chore: refactor IsNullOp and NotNullOp logic to make scalar ops generation easier (#1822)

* Refactor IsNullOp and NotNullOp logic

This change consolidates the definition and compilation logic for IsNullOp, isnull_op, NotNullOp, and notnull_op into a new, dedicated file: `bigframes/operations/isnull_op.py`.

Key changes include:
- Moved operator definitions from `generic_ops.py` to `isnull_op.py`.
- Moved Ibis scalar compilation logic from `scalar_op_compiler.py` to `isnull_op.py`.
- Moved Polars expression compilation logic from `polars/compiler.py` to `isnull_op.py`.
- Updated main compilers (`ScalarOpCompiler` and `PolarsExpressionCompiler`) to directly import and register the compilation functions from `isnull_op.py`.
- Ensured all internal references and naming conventions (`IsNullOp`, `isnull_op`, `NotNullOp`, `notnull_op`) are consistent with the refactored structure.

NOTE: I was unable to perform test validation (unit and system) due to missing project-specific dependencies, primarily `bigframes_vendored` and `test_utils.prefixer`. The changes are provided based on the completion of the refactoring steps as you requested.

* fix circular imports

* bad merge

* fix local pytest

* dont construct polars compiler if no polars

* limit scope to just splitting large files

* Update bigframes/core/compile/compiled.py

* revert unneeded circular import workaround

* combine null ops into generic_ops files

* revert expression change

* Update bigframes/core/compile/polars/operations/__init__.py

* skip polars test for old polars

* Update bigframes/core/compile/ibis_compiler/operations/__init__.py

* add minversion to skips

* more skips

* fix minimum polars version detection

* update colab constraints

* skip polars on 3.10

---------

Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com>

Author: tswast

bigframes/_importing.py

@@ -14,6 +14,7 @@
 import importlib
 from types import ModuleType
 
+import numpy
 from packaging import version
 
 # Keep this in sync with setup.py
@@ -22,9 +23,13 @@
 
 def import_polars() -> ModuleType:
     polars_module = importlib.import_module("polars")
-    imported_version = version.Version(polars_module.build_info()["version"])
-    if imported_version < POLARS_MIN_VERSION:
+    # Check for necessary methods instead of the version number because we
+    # can't trust the polars version until
+    # https://github.com/pola-rs/polars/issues/23940 is fixed.
+    try:
+        polars_module.lit(numpy.int64(100), dtype=polars_module.Int64())
+    except TypeError:
         raise ImportError(
-            f"Imported polars version: {imported_version} is below the minimum version: {POLARS_MIN_VERSION}"
+            f"Imported polars version is likely below the minimum version: {POLARS_MIN_VERSION}"
         )
     return polars_module

bigframes/core/compile/__init__.py

@@ -14,8 +14,8 @@
 from __future__ import annotations
 
 from bigframes.core.compile.api import test_only_ibis_inferred_schema
-from bigframes.core.compile.compiler import compile_sql
 from bigframes.core.compile.configs import CompileRequest, CompileResult
+from bigframes.core.compile.ibis_compiler.ibis_compiler import compile_sql
 
 __all__ = [
     "test_only_ibis_inferred_schema",

bigframes/core/compile/api.py

@@ -16,7 +16,7 @@
 from typing import TYPE_CHECKING
 
 from bigframes.core import rewrite
-from bigframes.core.compile import compiler
+from bigframes.core.compile.ibis_compiler import ibis_compiler
 
 if TYPE_CHECKING:
     import bigframes.core.nodes
@@ -26,9 +26,9 @@ def test_only_ibis_inferred_schema(node: bigframes.core.nodes.BigFrameNode):
     """Use only for testing paths to ensure ibis inferred schema does not diverge from bigframes inferred schema."""
     import bigframes.core.schema
 
-    node = compiler._replace_unsupported_ops(node)
+    node = ibis_compiler._replace_unsupported_ops(node)
     node = rewrite.bake_order(node)
-    ir = compiler.compile_node(node)
+    ir = ibis_compiler.compile_node(node)
     items = tuple(
         bigframes.core.schema.SchemaItem(name, ir.get_column_type(ibis_id))
         for name, ibis_id in zip(node.schema.names, ir.column_ids)

bigframes/core/compile/compiled.py

@@ -30,11 +30,10 @@
 import pyarrow as pa
 
 from bigframes.core import utils
-import bigframes.core.compile.aggregate_compiler as agg_compiler
 import bigframes.core.compile.googlesql
+import bigframes.core.compile.ibis_compiler.aggregate_compiler as agg_compiler
+import bigframes.core.compile.ibis_compiler.scalar_op_compiler as op_compilers
 import bigframes.core.compile.ibis_types
-import bigframes.core.compile.scalar_op_compiler as op_compilers
-import bigframes.core.compile.scalar_op_compiler as scalar_op_compiler
 import bigframes.core.expression as ex
 from bigframes.core.ordering import OrderingExpression
 import bigframes.core.sql
@@ -679,13 +678,15 @@ def _join_condition(
 
 
 def _as_groupable(value: ibis_types.Value):
+    from bigframes.core.compile.ibis_compiler import scalar_op_registry
+
     # Some types need to be converted to another type to enable groupby
     if value.type().is_float64():
         return value.cast(ibis_dtypes.str)
     elif value.type().is_geospatial():
         return typing.cast(ibis_types.GeoSpatialColumn, value).as_binary()
     elif value.type().is_json():
-        return scalar_op_compiler.to_json_string(value)
+        return scalar_op_registry.to_json_string(value)
     else:
         return value
 

tests/system/small/pandas/io/__init__.py …s/core/compile/ibis_compiler/__init__.pytests/system/small/pandas/io/__init__.py renamed to bigframes/core/compile/ibis_compiler/__init__.py tests/system/small/pandas/io/__init__.py renamed to bigframes/core/compile/ibis_compiler/__init__.py

@@ -11,3 +11,14 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+
+"""Compiler for BigFrames expression to Ibis expression.
+
+Make sure to import all ibis_compiler implementations here so that they get
+registered.
+"""
+
+from __future__ import annotations
+
+import bigframes.core.compile.ibis_compiler.operations.generic_ops  # noqa: F401
+import bigframes.core.compile.ibis_compiler.scalar_op_registry  # noqa: F401

…rames/core/compile/aggregate_compiler.py …pile/ibis_compiler/aggregate_compiler.pybigframes/core/compile/aggregate_compiler.py renamed to bigframes/core/compile/ibis_compiler/aggregate_compiler.py bigframes/core/compile/aggregate_compiler.py renamed to bigframes/core/compile/ibis_compiler/aggregate_compiler.py

@@ -27,8 +27,8 @@
 import pandas as pd
 
 from bigframes.core.compile import constants as compiler_constants
+import bigframes.core.compile.ibis_compiler.scalar_op_compiler as scalar_compilers
 import bigframes.core.compile.ibis_types as compile_ibis_types
-import bigframes.core.compile.scalar_op_compiler as scalar_compilers
 import bigframes.core.expression as ex
 import bigframes.core.window_spec as window_spec
 import bigframes.operations.aggregations as agg_ops

bigframes/core/compile/compiler.py …e/compile/ibis_compiler/ibis_compiler.pybigframes/core/compile/compiler.py renamed to bigframes/core/compile/ibis_compiler/ibis_compiler.py bigframes/core/compile/compiler.py renamed to bigframes/core/compile/ibis_compiler/ibis_compiler.py

@@ -29,7 +29,6 @@
 import bigframes.core.compile.concat as concat_impl
 import bigframes.core.compile.configs as configs
 import bigframes.core.compile.explode
-import bigframes.core.compile.scalar_op_compiler as compile_scalar
 import bigframes.core.nodes as nodes
 import bigframes.core.ordering as bf_ordering
 import bigframes.core.rewrite as rewrites
@@ -178,6 +177,8 @@ def compile_readlocal(node: nodes.ReadLocalNode, *args):
 
 @_compile_node.register
 def compile_readtable(node: nodes.ReadTableNode, *args):
+    from bigframes.core.compile.ibis_compiler import scalar_op_registry
+
     ibis_table = _table_to_ibis(
         node.source, scan_cols=[col.source_id for col in node.scan_list.items]
     )
@@ -188,7 +189,7 @@ def compile_readtable(node: nodes.ReadTableNode, *args):
             scan_item.dtype == dtypes.JSON_DTYPE
             and ibis_table[scan_item.source_id].type() == ibis_dtypes.string
         ):
-            json_column = compile_scalar.parse_json(
+            json_column = scalar_op_registry.parse_json(
                 ibis_table[scan_item.source_id]
             ).name(scan_item.source_id)
             ibis_table = ibis_table.mutate(json_column)

…s/system/small/pandas/io/api/__init__.py …ile/ibis_compiler/operations/__init__.pytests/system/small/pandas/io/api/__init__.py renamed to bigframes/core/compile/ibis_compiler/operations/__init__.py tests/system/small/pandas/io/api/__init__.py renamed to bigframes/core/compile/ibis_compiler/operations/__init__.py

@@ -11,3 +11,11 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+
+"""Operation implementations for the Ibis-based compiler.
+
+This directory structure should reflect the same layout as the
+`bigframes/operations` directory where the operations are defined.
+
+Prefer a few ops per file to keep file sizes manageable for text editors and LLMs.
+"""

bigframes/core/compile/ibis_compiler/operations/generic_ops.py

@@ -0,0 +1,38 @@
+# Copyright 2025 Google LLC
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+BigFrames -> Ibis compilation for the operations in bigframes.operations.generic_ops.
+
+Please keep implementations in sequential order by op name.
+"""
+
+from __future__ import annotations
+
+from bigframes_vendored.ibis.expr import types as ibis_types
+
+from bigframes.core.compile.ibis_compiler import scalar_op_compiler
+from bigframes.operations import generic_ops
+
+register_unary_op = scalar_op_compiler.scalar_op_compiler.register_unary_op
+
+
+@register_unary_op(generic_ops.notnull_op)
+def notnull_op_impl(x: ibis_types.Value):
+    return x.notnull()
+
+
+@register_unary_op(generic_ops.isnull_op)
+def isnull_op_impl(x: ibis_types.Value):
+    return x.isnull()

bigframes/core/compile/ibis_compiler/scalar_op_compiler.py

@@ -0,0 +1,207 @@
+# Copyright 2023 Google LLC
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""To avoid circular imports, this module should _not_ depend on any ops."""
+
+from __future__ import annotations
+
+import functools
+import typing
+from typing import TYPE_CHECKING
+
+import bigframes_vendored.ibis.expr.types as ibis_types
+
+import bigframes.core.compile.ibis_types
+import bigframes.core.expression as ex
+
+if TYPE_CHECKING:
+    import bigframes.operations as ops
+
+
+class ScalarOpCompiler:
+    # Mapping of operation name to implemenations
+    _registry: dict[
+        str,
+        typing.Callable[
+            [typing.Sequence[ibis_types.Value], ops.RowOp], ibis_types.Value
+        ],
+    ] = {}
+
+    @functools.singledispatchmethod
+    def compile_expression(
+        self,
+        expression: ex.Expression,
+        bindings: typing.Dict[str, ibis_types.Value],
+    ) -> ibis_types.Value:
+        raise NotImplementedError(f"Unrecognized expression: {expression}")
+
+    @compile_expression.register
+    def _(
+        self,
+        expression: ex.ScalarConstantExpression,
+        bindings: typing.Dict[str, ibis_types.Value],
+    ) -> ibis_types.Value:
+        return bigframes.core.compile.ibis_types.literal_to_ibis_scalar(
+            expression.value, expression.dtype
+        )
+
+    @compile_expression.register
+    def _(
+        self,
+        expression: ex.DerefOp,
+        bindings: typing.Dict[str, ibis_types.Value],
+    ) -> ibis_types.Value:
+        if expression.id.sql not in bindings:
+            raise ValueError(f"Could not resolve unbound variable {expression.id}")
+        else:
+            return bindings[expression.id.sql]
+
+    @compile_expression.register
+    def _(
+        self,
+        expression: ex.OpExpression,
+        bindings: typing.Dict[str, ibis_types.Value],
+    ) -> ibis_types.Value:
+        inputs = [
+            self.compile_expression(sub_expr, bindings)
+            for sub_expr in expression.inputs
+        ]
+        return self.compile_row_op(expression.op, inputs)
+
+    def compile_row_op(
+        self, op: ops.RowOp, inputs: typing.Sequence[ibis_types.Value]
+    ) -> ibis_types.Value:
+        impl = self._registry[op.name]
+        return impl(inputs, op)
+
+    def register_unary_op(
+        self,
+        op_ref: typing.Union[ops.UnaryOp, type[ops.UnaryOp]],
+        pass_op: bool = False,
+    ):
+        """
+        Decorator to register a unary op implementation.
+
+        Args:
+            op_ref (UnaryOp or UnaryOp type):
+                Class or instance of operator that is implemented by the decorated function.
+            pass_op (bool):
+                Set to true if implementation takes the operator object as the last argument.
+                This is needed for parameterized ops where parameters are part of op object.
+        """
+        key = typing.cast(str, op_ref.name)
+
+        def decorator(impl: typing.Callable[..., ibis_types.Value]):
+            def normalized_impl(args: typing.Sequence[ibis_types.Value], op: ops.RowOp):
+                if pass_op:
+                    return impl(args[0], op)
+                else:
+                    return impl(args[0])
+
+            self._register(key, normalized_impl)
+            return impl
+
+        return decorator
+
+    def register_binary_op(
+        self,
+        op_ref: typing.Union[ops.BinaryOp, type[ops.BinaryOp]],
+        pass_op: bool = False,
+    ):
+        """
+        Decorator to register a binary op implementation.
+
+        Args:
+            op_ref (BinaryOp or BinaryOp type):
+                Class or instance of operator that is implemented by the decorated function.
+            pass_op (bool):
+                Set to true if implementation takes the operator object as the last argument.
+                This is needed for parameterized ops where parameters are part of op object.
+        """
+        key = typing.cast(str, op_ref.name)
+
+        def decorator(impl: typing.Callable[..., ibis_types.Value]):
+            def normalized_impl(args: typing.Sequence[ibis_types.Value], op: ops.RowOp):
+                if pass_op:
+                    return impl(args[0], args[1], op)
+                else:
+                    return impl(args[0], args[1])
+
+            self._register(key, normalized_impl)
+            return impl
+
+        return decorator
+
+    def register_ternary_op(
+        self, op_ref: typing.Union[ops.TernaryOp, type[ops.TernaryOp]]
+    ):
+        """
+        Decorator to register a ternary op implementation.
+
+        Args:
+            op_ref (TernaryOp or TernaryOp type):
+                Class or instance of operator that is implemented by the decorated function.
+        """
+        key = typing.cast(str, op_ref.name)
+
+        def decorator(impl: typing.Callable[..., ibis_types.Value]):
+            def normalized_impl(args: typing.Sequence[ibis_types.Value], op: ops.RowOp):
+                return impl(args[0], args[1], args[2])
+
+            self._register(key, normalized_impl)
+            return impl
+
+        return decorator
+
+    def register_nary_op(
+        self, op_ref: typing.Union[ops.NaryOp, type[ops.NaryOp]], pass_op: bool = False
+    ):
+        """
+        Decorator to register a nary op implementation.
+
+        Args:
+            op_ref (NaryOp or NaryOp type):
+                Class or instance of operator that is implemented by the decorated function.
+            pass_op (bool):
+                Set to true if implementation takes the operator object as the last argument.
+                This is needed for parameterized ops where parameters are part of op object.
+        """
+        key = typing.cast(str, op_ref.name)
+
+        def decorator(impl: typing.Callable[..., ibis_types.Value]):
+            def normalized_impl(args: typing.Sequence[ibis_types.Value], op: ops.RowOp):
+                if pass_op:
+                    return impl(*args, op=op)
+                else:
+                    return impl(*args)
+
+            self._register(key, normalized_impl)
+            return impl
+
+        return decorator
+
+    def _register(
+        self,
+        op_name: str,
+        impl: typing.Callable[
+            [typing.Sequence[ibis_types.Value], ops.RowOp], ibis_types.Value
+        ],
+    ):
+        if op_name in self._registry:
+            raise ValueError(f"Operation name {op_name} already registered")
+        self._registry[op_name] = impl
+
+
+# Singleton compiler
+scalar_op_compiler = ScalarOpCompiler()