sqlglot.dataframe.sql

 21class SparkSession:
 22    DEFAULT_DIALECT = "spark"
 23    _instance = None
 24
 25    def __init__(self):
 26        if not hasattr(self, "known_ids"):
 27            self.known_ids = set()
 28            self.known_branch_ids = set()
 29            self.known_sequence_ids = set()
 30            self.name_to_sequence_id_mapping = defaultdict(list)
 31            self.incrementing_id = 1
 32            self.dialect = Dialect.get_or_raise(self.DEFAULT_DIALECT)
 33
 34    def __new__(cls, *args, **kwargs) -> SparkSession:
 35        if cls._instance is None:
 36            cls._instance = super().__new__(cls)
 37        return cls._instance
 38
 39    @property
 40    def read(self) -> DataFrameReader:
 41        return DataFrameReader(self)
 42
 43    def table(self, tableName: str) -> DataFrame:
 44        return self.read.table(tableName)
 45
 46    def createDataFrame(
 47        self,
 48        data: t.Sequence[t.Union[t.Dict[str, ColumnLiterals], t.List[ColumnLiterals], t.Tuple]],
 49        schema: t.Optional[SchemaInput] = None,
 50        samplingRatio: t.Optional[float] = None,
 51        verifySchema: bool = False,
 52    ) -> DataFrame:
 53        from sqlglot.dataframe.sql.dataframe import DataFrame
 54
 55        if samplingRatio is not None or verifySchema:
 56            raise NotImplementedError("Sampling Ratio and Verify Schema are not supported")
 57        if schema is not None and (
 58            not isinstance(schema, (StructType, str, list))
 59            or (isinstance(schema, list) and not isinstance(schema[0], str))
 60        ):
 61            raise NotImplementedError("Only schema of either list or string of list supported")
 62        if not data:
 63            raise ValueError("Must provide data to create into a DataFrame")
 64
 65        column_mapping: t.Dict[str, t.Optional[str]]
 66        if schema is not None:
 67            column_mapping = get_column_mapping_from_schema_input(schema)
 68        elif isinstance(data[0], dict):
 69            column_mapping = {col_name.strip(): None for col_name in data[0]}
 70        else:
 71            column_mapping = {f"_{i}": None for i in range(1, len(data[0]) + 1)}
 72
 73        data_expressions = [
 74            exp.Tuple(
 75                expressions=list(
 76                    map(
 77                        lambda x: F.lit(x).expression,
 78                        row if not isinstance(row, dict) else row.values(),
 79                    )
 80                )
 81            )
 82            for row in data
 83        ]
 84
 85        sel_columns = [
 86            F.col(name).cast(data_type).alias(name).expression
 87            if data_type is not None
 88            else F.col(name).expression
 89            for name, data_type in column_mapping.items()
 90        ]
 91
 92        select_kwargs = {
 93            "expressions": sel_columns,
 94            "from": exp.From(
 95                this=exp.Values(
 96                    expressions=data_expressions,
 97                    alias=exp.TableAlias(
 98                        this=exp.to_identifier(self._auto_incrementing_name),
 99                        columns=[exp.to_identifier(col_name) for col_name in column_mapping],
100                    ),
101                ),
102            ),
103        }
104
105        sel_expression = exp.Select(**select_kwargs)
106        return DataFrame(self, sel_expression)
107
108    def sql(self, sqlQuery: str) -> DataFrame:
109        expression = sqlglot.parse_one(sqlQuery, read=self.dialect)
110        if isinstance(expression, exp.Select):
111            df = DataFrame(self, expression)
112            df = df._convert_leaf_to_cte()
113        elif isinstance(expression, (exp.Create, exp.Insert)):
114            select_expression = expression.expression.copy()
115            if isinstance(expression, exp.Insert):
116                select_expression.set("with", expression.args.get("with"))
117                expression.set("with", None)
118            del expression.args["expression"]
119            df = DataFrame(self, select_expression, output_expression_container=expression)  # type: ignore
120            df = df._convert_leaf_to_cte()
121        else:
122            raise ValueError(
123                "Unknown expression type provided in the SQL. Please create an issue with the SQL."
124            )
125        return df
126
127    @property
128    def _auto_incrementing_name(self) -> str:
129        name = f"a{self.incrementing_id}"
130        self.incrementing_id += 1
131        return name
132
133    @property
134    def _random_branch_id(self) -> str:
135        id = self._random_id
136        self.known_branch_ids.add(id)
137        return id
138
139    @property
140    def _random_sequence_id(self):
141        id = self._random_id
142        self.known_sequence_ids.add(id)
143        return id
144
145    @property
146    def _random_id(self) -> str:
147        id = "r" + uuid.uuid4().hex
148        self.known_ids.add(id)
149        return id
150
151    @property
152    def _join_hint_names(self) -> t.Set[str]:
153        return {"BROADCAST", "MERGE", "SHUFFLE_HASH", "SHUFFLE_REPLICATE_NL"}
154
155    def _add_alias_to_mapping(self, name: str, sequence_id: str):
156        self.name_to_sequence_id_mapping[name].append(sequence_id)
157
158    class Builder:
159        SQLFRAME_DIALECT_KEY = "sqlframe.dialect"
160
161        def __init__(self):
162            self.dialect = "spark"
163
164        def __getattr__(self, item) -> SparkSession.Builder:
165            return self
166
167        def __call__(self, *args, **kwargs):
168            return self
169
170        def config(
171            self,
172            key: t.Optional[str] = None,
173            value: t.Optional[t.Any] = None,
174            *,
175            map: t.Optional[t.Dict[str, t.Any]] = None,
176            **kwargs: t.Any,
177        ) -> SparkSession.Builder:
178            if key == self.SQLFRAME_DIALECT_KEY:
179                self.dialect = value
180            elif map and self.SQLFRAME_DIALECT_KEY in map:
181                self.dialect = map[self.SQLFRAME_DIALECT_KEY]
182            return self
183
184        def getOrCreate(self) -> SparkSession:
185            spark = SparkSession()
186            spark.dialect = Dialect.get_or_raise(self.dialect)
187            return spark
188
189    @classproperty
190    def builder(cls) -> Builder:
191        return cls.Builder()

158    class Builder:
159        SQLFRAME_DIALECT_KEY = "sqlframe.dialect"
160
161        def __init__(self):
162            self.dialect = "spark"
163
164        def __getattr__(self, item) -> SparkSession.Builder:
165            return self
166
167        def __call__(self, *args, **kwargs):
168            return self
169
170        def config(
171            self,
172            key: t.Optional[str] = None,
173            value: t.Optional[t.Any] = None,
174            *,
175            map: t.Optional[t.Dict[str, t.Any]] = None,
176            **kwargs: t.Any,
177        ) -> SparkSession.Builder:
178            if key == self.SQLFRAME_DIALECT_KEY:
179                self.dialect = value
180            elif map and self.SQLFRAME_DIALECT_KEY in map:
181                self.dialect = map[self.SQLFRAME_DIALECT_KEY]
182            return self
183
184        def getOrCreate(self) -> SparkSession:
185            spark = SparkSession()
186            spark.dialect = Dialect.get_or_raise(self.dialect)
187            return spark

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