSFORMS |= {
     sge.LogicalOr: rename_func("max"),
     sge.LogicalAnd: rename_func("min"),

ibis/backends/tests/snapshots/test_generic/test_many_subqueries/impala/out.sql

@@ -1,12 +1,12 @@
 WITH `t1` AS (
   SELECT
     `t0`.`street`,
-    ROW_NUMBER() OVER (ORDER BY `t0`.`street` ASC NULLS LAST) - 1 AS `key`
+    ROW_NUMBER() OVER (ORDER BY `t0`.`street` ASC) - 1 AS `key`
   FROM `data` AS `t0`
 ), `t7` AS (
   SELECT
     `t6`.`street`,
-    ROW_NUMBER() OVER (ORDER BY `t6`.`street` ASC NULLS LAST) - 1 AS `key`
+    ROW_NUMBER() OVER (ORDER BY `t6`.`street` ASC) - 1 AS `key`
   FROM (
     SELECT
       `t3`.`street`,