Coverage for python/lsst/daf/butler/registry/queries/expressions/_predicate.py: 13%
209 statements
« prev ^ index » next coverage.py v7.5.1, created at 2024-05-11 03:16 -0700
« prev ^ index » next coverage.py v7.5.1, created at 2024-05-11 03:16 -0700
1# This file is part of daf_butler.
2#
3# Developed for the LSST Data Management System.
4# This product includes software developed by the LSST Project
5# (http://www.lsst.org).
6# See the COPYRIGHT file at the top-level directory of this distribution
7# for details of code ownership.
8#
9# This software is dual licensed under the GNU General Public License and also
10# under a 3-clause BSD license. Recipients may choose which of these licenses
11# to use; please see the files gpl-3.0.txt and/or bsd_license.txt,
12# respectively. If you choose the GPL option then the following text applies
13# (but note that there is still no warranty even if you opt for BSD instead):
14#
15# This program is free software: you can redistribute it and/or modify
16# it under the terms of the GNU General Public License as published by
17# the Free Software Foundation, either version 3 of the License, or
18# (at your option) any later version.
19#
20# This program is distributed in the hope that it will be useful,
21# but WITHOUT ANY WARRANTY; without even the implied warranty of
22# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
23# GNU General Public License for more details.
24#
25# You should have received a copy of the GNU General Public License
26# along with this program. If not, see <http://www.gnu.org/licenses/>.
27from __future__ import annotations
29__all__ = ("make_string_expression_predicate", "ExpressionTypeError")
31import builtins
32import datetime
33import types
34import warnings
35from collections.abc import Mapping, Set
36from typing import Any, cast
38import astropy.time
39import astropy.utils.exceptions
40from lsst.daf.relation import (
41 ColumnContainer,
42 ColumnExpression,
43 ColumnExpressionSequence,
44 ColumnLiteral,
45 ColumnTag,
46 Predicate,
47 sql,
48)
50# We import the some modules rather than types within them because match syntax
51# uses qualified names with periods to distinguish literals from captures.
52from .... import _timespan, timespan_database_representation
53from ...._column_tags import DatasetColumnTag, DimensionKeyColumnTag, DimensionRecordColumnTag
54from ...._column_type_info import ColumnTypeInfo
55from ....dimensions import DataCoordinate, Dimension, DimensionGroup, DimensionUniverse
56from ..._exceptions import UserExpressionError, UserExpressionSyntaxError
57from .categorize import ExpressionConstant, categorizeConstant, categorizeElementId
58from .check import CheckVisitor
59from .normalForm import NormalForm, NormalFormExpression
60from .parser import Node, TreeVisitor, parse_expression
62# As of astropy 4.2, the erfa interface is shipped independently and
63# ErfaWarning is no longer an AstropyWarning
64try:
65 import erfa
66except ImportError:
67 erfa = None
70class ExpressionTypeError(TypeError):
71 """Exception raised when the types in a query expression are not
72 compatible with the operators or other syntax.
73 """
76def make_string_expression_predicate(
77 string: str,
78 dimensions: DimensionGroup,
79 *,
80 column_types: ColumnTypeInfo,
81 bind: Mapping[str, Any] | None = None,
82 data_id: DataCoordinate | None = None,
83 defaults: DataCoordinate | None = None,
84 dataset_type_name: str | None = None,
85 allow_orphans: bool = False,
86) -> tuple[Predicate | None, Mapping[str, Set[str]]]:
87 """Create a predicate by parsing and analyzing a string expression.
89 Parameters
90 ----------
91 string : `str`
92 String to parse.
93 dimensions : `DimensionGroup`
94 The dimensions the query would include in the absence of this WHERE
95 expression.
96 column_types : `ColumnTypeInfo`
97 Information about column types.
98 bind : `~collections.abc.Mapping` [ `str`, `Any` ], optional
99 Literal values referenced in the expression.
100 data_id : `DataCoordinate`, optional
101 A fully-expanded data ID identifying dimensions known in advance.
102 If not provided, will be set to an empty data ID.
103 ``dataId.hasRecords()`` must return `True`.
104 defaults : `DataCoordinate`, optional
105 A data ID containing default for governor dimensions. Ignored
106 unless ``check=True``.
107 dataset_type_name : `str` or `None`, optional
108 The name of the dataset type to assume for unqualified dataset
109 columns, or `None` if there are no such identifiers.
110 allow_orphans : `bool`, optional
111 If `True`, permit expressions to refer to dimensions without
112 providing a value for their governor dimensions (e.g. referring to
113 a visit without an instrument). Should be left to default to
114 `False` in essentially all new code.
116 Returns
117 -------
118 predicate : `lsst.daf.relation.colum_expressions.Predicate` or `None`
119 New predicate derived from the string expression, or `None` if the
120 string is empty.
121 governor_constraints : `~collections.abc.Mapping` [ `str` , \
122 `~collections.abc.Set` ]
123 Constraints on dimension values derived from the expression and data
124 ID.
125 """
126 governor_constraints: dict[str, Set[str]] = {}
127 if data_id is None:
128 data_id = DataCoordinate.make_empty(dimensions.universe)
129 if not string:
130 for dimension in data_id.dimensions.governors:
131 governor_constraints[dimension] = {cast(str, data_id[dimension])}
132 return None, governor_constraints
133 try:
134 tree = parse_expression(string)
135 except Exception as exc:
136 raise UserExpressionSyntaxError(f"Failed to parse user expression {string!r}.") from exc
137 if bind is None:
138 bind = {}
139 if bind:
140 for identifier in bind:
141 if identifier in dimensions.universe.elements.names:
142 raise RuntimeError(f"Bind parameter key {identifier!r} conflicts with a dimension element.")
143 table, _, column = identifier.partition(".")
144 if column and table in dimensions.universe.elements.names:
145 raise RuntimeError(f"Bind parameter key {identifier!r} looks like a dimension column.")
146 if defaults is None:
147 defaults = DataCoordinate.make_empty(dimensions.universe)
148 # Convert the expression to disjunctive normal form (ORs of ANDs).
149 # That's potentially super expensive in the general case (where there's
150 # a ton of nesting of ANDs and ORs). That won't be the case for the
151 # expressions we expect, and we actually use disjunctive normal instead
152 # of conjunctive (i.e. ANDs of ORs) because I think the worst-case is
153 # a long list of OR'd-together data IDs, which is already in or very
154 # close to disjunctive normal form.
155 expr = NormalFormExpression.fromTree(tree, NormalForm.DISJUNCTIVE)
156 # Check the expression for consistency and completeness.
157 visitor = CheckVisitor(data_id, dimensions, bind, defaults, allow_orphans=allow_orphans)
158 try:
159 summary = expr.visit(visitor)
160 except UserExpressionError as err:
161 exprOriginal = str(tree)
162 exprNormal = str(expr.toTree())
163 if exprNormal == exprOriginal:
164 msg = f'Error in query expression "{exprOriginal}": {err}'
165 else:
166 msg = f'Error in query expression "{exprOriginal}" (normalized to "{exprNormal}"): {err}'
167 raise UserExpressionError(msg) from None
168 for dimension_name, values in summary.dimension_constraints.items():
169 if dimension_name in dimensions.universe.governor_dimensions.names:
170 governor_constraints[dimension_name] = cast(Set[str], values)
171 converter = PredicateConversionVisitor(bind, dataset_type_name, dimensions.universe, column_types)
172 predicate = tree.visit(converter)
173 return predicate, governor_constraints
176VisitorResult = Predicate | ColumnExpression | ColumnContainer
179class PredicateConversionVisitor(TreeVisitor[VisitorResult]):
180 def __init__(
181 self,
182 bind: Mapping[str, Any],
183 dataset_type_name: str | None,
184 universe: DimensionUniverse,
185 column_types: ColumnTypeInfo,
186 ):
187 self.bind = bind
188 self.dataset_type_name = dataset_type_name
189 self.universe = universe
190 self.column_types = column_types
192 OPERATOR_MAP = {
193 "=": "__eq__",
194 "!=": "__ne__",
195 "<": "__lt__",
196 ">": "__gt__",
197 "<=": "__le__",
198 ">=": "__ge__",
199 "+": "__add__",
200 "-": "__sub__",
201 "*": "__mul__",
202 }
204 def to_datetime(self, time: astropy.time.Time) -> datetime.datetime:
205 with warnings.catch_warnings():
206 warnings.simplefilter("ignore", category=astropy.utils.exceptions.AstropyWarning)
207 if erfa is not None:
208 warnings.simplefilter("ignore", category=erfa.ErfaWarning)
209 return time.to_datetime()
211 def visitBinaryOp(
212 self, operator: str, lhs: VisitorResult, rhs: VisitorResult, node: Node
213 ) -> VisitorResult:
214 # Docstring inherited.
215 b = builtins
216 match (operator, lhs, rhs):
217 case ["OR", Predicate() as lhs, Predicate() as rhs]:
218 return lhs.logical_or(rhs)
219 case ["AND", Predicate() as lhs, Predicate() as rhs]:
220 return lhs.logical_and(rhs)
221 # Allow all comparisons between expressions of the same type for
222 # sortable types.
223 case [
224 "=" | "!=" | "<" | ">" | "<=" | ">=",
225 ColumnExpression(
226 dtype=b.int | b.float | b.str | astropy.time.Time | datetime.datetime
227 ) as lhs,
228 ColumnExpression() as rhs,
229 ] if lhs.dtype is rhs.dtype:
230 return lhs.predicate_method(self.OPERATOR_MAP[operator], rhs)
231 # Allow comparisons between datetime expressions and
232 # astropy.time.Time literals/binds (only), by coercing the
233 # astropy.time.Time version to datetime.
234 case [
235 "=" | "!=" | "<" | ">" | "<=" | ">=",
236 ColumnLiteral(dtype=astropy.time.Time) as lhs,
237 ColumnExpression(dtype=datetime.datetime) as rhs,
238 ]:
239 lhs = ColumnLiteral(self.to_datetime(lhs.value), datetime.datetime)
240 return lhs.predicate_method(self.OPERATOR_MAP[operator], rhs)
241 case [
242 "=" | "!=" | "<" | ">" | "<=" | ">=",
243 ColumnExpression(dtype=datetime.datetime) as lhs,
244 ColumnLiteral(dtype=astropy.time.Time) as rhs,
245 ]:
246 rhs = ColumnLiteral(self.to_datetime(rhs.value), datetime.datetime)
247 return lhs.predicate_method(self.OPERATOR_MAP[operator], rhs)
248 # Allow comparisons between astropy.time.Time expressions and
249 # datetime literals/binds, by coercing the
250 # datetime literals to astropy.time.Time (in UTC scale).
251 case [
252 "=" | "!=" | "<" | ">" | "<=" | ">=",
253 ColumnLiteral(dtype=datetime.datetime) as lhs,
254 ColumnExpression(dtype=astropy.time.Time) as rhs,
255 ]:
256 lhs = ColumnLiteral(astropy.time.Time(lhs.value, scale="utc"), astropy.time.Time)
257 return lhs.predicate_method(self.OPERATOR_MAP[operator], rhs)
258 case [
259 "=" | "!=" | "<" | ">" | "<=" | ">=",
260 ColumnExpression(dtype=astropy.time.Time) as lhs,
261 ColumnLiteral(dtype=datetime.datetime) as rhs,
262 ]:
263 rhs = ColumnLiteral(astropy.time.Time(rhs.value, scale="utc"), astropy.time.Time)
264 return lhs.predicate_method(self.OPERATOR_MAP[operator], rhs)
265 # Allow equality comparisons with None/NULL. We don't have an 'IS'
266 # operator.
267 case ["=" | "!=", ColumnExpression(dtype=types.NoneType) as lhs, ColumnExpression() as rhs]:
268 return lhs.predicate_method(self.OPERATOR_MAP[operator], rhs)
269 case ["=" | "!=", ColumnExpression() as lhs, ColumnExpression(dtype=types.NoneType) as rhs]:
270 return lhs.predicate_method(self.OPERATOR_MAP[operator], rhs)
271 # Comparisions between Time and Timespan need have the Timespan on
272 # the lhs, since that (actually TimespanDatabaseRepresentation) is
273 # what actually has the methods.
274 case [
275 "<",
276 ColumnExpression(dtype=astropy.time.Time) as lhs,
277 ColumnExpression(dtype=_timespan.Timespan) as rhs,
278 ]:
279 return rhs.predicate_method(self.OPERATOR_MAP[">"], lhs)
280 case [
281 ">",
282 ColumnExpression(dtype=astropy.time.Time) as lhs,
283 ColumnExpression(dtype=_timespan.Timespan) as rhs,
284 ]:
285 return rhs.predicate_method(self.OPERATOR_MAP["<"], lhs)
286 # Enable other comparisons between times and Timespans (many of the
287 # combinations matched by this branch will have already been
288 # covered by a preceding branch).
289 case [
290 "<" | ">",
291 ColumnExpression(dtype=_timespan.Timespan | astropy.time.Time) as lhs,
292 ColumnExpression(dtype=_timespan.Timespan | astropy.time.Time) as rhs,
293 ]:
294 return lhs.predicate_method(self.OPERATOR_MAP[operator], rhs)
295 # Enable "overlaps" operations between timespans, and between times
296 # and timespans. The latter resolve to the `Timespan.contains` or
297 # `TimespanDatabaseRepresentation.contains` methods, but we use
298 # OVERLAPS in the string expression language to keep that simple.
299 case [
300 "OVERLAPS",
301 ColumnExpression(dtype=_timespan.Timespan) as lhs,
302 ColumnExpression(dtype=_timespan.Timespan) as rhs,
303 ]:
304 return lhs.predicate_method("overlaps", rhs)
305 case [
306 "OVERLAPS",
307 ColumnExpression(dtype=_timespan.Timespan) as lhs,
308 ColumnExpression(dtype=astropy.time.Time) as rhs,
309 ]:
310 return lhs.predicate_method("overlaps", rhs)
311 case [
312 "OVERLAPS",
313 ColumnExpression(dtype=astropy.time.Time) as lhs,
314 ColumnExpression(dtype=_timespan.Timespan) as rhs,
315 ]:
316 return rhs.predicate_method("overlaps", lhs)
317 # Enable arithmetic operators on numeric types, without any type
318 # coercion or broadening.
319 case [
320 "+" | "-" | "*",
321 ColumnExpression(dtype=b.int | b.float) as lhs,
322 ColumnExpression() as rhs,
323 ] if lhs.dtype is rhs.dtype:
324 return lhs.method(self.OPERATOR_MAP[operator], rhs, dtype=lhs.dtype)
325 case ["/", ColumnExpression(dtype=b.float) as lhs, ColumnExpression(dtype=b.float) as rhs]:
326 return lhs.method("__truediv__", rhs, dtype=b.float)
327 case ["/", ColumnExpression(dtype=b.int) as lhs, ColumnExpression(dtype=b.int) as rhs]:
328 # SQLAlchemy maps Python's '/' (__truediv__) operator directly
329 # to SQL's '/', despite those being defined differently for
330 # integers. Our expression language uses the SQL definition,
331 # and we only care about these expressions being evaluated in
332 # SQL right now, but we still want to guard against it being
333 # evaluated in Python and producing a surprising answer, so we
334 # mark it as being supported only by a SQL engine.
335 return lhs.method(
336 "__truediv__",
337 rhs,
338 dtype=b.int,
339 supporting_engine_types={sql.Engine},
340 )
341 case ["%", ColumnExpression(dtype=b.int) as lhs, ColumnExpression(dtype=b.int) as rhs]:
342 return lhs.method("__mod__", rhs, dtype=b.int)
343 assert (
344 lhs.dtype is not None and rhs.dtype is not None
345 ), "Expression converter should not yield untyped nodes."
346 raise ExpressionTypeError(
347 f"Invalid types {lhs.dtype.__name__}, {rhs.dtype.__name__} for binary operator {operator!r} "
348 f"in expression {node!s}."
349 )
351 def visitIdentifier(self, name: str, node: Node) -> VisitorResult:
352 # Docstring inherited.
353 if name in self.bind:
354 value = self.bind[name]
355 if isinstance(value, list | tuple | Set):
356 elements = []
357 all_dtypes = set()
358 for item in value:
359 dtype = type(item)
360 all_dtypes.add(dtype)
361 elements.append(ColumnExpression.literal(item, dtype=dtype))
362 if len(all_dtypes) > 1:
363 raise ExpressionTypeError(
364 f"Mismatched types in bind iterable: {value} has a mix of {all_dtypes}."
365 )
366 elif not elements:
367 # Empty container
368 return ColumnContainer.sequence([])
369 else:
370 (dtype,) = all_dtypes
371 return ColumnContainer.sequence(elements, dtype=dtype)
372 return ColumnExpression.literal(value, dtype=type(value))
373 tag: ColumnTag
374 match categorizeConstant(name):
375 case ExpressionConstant.INGEST_DATE:
376 assert self.dataset_type_name is not None
377 tag = DatasetColumnTag(self.dataset_type_name, "ingest_date")
378 return ColumnExpression.reference(tag, self.column_types.ingest_date_pytype)
379 case ExpressionConstant.NULL:
380 return ColumnExpression.literal(None, type(None))
381 case None:
382 pass
383 case _:
384 raise AssertionError("Check for enum values should be exhaustive.")
385 element, column = categorizeElementId(self.universe, name)
386 if column is not None:
387 tag = DimensionRecordColumnTag(element.name, column)
388 dtype = (
389 _timespan.Timespan
390 if column == timespan_database_representation.TimespanDatabaseRepresentation.NAME
391 else element.RecordClass.fields.standard[column].getPythonType()
392 )
393 return ColumnExpression.reference(tag, dtype)
394 else:
395 tag = DimensionKeyColumnTag(element.name)
396 assert isinstance(element, Dimension)
397 return ColumnExpression.reference(tag, element.primaryKey.getPythonType())
399 def visitIsIn(
400 self, lhs: VisitorResult, values: list[VisitorResult], not_in: bool, node: Node
401 ) -> VisitorResult:
402 # Docstring inherited.
403 clauses: list[Predicate] = []
404 items: list[ColumnExpression] = []
405 assert isinstance(lhs, ColumnExpression), "LHS of IN guaranteed to be scalar by parser."
406 for rhs_item in values:
407 match rhs_item:
408 case ColumnExpressionSequence(
409 items=rhs_items, dtype=rhs_dtype
410 ) if rhs_dtype is None or rhs_dtype == lhs.dtype:
411 items.extend(rhs_items)
412 case ColumnContainer(dtype=lhs.dtype):
413 clauses.append(rhs_item.contains(lhs))
414 case ColumnExpression(dtype=lhs.dtype):
415 items.append(rhs_item)
416 case _:
417 raise ExpressionTypeError(
418 f"Invalid type {rhs_item.dtype} for element in {lhs.dtype} IN expression '{node}'."
419 )
420 if items:
421 clauses.append(ColumnContainer.sequence(items, dtype=lhs.dtype).contains(lhs))
422 result = Predicate.logical_or(*clauses)
423 if not_in:
424 result = result.logical_not()
425 return result
427 def visitNumericLiteral(self, value: str, node: Node) -> VisitorResult:
428 # Docstring inherited.
429 try:
430 return ColumnExpression.literal(int(value), dtype=int)
431 except ValueError:
432 return ColumnExpression.literal(float(value), dtype=float)
434 def visitParens(self, expression: VisitorResult, node: Node) -> VisitorResult:
435 # Docstring inherited.
436 return expression
438 def visitPointNode(self, ra: VisitorResult, dec: VisitorResult, node: Node) -> VisitorResult:
439 # Docstring inherited.
441 # this is a placeholder for future extension, we enabled syntax but
442 # do not support actual use just yet.
443 raise NotImplementedError("POINT() function is not supported yet")
445 def visitRangeLiteral(self, start: int, stop: int, stride: int | None, node: Node) -> VisitorResult:
446 # Docstring inherited.
447 return ColumnContainer.range_literal(range(start, stop + 1, stride or 1))
449 def visitStringLiteral(self, value: str, node: Node) -> VisitorResult:
450 # Docstring inherited.
451 return ColumnExpression.literal(value, dtype=str)
453 def visitTimeLiteral(self, value: astropy.time.Time, node: Node) -> VisitorResult:
454 # Docstring inherited.
455 return ColumnExpression.literal(value, dtype=astropy.time.Time)
457 def visitTupleNode(self, items: tuple[VisitorResult, ...], node: Node) -> VisitorResult:
458 # Docstring inherited.
459 match items:
460 case [
461 ColumnLiteral(value=begin, dtype=astropy.time.Time | types.NoneType),
462 ColumnLiteral(value=end, dtype=astropy.time.Time | types.NoneType),
463 ]:
464 return ColumnExpression.literal(_timespan.Timespan(begin, end), dtype=_timespan.Timespan)
465 raise ExpressionTypeError(
466 f'Invalid type(s) ({items[0].dtype}, {items[1].dtype}) in timespan tuple "{node}" '
467 '(Note that date/time strings must be preceded by "T" to be recognized).'
468 )
470 def visitUnaryOp(self, operator: str, operand: VisitorResult, node: Node) -> VisitorResult:
471 # Docstring inherited.
472 match (operator, operand):
473 case ["NOT", Predicate() as operand]:
474 return operand.logical_not()
475 case ["+", ColumnExpression(dtype=builtins.int | builtins.float) as operand]:
476 return operand.method("__pos__")
477 case ["-", ColumnExpression(dtype=builtins.int | builtins.float) as operand]:
478 return operand.method("__neg__")
479 raise ExpressionTypeError(
480 f"Unary operator {operator!r} is not valid for operand of type {operand.dtype!s} in {node!s}."
481 )