sqlglot.optimizer.qualify_columns

  1from __future__ import annotations
  2
  3import itertools
  4import typing as t
  5
  6from sqlglot import alias, exp
  7from sqlglot._typing import E
  8from sqlglot.dialects.dialect import Dialect, DialectType
  9from sqlglot.errors import OptimizeError
 10from sqlglot.helper import seq_get
 11from sqlglot.optimizer.scope import Scope, traverse_scope, walk_in_scope
 12from sqlglot.schema import Schema, ensure_schema
 13
 14
 15def qualify_columns(
 16    expression: exp.Expression,
 17    schema: t.Dict | Schema,
 18    expand_alias_refs: bool = True,
 19    infer_schema: t.Optional[bool] = None,
 20) -> exp.Expression:
 21    """
 22    Rewrite sqlglot AST to have fully qualified columns.
 23
 24    Example:
 25        >>> import sqlglot
 26        >>> schema = {"tbl": {"col": "INT"}}
 27        >>> expression = sqlglot.parse_one("SELECT col FROM tbl")
 28        >>> qualify_columns(expression, schema).sql()
 29        'SELECT tbl.col AS col FROM tbl'
 30
 31    Args:
 32        expression: expression to qualify
 33        schema: Database schema
 34        expand_alias_refs: whether or not to expand references to aliases
 35        infer_schema: whether or not to infer the schema if missing
 36    Returns:
 37        sqlglot.Expression: qualified expression
 38    """
 39    schema = ensure_schema(schema)
 40    infer_schema = schema.empty if infer_schema is None else infer_schema
 41
 42    for scope in traverse_scope(expression):
 43        resolver = Resolver(scope, schema, infer_schema=infer_schema)
 44        _pop_table_column_aliases(scope.ctes)
 45        _pop_table_column_aliases(scope.derived_tables)
 46        using_column_tables = _expand_using(scope, resolver)
 47
 48        if schema.empty and expand_alias_refs:
 49            _expand_alias_refs(scope, resolver)
 50
 51        _qualify_columns(scope, resolver)
 52
 53        if not schema.empty and expand_alias_refs:
 54            _expand_alias_refs(scope, resolver)
 55
 56        if not isinstance(scope.expression, exp.UDTF):
 57            _expand_stars(scope, resolver, using_column_tables)
 58            _qualify_outputs(scope)
 59        _expand_group_by(scope)
 60        _expand_order_by(scope, resolver)
 61
 62    return expression
 63
 64
 65def validate_qualify_columns(expression: E) -> E:
 66    """Raise an `OptimizeError` if any columns aren't qualified"""
 67    unqualified_columns = []
 68    for scope in traverse_scope(expression):
 69        if isinstance(scope.expression, exp.Select):
 70            unqualified_columns.extend(scope.unqualified_columns)
 71            if scope.external_columns and not scope.is_correlated_subquery and not scope.pivots:
 72                column = scope.external_columns[0]
 73                raise OptimizeError(
 74                    f"""Column '{column}' could not be resolved{f" for table: '{column.table}'" if column.table else ''}"""
 75                )
 76
 77    if unqualified_columns:
 78        raise OptimizeError(f"Ambiguous columns: {unqualified_columns}")
 79    return expression
 80
 81
 82def _pop_table_column_aliases(derived_tables: t.List[exp.CTE | exp.Subquery]) -> None:
 83    """
 84    Remove table column aliases.
 85
 86    (e.g. SELECT ... FROM (SELECT ...) AS foo(col1, col2)
 87    """
 88    for derived_table in derived_tables:
 89        table_alias = derived_table.args.get("alias")
 90        if table_alias:
 91            table_alias.args.pop("columns", None)
 92
 93
 94def _expand_using(scope: Scope, resolver: Resolver) -> t.Dict[str, t.Any]:
 95    joins = list(scope.find_all(exp.Join))
 96    names = {join.alias_or_name for join in joins}
 97    ordered = [key for key in scope.selected_sources if key not in names]
 98
 99    # Mapping of automatically joined column names to an ordered set of source names (dict).
100    column_tables: t.Dict[str, t.Dict[str, t.Any]] = {}
101
102    for join in joins:
103        using = join.args.get("using")
104
105        if not using:
106            continue
107
108        join_table = join.alias_or_name
109
110        columns = {}
111
112        for k in scope.selected_sources:
113            if k in ordered:
114                for column in resolver.get_source_columns(k):
115                    if column not in columns:
116                        columns[column] = k
117
118        source_table = ordered[-1]
119        ordered.append(join_table)
120        join_columns = resolver.get_source_columns(join_table)
121        conditions = []
122
123        for identifier in using:
124            identifier = identifier.name
125            table = columns.get(identifier)
126
127            if not table or identifier not in join_columns:
128                if columns and join_columns:
129                    raise OptimizeError(f"Cannot automatically join: {identifier}")
130
131            table = table or source_table
132            conditions.append(
133                exp.condition(
134                    exp.EQ(
135                        this=exp.column(identifier, table=table),
136                        expression=exp.column(identifier, table=join_table),
137                    )
138                )
139            )
140
141            # Set all values in the dict to None, because we only care about the key ordering
142            tables = column_tables.setdefault(identifier, {})
143            if table not in tables:
144                tables[table] = None
145            if join_table not in tables:
146                tables[join_table] = None
147
148        join.args.pop("using")
149        join.set("on", exp.and_(*conditions, copy=False))
150
151    if column_tables:
152        for column in scope.columns:
153            if not column.table and column.name in column_tables:
154                tables = column_tables[column.name]
155                coalesce = [exp.column(column.name, table=table) for table in tables]
156                replacement = exp.Coalesce(this=coalesce[0], expressions=coalesce[1:])
157
158                # Ensure selects keep their output name
159                if isinstance(column.parent, exp.Select):
160                    replacement = alias(replacement, alias=column.name, copy=False)
161
162                scope.replace(column, replacement)
163
164    return column_tables
165
166
167def _expand_alias_refs(scope: Scope, resolver: Resolver) -> None:
168    expression = scope.expression
169
170    if not isinstance(expression, exp.Select):
171        return
172
173    alias_to_expression: t.Dict[str, t.Tuple[exp.Expression, int]] = {}
174
175    def replace_columns(
176        node: t.Optional[exp.Expression], resolve_table: bool = False, literal_index: bool = False
177    ) -> None:
178        if not node:
179            return
180
181        for column, *_ in walk_in_scope(node):
182            if not isinstance(column, exp.Column):
183                continue
184            table = resolver.get_table(column.name) if resolve_table and not column.table else None
185            alias_expr, i = alias_to_expression.get(column.name, (None, 1))
186            double_agg = (
187                (alias_expr.find(exp.AggFunc) and column.find_ancestor(exp.AggFunc))
188                if alias_expr
189                else False
190            )
191
192            if table and (not alias_expr or double_agg):
193                column.set("table", table)
194            elif not column.table and alias_expr and not double_agg:
195                if isinstance(alias_expr, exp.Literal) and (literal_index or resolve_table):
196                    if literal_index:
197                        column.replace(exp.Literal.number(i))
198                else:
199                    column.replace(alias_expr.copy())
200
201    for i, projection in enumerate(scope.expression.selects):
202        replace_columns(projection)
203
204        if isinstance(projection, exp.Alias):
205            alias_to_expression[projection.alias] = (projection.this, i + 1)
206
207    replace_columns(expression.args.get("where"))
208    replace_columns(expression.args.get("group"), literal_index=True)
209    replace_columns(expression.args.get("having"), resolve_table=True)
210    replace_columns(expression.args.get("qualify"), resolve_table=True)
211    scope.clear_cache()
212
213
214def _expand_group_by(scope: Scope):
215    expression = scope.expression
216    group = expression.args.get("group")
217    if not group:
218        return
219
220    group.set("expressions", _expand_positional_references(scope, group.expressions))
221    expression.set("group", group)
222
223
224def _expand_order_by(scope: Scope, resolver: Resolver):
225    order = scope.expression.args.get("order")
226    if not order:
227        return
228
229    ordereds = order.expressions
230    for ordered, new_expression in zip(
231        ordereds,
232        _expand_positional_references(scope, (o.this for o in ordereds)),
233    ):
234        for agg in ordered.find_all(exp.AggFunc):
235            for col in agg.find_all(exp.Column):
236                if not col.table:
237                    col.set("table", resolver.get_table(col.name))
238
239        ordered.set("this", new_expression)
240
241    if scope.expression.args.get("group"):
242        selects = {s.this: exp.column(s.alias_or_name) for s in scope.expression.selects}
243
244        for ordered in ordereds:
245            ordered = ordered.this
246
247            ordered.replace(
248                exp.to_identifier(_select_by_pos(scope, ordered).alias)
249                if ordered.is_int
250                else selects.get(ordered, ordered)
251            )
252
253
254def _expand_positional_references(scope: Scope, expressions: t.Iterable[E]) -> t.List[E]:
255    new_nodes = []
256    for node in expressions:
257        if node.is_int:
258            select = _select_by_pos(scope, t.cast(exp.Literal, node)).this
259
260            if isinstance(select, exp.Literal):
261                new_nodes.append(node)
262            else:
263                new_nodes.append(select.copy())
264                scope.clear_cache()
265        else:
266            new_nodes.append(node)
267
268    return new_nodes
269
270
271def _select_by_pos(scope: Scope, node: exp.Literal) -> exp.Alias:
272    try:
273        return scope.expression.selects[int(node.this) - 1].assert_is(exp.Alias)
274    except IndexError:
275        raise OptimizeError(f"Unknown output column: {node.name}")
276
277
278def _qualify_columns(scope: Scope, resolver: Resolver) -> None:
279    """Disambiguate columns, ensuring each column specifies a source"""
280    for column in scope.columns:
281        column_table = column.table
282        column_name = column.name
283
284        if column_table and column_table in scope.sources:
285            source_columns = resolver.get_source_columns(column_table)
286            if source_columns and column_name not in source_columns and "*" not in source_columns:
287                raise OptimizeError(f"Unknown column: {column_name}")
288
289        if not column_table:
290            if scope.pivots and not column.find_ancestor(exp.Pivot):
291                # If the column is under the Pivot expression, we need to qualify it
292                # using the name of the pivoted source instead of the pivot's alias
293                column.set("table", exp.to_identifier(scope.pivots[0].alias))
294                continue
295
296            column_table = resolver.get_table(column_name)
297
298            # column_table can be a '' because bigquery unnest has no table alias
299            if column_table:
300                column.set("table", column_table)
301        elif column_table not in scope.sources and (
302            not scope.parent or column_table not in scope.parent.sources
303        ):
304            # structs are used like tables (e.g. "struct"."field"), so they need to be qualified
305            # separately and represented as dot(dot(...(<table>.<column>, field1), field2, ...))
306
307            root, *parts = column.parts
308
309            if root.name in scope.sources:
310                # struct is already qualified, but we still need to change the AST representation
311                column_table = root
312                root, *parts = parts
313            else:
314                column_table = resolver.get_table(root.name)
315
316            if column_table:
317                column.replace(exp.Dot.build([exp.column(root, table=column_table), *parts]))
318
319    for pivot in scope.pivots:
320        for column in pivot.find_all(exp.Column):
321            if not column.table and column.name in resolver.all_columns:
322                column_table = resolver.get_table(column.name)
323                if column_table:
324                    column.set("table", column_table)
325
326
327def _expand_stars(
328    scope: Scope, resolver: Resolver, using_column_tables: t.Dict[str, t.Any]
329) -> None:
330    """Expand stars to lists of column selections"""
331
332    new_selections = []
333    except_columns: t.Dict[int, t.Set[str]] = {}
334    replace_columns: t.Dict[int, t.Dict[str, str]] = {}
335    coalesced_columns = set()
336
337    # TODO: handle optimization of multiple PIVOTs (and possibly UNPIVOTs) in the future
338    pivot_columns = None
339    pivot_output_columns = None
340    pivot = t.cast(t.Optional[exp.Pivot], seq_get(scope.pivots, 0))
341
342    has_pivoted_source = pivot and not pivot.args.get("unpivot")
343    if pivot and has_pivoted_source:
344        pivot_columns = set(col.output_name for col in pivot.find_all(exp.Column))
345
346        pivot_output_columns = [col.output_name for col in pivot.args.get("columns", [])]
347        if not pivot_output_columns:
348            pivot_output_columns = [col.alias_or_name for col in pivot.expressions]
349
350    for expression in scope.expression.selects:
351        if isinstance(expression, exp.Star):
352            tables = list(scope.selected_sources)
353            _add_except_columns(expression, tables, except_columns)
354            _add_replace_columns(expression, tables, replace_columns)
355        elif expression.is_star:
356            tables = [expression.table]
357            _add_except_columns(expression.this, tables, except_columns)
358            _add_replace_columns(expression.this, tables, replace_columns)
359        else:
360            new_selections.append(expression)
361            continue
362
363        for table in tables:
364            if table not in scope.sources:
365                raise OptimizeError(f"Unknown table: {table}")
366
367            columns = resolver.get_source_columns(table, only_visible=True)
368
369            # The _PARTITIONTIME and _PARTITIONDATE pseudo-columns are not returned by a SELECT * statement
370            # https://cloud.google.com/bigquery/docs/querying-partitioned-tables#query_an_ingestion-time_partitioned_table
371            if resolver.schema.dialect == "bigquery":
372                columns = [
373                    name
374                    for name in columns
375                    if name.upper() not in ("_PARTITIONTIME", "_PARTITIONDATE")
376                ]
377
378            if columns and "*" not in columns:
379                if pivot and has_pivoted_source and pivot_columns and pivot_output_columns:
380                    implicit_columns = [col for col in columns if col not in pivot_columns]
381                    new_selections.extend(
382                        exp.alias_(exp.column(name, table=pivot.alias), name, copy=False)
383                        for name in implicit_columns + pivot_output_columns
384                    )
385                    continue
386
387                table_id = id(table)
388                for name in columns:
389                    if name in using_column_tables and table in using_column_tables[name]:
390                        if name in coalesced_columns:
391                            continue
392
393                        coalesced_columns.add(name)
394                        tables = using_column_tables[name]
395                        coalesce = [exp.column(name, table=table) for table in tables]
396
397                        new_selections.append(
398                            alias(
399                                exp.Coalesce(this=coalesce[0], expressions=coalesce[1:]),
400                                alias=name,
401                                copy=False,
402                            )
403                        )
404                    elif name not in except_columns.get(table_id, set()):
405                        alias_ = replace_columns.get(table_id, {}).get(name, name)
406                        column = exp.column(name, table=table)
407                        new_selections.append(
408                            alias(column, alias_, copy=False) if alias_ != name else column
409                        )
410            else:
411                return
412
413    scope.expression.set("expressions", new_selections)
414
415
416def _add_except_columns(
417    expression: exp.Expression, tables, except_columns: t.Dict[int, t.Set[str]]
418) -> None:
419    except_ = expression.args.get("except")
420
421    if not except_:
422        return
423
424    columns = {e.name for e in except_}
425
426    for table in tables:
427        except_columns[id(table)] = columns
428
429
430def _add_replace_columns(
431    expression: exp.Expression, tables, replace_columns: t.Dict[int, t.Dict[str, str]]
432) -> None:
433    replace = expression.args.get("replace")
434
435    if not replace:
436        return
437
438    columns = {e.this.name: e.alias for e in replace}
439
440    for table in tables:
441        replace_columns[id(table)] = columns
442
443
444def _qualify_outputs(scope: Scope):
445    """Ensure all output columns are aliased"""
446    new_selections = []
447
448    for i, (selection, aliased_column) in enumerate(
449        itertools.zip_longest(scope.expression.selects, scope.outer_column_list)
450    ):
451        if isinstance(selection, exp.Subquery):
452            if not selection.output_name:
453                selection.set("alias", exp.TableAlias(this=exp.to_identifier(f"_col_{i}")))
454        elif not isinstance(selection, exp.Alias) and not selection.is_star:
455            selection = alias(
456                selection,
457                alias=selection.output_name or f"_col_{i}",
458            )
459        if aliased_column:
460            selection.set("alias", exp.to_identifier(aliased_column))
461
462        new_selections.append(selection)
463
464    scope.expression.set("expressions", new_selections)
465
466
467def quote_identifiers(expression: E, dialect: DialectType = None, identify: bool = True) -> E:
468    """Makes sure all identifiers that need to be quoted are quoted."""
469    return expression.transform(
470        Dialect.get_or_raise(dialect).quote_identifier, identify=identify, copy=False
471    )
472
473
474class Resolver:
475    """
476    Helper for resolving columns.
477
478    This is a class so we can lazily load some things and easily share them across functions.
479    """
480
481    def __init__(self, scope: Scope, schema: Schema, infer_schema: bool = True):
482        self.scope = scope
483        self.schema = schema
484        self._source_columns = None
485        self._unambiguous_columns: t.Optional[t.Dict[str, str]] = None
486        self._all_columns = None
487        self._infer_schema = infer_schema
488
489    def get_table(self, column_name: str) -> t.Optional[exp.Identifier]:
490        """
491        Get the table for a column name.
492
493        Args:
494            column_name: The column name to find the table for.
495        Returns:
496            The table name if it can be found/inferred.
497        """
498        if self._unambiguous_columns is None:
499            self._unambiguous_columns = self._get_unambiguous_columns(
500                self._get_all_source_columns()
501            )
502
503        table_name = self._unambiguous_columns.get(column_name)
504
505        if not table_name and self._infer_schema:
506            sources_without_schema = tuple(
507                source
508                for source, columns in self._get_all_source_columns().items()
509                if not columns or "*" in columns
510            )
511            if len(sources_without_schema) == 1:
512                table_name = sources_without_schema[0]
513
514        if table_name not in self.scope.selected_sources:
515            return exp.to_identifier(table_name)
516
517        node, _ = self.scope.selected_sources.get(table_name)
518
519        if isinstance(node, exp.Subqueryable):
520            while node and node.alias != table_name:
521                node = node.parent
522
523        node_alias = node.args.get("alias")
524        if node_alias:
525            return exp.to_identifier(node_alias.this)
526
527        return exp.to_identifier(table_name)
528
529    @property
530    def all_columns(self):
531        """All available columns of all sources in this scope"""
532        if self._all_columns is None:
533            self._all_columns = {
534                column for columns in self._get_all_source_columns().values() for column in columns
535            }
536        return self._all_columns
537
538    def get_source_columns(self, name, only_visible=False):
539        """Resolve the source columns for a given source `name`"""
540        if name not in self.scope.sources:
541            raise OptimizeError(f"Unknown table: {name}")
542
543        source = self.scope.sources[name]
544
545        # If referencing a table, return the columns from the schema
546        if isinstance(source, exp.Table):
547            return self.schema.column_names(source, only_visible)
548
549        if isinstance(source, Scope) and isinstance(source.expression, exp.Values):
550            return source.expression.alias_column_names
551
552        # Otherwise, if referencing another scope, return that scope's named selects
553        return source.expression.named_selects
554
555    def _get_all_source_columns(self):
556        if self._source_columns is None:
557            self._source_columns = {
558                k: self.get_source_columns(k)
559                for k in itertools.chain(self.scope.selected_sources, self.scope.lateral_sources)
560            }
561        return self._source_columns
562
563    def _get_unambiguous_columns(self, source_columns):
564        """
565        Find all the unambiguous columns in sources.
566
567        Args:
568            source_columns (dict): Mapping of names to source columns
569        Returns:
570            dict: Mapping of column name to source name
571        """
572        if not source_columns:
573            return {}
574
575        source_columns = list(source_columns.items())
576
577        first_table, first_columns = source_columns[0]
578        unambiguous_columns = {col: first_table for col in self._find_unique_columns(first_columns)}
579        all_columns = set(unambiguous_columns)
580
581        for table, columns in source_columns[1:]:
582            unique = self._find_unique_columns(columns)
583            ambiguous = set(all_columns).intersection(unique)
584            all_columns.update(columns)
585            for column in ambiguous:
586                unambiguous_columns.pop(column, None)
587            for column in unique.difference(ambiguous):
588                unambiguous_columns[column] = table
589
590        return unambiguous_columns
591
592    @staticmethod
593    def _find_unique_columns(columns):
594        """
595        Find the unique columns in a list of columns.
596
597        Example:
598            >>> sorted(Resolver._find_unique_columns(["a", "b", "b", "c"]))
599            ['a', 'c']
600
601        This is necessary because duplicate column names are ambiguous.
602        """
603        counts = {}
604        for column in columns:
605            counts[column] = counts.get(column, 0) + 1
606        return {column for column, count in counts.items() if count == 1}