sqlglot.dataframe.sql

 23class SparkSession:
 24    DEFAULT_DIALECT = "spark"
 25    _instance = None
 26
 27    def __init__(self):
 28        if not hasattr(self, "known_ids"):
 29            self.known_ids = set()
 30            self.known_branch_ids = set()
 31            self.known_sequence_ids = set()
 32            self.name_to_sequence_id_mapping = defaultdict(list)
 33            self.incrementing_id = 1
 34            self.dialect = Dialect.get_or_raise(self.DEFAULT_DIALECT)
 35
 36    def __new__(cls, *args, **kwargs) -> SparkSession:
 37        if cls._instance is None:
 38            cls._instance = super().__new__(cls)
 39        return cls._instance
 40
 41    @property
 42    def read(self) -> DataFrameReader:
 43        return DataFrameReader(self)
 44
 45    def table(self, tableName: str) -> DataFrame:
 46        return self.read.table(tableName)
 47
 48    def createDataFrame(
 49        self,
 50        data: t.Sequence[t.Union[t.Dict[str, ColumnLiterals], t.List[ColumnLiterals], t.Tuple]],
 51        schema: t.Optional[SchemaInput] = None,
 52        samplingRatio: t.Optional[float] = None,
 53        verifySchema: bool = False,
 54    ) -> DataFrame:
 55        from sqlglot.dataframe.sql.dataframe import DataFrame
 56
 57        if samplingRatio is not None or verifySchema:
 58            raise NotImplementedError("Sampling Ratio and Verify Schema are not supported")
 59        if schema is not None and (
 60            not isinstance(schema, (StructType, str, list))
 61            or (isinstance(schema, list) and not isinstance(schema[0], str))
 62        ):
 63            raise NotImplementedError("Only schema of either list or string of list supported")
 64        if not data:
 65            raise ValueError("Must provide data to create into a DataFrame")
 66
 67        column_mapping: t.Dict[str, t.Optional[str]]
 68        if schema is not None:
 69            column_mapping = get_column_mapping_from_schema_input(schema)
 70        elif isinstance(data[0], dict):
 71            column_mapping = {col_name.strip(): None for col_name in data[0]}
 72        else:
 73            column_mapping = {f"_{i}": None for i in range(1, len(data[0]) + 1)}
 74
 75        data_expressions = [
 76            exp.tuple_(
 77                *map(
 78                    lambda x: F.lit(x).expression,
 79                    row if not isinstance(row, dict) else row.values(),
 80                )
 81            )
 82            for row in data
 83        ]
 84
 85        sel_columns = [
 86            (
 87                F.col(name).cast(data_type).alias(name).expression
 88                if data_type is not None
 89                else F.col(name).expression
 90            )
 91            for name, data_type in column_mapping.items()
 92        ]
 93
 94        select_kwargs = {
 95            "expressions": sel_columns,
 96            "from": exp.From(
 97                this=exp.Values(
 98                    expressions=data_expressions,
 99                    alias=exp.TableAlias(
100                        this=exp.to_identifier(self._auto_incrementing_name),
101                        columns=[exp.to_identifier(col_name) for col_name in column_mapping],
102                    ),
103                ),
104            ),
105        }
106
107        sel_expression = exp.Select(**select_kwargs)
108        return DataFrame(self, sel_expression)
109
110    def _optimize(
111        self, expression: exp.Expression, dialect: t.Optional[Dialect] = None
112    ) -> exp.Expression:
113        dialect = dialect or self.dialect
114        quote_identifiers(expression, dialect=dialect)
115        return optimize(expression, dialect=dialect)
116
117    def sql(self, sqlQuery: str) -> DataFrame:
118        expression = self._optimize(sqlglot.parse_one(sqlQuery, read=self.dialect))
119        if isinstance(expression, exp.Select):
120            df = DataFrame(self, expression)
121            df = df._convert_leaf_to_cte()
122        elif isinstance(expression, (exp.Create, exp.Insert)):
123            select_expression = expression.expression.copy()
124            if isinstance(expression, exp.Insert):
125                select_expression.set("with", expression.args.get("with"))
126                expression.set("with", None)
127            del expression.args["expression"]
128            df = DataFrame(self, select_expression, output_expression_container=expression)  # type: ignore
129            df = df._convert_leaf_to_cte()
130        else:
131            raise ValueError(
132                "Unknown expression type provided in the SQL. Please create an issue with the SQL."
133            )
134        return df
135
136    @property
137    def _auto_incrementing_name(self) -> str:
138        name = f"a{self.incrementing_id}"
139        self.incrementing_id += 1
140        return name
141
142    @property
143    def _random_branch_id(self) -> str:
144        id = self._random_id
145        self.known_branch_ids.add(id)
146        return id
147
148    @property
149    def _random_sequence_id(self):
150        id = self._random_id
151        self.known_sequence_ids.add(id)
152        return id
153
154    @property
155    def _random_id(self) -> str:
156        id = "r" + uuid.uuid4().hex
157        self.known_ids.add(id)
158        return id
159
160    @property
161    def _join_hint_names(self) -> t.Set[str]:
162        return {"BROADCAST", "MERGE", "SHUFFLE_HASH", "SHUFFLE_REPLICATE_NL"}
163
164    def _add_alias_to_mapping(self, name: str, sequence_id: str):
165        self.name_to_sequence_id_mapping[name].append(sequence_id)
166
167    class Builder:
168        SQLFRAME_DIALECT_KEY = "sqlframe.dialect"
169
170        def __init__(self):
171            self.dialect = "spark"
172
173        def __getattr__(self, item) -> SparkSession.Builder:
174            return self
175
176        def __call__(self, *args, **kwargs):
177            return self
178
179        def config(
180            self,
181            key: t.Optional[str] = None,
182            value: t.Optional[t.Any] = None,
183            *,
184            map: t.Optional[t.Dict[str, t.Any]] = None,
185            **kwargs: t.Any,
186        ) -> SparkSession.Builder:
187            if key == self.SQLFRAME_DIALECT_KEY:
188                self.dialect = value
189            elif map and self.SQLFRAME_DIALECT_KEY in map:
190                self.dialect = map[self.SQLFRAME_DIALECT_KEY]
191            return self
192
193        def getOrCreate(self) -> SparkSession:
194            spark = SparkSession()
195            spark.dialect = Dialect.get_or_raise(self.dialect)
196            return spark
197
198    @classproperty
199    def builder(cls) -> Builder:
200        return cls.Builder()

167    class Builder:
168        SQLFRAME_DIALECT_KEY = "sqlframe.dialect"
169
170        def __init__(self):
171            self.dialect = "spark"
172
173        def __getattr__(self, item) -> SparkSession.Builder:
174            return self
175
176        def __call__(self, *args, **kwargs):
177            return self
178
179        def config(
180            self,
181            key: t.Optional[str] = None,
182            value: t.Optional[t.Any] = None,
183            *,
184            map: t.Optional[t.Dict[str, t.Any]] = None,
185            **kwargs: t.Any,
186        ) -> SparkSession.Builder:
187            if key == self.SQLFRAME_DIALECT_KEY:
188                self.dialect = value
189            elif map and self.SQLFRAME_DIALECT_KEY in map:
190                self.dialect = map[self.SQLFRAME_DIALECT_KEY]
191            return self
192
193        def getOrCreate(self) -> SparkSession:
194            spark = SparkSession()
195            spark.dialect = Dialect.get_or_raise(self.dialect)
196            return spark

 47class DataFrame:
 48    def __init__(
 49        self,
 50        spark: SparkSession,
 51        expression: exp.Select,
 52        branch_id: t.Optional[str] = None,
 53        sequence_id: t.Optional[str] = None,
 54        last_op: Operation = Operation.INIT,
 55        pending_hints: t.Optional[t.List[exp.Expression]] = None,
 56        output_expression_container: t.Optional[OutputExpressionContainer] = None,
 57        **kwargs,
 58    ):
 59        self.spark = spark
 60        self.expression = expression
 61        self.branch_id = branch_id or self.spark._random_branch_id
 62        self.sequence_id = sequence_id or self.spark._random_sequence_id
 63        self.last_op = last_op
 64        self.pending_hints = pending_hints or []
 65        self.output_expression_container = output_expression_container or exp.Select()
 66
 67    def __getattr__(self, column_name: str) -> Column:
 68        return self[column_name]
 69
 70    def __getitem__(self, column_name: str) -> Column:
 71        column_name = f"{self.branch_id}.{column_name}"
 72        return Column(column_name)
 73
 74    def __copy__(self):
 75        return self.copy()
 76
 77    @property
 78    def sparkSession(self):
 79        return self.spark
 80
 81    @property
 82    def write(self):
 83        return DataFrameWriter(self)
 84
 85    @property
 86    def latest_cte_name(self) -> str:
 87        if not self.expression.ctes:
 88            from_exp = self.expression.args["from"]
 89            if from_exp.alias_or_name:
 90                return from_exp.alias_or_name
 91            table_alias = from_exp.find(exp.TableAlias)
 92            if not table_alias:
 93                raise RuntimeError(
 94                    f"Could not find an alias name for this expression: {self.expression}"
 95                )
 96            return table_alias.alias_or_name
 97        return self.expression.ctes[-1].alias
 98
 99    @property
100    def pending_join_hints(self):
101        return [hint for hint in self.pending_hints if isinstance(hint, exp.JoinHint)]
102
103    @property
104    def pending_partition_hints(self):
105        return [hint for hint in self.pending_hints if isinstance(hint, exp.Anonymous)]
106
107    @property
108    def columns(self) -> t.List[str]:
109        return self.expression.named_selects
110
111    @property
112    def na(self) -> DataFrameNaFunctions:
113        return DataFrameNaFunctions(self)
114
115    def _replace_cte_names_with_hashes(self, expression: exp.Select):
116        replacement_mapping = {}
117        for cte in expression.ctes:
118            old_name_id = cte.args["alias"].this
119            new_hashed_id = exp.to_identifier(
120                self._create_hash_from_expression(cte.this), quoted=old_name_id.args["quoted"]
121            )
122            replacement_mapping[old_name_id] = new_hashed_id
123            expression = expression.transform(replace_id_value, replacement_mapping).assert_is(
124                exp.Select
125            )
126        return expression
127
128    def _create_cte_from_expression(
129        self,
130        expression: exp.Expression,
131        branch_id: t.Optional[str] = None,
132        sequence_id: t.Optional[str] = None,
133        **kwargs,
134    ) -> t.Tuple[exp.CTE, str]:
135        name = self._create_hash_from_expression(expression)
136        expression_to_cte = expression.copy()
137        expression_to_cte.set("with", None)
138        cte = exp.Select().with_(name, as_=expression_to_cte, **kwargs).ctes[0]
139        cte.set("branch_id", branch_id or self.branch_id)
140        cte.set("sequence_id", sequence_id or self.sequence_id)
141        return cte, name
142
143    @t.overload
144    def _ensure_list_of_columns(self, cols: t.Collection[ColumnOrLiteral]) -> t.List[Column]: ...
145
146    @t.overload
147    def _ensure_list_of_columns(self, cols: ColumnOrLiteral) -> t.List[Column]: ...
148
149    def _ensure_list_of_columns(self, cols):
150        return Column.ensure_cols(ensure_list(cols))
151
152    def _ensure_and_normalize_cols(self, cols, expression: t.Optional[exp.Select] = None):
153        cols = self._ensure_list_of_columns(cols)
154        normalize(self.spark, expression or self.expression, cols)
155        return cols
156
157    def _ensure_and_normalize_col(self, col):
158        col = Column.ensure_col(col)
159        normalize(self.spark, self.expression, col)
160        return col
161
162    def _convert_leaf_to_cte(self, sequence_id: t.Optional[str] = None) -> DataFrame:
163        df = self._resolve_pending_hints()
164        sequence_id = sequence_id or df.sequence_id
165        expression = df.expression.copy()
166        cte_expression, cte_name = df._create_cte_from_expression(
167            expression=expression, sequence_id=sequence_id
168        )
169        new_expression = df._add_ctes_to_expression(
170            exp.Select(), expression.ctes + [cte_expression]
171        )
172        sel_columns = df._get_outer_select_columns(cte_expression)
173        new_expression = new_expression.from_(cte_name).select(
174            *[x.alias_or_name for x in sel_columns]
175        )
176        return df.copy(expression=new_expression, sequence_id=sequence_id)
177
178    def _resolve_pending_hints(self) -> DataFrame:
179        df = self.copy()
180        if not self.pending_hints:
181            return df
182        expression = df.expression
183        hint_expression = expression.args.get("hint") or exp.Hint(expressions=[])
184        for hint in df.pending_partition_hints:
185            hint_expression.append("expressions", hint)
186            df.pending_hints.remove(hint)
187
188        join_aliases = {
189            join_table.alias_or_name
190            for join_table in get_tables_from_expression_with_join(expression)
191        }
192        if join_aliases:
193            for hint in df.pending_join_hints:
194                for sequence_id_expression in hint.expressions:
195                    sequence_id_or_name = sequence_id_expression.alias_or_name
196                    sequence_ids_to_match = [sequence_id_or_name]
197                    if sequence_id_or_name in df.spark.name_to_sequence_id_mapping:
198                        sequence_ids_to_match = df.spark.name_to_sequence_id_mapping[
199                            sequence_id_or_name
200                        ]
201                    matching_ctes = [
202                        cte
203                        for cte in reversed(expression.ctes)
204                        if cte.args["sequence_id"] in sequence_ids_to_match
205                    ]
206                    for matching_cte in matching_ctes:
207                        if matching_cte.alias_or_name in join_aliases:
208                            sequence_id_expression.set("this", matching_cte.args["alias"].this)
209                            df.pending_hints.remove(hint)
210                            break
211                hint_expression.append("expressions", hint)
212        if hint_expression.expressions:
213            expression.set("hint", hint_expression)
214        return df
215
216    def _hint(self, hint_name: str, args: t.List[Column]) -> DataFrame:
217        hint_name = hint_name.upper()
218        hint_expression = (
219            exp.JoinHint(
220                this=hint_name,
221                expressions=[exp.to_table(parameter.alias_or_name) for parameter in args],
222            )
223            if hint_name in JOIN_HINTS
224            else exp.Anonymous(
225                this=hint_name, expressions=[parameter.expression for parameter in args]
226            )
227        )
228        new_df = self.copy()
229        new_df.pending_hints.append(hint_expression)
230        return new_df
231
232    def _set_operation(self, klass: t.Callable, other: DataFrame, distinct: bool):
233        other_df = other._convert_leaf_to_cte()
234        base_expression = self.expression.copy()
235        base_expression = self._add_ctes_to_expression(base_expression, other_df.expression.ctes)
236        all_ctes = base_expression.ctes
237        other_df.expression.set("with", None)
238        base_expression.set("with", None)
239        operation = klass(this=base_expression, distinct=distinct, expression=other_df.expression)
240        operation.set("with", exp.With(expressions=all_ctes))
241        return self.copy(expression=operation)._convert_leaf_to_cte()
242
243    def _cache(self, storage_level: str):
244        df = self._convert_leaf_to_cte()
245        df.expression.ctes[-1].set("cache_storage_level", storage_level)
246        return df
247
248    @classmethod
249    def _add_ctes_to_expression(cls, expression: exp.Select, ctes: t.List[exp.CTE]) -> exp.Select:
250        expression = expression.copy()
251        with_expression = expression.args.get("with")
252        if with_expression:
253            existing_ctes = with_expression.expressions
254            existsing_cte_names = {x.alias_or_name for x in existing_ctes}
255            for cte in ctes:
256                if cte.alias_or_name not in existsing_cte_names:
257                    existing_ctes.append(cte)
258        else:
259            existing_ctes = ctes
260        expression.set("with", exp.With(expressions=existing_ctes))
261        return expression
262
263    @classmethod
264    def _get_outer_select_columns(cls, item: t.Union[exp.Expression, DataFrame]) -> t.List[Column]:
265        expression = item.expression if isinstance(item, DataFrame) else item
266        return [Column(x) for x in (expression.find(exp.Select) or exp.Select()).expressions]
267
268    @classmethod
269    def _create_hash_from_expression(cls, expression: exp.Expression) -> str:
270        from sqlglot.dataframe.sql.session import SparkSession
271
272        value = expression.sql(dialect=SparkSession().dialect).encode("utf-8")
273        return f"t{zlib.crc32(value)}"[:6]
274
275    def _get_select_expressions(
276        self,
277    ) -> t.List[t.Tuple[t.Union[t.Type[exp.Cache], OutputExpressionContainer], exp.Select]]:
278        select_expressions: t.List[
279            t.Tuple[t.Union[t.Type[exp.Cache], OutputExpressionContainer], exp.Select]
280        ] = []
281        main_select_ctes: t.List[exp.CTE] = []
282        for cte in self.expression.ctes:
283            cache_storage_level = cte.args.get("cache_storage_level")
284            if cache_storage_level:
285                select_expression = cte.this.copy()
286                select_expression.set("with", exp.With(expressions=copy(main_select_ctes)))
287                select_expression.set("cte_alias_name", cte.alias_or_name)
288                select_expression.set("cache_storage_level", cache_storage_level)
289                select_expressions.append((exp.Cache, select_expression))
290            else:
291                main_select_ctes.append(cte)
292        main_select = self.expression.copy()
293        if main_select_ctes:
294            main_select.set("with", exp.With(expressions=main_select_ctes))
295        expression_select_pair = (type(self.output_expression_container), main_select)
296        select_expressions.append(expression_select_pair)  # type: ignore
297        return select_expressions
298
299    def sql(self, dialect: DialectType = None, optimize: bool = True, **kwargs) -> t.List[str]:
300        from sqlglot.dataframe.sql.session import SparkSession
301
302        dialect = Dialect.get_or_raise(dialect or SparkSession().dialect)
303
304        df = self._resolve_pending_hints()
305        select_expressions = df._get_select_expressions()
306        output_expressions: t.List[t.Union[exp.Select, exp.Cache, exp.Drop]] = []
307        replacement_mapping: t.Dict[exp.Identifier, exp.Identifier] = {}
308
309        for expression_type, select_expression in select_expressions:
310            select_expression = select_expression.transform(
311                replace_id_value, replacement_mapping
312            ).assert_is(exp.Select)
313            if optimize:
314                select_expression = t.cast(
315                    exp.Select, self.spark._optimize(select_expression, dialect=dialect)
316                )
317
318            select_expression = df._replace_cte_names_with_hashes(select_expression)
319
320            expression: t.Union[exp.Select, exp.Cache, exp.Drop]
321            if expression_type == exp.Cache:
322                cache_table_name = df._create_hash_from_expression(select_expression)
323                cache_table = exp.to_table(cache_table_name)
324                original_alias_name = select_expression.args["cte_alias_name"]
325
326                replacement_mapping[exp.to_identifier(original_alias_name)] = exp.to_identifier(  # type: ignore
327                    cache_table_name
328                )
329                sqlglot.schema.add_table(
330                    cache_table_name,
331                    {
332                        expression.alias_or_name: expression.type.sql(dialect=dialect)
333                        for expression in select_expression.expressions
334                    },
335                    dialect=dialect,
336                )
337
338                cache_storage_level = select_expression.args["cache_storage_level"]
339                options = [
340                    exp.Literal.string("storageLevel"),
341                    exp.Literal.string(cache_storage_level),
342                ]
343                expression = exp.Cache(
344                    this=cache_table, expression=select_expression, lazy=True, options=options
345                )
346
347                # We will drop the "view" if it exists before running the cache table
348                output_expressions.append(exp.Drop(this=cache_table, exists=True, kind="VIEW"))
349            elif expression_type == exp.Create:
350                expression = df.output_expression_container.copy()
351                expression.set("expression", select_expression)
352            elif expression_type == exp.Insert:
353                expression = df.output_expression_container.copy()
354                select_without_ctes = select_expression.copy()
355                select_without_ctes.set("with", None)
356                expression.set("expression", select_without_ctes)
357
358                if select_expression.ctes:
359                    expression.set("with", exp.With(expressions=select_expression.ctes))
360            elif expression_type == exp.Select:
361                expression = select_expression
362            else:
363                raise ValueError(f"Invalid expression type: {expression_type}")
364
365            output_expressions.append(expression)
366
367        return [expression.sql(dialect=dialect, **kwargs) for expression in output_expressions]
368
369    def copy(self, **kwargs) -> DataFrame:
370        return DataFrame(**object_to_dict(self, **kwargs))
371
372    @operation(Operation.SELECT)
373    def select(self, *cols, **kwargs) -> DataFrame:
374        cols = self._ensure_and_normalize_cols(cols)
375        kwargs["append"] = kwargs.get("append", False)
376        if self.expression.args.get("joins"):
377            ambiguous_cols = [
378                col
379                for col in cols
380                if isinstance(col.column_expression, exp.Column) and not col.column_expression.table
381            ]
382            if ambiguous_cols:
383                join_table_identifiers = [
384                    x.this for x in get_tables_from_expression_with_join(self.expression)
385                ]
386                cte_names_in_join = [x.this for x in join_table_identifiers]
387                # If we have columns that resolve to multiple CTE expressions then we want to use each CTE left-to-right
388                # and therefore we allow multiple columns with the same name in the result. This matches the behavior
389                # of Spark.
390                resolved_column_position: t.Dict[Column, int] = {col: -1 for col in ambiguous_cols}
391                for ambiguous_col in ambiguous_cols:
392                    ctes_with_column = [
393                        cte
394                        for cte in self.expression.ctes
395                        if cte.alias_or_name in cte_names_in_join
396                        and ambiguous_col.alias_or_name in cte.this.named_selects
397                    ]
398                    # Check if there is a CTE with this column that we haven't used before. If so, use it. Otherwise,
399                    # use the same CTE we used before
400                    cte = seq_get(ctes_with_column, resolved_column_position[ambiguous_col] + 1)
401                    if cte:
402                        resolved_column_position[ambiguous_col] += 1
403                    else:
404                        cte = ctes_with_column[resolved_column_position[ambiguous_col]]
405                    ambiguous_col.expression.set("table", cte.alias_or_name)
406        return self.copy(
407            expression=self.expression.select(*[x.expression for x in cols], **kwargs), **kwargs
408        )
409
410    @operation(Operation.NO_OP)
411    def alias(self, name: str, **kwargs) -> DataFrame:
412        new_sequence_id = self.spark._random_sequence_id
413        df = self.copy()
414        for join_hint in df.pending_join_hints:
415            for expression in join_hint.expressions:
416                if expression.alias_or_name == self.sequence_id:
417                    expression.set("this", Column.ensure_col(new_sequence_id).expression)
418        df.spark._add_alias_to_mapping(name, new_sequence_id)
419        return df._convert_leaf_to_cte(sequence_id=new_sequence_id)
420
421    @operation(Operation.WHERE)
422    def where(self, column: t.Union[Column, bool], **kwargs) -> DataFrame:
423        col = self._ensure_and_normalize_col(column)
424        return self.copy(expression=self.expression.where(col.expression))
425
426    filter = where
427
428    @operation(Operation.GROUP_BY)
429    def groupBy(self, *cols, **kwargs) -> GroupedData:
430        columns = self._ensure_and_normalize_cols(cols)
431        return GroupedData(self, columns, self.last_op)
432
433    @operation(Operation.SELECT)
434    def agg(self, *exprs, **kwargs) -> DataFrame:
435        cols = self._ensure_and_normalize_cols(exprs)
436        return self.groupBy().agg(*cols)
437
438    @operation(Operation.FROM)
439    def join(
440        self,
441        other_df: DataFrame,
442        on: t.Union[str, t.List[str], Column, t.List[Column]],
443        how: str = "inner",
444        **kwargs,
445    ) -> DataFrame:
446        other_df = other_df._convert_leaf_to_cte()
447        join_columns = self._ensure_list_of_columns(on)
448        # We will determine actual "join on" expression later so we don't provide it at first
449        join_expression = self.expression.join(
450            other_df.latest_cte_name, join_type=how.replace("_", " ")
451        )
452        join_expression = self._add_ctes_to_expression(join_expression, other_df.expression.ctes)
453        self_columns = self._get_outer_select_columns(join_expression)
454        other_columns = self._get_outer_select_columns(other_df)
455        # Determines the join clause and select columns to be used passed on what type of columns were provided for
456        # the join. The columns returned changes based on how the on expression is provided.
457        if isinstance(join_columns[0].expression, exp.Column):
458            """
459            Unique characteristics of join on column names only:
460            * The column names are put at the front of the select list
461            * The column names are deduplicated across the entire select list and only the column names (other dups are allowed)
462            """
463            table_names = [
464                table.alias_or_name
465                for table in get_tables_from_expression_with_join(join_expression)
466            ]
467            potential_ctes = [
468                cte
469                for cte in join_expression.ctes
470                if cte.alias_or_name in table_names
471                and cte.alias_or_name != other_df.latest_cte_name
472            ]
473            # Determine the table to reference for the left side of the join by checking each of the left side
474            # tables and see if they have the column being referenced.
475            join_column_pairs = []
476            for join_column in join_columns:
477                num_matching_ctes = 0
478                for cte in potential_ctes:
479                    if join_column.alias_or_name in cte.this.named_selects:
480                        left_column = join_column.copy().set_table_name(cte.alias_or_name)
481                        right_column = join_column.copy().set_table_name(other_df.latest_cte_name)
482                        join_column_pairs.append((left_column, right_column))
483                        num_matching_ctes += 1
484                if num_matching_ctes > 1:
485                    raise ValueError(
486                        f"Column {join_column.alias_or_name} is ambiguous. Please specify the table name."
487                    )
488                elif num_matching_ctes == 0:
489                    raise ValueError(
490                        f"Column {join_column.alias_or_name} does not exist in any of the tables."
491                    )
492            join_clause = functools.reduce(
493                lambda x, y: x & y,
494                [left_column == right_column for left_column, right_column in join_column_pairs],
495            )
496            join_column_names = [left_col.alias_or_name for left_col, _ in join_column_pairs]
497            # To match spark behavior only the join clause gets deduplicated and it gets put in the front of the column list
498            select_column_names = [
499                (
500                    column.alias_or_name
501                    if not isinstance(column.expression.this, exp.Star)
502                    else column.sql()
503                )
504                for column in self_columns + other_columns
505            ]
506            select_column_names = [
507                column_name
508                for column_name in select_column_names
509                if column_name not in join_column_names
510            ]
511            select_column_names = join_column_names + select_column_names
512        else:
513            """
514            Unique characteristics of join on expressions:
515            * There is no deduplication of the results.
516            * The left join dataframe columns go first and right come after. No sort preference is given to join columns
517            """
518            join_columns = self._ensure_and_normalize_cols(join_columns, join_expression)
519            if len(join_columns) > 1:
520                join_columns = [functools.reduce(lambda x, y: x & y, join_columns)]
521            join_clause = join_columns[0]
522            select_column_names = [column.alias_or_name for column in self_columns + other_columns]
523
524        # Update the on expression with the actual join clause to replace the dummy one from before
525        join_expression.args["joins"][-1].set("on", join_clause.expression)
526        new_df = self.copy(expression=join_expression)
527        new_df.pending_join_hints.extend(self.pending_join_hints)
528        new_df.pending_hints.extend(other_df.pending_hints)
529        new_df = new_df.select.__wrapped__(new_df, *select_column_names)
530        return new_df
531
532    @operation(Operation.ORDER_BY)
533    def orderBy(
534        self,
535        *cols: t.Union[str, Column],
536        ascending: t.Optional[t.Union[t.Any, t.List[t.Any]]] = None,
537    ) -> DataFrame:
538        """
539        This implementation lets any ordered columns take priority over whatever is provided in `ascending`. Spark
540        has irregular behavior and can result in runtime errors. Users shouldn't be mixing the two anyways so this
541        is unlikely to come up.
542        """
543        columns = self._ensure_and_normalize_cols(cols)
544        pre_ordered_col_indexes = [
545            i for i, col in enumerate(columns) if isinstance(col.expression, exp.Ordered)
546        ]
547        if ascending is None:
548            ascending = [True] * len(columns)
549        elif not isinstance(ascending, list):
550            ascending = [ascending] * len(columns)
551        ascending = [bool(x) for i, x in enumerate(ascending)]
552        assert len(columns) == len(
553            ascending
554        ), "The length of items in ascending must equal the number of columns provided"
555        col_and_ascending = list(zip(columns, ascending))
556        order_by_columns = [
557            (
558                exp.Ordered(this=col.expression, desc=not asc)
559                if i not in pre_ordered_col_indexes
560                else columns[i].column_expression
561            )
562            for i, (col, asc) in enumerate(col_and_ascending)
563        ]
564        return self.copy(expression=self.expression.order_by(*order_by_columns))
565
566    sort = orderBy
567
568    @operation(Operation.FROM)
569    def union(self, other: DataFrame) -> DataFrame:
570        return self._set_operation(exp.Union, other, False)
571
572    unionAll = union
573
574    @operation(Operation.FROM)
575    def unionByName(self, other: DataFrame, allowMissingColumns: bool = False):
576        l_columns = self.columns
577        r_columns = other.columns
578        if not allowMissingColumns:
579            l_expressions = l_columns
580            r_expressions = l_columns
581        else:
582            l_expressions = []
583            r_expressions = []
584            r_columns_unused = copy(r_columns)
585            for l_column in l_columns:
586                l_expressions.append(l_column)
587                if l_column in r_columns:
588                    r_expressions.append(l_column)
589                    r_columns_unused.remove(l_column)
590                else:
591                    r_expressions.append(exp.alias_(exp.Null(), l_column, copy=False))
592            for r_column in r_columns_unused:
593                l_expressions.append(exp.alias_(exp.Null(), r_column, copy=False))
594                r_expressions.append(r_column)
595        r_df = (
596            other.copy()._convert_leaf_to_cte().select(*self._ensure_list_of_columns(r_expressions))
597        )
598        l_df = self.copy()
599        if allowMissingColumns:
600            l_df = l_df._convert_leaf_to_cte().select(*self._ensure_list_of_columns(l_expressions))
601        return l_df._set_operation(exp.Union, r_df, False)
602
603    @operation(Operation.FROM)
604    def intersect(self, other: DataFrame) -> DataFrame:
605        return self._set_operation(exp.Intersect, other, True)
606
607    @operation(Operation.FROM)
608    def intersectAll(self, other: DataFrame) -> DataFrame:
609        return self._set_operation(exp.Intersect, other, False)
610
611    @operation(Operation.FROM)
612    def exceptAll(self, other: DataFrame) -> DataFrame:
613        return self._set_operation(exp.Except, other, False)
614
615    @operation(Operation.SELECT)
616    def distinct(self) -> DataFrame:
617        return self.copy(expression=self.expression.distinct())
618
619    @operation(Operation.SELECT)
620    def dropDuplicates(self, subset: t.Optional[t.List[str]] = None):
621        if not subset:
622            return self.distinct()
623        column_names = ensure_list(subset)
624        window = Window.partitionBy(*column_names).orderBy(*column_names)
625        return (
626            self.copy()
627            .withColumn("row_num", F.row_number().over(window))
628            .where(F.col("row_num") == F.lit(1))
629            .drop("row_num")
630        )
631
632    @operation(Operation.FROM)
633    def dropna(
634        self,
635        how: str = "any",
636        thresh: t.Optional[int] = None,
637        subset: t.Optional[t.Union[str, t.Tuple[str, ...], t.List[str]]] = None,
638    ) -> DataFrame:
639        minimum_non_null = thresh or 0  # will be determined later if thresh is null
640        new_df = self.copy()
641        all_columns = self._get_outer_select_columns(new_df.expression)
642        if subset:
643            null_check_columns = self._ensure_and_normalize_cols(subset)
644        else:
645            null_check_columns = all_columns
646        if thresh is None:
647            minimum_num_nulls = 1 if how == "any" else len(null_check_columns)
648        else:
649            minimum_num_nulls = len(null_check_columns) - minimum_non_null + 1
650        if minimum_num_nulls > len(null_check_columns):
651            raise RuntimeError(
652                f"The minimum num nulls for dropna must be less than or equal to the number of columns. "
653                f"Minimum num nulls: {minimum_num_nulls}, Num Columns: {len(null_check_columns)}"
654            )
655        if_null_checks = [
656            F.when(column.isNull(), F.lit(1)).otherwise(F.lit(0)) for column in null_check_columns
657        ]
658        nulls_added_together = functools.reduce(lambda x, y: x + y, if_null_checks)
659        num_nulls = nulls_added_together.alias("num_nulls")
660        new_df = new_df.select(num_nulls, append=True)
661        filtered_df = new_df.where(F.col("num_nulls") < F.lit(minimum_num_nulls))
662        final_df = filtered_df.select(*all_columns)
663        return final_df
664
665    @operation(Operation.FROM)
666    def fillna(
667        self,
668        value: t.Union[ColumnLiterals],
669        subset: t.Optional[t.Union[str, t.Tuple[str, ...], t.List[str]]] = None,
670    ) -> DataFrame:
671        """
672        Functionality Difference: If you provide a value to replace a null and that type conflicts
673        with the type of the column then PySpark will just ignore your replacement.
674        This will try to cast them to be the same in some cases. So they won't always match.
675        Best to not mix types so make sure replacement is the same type as the column
676
677        Possibility for improvement: Use `typeof` function to get the type of the column
678        and check if it matches the type of the value provided. If not then make it null.
679        """
680        from sqlglot.dataframe.sql.functions import lit
681
682        values = None
683        columns = None
684        new_df = self.copy()
685        all_columns = self._get_outer_select_columns(new_df.expression)
686        all_column_mapping = {column.alias_or_name: column for column in all_columns}
687        if isinstance(value, dict):
688            values = list(value.values())
689            columns = self._ensure_and_normalize_cols(list(value))
690        if not columns:
691            columns = self._ensure_and_normalize_cols(subset) if subset else all_columns
692        if not values:
693            values = [value] * len(columns)
694        value_columns = [lit(value) for value in values]
695
696        null_replacement_mapping = {
697            column.alias_or_name: (
698                F.when(column.isNull(), value).otherwise(column).alias(column.alias_or_name)
699            )
700            for column, value in zip(columns, value_columns)
701        }
702        null_replacement_mapping = {**all_column_mapping, **null_replacement_mapping}
703        null_replacement_columns = [
704            null_replacement_mapping[column.alias_or_name] for column in all_columns
705        ]
706        new_df = new_df.select(*null_replacement_columns)
707        return new_df
708
709    @operation(Operation.FROM)
710    def replace(
711        self,
712        to_replace: t.Union[bool, int, float, str, t.List, t.Dict],
713        value: t.Optional[t.Union[bool, int, float, str, t.List]] = None,
714        subset: t.Optional[t.Collection[ColumnOrName] | ColumnOrName] = None,
715    ) -> DataFrame:
716        from sqlglot.dataframe.sql.functions import lit
717
718        old_values = None
719        new_df = self.copy()
720        all_columns = self._get_outer_select_columns(new_df.expression)
721        all_column_mapping = {column.alias_or_name: column for column in all_columns}
722
723        columns = self._ensure_and_normalize_cols(subset) if subset else all_columns
724        if isinstance(to_replace, dict):
725            old_values = list(to_replace)
726            new_values = list(to_replace.values())
727        elif not old_values and isinstance(to_replace, list):
728            assert isinstance(value, list), "value must be a list since the replacements are a list"
729            assert len(to_replace) == len(
730                value
731            ), "the replacements and values must be the same length"
732            old_values = to_replace
733            new_values = value
734        else:
735            old_values = [to_replace] * len(columns)
736            new_values = [value] * len(columns)
737        old_values = [lit(value) for value in old_values]
738        new_values = [lit(value) for value in new_values]
739
740        replacement_mapping = {}
741        for column in columns:
742            expression = Column(None)
743            for i, (old_value, new_value) in enumerate(zip(old_values, new_values)):
744                if i == 0:
745                    expression = F.when(column == old_value, new_value)
746                else:
747                    expression = expression.when(column == old_value, new_value)  # type: ignore
748            replacement_mapping[column.alias_or_name] = expression.otherwise(column).alias(
749                column.expression.alias_or_name
750            )
751
752        replacement_mapping = {**all_column_mapping, **replacement_mapping}
753        replacement_columns = [replacement_mapping[column.alias_or_name] for column in all_columns]
754        new_df = new_df.select(*replacement_columns)
755        return new_df
756
757    @operation(Operation.SELECT)
758    def withColumn(self, colName: str, col: Column) -> DataFrame:
759        col = self._ensure_and_normalize_col(col)
760        existing_col_names = self.expression.named_selects
761        existing_col_index = (
762            existing_col_names.index(colName) if colName in existing_col_names else None
763        )
764        if existing_col_index:
765            expression = self.expression.copy()
766            expression.expressions[existing_col_index] = col.expression
767            return self.copy(expression=expression)
768        return self.copy().select(col.alias(colName), append=True)
769
770    @operation(Operation.SELECT)
771    def withColumnRenamed(self, existing: str, new: str):
772        expression = self.expression.copy()
773        existing_columns = [
774            expression
775            for expression in expression.expressions
776            if expression.alias_or_name == existing
777        ]
778        if not existing_columns:
779            raise ValueError("Tried to rename a column that doesn't exist")
780        for existing_column in existing_columns:
781            if isinstance(existing_column, exp.Column):
782                existing_column.replace(exp.alias_(existing_column, new))
783            else:
784                existing_column.set("alias", exp.to_identifier(new))
785        return self.copy(expression=expression)
786
787    @operation(Operation.SELECT)
788    def drop(self, *cols: t.Union[str, Column]) -> DataFrame:
789        all_columns = self._get_outer_select_columns(self.expression)
790        drop_cols = self._ensure_and_normalize_cols(cols)
791        new_columns = [
792            col
793            for col in all_columns
794            if col.alias_or_name not in [drop_column.alias_or_name for drop_column in drop_cols]
795        ]
796        return self.copy().select(*new_columns, append=False)
797
798    @operation(Operation.LIMIT)
799    def limit(self, num: int) -> DataFrame:
800        return self.copy(expression=self.expression.limit(num))
801
802    @operation(Operation.NO_OP)
803    def hint(self, name: str, *parameters: t.Optional[t.Union[str, int]]) -> DataFrame:
804        parameter_list = ensure_list(parameters)
805        parameter_columns = (
806            self._ensure_list_of_columns(parameter_list)
807            if parameters
808            else Column.ensure_cols([self.sequence_id])
809        )
810        return self._hint(name, parameter_columns)
811
812    @operation(Operation.NO_OP)
813    def repartition(
814        self, numPartitions: t.Union[int, ColumnOrName], *cols: ColumnOrName
815    ) -> DataFrame:
816        num_partition_cols = self._ensure_list_of_columns(numPartitions)
817        columns = self._ensure_and_normalize_cols(cols)
818        args = num_partition_cols + columns
819        return self._hint("repartition", args)
820
821    @operation(Operation.NO_OP)
822    def coalesce(self, numPartitions: int) -> DataFrame:
823        num_partitions = Column.ensure_cols([numPartitions])
824        return self._hint("coalesce", num_partitions)
825
826    @operation(Operation.NO_OP)
827    def cache(self) -> DataFrame:
828        return self._cache(storage_level="MEMORY_AND_DISK")
829
830    @operation(Operation.NO_OP)
831    def persist(self, storageLevel: str = "MEMORY_AND_DISK_SER") -> DataFrame:
832        """
833        Storage Level Options: https://spark.apache.org/docs/3.0.0-preview/sql-ref-syntax-aux-cache-cache-table.html
834        """
835        return self._cache(storageLevel)