Fix Literal partitioning in cudf-polars (#19160)

Closes #19147

When a `Select` does not depend on any columns from the child, we must avoid using multiple partitions.

Authors:
  - Richard (Rick) Zamora (https://github.com/rjzamora)

Approvers:
  - Tom Augspurger (https://github.com/TomAugspurger)

URL: #19160

Author: rjzamora

python/cudf_polars/cudf_polars/experimental/select.py

@@ -9,7 +9,8 @@
 import pylibcudf as plc
 
 from cudf_polars.dsl import expr
-from cudf_polars.dsl.ir import HConcat, Scan, Select, Union
+from cudf_polars.dsl.expr import Col, Len
+from cudf_polars.dsl.ir import Empty, HConcat, Scan, Select, Union
 from cudf_polars.dsl.traversal import traversal
 from cudf_polars.experimental.base import PartitionInfo
 from cudf_polars.experimental.dispatch import lower_ir_node
@@ -116,6 +117,7 @@ def _(
         and isinstance(child.children[0], Scan)
         and child.children[0].predicate is None
     ):
+        # Special Case: Fast count.
         scan = child.children[0]
         count = scan.fast_count()
         dtype = ir.exprs[0].value.dtype
@@ -135,9 +137,18 @@ def _(
         partition_info[new_node] = PartitionInfo(count=1)
         return new_node, partition_info
 
+    if not any(
+        isinstance(expr, (Col, Len)) for expr in traversal([e.value for e in ir.exprs])
+    ):
+        # Special Case: Selection does not depend on any columns.
+        new_node = ir.reconstruct([input_ir := Empty()])
+        partition_info[input_ir] = partition_info[new_node] = PartitionInfo(count=1)
+        return new_node, partition_info
+
     if pi.count > 1 and not all(
         expr.is_pointwise for expr in traversal([e.value for e in ir.exprs])
     ):
+        # Special Case: Multiple partitions with 1+ non-pointwise expressions.
         try:
             # Try decomposing the underlying expressions
             return decompose_select(

python/cudf_polars/tests/experimental/test_select.py

@@ -124,3 +124,10 @@ def test_select_parquet_fast_count(request, tmp_path, df, engine):
     df.collect().write_parquet(file)
     q = pl.scan_parquet(file).select(pl.len())
     assert_gpu_result_equal(q, engine=engine)
+
+
+def test_select_literal(engine):
+    # See: https://github.com/rapidsai/cudf/issues/19147
+    ldf = pl.LazyFrame({"a": list(range(10))})
+    q = ldf.select(pl.lit(2).pow(pl.lit(-3, dtype=pl.Float32)))
+    assert_gpu_result_equal(q, engine=engine)