sqlglot.dialects.presto

  1from __future__ import annotations
  2
  3import typing as t
  4
  5from sqlglot import exp, generator, parser, tokens, transforms
  6from sqlglot.dialects.dialect import (
  7    Dialect,
  8    NormalizationStrategy,
  9    binary_from_function,
 10    bool_xor_sql,
 11    date_trunc_to_time,
 12    datestrtodate_sql,
 13    encode_decode_sql,
 14    build_formatted_time,
 15    if_sql,
 16    left_to_substring_sql,
 17    no_ilike_sql,
 18    no_pivot_sql,
 19    no_timestamp_sql,
 20    regexp_extract_sql,
 21    rename_func,
 22    right_to_substring_sql,
 23    sha256_sql,
 24    strposition_sql,
 25    struct_extract_sql,
 26    timestamptrunc_sql,
 27    timestrtotime_sql,
 28    ts_or_ds_add_cast,
 29    unit_to_str,
 30    sequence_sql,
 31    build_regexp_extract,
 32    explode_to_unnest_sql,
 33)
 34from sqlglot.dialects.hive import Hive
 35from sqlglot.dialects.mysql import MySQL
 36from sqlglot.helper import apply_index_offset, seq_get
 37from sqlglot.optimizer.scope import find_all_in_scope
 38from sqlglot.tokens import TokenType
 39from sqlglot.transforms import unqualify_columns
 40from sqlglot.generator import unsupported_args
 41
 42DATE_ADD_OR_SUB = t.Union[exp.DateAdd, exp.TimestampAdd, exp.DateSub]
 43
 44
 45def _initcap_sql(self: Presto.Generator, expression: exp.Initcap) -> str:
 46    regex = r"(\w)(\w*)"
 47    return f"REGEXP_REPLACE({self.sql(expression, 'this')}, '{regex}', x -> UPPER(x[1]) || LOWER(x[2]))"
 48
 49
 50def _no_sort_array(self: Presto.Generator, expression: exp.SortArray) -> str:
 51    if expression.args.get("asc") == exp.false():
 52        comparator = "(a, b) -> CASE WHEN a < b THEN 1 WHEN a > b THEN -1 ELSE 0 END"
 53    else:
 54        comparator = None
 55    return self.func("ARRAY_SORT", expression.this, comparator)
 56
 57
 58def _schema_sql(self: Presto.Generator, expression: exp.Schema) -> str:
 59    if isinstance(expression.parent, exp.PartitionedByProperty):
 60        # Any columns in the ARRAY[] string literals should not be quoted
 61        expression.transform(lambda n: n.name if isinstance(n, exp.Identifier) else n, copy=False)
 62
 63        partition_exprs = [
 64            self.sql(c) if isinstance(c, (exp.Func, exp.Property)) else self.sql(c, "this")
 65            for c in expression.expressions
 66        ]
 67        return self.sql(exp.Array(expressions=[exp.Literal.string(c) for c in partition_exprs]))
 68
 69    if expression.parent:
 70        for schema in expression.parent.find_all(exp.Schema):
 71            if schema is expression:
 72                continue
 73
 74            column_defs = schema.find_all(exp.ColumnDef)
 75            if column_defs and isinstance(schema.parent, exp.Property):
 76                expression.expressions.extend(column_defs)
 77
 78    return self.schema_sql(expression)
 79
 80
 81def _quantile_sql(self: Presto.Generator, expression: exp.Quantile) -> str:
 82    self.unsupported("Presto does not support exact quantiles")
 83    return self.func("APPROX_PERCENTILE", expression.this, expression.args.get("quantile"))
 84
 85
 86def _str_to_time_sql(
 87    self: Presto.Generator, expression: exp.StrToDate | exp.StrToTime | exp.TsOrDsToDate
 88) -> str:
 89    return self.func("DATE_PARSE", expression.this, self.format_time(expression))
 90
 91
 92def _ts_or_ds_to_date_sql(self: Presto.Generator, expression: exp.TsOrDsToDate) -> str:
 93    time_format = self.format_time(expression)
 94    if time_format and time_format not in (Presto.TIME_FORMAT, Presto.DATE_FORMAT):
 95        return self.sql(exp.cast(_str_to_time_sql(self, expression), exp.DataType.Type.DATE))
 96    return self.sql(
 97        exp.cast(exp.cast(expression.this, exp.DataType.Type.TIMESTAMP), exp.DataType.Type.DATE)
 98    )
 99
100
101def _ts_or_ds_add_sql(self: Presto.Generator, expression: exp.TsOrDsAdd) -> str:
102    expression = ts_or_ds_add_cast(expression)
103    unit = unit_to_str(expression)
104    return self.func("DATE_ADD", unit, expression.expression, expression.this)
105
106
107def _ts_or_ds_diff_sql(self: Presto.Generator, expression: exp.TsOrDsDiff) -> str:
108    this = exp.cast(expression.this, exp.DataType.Type.TIMESTAMP)
109    expr = exp.cast(expression.expression, exp.DataType.Type.TIMESTAMP)
110    unit = unit_to_str(expression)
111    return self.func("DATE_DIFF", unit, expr, this)
112
113
114def _build_approx_percentile(args: t.List) -> exp.Expression:
115    if len(args) == 4:
116        return exp.ApproxQuantile(
117            this=seq_get(args, 0),
118            weight=seq_get(args, 1),
119            quantile=seq_get(args, 2),
120            accuracy=seq_get(args, 3),
121        )
122    if len(args) == 3:
123        return exp.ApproxQuantile(
124            this=seq_get(args, 0), quantile=seq_get(args, 1), accuracy=seq_get(args, 2)
125        )
126    return exp.ApproxQuantile.from_arg_list(args)
127
128
129def _build_from_unixtime(args: t.List) -> exp.Expression:
130    if len(args) == 3:
131        return exp.UnixToTime(
132            this=seq_get(args, 0),
133            hours=seq_get(args, 1),
134            minutes=seq_get(args, 2),
135        )
136    if len(args) == 2:
137        return exp.UnixToTime(this=seq_get(args, 0), zone=seq_get(args, 1))
138
139    return exp.UnixToTime.from_arg_list(args)
140
141
142def _first_last_sql(self: Presto.Generator, expression: exp.Func) -> str:
143    """
144    Trino doesn't support FIRST / LAST as functions, but they're valid in the context
145    of MATCH_RECOGNIZE, so we need to preserve them in that case. In all other cases
146    they're converted into an ARBITRARY call.
147
148    Reference: https://trino.io/docs/current/sql/match-recognize.html#logical-navigation-functions
149    """
150    if isinstance(expression.find_ancestor(exp.MatchRecognize, exp.Select), exp.MatchRecognize):
151        return self.function_fallback_sql(expression)
152
153    return rename_func("ARBITRARY")(self, expression)
154
155
156def _unix_to_time_sql(self: Presto.Generator, expression: exp.UnixToTime) -> str:
157    scale = expression.args.get("scale")
158    timestamp = self.sql(expression, "this")
159    if scale in (None, exp.UnixToTime.SECONDS):
160        return rename_func("FROM_UNIXTIME")(self, expression)
161
162    return f"FROM_UNIXTIME(CAST({timestamp} AS DOUBLE) / POW(10, {scale}))"
163
164
165def _to_int(self: Presto.Generator, expression: exp.Expression) -> exp.Expression:
166    if not expression.type:
167        from sqlglot.optimizer.annotate_types import annotate_types
168
169        annotate_types(expression, dialect=self.dialect)
170    if expression.type and expression.type.this not in exp.DataType.INTEGER_TYPES:
171        return exp.cast(expression, to=exp.DataType.Type.BIGINT)
172    return expression
173
174
175def _build_to_char(args: t.List) -> exp.TimeToStr:
176    fmt = seq_get(args, 1)
177    if isinstance(fmt, exp.Literal):
178        # We uppercase this to match Teradata's format mapping keys
179        fmt.set("this", fmt.this.upper())
180
181    # We use "teradata" on purpose here, because the time formats are different in Presto.
182    # See https://prestodb.io/docs/current/functions/teradata.html?highlight=to_char#to_char
183    return build_formatted_time(exp.TimeToStr, "teradata")(args)
184
185
186def _date_delta_sql(
187    name: str, negate_interval: bool = False
188) -> t.Callable[[Presto.Generator, DATE_ADD_OR_SUB], str]:
189    def _delta_sql(self: Presto.Generator, expression: DATE_ADD_OR_SUB) -> str:
190        interval = _to_int(self, expression.expression)
191        return self.func(
192            name,
193            unit_to_str(expression),
194            interval * (-1) if negate_interval else interval,
195            expression.this,
196        )
197
198    return _delta_sql
199
200
201def _explode_to_unnest_sql(self: Presto.Generator, expression: exp.Lateral) -> str:
202    explode = expression.this
203    if isinstance(explode, exp.Explode):
204        exploded_type = explode.this.type
205        alias = expression.args.get("alias")
206
207        # This attempts a best-effort transpilation of LATERAL VIEW EXPLODE on a struct array
208        if (
209            isinstance(alias, exp.TableAlias)
210            and isinstance(exploded_type, exp.DataType)
211            and exploded_type.is_type(exp.DataType.Type.ARRAY)
212            and exploded_type.expressions
213            and exploded_type.expressions[0].is_type(exp.DataType.Type.STRUCT)
214        ):
215            # When unnesting a ROW in Presto, it produces N columns, so we need to fix the alias
216            alias.set("columns", [c.this.copy() for c in exploded_type.expressions[0].expressions])
217    elif isinstance(explode, exp.Inline):
218        explode.replace(exp.Explode(this=explode.this.copy()))
219
220    return explode_to_unnest_sql(self, expression)
221
222
223def _amend_exploded_column_table(expression: exp.Expression) -> exp.Expression:
224    # We check for expression.type because the columns can be amended only if types were inferred
225    if isinstance(expression, exp.Select) and expression.type:
226        for lateral in expression.args.get("laterals") or []:
227            alias = lateral.args.get("alias")
228            if (
229                not isinstance(lateral.this, exp.Explode)
230                or not isinstance(alias, exp.TableAlias)
231                or len(alias.columns) != 1
232            ):
233                continue
234
235            new_table = alias.this
236            old_table = alias.columns[0].name.lower()
237
238            # When transpiling a LATERAL VIEW EXPLODE Spark query, the exploded fields may be qualified
239            # with the struct column, resulting in invalid Presto references that need to be amended
240            for column in find_all_in_scope(expression, exp.Column):
241                if column.db.lower() == old_table:
242                    column.set("table", column.args["db"].pop())
243                elif column.table.lower() == old_table:
244                    column.set("table", new_table.copy())
245                elif column.name.lower() == old_table and isinstance(column.parent, exp.Dot):
246                    column.parent.replace(exp.column(column.parent.expression, table=new_table))
247
248    return expression
249
250
251class Presto(Dialect):
252    INDEX_OFFSET = 1
253    NULL_ORDERING = "nulls_are_last"
254    TIME_FORMAT = MySQL.TIME_FORMAT
255    STRICT_STRING_CONCAT = True
256    SUPPORTS_SEMI_ANTI_JOIN = False
257    TYPED_DIVISION = True
258    TABLESAMPLE_SIZE_IS_PERCENT = True
259    LOG_BASE_FIRST: t.Optional[bool] = None
260    SUPPORTS_VALUES_DEFAULT = False
261
262    TIME_MAPPING = MySQL.TIME_MAPPING
263
264    # https://github.com/trinodb/trino/issues/17
265    # https://github.com/trinodb/trino/issues/12289
266    # https://github.com/prestodb/presto/issues/2863
267    NORMALIZATION_STRATEGY = NormalizationStrategy.CASE_INSENSITIVE
268
269    # The result of certain math functions in Presto/Trino is of type
270    # equal to the input type e.g: FLOOR(5.5/2) -> DECIMAL, FLOOR(5/2) -> BIGINT
271    ANNOTATORS = {
272        **Dialect.ANNOTATORS,
273        exp.Floor: lambda self, e: self._annotate_by_args(e, "this"),
274        exp.Ceil: lambda self, e: self._annotate_by_args(e, "this"),
275        exp.Mod: lambda self, e: self._annotate_by_args(e, "this", "expression"),
276        exp.Round: lambda self, e: self._annotate_by_args(e, "this"),
277        exp.Sign: lambda self, e: self._annotate_by_args(e, "this"),
278        exp.Abs: lambda self, e: self._annotate_by_args(e, "this"),
279        exp.Rand: lambda self, e: self._annotate_by_args(e, "this")
280        if e.this
281        else self._set_type(e, exp.DataType.Type.DOUBLE),
282    }
283
284    class Tokenizer(tokens.Tokenizer):
285        HEX_STRINGS = [("x'", "'"), ("X'", "'")]
286        UNICODE_STRINGS = [
287            (prefix + q, q)
288            for q in t.cast(t.List[str], tokens.Tokenizer.QUOTES)
289            for prefix in ("U&", "u&")
290        ]
291
292        KEYWORDS = {
293            **tokens.Tokenizer.KEYWORDS,
294            "DEALLOCATE PREPARE": TokenType.COMMAND,
295            "DESCRIBE INPUT": TokenType.COMMAND,
296            "DESCRIBE OUTPUT": TokenType.COMMAND,
297            "RESET SESSION": TokenType.COMMAND,
298            "START": TokenType.BEGIN,
299            "MATCH_RECOGNIZE": TokenType.MATCH_RECOGNIZE,
300            "ROW": TokenType.STRUCT,
301            "IPADDRESS": TokenType.IPADDRESS,
302            "IPPREFIX": TokenType.IPPREFIX,
303            "TDIGEST": TokenType.TDIGEST,
304            "HYPERLOGLOG": TokenType.HLLSKETCH,
305        }
306        KEYWORDS.pop("/*+")
307        KEYWORDS.pop("QUALIFY")
308
309    class Parser(parser.Parser):
310        VALUES_FOLLOWED_BY_PAREN = False
311
312        FUNCTIONS = {
313            **parser.Parser.FUNCTIONS,
314            "ARBITRARY": exp.AnyValue.from_arg_list,
315            "APPROX_DISTINCT": exp.ApproxDistinct.from_arg_list,
316            "APPROX_PERCENTILE": _build_approx_percentile,
317            "BITWISE_AND": binary_from_function(exp.BitwiseAnd),
318            "BITWISE_NOT": lambda args: exp.BitwiseNot(this=seq_get(args, 0)),
319            "BITWISE_OR": binary_from_function(exp.BitwiseOr),
320            "BITWISE_XOR": binary_from_function(exp.BitwiseXor),
321            "CARDINALITY": exp.ArraySize.from_arg_list,
322            "CONTAINS": exp.ArrayContains.from_arg_list,
323            "DATE_ADD": lambda args: exp.DateAdd(
324                this=seq_get(args, 2), expression=seq_get(args, 1), unit=seq_get(args, 0)
325            ),
326            "DATE_DIFF": lambda args: exp.DateDiff(
327                this=seq_get(args, 2), expression=seq_get(args, 1), unit=seq_get(args, 0)
328            ),
329            "DATE_FORMAT": build_formatted_time(exp.TimeToStr, "presto"),
330            "DATE_PARSE": build_formatted_time(exp.StrToTime, "presto"),
331            "DATE_TRUNC": date_trunc_to_time,
332            "DAY_OF_WEEK": exp.DayOfWeekIso.from_arg_list,
333            "ELEMENT_AT": lambda args: exp.Bracket(
334                this=seq_get(args, 0), expressions=[seq_get(args, 1)], offset=1, safe=True
335            ),
336            "FROM_HEX": exp.Unhex.from_arg_list,
337            "FROM_UNIXTIME": _build_from_unixtime,
338            "FROM_UTF8": lambda args: exp.Decode(
339                this=seq_get(args, 0), replace=seq_get(args, 1), charset=exp.Literal.string("utf-8")
340            ),
341            "JSON_FORMAT": lambda args: exp.JSONFormat(
342                this=seq_get(args, 0), options=seq_get(args, 1), is_json=True
343            ),
344            "LEVENSHTEIN_DISTANCE": exp.Levenshtein.from_arg_list,
345            "NOW": exp.CurrentTimestamp.from_arg_list,
346            "REGEXP_EXTRACT": build_regexp_extract(exp.RegexpExtract),
347            "REGEXP_EXTRACT_ALL": build_regexp_extract(exp.RegexpExtractAll),
348            "REGEXP_REPLACE": lambda args: exp.RegexpReplace(
349                this=seq_get(args, 0),
350                expression=seq_get(args, 1),
351                replacement=seq_get(args, 2) or exp.Literal.string(""),
352            ),
353            "ROW": exp.Struct.from_arg_list,
354            "SEQUENCE": exp.GenerateSeries.from_arg_list,
355            "SET_AGG": exp.ArrayUniqueAgg.from_arg_list,
356            "SPLIT_TO_MAP": exp.StrToMap.from_arg_list,
357            "STRPOS": lambda args: exp.StrPosition(
358                this=seq_get(args, 0), substr=seq_get(args, 1), occurrence=seq_get(args, 2)
359            ),
360            "TO_CHAR": _build_to_char,
361            "TO_UNIXTIME": exp.TimeToUnix.from_arg_list,
362            "TO_UTF8": lambda args: exp.Encode(
363                this=seq_get(args, 0), charset=exp.Literal.string("utf-8")
364            ),
365            "MD5": exp.MD5Digest.from_arg_list,
366            "SHA256": lambda args: exp.SHA2(this=seq_get(args, 0), length=exp.Literal.number(256)),
367            "SHA512": lambda args: exp.SHA2(this=seq_get(args, 0), length=exp.Literal.number(512)),
368        }
369
370        FUNCTION_PARSERS = parser.Parser.FUNCTION_PARSERS.copy()
371        FUNCTION_PARSERS.pop("TRIM")
372
373    class Generator(generator.Generator):
374        INTERVAL_ALLOWS_PLURAL_FORM = False
375        JOIN_HINTS = False
376        TABLE_HINTS = False
377        QUERY_HINTS = False
378        IS_BOOL_ALLOWED = False
379        TZ_TO_WITH_TIME_ZONE = True
380        NVL2_SUPPORTED = False
381        STRUCT_DELIMITER = ("(", ")")
382        LIMIT_ONLY_LITERALS = True
383        SUPPORTS_SINGLE_ARG_CONCAT = False
384        LIKE_PROPERTY_INSIDE_SCHEMA = True
385        MULTI_ARG_DISTINCT = False
386        SUPPORTS_TO_NUMBER = False
387        HEX_FUNC = "TO_HEX"
388        PARSE_JSON_NAME = "JSON_PARSE"
389        PAD_FILL_PATTERN_IS_REQUIRED = True
390        EXCEPT_INTERSECT_SUPPORT_ALL_CLAUSE = False
391        SUPPORTS_MEDIAN = False
392        ARRAY_SIZE_NAME = "CARDINALITY"
393
394        PROPERTIES_LOCATION = {
395            **generator.Generator.PROPERTIES_LOCATION,
396            exp.LocationProperty: exp.Properties.Location.UNSUPPORTED,
397            exp.VolatileProperty: exp.Properties.Location.UNSUPPORTED,
398        }
399
400        TYPE_MAPPING = {
401            **generator.Generator.TYPE_MAPPING,
402            exp.DataType.Type.BINARY: "VARBINARY",
403            exp.DataType.Type.BIT: "BOOLEAN",
404            exp.DataType.Type.DATETIME: "TIMESTAMP",
405            exp.DataType.Type.DATETIME64: "TIMESTAMP",
406            exp.DataType.Type.FLOAT: "REAL",
407            exp.DataType.Type.HLLSKETCH: "HYPERLOGLOG",
408            exp.DataType.Type.INT: "INTEGER",
409            exp.DataType.Type.STRUCT: "ROW",
410            exp.DataType.Type.TEXT: "VARCHAR",
411            exp.DataType.Type.TIMESTAMPTZ: "TIMESTAMP",
412            exp.DataType.Type.TIMESTAMPNTZ: "TIMESTAMP",
413            exp.DataType.Type.TIMETZ: "TIME",
414        }
415
416        TRANSFORMS = {
417            **generator.Generator.TRANSFORMS,
418            exp.AnyValue: rename_func("ARBITRARY"),
419            exp.ApproxQuantile: rename_func("APPROX_PERCENTILE"),
420            exp.ArgMax: rename_func("MAX_BY"),
421            exp.ArgMin: rename_func("MIN_BY"),
422            exp.Array: lambda self, e: f"ARRAY[{self.expressions(e, flat=True)}]",
423            exp.ArrayAny: rename_func("ANY_MATCH"),
424            exp.ArrayConcat: rename_func("CONCAT"),
425            exp.ArrayContains: rename_func("CONTAINS"),
426            exp.ArrayToString: rename_func("ARRAY_JOIN"),
427            exp.ArrayUniqueAgg: rename_func("SET_AGG"),
428            exp.AtTimeZone: rename_func("AT_TIMEZONE"),
429            exp.BitwiseAnd: lambda self, e: self.func("BITWISE_AND", e.this, e.expression),
430            exp.BitwiseLeftShift: lambda self, e: self.func(
431                "BITWISE_ARITHMETIC_SHIFT_LEFT", e.this, e.expression
432            ),
433            exp.BitwiseNot: lambda self, e: self.func("BITWISE_NOT", e.this),
434            exp.BitwiseOr: lambda self, e: self.func("BITWISE_OR", e.this, e.expression),
435            exp.BitwiseRightShift: lambda self, e: self.func(
436                "BITWISE_ARITHMETIC_SHIFT_RIGHT", e.this, e.expression
437            ),
438            exp.BitwiseXor: lambda self, e: self.func("BITWISE_XOR", e.this, e.expression),
439            exp.Cast: transforms.preprocess([transforms.epoch_cast_to_ts]),
440            exp.CurrentTime: lambda *_: "CURRENT_TIME",
441            exp.CurrentTimestamp: lambda *_: "CURRENT_TIMESTAMP",
442            exp.CurrentUser: lambda *_: "CURRENT_USER",
443            exp.DateAdd: _date_delta_sql("DATE_ADD"),
444            exp.DateDiff: lambda self, e: self.func(
445                "DATE_DIFF", unit_to_str(e), e.expression, e.this
446            ),
447            exp.DateStrToDate: datestrtodate_sql,
448            exp.DateToDi: lambda self,
449            e: f"CAST(DATE_FORMAT({self.sql(e, 'this')}, {Presto.DATEINT_FORMAT}) AS INT)",
450            exp.DateSub: _date_delta_sql("DATE_ADD", negate_interval=True),
451            exp.DayOfWeek: lambda self, e: f"(({self.func('DAY_OF_WEEK', e.this)} % 7) + 1)",
452            exp.DayOfWeekIso: rename_func("DAY_OF_WEEK"),
453            exp.Decode: lambda self, e: encode_decode_sql(self, e, "FROM_UTF8"),
454            exp.DiToDate: lambda self,
455            e: f"CAST(DATE_PARSE(CAST({self.sql(e, 'this')} AS VARCHAR), {Presto.DATEINT_FORMAT}) AS DATE)",
456            exp.Encode: lambda self, e: encode_decode_sql(self, e, "TO_UTF8"),
457            exp.FileFormatProperty: lambda self, e: f"FORMAT='{e.name.upper()}'",
458            exp.First: _first_last_sql,
459            exp.FromTimeZone: lambda self,
460            e: f"WITH_TIMEZONE({self.sql(e, 'this')}, {self.sql(e, 'zone')}) AT TIME ZONE 'UTC'",
461            exp.GenerateSeries: sequence_sql,
462            exp.GenerateDateArray: sequence_sql,
463            exp.Group: transforms.preprocess([transforms.unalias_group]),
464            exp.If: if_sql(),
465            exp.ILike: no_ilike_sql,
466            exp.Initcap: _initcap_sql,
467            exp.Last: _first_last_sql,
468            exp.LastDay: lambda self, e: self.func("LAST_DAY_OF_MONTH", e.this),
469            exp.Lateral: _explode_to_unnest_sql,
470            exp.Left: left_to_substring_sql,
471            exp.Levenshtein: unsupported_args("ins_cost", "del_cost", "sub_cost", "max_dist")(
472                rename_func("LEVENSHTEIN_DISTANCE")
473            ),
474            exp.LogicalAnd: rename_func("BOOL_AND"),
475            exp.LogicalOr: rename_func("BOOL_OR"),
476            exp.Pivot: no_pivot_sql,
477            exp.Quantile: _quantile_sql,
478            exp.RegexpExtract: regexp_extract_sql,
479            exp.RegexpExtractAll: regexp_extract_sql,
480            exp.Right: right_to_substring_sql,
481            exp.Schema: _schema_sql,
482            exp.SchemaCommentProperty: lambda self, e: self.naked_property(e),
483            exp.Select: transforms.preprocess(
484                [
485                    transforms.eliminate_qualify,
486                    transforms.eliminate_distinct_on,
487                    transforms.explode_projection_to_unnest(1),
488                    transforms.eliminate_semi_and_anti_joins,
489                    _amend_exploded_column_table,
490                ]
491            ),
492            exp.SortArray: _no_sort_array,
493            exp.StrPosition: lambda self, e: strposition_sql(self, e, supports_occurrence=True),
494            exp.StrToDate: lambda self, e: f"CAST({_str_to_time_sql(self, e)} AS DATE)",
495            exp.StrToMap: rename_func("SPLIT_TO_MAP"),
496            exp.StrToTime: _str_to_time_sql,
497            exp.StructExtract: struct_extract_sql,
498            exp.Table: transforms.preprocess([transforms.unnest_generate_series]),
499            exp.Timestamp: no_timestamp_sql,
500            exp.TimestampAdd: _date_delta_sql("DATE_ADD"),
501            exp.TimestampTrunc: timestamptrunc_sql(),
502            exp.TimeStrToDate: timestrtotime_sql,
503            exp.TimeStrToTime: timestrtotime_sql,
504            exp.TimeStrToUnix: lambda self, e: self.func(
505                "TO_UNIXTIME", self.func("DATE_PARSE", e.this, Presto.TIME_FORMAT)
506            ),
507            exp.TimeToStr: lambda self, e: self.func("DATE_FORMAT", e.this, self.format_time(e)),
508            exp.TimeToUnix: rename_func("TO_UNIXTIME"),
509            exp.ToChar: lambda self, e: self.func("DATE_FORMAT", e.this, self.format_time(e)),
510            exp.TryCast: transforms.preprocess([transforms.epoch_cast_to_ts]),
511            exp.TsOrDiToDi: lambda self,
512            e: f"CAST(SUBSTR(REPLACE(CAST({self.sql(e, 'this')} AS VARCHAR), '-', ''), 1, 8) AS INT)",
513            exp.TsOrDsAdd: _ts_or_ds_add_sql,
514            exp.TsOrDsDiff: _ts_or_ds_diff_sql,
515            exp.TsOrDsToDate: _ts_or_ds_to_date_sql,
516            exp.Unhex: rename_func("FROM_HEX"),
517            exp.UnixToStr: lambda self,
518            e: f"DATE_FORMAT(FROM_UNIXTIME({self.sql(e, 'this')}), {self.format_time(e)})",
519            exp.UnixToTime: _unix_to_time_sql,
520            exp.UnixToTimeStr: lambda self,
521            e: f"CAST(FROM_UNIXTIME({self.sql(e, 'this')}) AS VARCHAR)",
522            exp.VariancePop: rename_func("VAR_POP"),
523            exp.With: transforms.preprocess([transforms.add_recursive_cte_column_names]),
524            exp.WithinGroup: transforms.preprocess(
525                [transforms.remove_within_group_for_percentiles]
526            ),
527            exp.Xor: bool_xor_sql,
528            exp.MD5Digest: rename_func("MD5"),
529            exp.SHA: rename_func("SHA1"),
530            exp.SHA2: sha256_sql,
531        }
532
533        RESERVED_KEYWORDS = {
534            "alter",
535            "and",
536            "as",
537            "between",
538            "by",
539            "case",
540            "cast",
541            "constraint",
542            "create",
543            "cross",
544            "current_time",
545            "current_timestamp",
546            "deallocate",
547            "delete",
548            "describe",
549            "distinct",
550            "drop",
551            "else",
552            "end",
553            "escape",
554            "except",
555            "execute",
556            "exists",
557            "extract",
558            "false",
559            "for",
560            "from",
561            "full",
562            "group",
563            "having",
564            "in",
565            "inner",
566            "insert",
567            "intersect",
568            "into",
569            "is",
570            "join",
571            "left",
572            "like",
573            "natural",
574            "not",
575            "null",
576            "on",
577            "or",
578            "order",
579            "outer",
580            "prepare",
581            "right",
582            "select",
583            "table",
584            "then",
585            "true",
586            "union",
587            "using",
588            "values",
589            "when",
590            "where",
591            "with",
592        }
593
594        def jsonformat_sql(self, expression: exp.JSONFormat) -> str:
595            this = expression.this
596            is_json = expression.args.get("is_json")
597
598            if this and not (is_json or this.type):
599                from sqlglot.optimizer.annotate_types import annotate_types
600
601                this = annotate_types(this, dialect=self.dialect)
602
603            if not (is_json or this.is_type(exp.DataType.Type.JSON)):
604                this.replace(exp.cast(this, exp.DataType.Type.JSON))
605
606            return self.function_fallback_sql(expression)
607
608        def md5_sql(self, expression: exp.MD5) -> str:
609            this = expression.this
610
611            if not this.type:
612                from sqlglot.optimizer.annotate_types import annotate_types
613
614                this = annotate_types(this, dialect=self.dialect)
615
616            if this.is_type(*exp.DataType.TEXT_TYPES):
617                this = exp.Encode(this=this, charset=exp.Literal.string("utf-8"))
618
619            return self.func("LOWER", self.func("TO_HEX", self.func("MD5", self.sql(this))))
620
621        def strtounix_sql(self, expression: exp.StrToUnix) -> str:
622            # Since `TO_UNIXTIME` requires a `TIMESTAMP`, we need to parse the argument into one.
623            # To do this, we first try to `DATE_PARSE` it, but since this can fail when there's a
624            # timezone involved, we wrap it in a `TRY` call and use `PARSE_DATETIME` as a fallback,
625            # which seems to be using the same time mapping as Hive, as per:
626            # https://joda-time.sourceforge.net/apidocs/org/joda/time/format/DateTimeFormat.html
627            this = expression.this
628            value_as_text = exp.cast(this, exp.DataType.Type.TEXT)
629            value_as_timestamp = (
630                exp.cast(this, exp.DataType.Type.TIMESTAMP) if this.is_string else this
631            )
632
633            parse_without_tz = self.func("DATE_PARSE", value_as_text, self.format_time(expression))
634
635            formatted_value = self.func(
636                "DATE_FORMAT", value_as_timestamp, self.format_time(expression)
637            )
638            parse_with_tz = self.func(
639                "PARSE_DATETIME",
640                formatted_value,
641                self.format_time(expression, Hive.INVERSE_TIME_MAPPING, Hive.INVERSE_TIME_TRIE),
642            )
643            coalesced = self.func("COALESCE", self.func("TRY", parse_without_tz), parse_with_tz)
644            return self.func("TO_UNIXTIME", coalesced)
645
646        def bracket_sql(self, expression: exp.Bracket) -> str:
647            if expression.args.get("safe"):
648                return self.func(
649                    "ELEMENT_AT",
650                    expression.this,
651                    seq_get(
652                        apply_index_offset(
653                            expression.this,
654                            expression.expressions,
655                            1 - expression.args.get("offset", 0),
656                            dialect=self.dialect,
657                        ),
658                        0,
659                    ),
660                )
661            return super().bracket_sql(expression)
662
663        def struct_sql(self, expression: exp.Struct) -> str:
664            from sqlglot.optimizer.annotate_types import annotate_types
665
666            expression = annotate_types(expression, dialect=self.dialect)
667            values: t.List[str] = []
668            schema: t.List[str] = []
669            unknown_type = False
670
671            for e in expression.expressions:
672                if isinstance(e, exp.PropertyEQ):
673                    if e.type and e.type.is_type(exp.DataType.Type.UNKNOWN):
674                        unknown_type = True
675                    else:
676                        schema.append(f"{self.sql(e, 'this')} {self.sql(e.type)}")
677                    values.append(self.sql(e, "expression"))
678                else:
679                    values.append(self.sql(e))
680
681            size = len(expression.expressions)
682
683            if not size or len(schema) != size:
684                if unknown_type:
685                    self.unsupported(
686                        "Cannot convert untyped key-value definitions (try annotate_types)."
687                    )
688                return self.func("ROW", *values)
689            return f"CAST(ROW({', '.join(values)}) AS ROW({', '.join(schema)}))"
690
691        def interval_sql(self, expression: exp.Interval) -> str:
692            if expression.this and expression.text("unit").upper().startswith("WEEK"):
693                return f"({expression.this.name} * INTERVAL '7' DAY)"
694            return super().interval_sql(expression)
695
696        def transaction_sql(self, expression: exp.Transaction) -> str:
697            modes = expression.args.get("modes")
698            modes = f" {', '.join(modes)}" if modes else ""
699            return f"START TRANSACTION{modes}"
700
701        def offset_limit_modifiers(
702            self, expression: exp.Expression, fetch: bool, limit: t.Optional[exp.Fetch | exp.Limit]
703        ) -> t.List[str]:
704            return [
705                self.sql(expression, "offset"),
706                self.sql(limit),
707            ]
708
709        def create_sql(self, expression: exp.Create) -> str:
710            """
711            Presto doesn't support CREATE VIEW with expressions (ex: `CREATE VIEW x (cola)` then `(cola)` is the expression),
712            so we need to remove them
713            """
714            kind = expression.args["kind"]
715            schema = expression.this
716            if kind == "VIEW" and schema.expressions:
717                expression.this.set("expressions", None)
718            return super().create_sql(expression)
719
720        def delete_sql(self, expression: exp.Delete) -> str:
721            """
722            Presto only supports DELETE FROM for a single table without an alias, so we need
723            to remove the unnecessary parts. If the original DELETE statement contains more
724            than one table to be deleted, we can't safely map it 1-1 to a Presto statement.
725            """
726            tables = expression.args.get("tables") or [expression.this]
727            if len(tables) > 1:
728                return super().delete_sql(expression)
729
730            table = tables[0]
731            expression.set("this", table)
732            expression.set("tables", None)
733
734            if isinstance(table, exp.Table):
735                table_alias = table.args.get("alias")
736                if table_alias:
737                    table_alias.pop()
738                    expression = t.cast(exp.Delete, expression.transform(unqualify_columns))
739
740            return super().delete_sql(expression)
741
742        def jsonextract_sql(self, expression: exp.JSONExtract) -> str:
743            is_json_extract = self.dialect.settings.get("variant_extract_is_json_extract", True)
744
745            # Generate JSON_EXTRACT unless the user has configured that a Snowflake / Databricks
746            # VARIANT extract (e.g. col:x.y) should map to dot notation (i.e ROW access) in Presto/Trino
747            if not expression.args.get("variant_extract") or is_json_extract:
748                return self.func(
749                    "JSON_EXTRACT", expression.this, expression.expression, *expression.expressions
750                )
751
752            this = self.sql(expression, "this")
753
754            # Convert the JSONPath extraction `JSON_EXTRACT(col, '$.x.y) to a ROW access col.x.y
755            segments = []
756            for path_key in expression.expression.expressions[1:]:
757                if not isinstance(path_key, exp.JSONPathKey):
758                    # Cannot transpile subscripts, wildcards etc to dot notation
759                    self.unsupported(
760                        f"Cannot transpile JSONPath segment '{path_key}' to ROW access"
761                    )
762                    continue
763                key = path_key.this
764                if not exp.SAFE_IDENTIFIER_RE.match(key):
765                    key = f'"{key}"'
766                segments.append(f".{key}")
767
768            expr = "".join(segments)
769
770            return f"{this}{expr}"
771
772        def groupconcat_sql(self, expression: exp.GroupConcat) -> str:
773            return self.func(
774                "ARRAY_JOIN",
775                self.func("ARRAY_AGG", expression.this),
776                expression.args.get("separator"),
777            )