sqlglot.dialects.duckdb

   1from __future__ import annotations
   2
   3import typing as t
   4
   5from sqlglot import exp, generator, parser, tokens, transforms
   6from sqlglot.expressions import DATA_TYPE
   7from sqlglot.dialects.dialect import (
   8    Dialect,
   9    JSON_EXTRACT_TYPE,
  10    NormalizationStrategy,
  11    Version,
  12    approx_count_distinct_sql,
  13    arrow_json_extract_sql,
  14    binary_from_function,
  15    bool_xor_sql,
  16    build_default_decimal_type,
  17    count_if_to_sum,
  18    date_trunc_to_time,
  19    datestrtodate_sql,
  20    no_datetime_sql,
  21    encode_decode_sql,
  22    build_formatted_time,
  23    inline_array_unless_query,
  24    no_comment_column_constraint_sql,
  25    no_time_sql,
  26    no_timestamp_sql,
  27    pivot_column_names,
  28    rename_func,
  29    remove_from_array_using_filter,
  30    strposition_sql,
  31    str_to_time_sql,
  32    timestamptrunc_sql,
  33    timestrtotime_sql,
  34    unit_to_var,
  35    unit_to_str,
  36    sha256_sql,
  37    build_regexp_extract,
  38    explode_to_unnest_sql,
  39    no_make_interval_sql,
  40    groupconcat_sql,
  41)
  42from sqlglot.generator import unsupported_args
  43from sqlglot.helper import seq_get
  44from sqlglot.tokens import TokenType
  45from sqlglot.parser import binary_range_parser
  46
  47DATETIME_DELTA = t.Union[
  48    exp.DateAdd, exp.TimeAdd, exp.DatetimeAdd, exp.TsOrDsAdd, exp.DateSub, exp.DatetimeSub
  49]
  50
  51
  52def _date_delta_sql(self: DuckDB.Generator, expression: DATETIME_DELTA) -> str:
  53    this = expression.this
  54    unit = unit_to_var(expression)
  55    op = (
  56        "+"
  57        if isinstance(expression, (exp.DateAdd, exp.TimeAdd, exp.DatetimeAdd, exp.TsOrDsAdd))
  58        else "-"
  59    )
  60
  61    to_type: t.Optional[DATA_TYPE] = None
  62    if isinstance(expression, exp.TsOrDsAdd):
  63        to_type = expression.return_type
  64    elif this.is_string:
  65        # Cast string literals (i.e function parameters) to the appropriate type for +/- interval to work
  66        to_type = (
  67            exp.DataType.Type.DATETIME
  68            if isinstance(expression, (exp.DatetimeAdd, exp.DatetimeSub))
  69            else exp.DataType.Type.DATE
  70        )
  71
  72    this = exp.cast(this, to_type) if to_type else this
  73
  74    expr = expression.expression
  75    interval = expr if isinstance(expr, exp.Interval) else exp.Interval(this=expr, unit=unit)
  76
  77    return f"{self.sql(this)} {op} {self.sql(interval)}"
  78
  79
  80# BigQuery -> DuckDB conversion for the DATE function
  81def _date_sql(self: DuckDB.Generator, expression: exp.Date) -> str:
  82    result = f"CAST({self.sql(expression, 'this')} AS DATE)"
  83    zone = self.sql(expression, "zone")
  84
  85    if zone:
  86        date_str = self.func("STRFTIME", result, "'%d/%m/%Y'")
  87        date_str = f"{date_str} || ' ' || {zone}"
  88
  89        # This will create a TIMESTAMP with time zone information
  90        result = self.func("STRPTIME", date_str, "'%d/%m/%Y %Z'")
  91
  92    return result
  93
  94
  95# BigQuery -> DuckDB conversion for the TIME_DIFF function
  96def _timediff_sql(self: DuckDB.Generator, expression: exp.TimeDiff) -> str:
  97    this = exp.cast(expression.this, exp.DataType.Type.TIME)
  98    expr = exp.cast(expression.expression, exp.DataType.Type.TIME)
  99
 100    # Although the 2 dialects share similar signatures, BQ seems to inverse
 101    # the sign of the result so the start/end time operands are flipped
 102    return self.func("DATE_DIFF", unit_to_str(expression), expr, this)
 103
 104
 105@unsupported_args(("expression", "DuckDB's ARRAY_SORT does not support a comparator."))
 106def _array_sort_sql(self: DuckDB.Generator, expression: exp.ArraySort) -> str:
 107    return self.func("ARRAY_SORT", expression.this)
 108
 109
 110def _sort_array_sql(self: DuckDB.Generator, expression: exp.SortArray) -> str:
 111    name = "ARRAY_REVERSE_SORT" if expression.args.get("asc") == exp.false() else "ARRAY_SORT"
 112    return self.func(name, expression.this)
 113
 114
 115def _build_sort_array_desc(args: t.List) -> exp.Expression:
 116    return exp.SortArray(this=seq_get(args, 0), asc=exp.false())
 117
 118
 119def _build_date_diff(args: t.List) -> exp.Expression:
 120    return exp.DateDiff(this=seq_get(args, 2), expression=seq_get(args, 1), unit=seq_get(args, 0))
 121
 122
 123def _build_generate_series(end_exclusive: bool = False) -> t.Callable[[t.List], exp.GenerateSeries]:
 124    def _builder(args: t.List) -> exp.GenerateSeries:
 125        # Check https://duckdb.org/docs/sql/functions/nested.html#range-functions
 126        if len(args) == 1:
 127            # DuckDB uses 0 as a default for the series' start when it's omitted
 128            args.insert(0, exp.Literal.number("0"))
 129
 130        gen_series = exp.GenerateSeries.from_arg_list(args)
 131        gen_series.set("is_end_exclusive", end_exclusive)
 132
 133        return gen_series
 134
 135    return _builder
 136
 137
 138def _build_make_timestamp(args: t.List) -> exp.Expression:
 139    if len(args) == 1:
 140        return exp.UnixToTime(this=seq_get(args, 0), scale=exp.UnixToTime.MICROS)
 141
 142    return exp.TimestampFromParts(
 143        year=seq_get(args, 0),
 144        month=seq_get(args, 1),
 145        day=seq_get(args, 2),
 146        hour=seq_get(args, 3),
 147        min=seq_get(args, 4),
 148        sec=seq_get(args, 5),
 149    )
 150
 151
 152def _show_parser(*args: t.Any, **kwargs: t.Any) -> t.Callable[[DuckDB.Parser], exp.Show]:
 153    def _parse(self: DuckDB.Parser) -> exp.Show:
 154        return self._parse_show_duckdb(*args, **kwargs)
 155
 156    return _parse
 157
 158
 159def _struct_sql(self: DuckDB.Generator, expression: exp.Struct) -> str:
 160    args: t.List[str] = []
 161
 162    # BigQuery allows inline construction such as "STRUCT<a STRING, b INTEGER>('str', 1)" which is
 163    # canonicalized to "ROW('str', 1) AS STRUCT(a TEXT, b INT)" in DuckDB
 164    # The transformation to ROW will take place if:
 165    #  1. The STRUCT itself does not have proper fields (key := value) as a "proper" STRUCT would
 166    #  2. A cast to STRUCT / ARRAY of STRUCTs is found
 167    ancestor_cast = expression.find_ancestor(exp.Cast)
 168    is_bq_inline_struct = (
 169        (expression.find(exp.PropertyEQ) is None)
 170        and ancestor_cast
 171        and any(
 172            casted_type.is_type(exp.DataType.Type.STRUCT)
 173            for casted_type in ancestor_cast.find_all(exp.DataType)
 174        )
 175    )
 176
 177    for i, expr in enumerate(expression.expressions):
 178        is_property_eq = isinstance(expr, exp.PropertyEQ)
 179        value = expr.expression if is_property_eq else expr
 180
 181        if is_bq_inline_struct:
 182            args.append(self.sql(value))
 183        else:
 184            key = expr.name if is_property_eq else f"_{i}"
 185            args.append(f"{self.sql(exp.Literal.string(key))}: {self.sql(value)}")
 186
 187    csv_args = ", ".join(args)
 188
 189    return f"ROW({csv_args})" if is_bq_inline_struct else f"{{{csv_args}}}"
 190
 191
 192def _datatype_sql(self: DuckDB.Generator, expression: exp.DataType) -> str:
 193    if expression.is_type("array"):
 194        return f"{self.expressions(expression, flat=True)}[{self.expressions(expression, key='values', flat=True)}]"
 195
 196    # Modifiers are not supported for TIME, [TIME | TIMESTAMP] WITH TIME ZONE
 197    if expression.is_type(
 198        exp.DataType.Type.TIME, exp.DataType.Type.TIMETZ, exp.DataType.Type.TIMESTAMPTZ
 199    ):
 200        return expression.this.value
 201
 202    return self.datatype_sql(expression)
 203
 204
 205def _json_format_sql(self: DuckDB.Generator, expression: exp.JSONFormat) -> str:
 206    sql = self.func("TO_JSON", expression.this, expression.args.get("options"))
 207    return f"CAST({sql} AS TEXT)"
 208
 209
 210def _unix_to_time_sql(self: DuckDB.Generator, expression: exp.UnixToTime) -> str:
 211    scale = expression.args.get("scale")
 212    timestamp = expression.this
 213
 214    if scale in (None, exp.UnixToTime.SECONDS):
 215        return self.func("TO_TIMESTAMP", timestamp)
 216    if scale == exp.UnixToTime.MILLIS:
 217        return self.func("EPOCH_MS", timestamp)
 218    if scale == exp.UnixToTime.MICROS:
 219        return self.func("MAKE_TIMESTAMP", timestamp)
 220
 221    return self.func("TO_TIMESTAMP", exp.Div(this=timestamp, expression=exp.func("POW", 10, scale)))
 222
 223
 224WRAPPED_JSON_EXTRACT_EXPRESSIONS = (exp.Binary, exp.Bracket, exp.In)
 225
 226
 227def _arrow_json_extract_sql(self: DuckDB.Generator, expression: JSON_EXTRACT_TYPE) -> str:
 228    arrow_sql = arrow_json_extract_sql(self, expression)
 229    if not expression.same_parent and isinstance(
 230        expression.parent, WRAPPED_JSON_EXTRACT_EXPRESSIONS
 231    ):
 232        arrow_sql = self.wrap(arrow_sql)
 233    return arrow_sql
 234
 235
 236def _implicit_datetime_cast(
 237    arg: t.Optional[exp.Expression], type: exp.DataType.Type = exp.DataType.Type.DATE
 238) -> t.Optional[exp.Expression]:
 239    return exp.cast(arg, type) if isinstance(arg, exp.Literal) else arg
 240
 241
 242def _date_diff_sql(self: DuckDB.Generator, expression: exp.DateDiff) -> str:
 243    this = _implicit_datetime_cast(expression.this)
 244    expr = _implicit_datetime_cast(expression.expression)
 245
 246    return self.func("DATE_DIFF", unit_to_str(expression), expr, this)
 247
 248
 249def _generate_datetime_array_sql(
 250    self: DuckDB.Generator, expression: t.Union[exp.GenerateDateArray, exp.GenerateTimestampArray]
 251) -> str:
 252    is_generate_date_array = isinstance(expression, exp.GenerateDateArray)
 253
 254    type = exp.DataType.Type.DATE if is_generate_date_array else exp.DataType.Type.TIMESTAMP
 255    start = _implicit_datetime_cast(expression.args.get("start"), type=type)
 256    end = _implicit_datetime_cast(expression.args.get("end"), type=type)
 257
 258    # BQ's GENERATE_DATE_ARRAY & GENERATE_TIMESTAMP_ARRAY are transformed to DuckDB'S GENERATE_SERIES
 259    gen_series: t.Union[exp.GenerateSeries, exp.Cast] = exp.GenerateSeries(
 260        start=start, end=end, step=expression.args.get("step")
 261    )
 262
 263    if is_generate_date_array:
 264        # The GENERATE_SERIES result type is TIMESTAMP array, so to match BQ's semantics for
 265        # GENERATE_DATE_ARRAY we must cast it back to DATE array
 266        gen_series = exp.cast(gen_series, exp.DataType.build("ARRAY<DATE>"))
 267
 268    return self.sql(gen_series)
 269
 270
 271def _json_extract_value_array_sql(
 272    self: DuckDB.Generator, expression: exp.JSONValueArray | exp.JSONExtractArray
 273) -> str:
 274    json_extract = exp.JSONExtract(this=expression.this, expression=expression.expression)
 275    data_type = "ARRAY<STRING>" if isinstance(expression, exp.JSONValueArray) else "ARRAY<JSON>"
 276    return self.sql(exp.cast(json_extract, to=exp.DataType.build(data_type)))
 277
 278
 279class DuckDB(Dialect):
 280    NULL_ORDERING = "nulls_are_last"
 281    SUPPORTS_USER_DEFINED_TYPES = True
 282    SAFE_DIVISION = True
 283    INDEX_OFFSET = 1
 284    CONCAT_COALESCE = True
 285    SUPPORTS_ORDER_BY_ALL = True
 286    SUPPORTS_FIXED_SIZE_ARRAYS = True
 287    STRICT_JSON_PATH_SYNTAX = False
 288    NUMBERS_CAN_BE_UNDERSCORE_SEPARATED = True
 289
 290    # https://duckdb.org/docs/sql/introduction.html#creating-a-new-table
 291    NORMALIZATION_STRATEGY = NormalizationStrategy.CASE_INSENSITIVE
 292
 293    def to_json_path(self, path: t.Optional[exp.Expression]) -> t.Optional[exp.Expression]:
 294        if isinstance(path, exp.Literal):
 295            # DuckDB also supports the JSON pointer syntax, where every path starts with a `/`.
 296            # Additionally, it allows accessing the back of lists using the `[#-i]` syntax.
 297            # This check ensures we'll avoid trying to parse these as JSON paths, which can
 298            # either result in a noisy warning or in an invalid representation of the path.
 299            path_text = path.name
 300            if path_text.startswith("/") or "[#" in path_text:
 301                return path
 302
 303        return super().to_json_path(path)
 304
 305    class Tokenizer(tokens.Tokenizer):
 306        BYTE_STRINGS = [("e'", "'"), ("E'", "'")]
 307        HEREDOC_STRINGS = ["$"]
 308
 309        HEREDOC_TAG_IS_IDENTIFIER = True
 310        HEREDOC_STRING_ALTERNATIVE = TokenType.PARAMETER
 311
 312        KEYWORDS = {
 313            **tokens.Tokenizer.KEYWORDS,
 314            "//": TokenType.DIV,
 315            "**": TokenType.DSTAR,
 316            "^@": TokenType.CARET_AT,
 317            "@>": TokenType.AT_GT,
 318            "<@": TokenType.LT_AT,
 319            "ATTACH": TokenType.ATTACH,
 320            "BINARY": TokenType.VARBINARY,
 321            "BITSTRING": TokenType.BIT,
 322            "BPCHAR": TokenType.TEXT,
 323            "CHAR": TokenType.TEXT,
 324            "DATETIME": TokenType.TIMESTAMPNTZ,
 325            "DETACH": TokenType.DETACH,
 326            "EXCLUDE": TokenType.EXCEPT,
 327            "LOGICAL": TokenType.BOOLEAN,
 328            "ONLY": TokenType.ONLY,
 329            "PIVOT_WIDER": TokenType.PIVOT,
 330            "POSITIONAL": TokenType.POSITIONAL,
 331            "SIGNED": TokenType.INT,
 332            "STRING": TokenType.TEXT,
 333            "SUMMARIZE": TokenType.SUMMARIZE,
 334            "TIMESTAMP": TokenType.TIMESTAMPNTZ,
 335            "TIMESTAMP_S": TokenType.TIMESTAMP_S,
 336            "TIMESTAMP_MS": TokenType.TIMESTAMP_MS,
 337            "TIMESTAMP_NS": TokenType.TIMESTAMP_NS,
 338            "TIMESTAMP_US": TokenType.TIMESTAMP,
 339            "UBIGINT": TokenType.UBIGINT,
 340            "UINTEGER": TokenType.UINT,
 341            "USMALLINT": TokenType.USMALLINT,
 342            "UTINYINT": TokenType.UTINYINT,
 343            "VARCHAR": TokenType.TEXT,
 344        }
 345        KEYWORDS.pop("/*+")
 346
 347        SINGLE_TOKENS = {
 348            **tokens.Tokenizer.SINGLE_TOKENS,
 349            "$": TokenType.PARAMETER,
 350        }
 351
 352        COMMANDS = tokens.Tokenizer.COMMANDS - {TokenType.SHOW}
 353
 354    class Parser(parser.Parser):
 355        BITWISE = {
 356            **parser.Parser.BITWISE,
 357            TokenType.TILDA: exp.RegexpLike,
 358        }
 359        BITWISE.pop(TokenType.CARET)
 360
 361        RANGE_PARSERS = {
 362            **parser.Parser.RANGE_PARSERS,
 363            TokenType.DAMP: binary_range_parser(exp.ArrayOverlaps),
 364            TokenType.CARET_AT: binary_range_parser(exp.StartsWith),
 365        }
 366
 367        EXPONENT = {
 368            **parser.Parser.EXPONENT,
 369            TokenType.CARET: exp.Pow,
 370            TokenType.DSTAR: exp.Pow,
 371        }
 372
 373        FUNCTIONS_WITH_ALIASED_ARGS = {*parser.Parser.FUNCTIONS_WITH_ALIASED_ARGS, "STRUCT_PACK"}
 374
 375        SHOW_PARSERS = {
 376            "TABLES": _show_parser("TABLES"),
 377            "ALL TABLES": _show_parser("ALL TABLES"),
 378        }
 379
 380        FUNCTIONS = {
 381            **parser.Parser.FUNCTIONS,
 382            "ARRAY_REVERSE_SORT": _build_sort_array_desc,
 383            "ARRAY_SORT": exp.SortArray.from_arg_list,
 384            "DATEDIFF": _build_date_diff,
 385            "DATE_DIFF": _build_date_diff,
 386            "DATE_TRUNC": date_trunc_to_time,
 387            "DATETRUNC": date_trunc_to_time,
 388            "DECODE": lambda args: exp.Decode(
 389                this=seq_get(args, 0), charset=exp.Literal.string("utf-8")
 390            ),
 391            "EDITDIST3": exp.Levenshtein.from_arg_list,
 392            "ENCODE": lambda args: exp.Encode(
 393                this=seq_get(args, 0), charset=exp.Literal.string("utf-8")
 394            ),
 395            "EPOCH": exp.TimeToUnix.from_arg_list,
 396            "EPOCH_MS": lambda args: exp.UnixToTime(
 397                this=seq_get(args, 0), scale=exp.UnixToTime.MILLIS
 398            ),
 399            "GENERATE_SERIES": _build_generate_series(),
 400            "JSON": exp.ParseJSON.from_arg_list,
 401            "JSON_EXTRACT_PATH": parser.build_extract_json_with_path(exp.JSONExtract),
 402            "JSON_EXTRACT_STRING": parser.build_extract_json_with_path(exp.JSONExtractScalar),
 403            "LIST_HAS": exp.ArrayContains.from_arg_list,
 404            "LIST_REVERSE_SORT": _build_sort_array_desc,
 405            "LIST_SORT": exp.SortArray.from_arg_list,
 406            "LIST_VALUE": lambda args: exp.Array(expressions=args),
 407            "MAKE_TIME": exp.TimeFromParts.from_arg_list,
 408            "MAKE_TIMESTAMP": _build_make_timestamp,
 409            "QUANTILE_CONT": exp.PercentileCont.from_arg_list,
 410            "QUANTILE_DISC": exp.PercentileDisc.from_arg_list,
 411            "RANGE": _build_generate_series(end_exclusive=True),
 412            "REGEXP_EXTRACT": build_regexp_extract(exp.RegexpExtract),
 413            "REGEXP_EXTRACT_ALL": build_regexp_extract(exp.RegexpExtractAll),
 414            "REGEXP_MATCHES": exp.RegexpLike.from_arg_list,
 415            "REGEXP_REPLACE": lambda args: exp.RegexpReplace(
 416                this=seq_get(args, 0),
 417                expression=seq_get(args, 1),
 418                replacement=seq_get(args, 2),
 419                modifiers=seq_get(args, 3),
 420            ),
 421            "SHA256": lambda args: exp.SHA2(this=seq_get(args, 0), length=exp.Literal.number(256)),
 422            "STRFTIME": build_formatted_time(exp.TimeToStr, "duckdb"),
 423            "STRING_SPLIT": exp.Split.from_arg_list,
 424            "STRING_SPLIT_REGEX": exp.RegexpSplit.from_arg_list,
 425            "STRING_TO_ARRAY": exp.Split.from_arg_list,
 426            "STRPTIME": build_formatted_time(exp.StrToTime, "duckdb"),
 427            "STRUCT_PACK": exp.Struct.from_arg_list,
 428            "STR_SPLIT": exp.Split.from_arg_list,
 429            "STR_SPLIT_REGEX": exp.RegexpSplit.from_arg_list,
 430            "TIME_BUCKET": exp.DateBin.from_arg_list,
 431            "TO_TIMESTAMP": exp.UnixToTime.from_arg_list,
 432            "UNNEST": exp.Explode.from_arg_list,
 433            "XOR": binary_from_function(exp.BitwiseXor),
 434        }
 435
 436        FUNCTIONS.pop("DATE_SUB")
 437        FUNCTIONS.pop("GLOB")
 438
 439        FUNCTION_PARSERS = {
 440            **parser.Parser.FUNCTION_PARSERS,
 441            **dict.fromkeys(
 442                ("GROUP_CONCAT", "LISTAGG", "STRINGAGG"), lambda self: self._parse_string_agg()
 443            ),
 444        }
 445        FUNCTION_PARSERS.pop("DECODE")
 446
 447        NO_PAREN_FUNCTION_PARSERS = {
 448            **parser.Parser.NO_PAREN_FUNCTION_PARSERS,
 449            "MAP": lambda self: self._parse_map(),
 450            "@": lambda self: exp.Abs(this=self._parse_bitwise()),
 451        }
 452
 453        TABLE_ALIAS_TOKENS = parser.Parser.TABLE_ALIAS_TOKENS - {
 454            TokenType.SEMI,
 455            TokenType.ANTI,
 456        }
 457
 458        PLACEHOLDER_PARSERS = {
 459            **parser.Parser.PLACEHOLDER_PARSERS,
 460            TokenType.PARAMETER: lambda self: (
 461                self.expression(exp.Placeholder, this=self._prev.text)
 462                if self._match(TokenType.NUMBER) or self._match_set(self.ID_VAR_TOKENS)
 463                else None
 464            ),
 465        }
 466
 467        TYPE_CONVERTERS = {
 468            # https://duckdb.org/docs/sql/data_types/numeric
 469            exp.DataType.Type.DECIMAL: build_default_decimal_type(precision=18, scale=3),
 470            # https://duckdb.org/docs/sql/data_types/text
 471            exp.DataType.Type.TEXT: lambda dtype: exp.DataType.build("TEXT"),
 472        }
 473
 474        STATEMENT_PARSERS = {
 475            **parser.Parser.STATEMENT_PARSERS,
 476            TokenType.ATTACH: lambda self: self._parse_attach_detach(),
 477            TokenType.DETACH: lambda self: self._parse_attach_detach(is_attach=False),
 478            TokenType.SHOW: lambda self: self._parse_show(),
 479        }
 480
 481        def _parse_expression(self) -> t.Optional[exp.Expression]:
 482            # DuckDB supports prefix aliases, e.g. foo: 1
 483            if self._next and self._next.token_type == TokenType.COLON:
 484                alias = self._parse_id_var(tokens=self.ALIAS_TOKENS)
 485                self._match(TokenType.COLON)
 486                comments = self._prev_comments or []
 487
 488                this = self._parse_assignment()
 489                if isinstance(this, exp.Expression):
 490                    # Moves the comment next to the alias in `alias: expr /* comment */`
 491                    comments += this.pop_comments() or []
 492
 493                return self.expression(exp.Alias, comments=comments, this=this, alias=alias)
 494
 495            return super()._parse_expression()
 496
 497        def _parse_table(
 498            self,
 499            schema: bool = False,
 500            joins: bool = False,
 501            alias_tokens: t.Optional[t.Collection[TokenType]] = None,
 502            parse_bracket: bool = False,
 503            is_db_reference: bool = False,
 504            parse_partition: bool = False,
 505        ) -> t.Optional[exp.Expression]:
 506            # DuckDB supports prefix aliases, e.g. FROM foo: bar
 507            if self._next and self._next.token_type == TokenType.COLON:
 508                alias = self._parse_table_alias(
 509                    alias_tokens=alias_tokens or self.TABLE_ALIAS_TOKENS
 510                )
 511                self._match(TokenType.COLON)
 512                comments = self._prev_comments or []
 513            else:
 514                alias = None
 515                comments = []
 516
 517            table = super()._parse_table(
 518                schema=schema,
 519                joins=joins,
 520                alias_tokens=alias_tokens,
 521                parse_bracket=parse_bracket,
 522                is_db_reference=is_db_reference,
 523                parse_partition=parse_partition,
 524            )
 525            if isinstance(table, exp.Expression) and isinstance(alias, exp.TableAlias):
 526                # Moves the comment next to the alias in `alias: table /* comment */`
 527                comments += table.pop_comments() or []
 528                alias.comments = alias.pop_comments() + comments
 529                table.set("alias", alias)
 530
 531            return table
 532
 533        def _parse_table_sample(self, as_modifier: bool = False) -> t.Optional[exp.TableSample]:
 534            # https://duckdb.org/docs/sql/samples.html
 535            sample = super()._parse_table_sample(as_modifier=as_modifier)
 536            if sample and not sample.args.get("method"):
 537                if sample.args.get("size"):
 538                    sample.set("method", exp.var("RESERVOIR"))
 539                else:
 540                    sample.set("method", exp.var("SYSTEM"))
 541
 542            return sample
 543
 544        def _parse_bracket(
 545            self, this: t.Optional[exp.Expression] = None
 546        ) -> t.Optional[exp.Expression]:
 547            bracket = super()._parse_bracket(this)
 548
 549            if self.dialect.version < Version("1.2.0") and isinstance(bracket, exp.Bracket):
 550                # https://duckdb.org/2025/02/05/announcing-duckdb-120.html#breaking-changes
 551                bracket.set("returns_list_for_maps", True)
 552
 553            return bracket
 554
 555        def _parse_map(self) -> exp.ToMap | exp.Map:
 556            if self._match(TokenType.L_BRACE, advance=False):
 557                return self.expression(exp.ToMap, this=self._parse_bracket())
 558
 559            args = self._parse_wrapped_csv(self._parse_assignment)
 560            return self.expression(exp.Map, keys=seq_get(args, 0), values=seq_get(args, 1))
 561
 562        def _parse_struct_types(self, type_required: bool = False) -> t.Optional[exp.Expression]:
 563            return self._parse_field_def()
 564
 565        def _pivot_column_names(self, aggregations: t.List[exp.Expression]) -> t.List[str]:
 566            if len(aggregations) == 1:
 567                return super()._pivot_column_names(aggregations)
 568            return pivot_column_names(aggregations, dialect="duckdb")
 569
 570        def _parse_attach_detach(self, is_attach=True) -> exp.Attach | exp.Detach:
 571            def _parse_attach_option() -> exp.AttachOption:
 572                return self.expression(
 573                    exp.AttachOption,
 574                    this=self._parse_var(any_token=True),
 575                    expression=self._parse_field(any_token=True),
 576                )
 577
 578            self._match(TokenType.DATABASE)
 579            exists = self._parse_exists(not_=is_attach)
 580            this = self._parse_alias(self._parse_primary_or_var(), explicit=True)
 581
 582            if self._match(TokenType.L_PAREN, advance=False):
 583                expressions = self._parse_wrapped_csv(_parse_attach_option)
 584            else:
 585                expressions = None
 586
 587            return (
 588                self.expression(exp.Attach, this=this, exists=exists, expressions=expressions)
 589                if is_attach
 590                else self.expression(exp.Detach, this=this, exists=exists)
 591            )
 592
 593        def _parse_show_duckdb(self, this: str) -> exp.Show:
 594            return self.expression(exp.Show, this=this)
 595
 596    class Generator(generator.Generator):
 597        PARAMETER_TOKEN = "$"
 598        NAMED_PLACEHOLDER_TOKEN = "$"
 599        JOIN_HINTS = False
 600        TABLE_HINTS = False
 601        QUERY_HINTS = False
 602        LIMIT_FETCH = "LIMIT"
 603        STRUCT_DELIMITER = ("(", ")")
 604        RENAME_TABLE_WITH_DB = False
 605        NVL2_SUPPORTED = False
 606        SEMI_ANTI_JOIN_WITH_SIDE = False
 607        TABLESAMPLE_KEYWORDS = "USING SAMPLE"
 608        TABLESAMPLE_SEED_KEYWORD = "REPEATABLE"
 609        LAST_DAY_SUPPORTS_DATE_PART = False
 610        JSON_KEY_VALUE_PAIR_SEP = ","
 611        IGNORE_NULLS_IN_FUNC = True
 612        JSON_PATH_BRACKETED_KEY_SUPPORTED = False
 613        SUPPORTS_CREATE_TABLE_LIKE = False
 614        MULTI_ARG_DISTINCT = False
 615        CAN_IMPLEMENT_ARRAY_ANY = True
 616        SUPPORTS_TO_NUMBER = False
 617        SUPPORTS_WINDOW_EXCLUDE = True
 618        COPY_HAS_INTO_KEYWORD = False
 619        STAR_EXCEPT = "EXCLUDE"
 620        PAD_FILL_PATTERN_IS_REQUIRED = True
 621        ARRAY_CONCAT_IS_VAR_LEN = False
 622        ARRAY_SIZE_DIM_REQUIRED = False
 623
 624        TRANSFORMS = {
 625            **generator.Generator.TRANSFORMS,
 626            exp.ApproxDistinct: approx_count_distinct_sql,
 627            exp.Array: inline_array_unless_query,
 628            exp.ArrayFilter: rename_func("LIST_FILTER"),
 629            exp.ArrayRemove: remove_from_array_using_filter,
 630            exp.ArraySort: _array_sort_sql,
 631            exp.ArraySum: rename_func("LIST_SUM"),
 632            exp.BitwiseXor: rename_func("XOR"),
 633            exp.CommentColumnConstraint: no_comment_column_constraint_sql,
 634            exp.CurrentDate: lambda *_: "CURRENT_DATE",
 635            exp.CurrentTime: lambda *_: "CURRENT_TIME",
 636            exp.CurrentTimestamp: lambda *_: "CURRENT_TIMESTAMP",
 637            exp.DayOfMonth: rename_func("DAYOFMONTH"),
 638            exp.DayOfWeek: rename_func("DAYOFWEEK"),
 639            exp.DayOfWeekIso: rename_func("ISODOW"),
 640            exp.DayOfYear: rename_func("DAYOFYEAR"),
 641            exp.DataType: _datatype_sql,
 642            exp.Date: _date_sql,
 643            exp.DateAdd: _date_delta_sql,
 644            exp.DateFromParts: rename_func("MAKE_DATE"),
 645            exp.DateSub: _date_delta_sql,
 646            exp.DateDiff: _date_diff_sql,
 647            exp.DateStrToDate: datestrtodate_sql,
 648            exp.Datetime: no_datetime_sql,
 649            exp.DatetimeSub: _date_delta_sql,
 650            exp.DatetimeAdd: _date_delta_sql,
 651            exp.DateToDi: lambda self,
 652            e: f"CAST(STRFTIME({self.sql(e, 'this')}, {DuckDB.DATEINT_FORMAT}) AS INT)",
 653            exp.Decode: lambda self, e: encode_decode_sql(self, e, "DECODE", replace=False),
 654            exp.DiToDate: lambda self,
 655            e: f"CAST(STRPTIME(CAST({self.sql(e, 'this')} AS TEXT), {DuckDB.DATEINT_FORMAT}) AS DATE)",
 656            exp.Encode: lambda self, e: encode_decode_sql(self, e, "ENCODE", replace=False),
 657            exp.GenerateDateArray: _generate_datetime_array_sql,
 658            exp.GenerateTimestampArray: _generate_datetime_array_sql,
 659            exp.GroupConcat: lambda self, e: groupconcat_sql(self, e, within_group=False),
 660            exp.HexString: lambda self, e: self.hexstring_sql(e, binary_function_repr="FROM_HEX"),
 661            exp.Explode: rename_func("UNNEST"),
 662            exp.IntDiv: lambda self, e: self.binary(e, "//"),
 663            exp.IsInf: rename_func("ISINF"),
 664            exp.IsNan: rename_func("ISNAN"),
 665            exp.JSONBExists: rename_func("JSON_EXISTS"),
 666            exp.JSONExtract: _arrow_json_extract_sql,
 667            exp.JSONExtractArray: _json_extract_value_array_sql,
 668            exp.JSONExtractScalar: _arrow_json_extract_sql,
 669            exp.JSONFormat: _json_format_sql,
 670            exp.JSONValueArray: _json_extract_value_array_sql,
 671            exp.Lateral: explode_to_unnest_sql,
 672            exp.LogicalOr: rename_func("BOOL_OR"),
 673            exp.LogicalAnd: rename_func("BOOL_AND"),
 674            exp.MakeInterval: lambda self, e: no_make_interval_sql(self, e, sep=" "),
 675            exp.MD5Digest: lambda self, e: self.func("UNHEX", self.func("MD5", e.this)),
 676            exp.MonthsBetween: lambda self, e: self.func(
 677                "DATEDIFF",
 678                "'month'",
 679                exp.cast(e.expression, exp.DataType.Type.TIMESTAMP, copy=True),
 680                exp.cast(e.this, exp.DataType.Type.TIMESTAMP, copy=True),
 681            ),
 682            exp.PercentileCont: rename_func("QUANTILE_CONT"),
 683            exp.PercentileDisc: rename_func("QUANTILE_DISC"),
 684            # DuckDB doesn't allow qualified columns inside of PIVOT expressions.
 685            # See: https://github.com/duckdb/duckdb/blob/671faf92411182f81dce42ac43de8bfb05d9909e/src/planner/binder/tableref/bind_pivot.cpp#L61-L62
 686            exp.Pivot: transforms.preprocess([transforms.unqualify_columns]),
 687            exp.RegexpReplace: lambda self, e: self.func(
 688                "REGEXP_REPLACE",
 689                e.this,
 690                e.expression,
 691                e.args.get("replacement"),
 692                e.args.get("modifiers"),
 693            ),
 694            exp.RegexpLike: rename_func("REGEXP_MATCHES"),
 695            exp.RegexpILike: lambda self, e: self.func(
 696                "REGEXP_MATCHES", e.this, e.expression, exp.Literal.string("i")
 697            ),
 698            exp.RegexpSplit: rename_func("STR_SPLIT_REGEX"),
 699            exp.Return: lambda self, e: self.sql(e, "this"),
 700            exp.ReturnsProperty: lambda self, e: "TABLE" if isinstance(e.this, exp.Schema) else "",
 701            exp.Rand: rename_func("RANDOM"),
 702            exp.SHA: rename_func("SHA1"),
 703            exp.SHA2: sha256_sql,
 704            exp.Split: rename_func("STR_SPLIT"),
 705            exp.SortArray: _sort_array_sql,
 706            exp.StrPosition: strposition_sql,
 707            exp.StrToUnix: lambda self, e: self.func(
 708                "EPOCH", self.func("STRPTIME", e.this, self.format_time(e))
 709            ),
 710            exp.Struct: _struct_sql,
 711            exp.Transform: rename_func("LIST_TRANSFORM"),
 712            exp.TimeAdd: _date_delta_sql,
 713            exp.Time: no_time_sql,
 714            exp.TimeDiff: _timediff_sql,
 715            exp.Timestamp: no_timestamp_sql,
 716            exp.TimestampDiff: lambda self, e: self.func(
 717                "DATE_DIFF", exp.Literal.string(e.unit), e.expression, e.this
 718            ),
 719            exp.TimestampTrunc: timestamptrunc_sql(),
 720            exp.TimeStrToDate: lambda self, e: self.sql(exp.cast(e.this, exp.DataType.Type.DATE)),
 721            exp.TimeStrToTime: timestrtotime_sql,
 722            exp.TimeStrToUnix: lambda self, e: self.func(
 723                "EPOCH", exp.cast(e.this, exp.DataType.Type.TIMESTAMP)
 724            ),
 725            exp.TimeToStr: lambda self, e: self.func("STRFTIME", e.this, self.format_time(e)),
 726            exp.TimeToUnix: rename_func("EPOCH"),
 727            exp.TsOrDiToDi: lambda self,
 728            e: f"CAST(SUBSTR(REPLACE(CAST({self.sql(e, 'this')} AS TEXT), '-', ''), 1, 8) AS INT)",
 729            exp.TsOrDsAdd: _date_delta_sql,
 730            exp.TsOrDsDiff: lambda self, e: self.func(
 731                "DATE_DIFF",
 732                f"'{e.args.get('unit') or 'DAY'}'",
 733                exp.cast(e.expression, exp.DataType.Type.TIMESTAMP),
 734                exp.cast(e.this, exp.DataType.Type.TIMESTAMP),
 735            ),
 736            exp.UnixToStr: lambda self, e: self.func(
 737                "STRFTIME", self.func("TO_TIMESTAMP", e.this), self.format_time(e)
 738            ),
 739            exp.DatetimeTrunc: lambda self, e: self.func(
 740                "DATE_TRUNC", unit_to_str(e), exp.cast(e.this, exp.DataType.Type.DATETIME)
 741            ),
 742            exp.UnixToTime: _unix_to_time_sql,
 743            exp.UnixToTimeStr: lambda self, e: f"CAST(TO_TIMESTAMP({self.sql(e, 'this')}) AS TEXT)",
 744            exp.VariancePop: rename_func("VAR_POP"),
 745            exp.WeekOfYear: rename_func("WEEKOFYEAR"),
 746            exp.Xor: bool_xor_sql,
 747            exp.Levenshtein: unsupported_args("ins_cost", "del_cost", "sub_cost", "max_dist")(
 748                rename_func("LEVENSHTEIN")
 749            ),
 750            exp.JSONObjectAgg: rename_func("JSON_GROUP_OBJECT"),
 751            exp.JSONBObjectAgg: rename_func("JSON_GROUP_OBJECT"),
 752            exp.DateBin: rename_func("TIME_BUCKET"),
 753        }
 754
 755        SUPPORTED_JSON_PATH_PARTS = {
 756            exp.JSONPathKey,
 757            exp.JSONPathRoot,
 758            exp.JSONPathSubscript,
 759            exp.JSONPathWildcard,
 760        }
 761
 762        TYPE_MAPPING = {
 763            **generator.Generator.TYPE_MAPPING,
 764            exp.DataType.Type.BINARY: "BLOB",
 765            exp.DataType.Type.BPCHAR: "TEXT",
 766            exp.DataType.Type.CHAR: "TEXT",
 767            exp.DataType.Type.DATETIME: "TIMESTAMP",
 768            exp.DataType.Type.FLOAT: "REAL",
 769            exp.DataType.Type.JSONB: "JSON",
 770            exp.DataType.Type.NCHAR: "TEXT",
 771            exp.DataType.Type.NVARCHAR: "TEXT",
 772            exp.DataType.Type.UINT: "UINTEGER",
 773            exp.DataType.Type.VARBINARY: "BLOB",
 774            exp.DataType.Type.ROWVERSION: "BLOB",
 775            exp.DataType.Type.VARCHAR: "TEXT",
 776            exp.DataType.Type.TIMESTAMPNTZ: "TIMESTAMP",
 777            exp.DataType.Type.TIMESTAMP_S: "TIMESTAMP_S",
 778            exp.DataType.Type.TIMESTAMP_MS: "TIMESTAMP_MS",
 779            exp.DataType.Type.TIMESTAMP_NS: "TIMESTAMP_NS",
 780        }
 781
 782        # https://github.com/duckdb/duckdb/blob/ff7f24fd8e3128d94371827523dae85ebaf58713/third_party/libpg_query/grammar/keywords/reserved_keywords.list#L1-L77
 783        RESERVED_KEYWORDS = {
 784            "array",
 785            "analyse",
 786            "union",
 787            "all",
 788            "when",
 789            "in_p",
 790            "default",
 791            "create_p",
 792            "window",
 793            "asymmetric",
 794            "to",
 795            "else",
 796            "localtime",
 797            "from",
 798            "end_p",
 799            "select",
 800            "current_date",
 801            "foreign",
 802            "with",
 803            "grant",
 804            "session_user",
 805            "or",
 806            "except",
 807            "references",
 808            "fetch",
 809            "limit",
 810            "group_p",
 811            "leading",
 812            "into",
 813            "collate",
 814            "offset",
 815            "do",
 816            "then",
 817            "localtimestamp",
 818            "check_p",
 819            "lateral_p",
 820            "current_role",
 821            "where",
 822            "asc_p",
 823            "placing",
 824            "desc_p",
 825            "user",
 826            "unique",
 827            "initially",
 828            "column",
 829            "both",
 830            "some",
 831            "as",
 832            "any",
 833            "only",
 834            "deferrable",
 835            "null_p",
 836            "current_time",
 837            "true_p",
 838            "table",
 839            "case",
 840            "trailing",
 841            "variadic",
 842            "for",
 843            "on",
 844            "distinct",
 845            "false_p",
 846            "not",
 847            "constraint",
 848            "current_timestamp",
 849            "returning",
 850            "primary",
 851            "intersect",
 852            "having",
 853            "analyze",
 854            "current_user",
 855            "and",
 856            "cast",
 857            "symmetric",
 858            "using",
 859            "order",
 860            "current_catalog",
 861        }
 862
 863        UNWRAPPED_INTERVAL_VALUES = (exp.Literal, exp.Paren)
 864
 865        # DuckDB doesn't generally support CREATE TABLE .. properties
 866        # https://duckdb.org/docs/sql/statements/create_table.html
 867        PROPERTIES_LOCATION = {
 868            prop: exp.Properties.Location.UNSUPPORTED
 869            for prop in generator.Generator.PROPERTIES_LOCATION
 870        }
 871
 872        # There are a few exceptions (e.g. temporary tables) which are supported or
 873        # can be transpiled to DuckDB, so we explicitly override them accordingly
 874        PROPERTIES_LOCATION[exp.LikeProperty] = exp.Properties.Location.POST_SCHEMA
 875        PROPERTIES_LOCATION[exp.TemporaryProperty] = exp.Properties.Location.POST_CREATE
 876        PROPERTIES_LOCATION[exp.ReturnsProperty] = exp.Properties.Location.POST_ALIAS
 877
 878        IGNORE_RESPECT_NULLS_WINDOW_FUNCTIONS = (
 879            exp.FirstValue,
 880            exp.Lag,
 881            exp.LastValue,
 882            exp.Lead,
 883            exp.NthValue,
 884        )
 885
 886        def show_sql(self, expression: exp.Show) -> str:
 887            return f"SHOW {expression.name}"
 888
 889        def fromiso8601timestamp_sql(self, expression: exp.FromISO8601Timestamp) -> str:
 890            return self.sql(exp.cast(expression.this, exp.DataType.Type.TIMESTAMPTZ))
 891
 892        def strtotime_sql(self, expression: exp.StrToTime) -> str:
 893            if expression.args.get("safe"):
 894                formatted_time = self.format_time(expression)
 895                return f"CAST({self.func('TRY_STRPTIME', expression.this, formatted_time)} AS TIMESTAMP)"
 896            return str_to_time_sql(self, expression)
 897
 898        def strtodate_sql(self, expression: exp.StrToDate) -> str:
 899            if expression.args.get("safe"):
 900                formatted_time = self.format_time(expression)
 901                return f"CAST({self.func('TRY_STRPTIME', expression.this, formatted_time)} AS DATE)"
 902            return f"CAST({str_to_time_sql(self, expression)} AS DATE)"
 903
 904        def parsejson_sql(self, expression: exp.ParseJSON) -> str:
 905            arg = expression.this
 906            if expression.args.get("safe"):
 907                return self.sql(exp.case().when(exp.func("json_valid", arg), arg).else_(exp.null()))
 908            return self.func("JSON", arg)
 909
 910        def timefromparts_sql(self, expression: exp.TimeFromParts) -> str:
 911            nano = expression.args.get("nano")
 912            if nano is not None:
 913                expression.set(
 914                    "sec", expression.args["sec"] + nano.pop() / exp.Literal.number(1000000000.0)
 915                )
 916
 917            return rename_func("MAKE_TIME")(self, expression)
 918
 919        def timestampfromparts_sql(self, expression: exp.TimestampFromParts) -> str:
 920            sec = expression.args["sec"]
 921
 922            milli = expression.args.get("milli")
 923            if milli is not None:
 924                sec += milli.pop() / exp.Literal.number(1000.0)
 925
 926            nano = expression.args.get("nano")
 927            if nano is not None:
 928                sec += nano.pop() / exp.Literal.number(1000000000.0)
 929
 930            if milli or nano:
 931                expression.set("sec", sec)
 932
 933            return rename_func("MAKE_TIMESTAMP")(self, expression)
 934
 935        def tablesample_sql(
 936            self,
 937            expression: exp.TableSample,
 938            tablesample_keyword: t.Optional[str] = None,
 939        ) -> str:
 940            if not isinstance(expression.parent, exp.Select):
 941                # This sample clause only applies to a single source, not the entire resulting relation
 942                tablesample_keyword = "TABLESAMPLE"
 943
 944            if expression.args.get("size"):
 945                method = expression.args.get("method")
 946                if method and method.name.upper() != "RESERVOIR":
 947                    self.unsupported(
 948                        f"Sampling method {method} is not supported with a discrete sample count, "
 949                        "defaulting to reservoir sampling"
 950                    )
 951                    expression.set("method", exp.var("RESERVOIR"))
 952
 953            return super().tablesample_sql(expression, tablesample_keyword=tablesample_keyword)
 954
 955        def interval_sql(self, expression: exp.Interval) -> str:
 956            multiplier: t.Optional[int] = None
 957            unit = expression.text("unit").lower()
 958
 959            if unit.startswith("week"):
 960                multiplier = 7
 961            if unit.startswith("quarter"):
 962                multiplier = 90
 963
 964            if multiplier:
 965                return f"({multiplier} * {super().interval_sql(exp.Interval(this=expression.this, unit=exp.var('DAY')))})"
 966
 967            return super().interval_sql(expression)
 968
 969        def columndef_sql(self, expression: exp.ColumnDef, sep: str = " ") -> str:
 970            if isinstance(expression.parent, exp.UserDefinedFunction):
 971                return self.sql(expression, "this")
 972            return super().columndef_sql(expression, sep)
 973
 974        def join_sql(self, expression: exp.Join) -> str:
 975            if (
 976                expression.side == "LEFT"
 977                and not expression.args.get("on")
 978                and isinstance(expression.this, exp.Unnest)
 979            ):
 980                # Some dialects support `LEFT JOIN UNNEST(...)` without an explicit ON clause
 981                # DuckDB doesn't, but we can just add a dummy ON clause that is always true
 982                return super().join_sql(expression.on(exp.true()))
 983
 984            return super().join_sql(expression)
 985
 986        def generateseries_sql(self, expression: exp.GenerateSeries) -> str:
 987            # GENERATE_SERIES(a, b) -> [a, b], RANGE(a, b) -> [a, b)
 988            if expression.args.get("is_end_exclusive"):
 989                return rename_func("RANGE")(self, expression)
 990
 991            return self.function_fallback_sql(expression)
 992
 993        def countif_sql(self, expression: exp.CountIf) -> str:
 994            if self.dialect.version >= Version("1.2"):
 995                return self.function_fallback_sql(expression)
 996
 997            # https://github.com/tobymao/sqlglot/pull/4749
 998            return count_if_to_sum(self, expression)
 999
1000        def bracket_sql(self, expression: exp.Bracket) -> str:
1001            if self.dialect.version >= Version("1.2"):
1002                return super().bracket_sql(expression)
1003
1004            # https://duckdb.org/2025/02/05/announcing-duckdb-120.html#breaking-changes
1005            this = expression.this
1006            if isinstance(this, exp.Array):
1007                this.replace(exp.paren(this))
1008
1009            bracket = super().bracket_sql(expression)
1010
1011            if not expression.args.get("returns_list_for_maps"):
1012                if not this.type:
1013                    from sqlglot.optimizer.annotate_types import annotate_types
1014
1015                    this = annotate_types(this, dialect=self.dialect)
1016
1017                if this.is_type(exp.DataType.Type.MAP):
1018                    bracket = f"({bracket})[1]"
1019
1020            return bracket
1021
1022        def withingroup_sql(self, expression: exp.WithinGroup) -> str:
1023            expression_sql = self.sql(expression, "expression")
1024
1025            func = expression.this
1026            if isinstance(func, exp.PERCENTILES):
1027                # Make the order key the first arg and slide the fraction to the right
1028                # https://duckdb.org/docs/sql/aggregates#ordered-set-aggregate-functions
1029                order_col = expression.find(exp.Ordered)
1030                if order_col:
1031                    func.set("expression", func.this)
1032                    func.set("this", order_col.this)
1033
1034            this = self.sql(expression, "this").rstrip(")")
1035
1036            return f"{this}{expression_sql})"
1037
1038        def length_sql(self, expression: exp.Length) -> str:
1039            arg = expression.this
1040
1041            # Dialects like BQ and Snowflake also accept binary values as args, so
1042            # DDB will attempt to infer the type or resort to case/when resolution
1043            if not expression.args.get("binary") or arg.is_string:
1044                return self.func("LENGTH", arg)
1045
1046            if not arg.type:
1047                from sqlglot.optimizer.annotate_types import annotate_types
1048
1049                arg = annotate_types(arg, dialect=self.dialect)
1050
1051            if arg.is_type(*exp.DataType.TEXT_TYPES):
1052                return self.func("LENGTH", arg)
1053
1054            # We need these casts to make duckdb's static type checker happy
1055            blob = exp.cast(arg, exp.DataType.Type.VARBINARY)
1056            varchar = exp.cast(arg, exp.DataType.Type.VARCHAR)
1057
1058            case = (
1059                exp.case(self.func("TYPEOF", arg))
1060                .when("'BLOB'", self.func("OCTET_LENGTH", blob))
1061                .else_(
1062                    exp.Anonymous(this="LENGTH", expressions=[varchar])
1063                )  # anonymous to break length_sql recursion
1064            )
1065
1066            return self.sql(case)
1067
1068        def objectinsert_sql(self, expression: exp.ObjectInsert) -> str:
1069            this = expression.this
1070            key = expression.args.get("key")
1071            key_sql = key.name if isinstance(key, exp.Expression) else ""
1072            value_sql = self.sql(expression, "value")
1073
1074            kv_sql = f"{key_sql} := {value_sql}"
1075
1076            # If the input struct is empty e.g. transpiling OBJECT_INSERT(OBJECT_CONSTRUCT(), key, value) from Snowflake
1077            # then we can generate STRUCT_PACK which will build it since STRUCT_INSERT({}, key := value) is not valid DuckDB
1078            if isinstance(this, exp.Struct) and not this.expressions:
1079                return self.func("STRUCT_PACK", kv_sql)
1080
1081            return self.func("STRUCT_INSERT", this, kv_sql)
1082
1083        def unnest_sql(self, expression: exp.Unnest) -> str:
1084            explode_array = expression.args.get("explode_array")
1085            if explode_array:
1086                # In BigQuery, UNNESTing a nested array leads to explosion of the top-level array & struct
1087                # This is transpiled to DDB by transforming "FROM UNNEST(...)" to "FROM (SELECT UNNEST(..., max_depth => 2))"
1088                expression.expressions.append(
1089                    exp.Kwarg(this=exp.var("max_depth"), expression=exp.Literal.number(2))
1090                )
1091
1092                # If BQ's UNNEST is aliased, we transform it from a column alias to a table alias in DDB
1093                alias = expression.args.get("alias")
1094                if alias:
1095                    expression.set("alias", None)
1096                    alias = exp.TableAlias(this=seq_get(alias.args.get("columns"), 0))
1097
1098                unnest_sql = super().unnest_sql(expression)
1099                select = exp.Select(expressions=[unnest_sql]).subquery(alias)
1100                return self.sql(select)
1101
1102            return super().unnest_sql(expression)
1103
1104        def ignorenulls_sql(self, expression: exp.IgnoreNulls) -> str:
1105            if isinstance(expression.this, self.IGNORE_RESPECT_NULLS_WINDOW_FUNCTIONS):
1106                # DuckDB should render IGNORE NULLS only for the general-purpose
1107                # window functions that accept it e.g. FIRST_VALUE(... IGNORE NULLS) OVER (...)
1108                return super().ignorenulls_sql(expression)
1109
1110            self.unsupported("IGNORE NULLS is not supported for non-window functions.")
1111            return self.sql(expression, "this")
1112
1113        def respectnulls_sql(self, expression: exp.RespectNulls) -> str:
1114            if isinstance(expression.this, self.IGNORE_RESPECT_NULLS_WINDOW_FUNCTIONS):
1115                # DuckDB should render RESPECT NULLS only for the general-purpose
1116                # window functions that accept it e.g. FIRST_VALUE(... RESPECT NULLS) OVER (...)
1117                return super().respectnulls_sql(expression)
1118
1119            self.unsupported("RESPECT NULLS is not supported for non-window functions.")
1120            return self.sql(expression, "this")
1121
1122        def arraytostring_sql(self, expression: exp.ArrayToString) -> str:
1123            this = self.sql(expression, "this")
1124            null_text = self.sql(expression, "null")
1125
1126            if null_text:
1127                this = f"LIST_TRANSFORM({this}, x -> COALESCE(x, {null_text}))"
1128
1129            return self.func("ARRAY_TO_STRING", this, expression.expression)
1130
1131        @unsupported_args("position", "occurrence")
1132        def regexpextract_sql(self, expression: exp.RegexpExtract) -> str:
1133            group = expression.args.get("group")
1134            params = expression.args.get("parameters")
1135
1136            # Do not render group if there is no following argument,
1137            # and it's the default value for this dialect
1138            if (
1139                not params
1140                and group
1141                and group.name == str(self.dialect.REGEXP_EXTRACT_DEFAULT_GROUP)
1142            ):
1143                group = None
1144            return self.func(
1145                "REGEXP_EXTRACT", expression.this, expression.expression, group, params
1146            )
1147
1148        @unsupported_args("culture")
1149        def numbertostr_sql(self, expression: exp.NumberToStr) -> str:
1150            fmt = expression.args.get("format")
1151            if fmt and fmt.is_int:
1152                return self.func("FORMAT", f"'{{:,.{fmt.name}f}}'", expression.this)
1153
1154            self.unsupported("Only integer formats are supported by NumberToStr")
1155            return self.function_fallback_sql(expression)
1156
1157        def autoincrementcolumnconstraint_sql(self, _) -> str:
1158            self.unsupported("The AUTOINCREMENT column constraint is not supported by DuckDB")
1159            return ""