|
| 1 | +# Copyright 2025 Google LLC |
| 2 | +# |
| 3 | +# Licensed under the Apache License, Version 2.0 (the "License"); |
| 4 | +# you may not use this file except in compliance with the License. |
| 5 | +# You may obtain a copy of the License at |
| 6 | +# |
| 7 | +# http://www.apache.org/licenses/LICENSE-2.0 |
| 8 | +# |
| 9 | +# Unless required by applicable law or agreed to in writing, software |
| 10 | +# distributed under the License is distributed on an "AS IS" BASIS, |
| 11 | +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 12 | +# See the License for the specific language governing permissions and |
| 13 | +# limitations under the License. |
| 14 | +from __future__ import annotations |
| 15 | + |
| 16 | +import typing |
| 17 | + |
| 18 | +import sqlglot.expressions as sge |
| 19 | + |
| 20 | +from bigframes.core import utils, window_spec |
| 21 | +import bigframes.core.compile.sqlglot.scalar_compiler as scalar_compiler |
| 22 | +import bigframes.core.ordering as ordering_spec |
| 23 | + |
| 24 | + |
| 25 | +def apply_window_if_present( |
| 26 | + value: sge.Expression, |
| 27 | + window: typing.Optional[window_spec.WindowSpec] = None, |
| 28 | +) -> sge.Expression: |
| 29 | + if window is None: |
| 30 | + return value |
| 31 | + |
| 32 | + if window.is_row_bounded and not window.ordering: |
| 33 | + raise ValueError("No ordering provided for ordered analytic function") |
| 34 | + elif ( |
| 35 | + not window.is_row_bounded |
| 36 | + and not window.is_range_bounded |
| 37 | + and not window.ordering |
| 38 | + ): |
| 39 | + # Unbound grouping window. |
| 40 | + order_by = None |
| 41 | + elif window.is_range_bounded: |
| 42 | + # Note that, when the window is range-bounded, we only need one ordering key. |
| 43 | + # There are two reasons: |
| 44 | + # 1. Manipulating null positions requires more than one ordering key, which |
| 45 | + # is forbidden by SQL window syntax for range rolling. |
| 46 | + # 2. Pandas does not allow range rolling on timeseries with nulls. |
| 47 | + order_by = get_window_order_by((window.ordering[0],), override_null_order=False) |
| 48 | + else: |
| 49 | + order_by = get_window_order_by(window.ordering, override_null_order=True) |
| 50 | + |
| 51 | + order = sge.Order(expressions=order_by) if order_by else None |
| 52 | + |
| 53 | + group_by = ( |
| 54 | + [scalar_compiler.compile_scalar_expression(key) for key in window.grouping_keys] |
| 55 | + if window.grouping_keys |
| 56 | + else None |
| 57 | + ) |
| 58 | + |
| 59 | + # This is the key change. Don't create a spec for the default window frame |
| 60 | + # if there's no ordering. This avoids generating an `ORDER BY NULL` clause. |
| 61 | + if not window.bounds and not order: |
| 62 | + return sge.Window(this=value, partition_by=group_by) |
| 63 | + |
| 64 | + kind = ( |
| 65 | + "ROWS" if isinstance(window.bounds, window_spec.RowsWindowBounds) else "RANGE" |
| 66 | + ) |
| 67 | + |
| 68 | + start: typing.Union[int, float, None] = None |
| 69 | + end: typing.Union[int, float, None] = None |
| 70 | + if isinstance(window.bounds, window_spec.RangeWindowBounds): |
| 71 | + if window.bounds.start is not None: |
| 72 | + start = utils.timedelta_to_micros(window.bounds.start) |
| 73 | + if window.bounds.end is not None: |
| 74 | + end = utils.timedelta_to_micros(window.bounds.end) |
| 75 | + elif window.bounds: |
| 76 | + start = window.bounds.start |
| 77 | + end = window.bounds.end |
| 78 | + |
| 79 | + start_value, start_side = _get_window_bounds(start, is_preceding=True) |
| 80 | + end_value, end_side = _get_window_bounds(end, is_preceding=False) |
| 81 | + |
| 82 | + spec = sge.WindowSpec( |
| 83 | + kind=kind, |
| 84 | + start=start_value, |
| 85 | + start_side=start_side, |
| 86 | + end=end_value, |
| 87 | + end_side=end_side, |
| 88 | + over="OVER", |
| 89 | + ) |
| 90 | + |
| 91 | + return sge.Window(this=value, partition_by=group_by, order=order, spec=spec) |
| 92 | + |
| 93 | + |
| 94 | +def get_window_order_by( |
| 95 | + ordering: typing.Tuple[ordering_spec.OrderingExpression, ...], |
| 96 | + override_null_order: bool = False, |
| 97 | +) -> typing.Optional[tuple[sge.Ordered, ...]]: |
| 98 | + """Returns the SQL order by clause for a window specification.""" |
| 99 | + if not ordering: |
| 100 | + return None |
| 101 | + |
| 102 | + order_by = [] |
| 103 | + for ordering_spec_item in ordering: |
| 104 | + expr = scalar_compiler.compile_scalar_expression( |
| 105 | + ordering_spec_item.scalar_expression |
| 106 | + ) |
| 107 | + desc = not ordering_spec_item.direction.is_ascending |
| 108 | + nulls_first = not ordering_spec_item.na_last |
| 109 | + |
| 110 | + if override_null_order: |
| 111 | + # Bigquery SQL considers NULLS to be "smallest" values, but we need |
| 112 | + # to override in these cases. |
| 113 | + is_null_expr = sge.Is(this=expr, expression=sge.Null()) |
| 114 | + if nulls_first and desc: |
| 115 | + order_by.append( |
| 116 | + sge.Ordered( |
| 117 | + this=is_null_expr, |
| 118 | + desc=desc, |
| 119 | + nulls_first=nulls_first, |
| 120 | + ) |
| 121 | + ) |
| 122 | + elif not nulls_first and not desc: |
| 123 | + order_by.append( |
| 124 | + sge.Ordered( |
| 125 | + this=is_null_expr, |
| 126 | + desc=desc, |
| 127 | + nulls_first=nulls_first, |
| 128 | + ) |
| 129 | + ) |
| 130 | + |
| 131 | + order_by.append( |
| 132 | + sge.Ordered( |
| 133 | + this=expr, |
| 134 | + desc=desc, |
| 135 | + nulls_first=nulls_first, |
| 136 | + ) |
| 137 | + ) |
| 138 | + return tuple(order_by) |
| 139 | + |
| 140 | + |
| 141 | +def _get_window_bounds( |
| 142 | + value, is_preceding: bool |
| 143 | +) -> tuple[typing.Union[str, sge.Expression], typing.Optional[str]]: |
| 144 | + """Compiles a single boundary value into its SQL components.""" |
| 145 | + if value is None: |
| 146 | + side = "PRECEDING" if is_preceding else "FOLLOWING" |
| 147 | + return "UNBOUNDED", side |
| 148 | + |
| 149 | + if value == 0: |
| 150 | + return "CURRENT ROW", None |
| 151 | + |
| 152 | + side = "PRECEDING" if value < 0 else "FOLLOWING" |
| 153 | + return sge.convert(abs(value)), side |
0 commit comments