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    approx_count_distinct_sql,
 12    arg_max_or_min_no_count,
 13    arrow_json_extract_sql,
 14    binary_from_function,
 15    bool_xor_sql,
 16    build_default_decimal_type,
 17    date_trunc_to_time,
 18    datestrtodate_sql,
 19    no_datetime_sql,
 20    encode_decode_sql,
 21    build_formatted_time,
 22    inline_array_unless_query,
 23    no_comment_column_constraint_sql,
 24    no_safe_divide_sql,
 25    no_time_sql,
 26    no_timestamp_sql,
 27    pivot_column_names,
 28    regexp_extract_sql,
 29    rename_func,
 30    str_position_sql,
 31    str_to_time_sql,
 32    timestamptrunc_sql,
 33    timestrtotime_sql,
 34    unit_to_var,
 35    unit_to_str,
 36)
 37from sqlglot.helper import seq_get
 38from sqlglot.tokens import TokenType
 39
 40DATETIME_DELTA = t.Union[
 41    exp.DateAdd, exp.TimeAdd, exp.DatetimeAdd, exp.TsOrDsAdd, exp.DateSub, exp.DatetimeSub
 42]
 43
 44
 45def _date_delta_sql(self: DuckDB.Generator, expression: DATETIME_DELTA) -> str:
 46    this = expression.this
 47    unit = unit_to_var(expression)
 48    op = (
 49        "+"
 50        if isinstance(expression, (exp.DateAdd, exp.TimeAdd, exp.DatetimeAdd, exp.TsOrDsAdd))
 51        else "-"
 52    )
 53
 54    to_type: t.Optional[DATA_TYPE] = None
 55    if isinstance(expression, exp.TsOrDsAdd):
 56        to_type = expression.return_type
 57    elif this.is_string:
 58        # Cast string literals (i.e function parameters) to the appropriate type for +/- interval to work
 59        to_type = (
 60            exp.DataType.Type.DATETIME
 61            if isinstance(expression, (exp.DatetimeAdd, exp.DatetimeSub))
 62            else exp.DataType.Type.DATE
 63        )
 64
 65    this = exp.cast(this, to_type) if to_type else this
 66
 67    return f"{self.sql(this)} {op} {self.sql(exp.Interval(this=expression.expression, unit=unit))}"
 68
 69
 70# BigQuery -> DuckDB conversion for the DATE function
 71def _date_sql(self: DuckDB.Generator, expression: exp.Date) -> str:
 72    result = f"CAST({self.sql(expression, 'this')} AS DATE)"
 73    zone = self.sql(expression, "zone")
 74
 75    if zone:
 76        date_str = self.func("STRFTIME", result, "'%d/%m/%Y'")
 77        date_str = f"{date_str} || ' ' || {zone}"
 78
 79        # This will create a TIMESTAMP with time zone information
 80        result = self.func("STRPTIME", date_str, "'%d/%m/%Y %Z'")
 81
 82    return result
 83
 84
 85# BigQuery -> DuckDB conversion for the TIME_DIFF function
 86def _timediff_sql(self: DuckDB.Generator, expression: exp.TimeDiff) -> str:
 87    this = exp.cast(expression.this, exp.DataType.Type.TIME)
 88    expr = exp.cast(expression.expression, exp.DataType.Type.TIME)
 89
 90    # Although the 2 dialects share similar signatures, BQ seems to inverse
 91    # the sign of the result so the start/end time operands are flipped
 92    return self.func("DATE_DIFF", unit_to_str(expression), expr, this)
 93
 94
 95def _array_sort_sql(self: DuckDB.Generator, expression: exp.ArraySort) -> str:
 96    if expression.expression:
 97        self.unsupported("DuckDB ARRAY_SORT does not support a comparator")
 98    return self.func("ARRAY_SORT", expression.this)
 99
100
101def _sort_array_sql(self: DuckDB.Generator, expression: exp.SortArray) -> str:
102    name = "ARRAY_REVERSE_SORT" if expression.args.get("asc") == exp.false() else "ARRAY_SORT"
103    return self.func(name, expression.this)
104
105
106def _build_sort_array_desc(args: t.List) -> exp.Expression:
107    return exp.SortArray(this=seq_get(args, 0), asc=exp.false())
108
109
110def _build_date_diff(args: t.List) -> exp.Expression:
111    return exp.DateDiff(this=seq_get(args, 2), expression=seq_get(args, 1), unit=seq_get(args, 0))
112
113
114def _build_generate_series(end_exclusive: bool = False) -> t.Callable[[t.List], exp.GenerateSeries]:
115    def _builder(args: t.List) -> exp.GenerateSeries:
116        # Check https://duckdb.org/docs/sql/functions/nested.html#range-functions
117        if len(args) == 1:
118            # DuckDB uses 0 as a default for the series' start when it's omitted
119            args.insert(0, exp.Literal.number("0"))
120
121        gen_series = exp.GenerateSeries.from_arg_list(args)
122        gen_series.set("is_end_exclusive", end_exclusive)
123
124        return gen_series
125
126    return _builder
127
128
129def _build_make_timestamp(args: t.List) -> exp.Expression:
130    if len(args) == 1:
131        return exp.UnixToTime(this=seq_get(args, 0), scale=exp.UnixToTime.MICROS)
132
133    return exp.TimestampFromParts(
134        year=seq_get(args, 0),
135        month=seq_get(args, 1),
136        day=seq_get(args, 2),
137        hour=seq_get(args, 3),
138        min=seq_get(args, 4),
139        sec=seq_get(args, 5),
140    )
141
142
143def _struct_sql(self: DuckDB.Generator, expression: exp.Struct) -> str:
144    args: t.List[str] = []
145
146    # BigQuery allows inline construction such as "STRUCT<a STRING, b INTEGER>('str', 1)" which is
147    # canonicalized to "ROW('str', 1) AS STRUCT(a TEXT, b INT)" in DuckDB
148    # The transformation to ROW will take place if a cast to STRUCT / ARRAY of STRUCTs is found
149    ancestor_cast = expression.find_ancestor(exp.Cast)
150    is_struct_cast = ancestor_cast and any(
151        casted_type.is_type(exp.DataType.Type.STRUCT)
152        for casted_type in ancestor_cast.find_all(exp.DataType)
153    )
154
155    for i, expr in enumerate(expression.expressions):
156        is_property_eq = isinstance(expr, exp.PropertyEQ)
157        value = expr.expression if is_property_eq else expr
158
159        if is_struct_cast:
160            args.append(self.sql(value))
161        else:
162            key = expr.name if is_property_eq else f"_{i}"
163            args.append(f"{self.sql(exp.Literal.string(key))}: {self.sql(value)}")
164
165    csv_args = ", ".join(args)
166
167    return f"ROW({csv_args})" if is_struct_cast else f"{{{csv_args}}}"
168
169
170def _datatype_sql(self: DuckDB.Generator, expression: exp.DataType) -> str:
171    if expression.is_type("array"):
172        return f"{self.expressions(expression, flat=True)}[{self.expressions(expression, key='values', flat=True)}]"
173
174    # Modifiers are not supported for TIME, [TIME | TIMESTAMP] WITH TIME ZONE
175    if expression.is_type(
176        exp.DataType.Type.TIME, exp.DataType.Type.TIMETZ, exp.DataType.Type.TIMESTAMPTZ
177    ):
178        return expression.this.value
179
180    return self.datatype_sql(expression)
181
182
183def _json_format_sql(self: DuckDB.Generator, expression: exp.JSONFormat) -> str:
184    sql = self.func("TO_JSON", expression.this, expression.args.get("options"))
185    return f"CAST({sql} AS TEXT)"
186
187
188def _unix_to_time_sql(self: DuckDB.Generator, expression: exp.UnixToTime) -> str:
189    scale = expression.args.get("scale")
190    timestamp = expression.this
191
192    if scale in (None, exp.UnixToTime.SECONDS):
193        return self.func("TO_TIMESTAMP", timestamp)
194    if scale == exp.UnixToTime.MILLIS:
195        return self.func("EPOCH_MS", timestamp)
196    if scale == exp.UnixToTime.MICROS:
197        return self.func("MAKE_TIMESTAMP", timestamp)
198
199    return self.func("TO_TIMESTAMP", exp.Div(this=timestamp, expression=exp.func("POW", 10, scale)))
200
201
202WRAPPED_JSON_EXTRACT_EXPRESSIONS = (exp.Binary, exp.Bracket, exp.In)
203
204
205def _arrow_json_extract_sql(self: DuckDB.Generator, expression: JSON_EXTRACT_TYPE) -> str:
206    arrow_sql = arrow_json_extract_sql(self, expression)
207    if not expression.same_parent and isinstance(
208        expression.parent, WRAPPED_JSON_EXTRACT_EXPRESSIONS
209    ):
210        arrow_sql = self.wrap(arrow_sql)
211    return arrow_sql
212
213
214def _implicit_datetime_cast(
215    arg: t.Optional[exp.Expression], type: exp.DataType.Type = exp.DataType.Type.DATE
216) -> t.Optional[exp.Expression]:
217    return exp.cast(arg, type) if isinstance(arg, exp.Literal) else arg
218
219
220def _date_diff_sql(self: DuckDB.Generator, expression: exp.DateDiff) -> str:
221    this = _implicit_datetime_cast(expression.this)
222    expr = _implicit_datetime_cast(expression.expression)
223
224    return self.func("DATE_DIFF", unit_to_str(expression), expr, this)
225
226
227def _generate_datetime_array_sql(
228    self: DuckDB.Generator, expression: t.Union[exp.GenerateDateArray, exp.GenerateTimestampArray]
229) -> str:
230    is_generate_date_array = isinstance(expression, exp.GenerateDateArray)
231
232    type = exp.DataType.Type.DATE if is_generate_date_array else exp.DataType.Type.TIMESTAMP
233    start = _implicit_datetime_cast(expression.args.get("start"), type=type)
234    end = _implicit_datetime_cast(expression.args.get("end"), type=type)
235
236    # BQ's GENERATE_DATE_ARRAY & GENERATE_TIMESTAMP_ARRAY are transformed to DuckDB'S GENERATE_SERIES
237    gen_series: t.Union[exp.GenerateSeries, exp.Cast] = exp.GenerateSeries(
238        start=start, end=end, step=expression.args.get("step")
239    )
240
241    if is_generate_date_array:
242        # The GENERATE_SERIES result type is TIMESTAMP array, so to match BQ's semantics for
243        # GENERATE_DATE_ARRAY we must cast it back to DATE array
244        gen_series = exp.cast(gen_series, exp.DataType.build("ARRAY<DATE>"))
245
246    return self.sql(gen_series)
247
248
249class DuckDB(Dialect):
250    NULL_ORDERING = "nulls_are_last"
251    SUPPORTS_USER_DEFINED_TYPES = False
252    SAFE_DIVISION = True
253    INDEX_OFFSET = 1
254    CONCAT_COALESCE = True
255    SUPPORTS_ORDER_BY_ALL = True
256    SUPPORTS_FIXED_SIZE_ARRAYS = True
257
258    # https://duckdb.org/docs/sql/introduction.html#creating-a-new-table
259    NORMALIZATION_STRATEGY = NormalizationStrategy.CASE_INSENSITIVE
260
261    def to_json_path(self, path: t.Optional[exp.Expression]) -> t.Optional[exp.Expression]:
262        if isinstance(path, exp.Literal):
263            # DuckDB also supports the JSON pointer syntax, where every path starts with a `/`.
264            # Additionally, it allows accessing the back of lists using the `[#-i]` syntax.
265            # This check ensures we'll avoid trying to parse these as JSON paths, which can
266            # either result in a noisy warning or in an invalid representation of the path.
267            path_text = path.name
268            if path_text.startswith("/") or "[#" in path_text:
269                return path
270
271        return super().to_json_path(path)
272
273    class Tokenizer(tokens.Tokenizer):
274        HEREDOC_STRINGS = ["$"]
275
276        HEREDOC_TAG_IS_IDENTIFIER = True
277        HEREDOC_STRING_ALTERNATIVE = TokenType.PARAMETER
278
279        KEYWORDS = {
280            **tokens.Tokenizer.KEYWORDS,
281            "//": TokenType.DIV,
282            "ATTACH": TokenType.COMMAND,
283            "BINARY": TokenType.VARBINARY,
284            "BITSTRING": TokenType.BIT,
285            "BPCHAR": TokenType.TEXT,
286            "CHAR": TokenType.TEXT,
287            "CHARACTER VARYING": TokenType.TEXT,
288            "EXCLUDE": TokenType.EXCEPT,
289            "LOGICAL": TokenType.BOOLEAN,
290            "ONLY": TokenType.ONLY,
291            "PIVOT_WIDER": TokenType.PIVOT,
292            "POSITIONAL": TokenType.POSITIONAL,
293            "SIGNED": TokenType.INT,
294            "STRING": TokenType.TEXT,
295            "SUMMARIZE": TokenType.SUMMARIZE,
296            "TIMESTAMP_S": TokenType.TIMESTAMP_S,
297            "TIMESTAMP_MS": TokenType.TIMESTAMP_MS,
298            "TIMESTAMP_NS": TokenType.TIMESTAMP_NS,
299            "TIMESTAMP_US": TokenType.TIMESTAMP,
300            "UBIGINT": TokenType.UBIGINT,
301            "UINTEGER": TokenType.UINT,
302            "USMALLINT": TokenType.USMALLINT,
303            "UTINYINT": TokenType.UTINYINT,
304            "VARCHAR": TokenType.TEXT,
305        }
306        KEYWORDS.pop("/*+")
307
308        SINGLE_TOKENS = {
309            **tokens.Tokenizer.SINGLE_TOKENS,
310            "$": TokenType.PARAMETER,
311        }
312
313    class Parser(parser.Parser):
314        BITWISE = {
315            **parser.Parser.BITWISE,
316            TokenType.TILDA: exp.RegexpLike,
317        }
318
319        FUNCTIONS_WITH_ALIASED_ARGS = {*parser.Parser.FUNCTIONS_WITH_ALIASED_ARGS, "STRUCT_PACK"}
320
321        FUNCTIONS = {
322            **parser.Parser.FUNCTIONS,
323            "ARRAY_REVERSE_SORT": _build_sort_array_desc,
324            "ARRAY_SORT": exp.SortArray.from_arg_list,
325            "DATEDIFF": _build_date_diff,
326            "DATE_DIFF": _build_date_diff,
327            "DATE_TRUNC": date_trunc_to_time,
328            "DATETRUNC": date_trunc_to_time,
329            "DECODE": lambda args: exp.Decode(
330                this=seq_get(args, 0), charset=exp.Literal.string("utf-8")
331            ),
332            "ENCODE": lambda args: exp.Encode(
333                this=seq_get(args, 0), charset=exp.Literal.string("utf-8")
334            ),
335            "EPOCH": exp.TimeToUnix.from_arg_list,
336            "EPOCH_MS": lambda args: exp.UnixToTime(
337                this=seq_get(args, 0), scale=exp.UnixToTime.MILLIS
338            ),
339            "JSON": exp.ParseJSON.from_arg_list,
340            "JSON_EXTRACT_PATH": parser.build_extract_json_with_path(exp.JSONExtract),
341            "JSON_EXTRACT_STRING": parser.build_extract_json_with_path(exp.JSONExtractScalar),
342            "LIST_HAS": exp.ArrayContains.from_arg_list,
343            "LIST_REVERSE_SORT": _build_sort_array_desc,
344            "LIST_SORT": exp.SortArray.from_arg_list,
345            "LIST_VALUE": lambda args: exp.Array(expressions=args),
346            "MAKE_TIME": exp.TimeFromParts.from_arg_list,
347            "MAKE_TIMESTAMP": _build_make_timestamp,
348            "MEDIAN": lambda args: exp.PercentileCont(
349                this=seq_get(args, 0), expression=exp.Literal.number(0.5)
350            ),
351            "QUANTILE_CONT": exp.PercentileCont.from_arg_list,
352            "QUANTILE_DISC": exp.PercentileDisc.from_arg_list,
353            "REGEXP_EXTRACT": lambda args: exp.RegexpExtract(
354                this=seq_get(args, 0), expression=seq_get(args, 1), group=seq_get(args, 2)
355            ),
356            "REGEXP_MATCHES": exp.RegexpLike.from_arg_list,
357            "REGEXP_REPLACE": lambda args: exp.RegexpReplace(
358                this=seq_get(args, 0),
359                expression=seq_get(args, 1),
360                replacement=seq_get(args, 2),
361                modifiers=seq_get(args, 3),
362            ),
363            "STRFTIME": build_formatted_time(exp.TimeToStr, "duckdb"),
364            "STRING_SPLIT": exp.Split.from_arg_list,
365            "STRING_SPLIT_REGEX": exp.RegexpSplit.from_arg_list,
366            "STRING_TO_ARRAY": exp.Split.from_arg_list,
367            "STRPTIME": build_formatted_time(exp.StrToTime, "duckdb"),
368            "STRUCT_PACK": exp.Struct.from_arg_list,
369            "STR_SPLIT": exp.Split.from_arg_list,
370            "STR_SPLIT_REGEX": exp.RegexpSplit.from_arg_list,
371            "TO_TIMESTAMP": exp.UnixToTime.from_arg_list,
372            "UNNEST": exp.Explode.from_arg_list,
373            "XOR": binary_from_function(exp.BitwiseXor),
374            "GENERATE_SERIES": _build_generate_series(),
375            "RANGE": _build_generate_series(end_exclusive=True),
376        }
377
378        FUNCTIONS.pop("DATE_SUB")
379
380        FUNCTION_PARSERS = parser.Parser.FUNCTION_PARSERS.copy()
381        FUNCTION_PARSERS.pop("DECODE")
382
383        NO_PAREN_FUNCTION_PARSERS = {
384            **parser.Parser.NO_PAREN_FUNCTION_PARSERS,
385            "MAP": lambda self: self._parse_map(),
386        }
387
388        TABLE_ALIAS_TOKENS = parser.Parser.TABLE_ALIAS_TOKENS - {
389            TokenType.SEMI,
390            TokenType.ANTI,
391        }
392
393        PLACEHOLDER_PARSERS = {
394            **parser.Parser.PLACEHOLDER_PARSERS,
395            TokenType.PARAMETER: lambda self: (
396                self.expression(exp.Placeholder, this=self._prev.text)
397                if self._match(TokenType.NUMBER) or self._match_set(self.ID_VAR_TOKENS)
398                else None
399            ),
400        }
401
402        TYPE_CONVERTERS = {
403            # https://duckdb.org/docs/sql/data_types/numeric
404            exp.DataType.Type.DECIMAL: build_default_decimal_type(precision=18, scale=3),
405            # https://duckdb.org/docs/sql/data_types/text
406            exp.DataType.Type.TEXT: lambda dtype: exp.DataType.build("TEXT"),
407        }
408
409        def _parse_table_sample(self, as_modifier: bool = False) -> t.Optional[exp.TableSample]:
410            # https://duckdb.org/docs/sql/samples.html
411            sample = super()._parse_table_sample(as_modifier=as_modifier)
412            if sample and not sample.args.get("method"):
413                if sample.args.get("size"):
414                    sample.set("method", exp.var("RESERVOIR"))
415                else:
416                    sample.set("method", exp.var("SYSTEM"))
417
418            return sample
419
420        def _parse_bracket(
421            self, this: t.Optional[exp.Expression] = None
422        ) -> t.Optional[exp.Expression]:
423            bracket = super()._parse_bracket(this)
424            if isinstance(bracket, exp.Bracket):
425                bracket.set("returns_list_for_maps", True)
426
427            return bracket
428
429        def _parse_map(self) -> exp.ToMap | exp.Map:
430            if self._match(TokenType.L_BRACE, advance=False):
431                return self.expression(exp.ToMap, this=self._parse_bracket())
432
433            args = self._parse_wrapped_csv(self._parse_assignment)
434            return self.expression(exp.Map, keys=seq_get(args, 0), values=seq_get(args, 1))
435
436        def _parse_struct_types(self, type_required: bool = False) -> t.Optional[exp.Expression]:
437            return self._parse_field_def()
438
439        def _pivot_column_names(self, aggregations: t.List[exp.Expression]) -> t.List[str]:
440            if len(aggregations) == 1:
441                return super()._pivot_column_names(aggregations)
442            return pivot_column_names(aggregations, dialect="duckdb")
443
444    class Generator(generator.Generator):
445        PARAMETER_TOKEN = "$"
446        NAMED_PLACEHOLDER_TOKEN = "$"
447        JOIN_HINTS = False
448        TABLE_HINTS = False
449        QUERY_HINTS = False
450        LIMIT_FETCH = "LIMIT"
451        STRUCT_DELIMITER = ("(", ")")
452        RENAME_TABLE_WITH_DB = False
453        NVL2_SUPPORTED = False
454        SEMI_ANTI_JOIN_WITH_SIDE = False
455        TABLESAMPLE_KEYWORDS = "USING SAMPLE"
456        TABLESAMPLE_SEED_KEYWORD = "REPEATABLE"
457        LAST_DAY_SUPPORTS_DATE_PART = False
458        JSON_KEY_VALUE_PAIR_SEP = ","
459        IGNORE_NULLS_IN_FUNC = True
460        JSON_PATH_BRACKETED_KEY_SUPPORTED = False
461        SUPPORTS_CREATE_TABLE_LIKE = False
462        MULTI_ARG_DISTINCT = False
463        CAN_IMPLEMENT_ARRAY_ANY = True
464        SUPPORTS_TO_NUMBER = False
465        COPY_HAS_INTO_KEYWORD = False
466        STAR_EXCEPT = "EXCLUDE"
467        PAD_FILL_PATTERN_IS_REQUIRED = True
468        ARRAY_CONCAT_IS_VAR_LEN = False
469
470        TRANSFORMS = {
471            **generator.Generator.TRANSFORMS,
472            exp.ApproxDistinct: approx_count_distinct_sql,
473            exp.Array: inline_array_unless_query,
474            exp.ArrayContainsAll: rename_func("ARRAY_HAS_ALL"),
475            exp.ArrayFilter: rename_func("LIST_FILTER"),
476            exp.ArraySize: rename_func("ARRAY_LENGTH"),
477            exp.ArgMax: arg_max_or_min_no_count("ARG_MAX"),
478            exp.ArgMin: arg_max_or_min_no_count("ARG_MIN"),
479            exp.ArraySort: _array_sort_sql,
480            exp.ArraySum: rename_func("LIST_SUM"),
481            exp.BitwiseXor: rename_func("XOR"),
482            exp.CommentColumnConstraint: no_comment_column_constraint_sql,
483            exp.CurrentDate: lambda *_: "CURRENT_DATE",
484            exp.CurrentTime: lambda *_: "CURRENT_TIME",
485            exp.CurrentTimestamp: lambda *_: "CURRENT_TIMESTAMP",
486            exp.DayOfMonth: rename_func("DAYOFMONTH"),
487            exp.DayOfWeek: rename_func("DAYOFWEEK"),
488            exp.DayOfYear: rename_func("DAYOFYEAR"),
489            exp.DataType: _datatype_sql,
490            exp.Date: _date_sql,
491            exp.DateAdd: _date_delta_sql,
492            exp.DateFromParts: rename_func("MAKE_DATE"),
493            exp.DateSub: _date_delta_sql,
494            exp.DateDiff: _date_diff_sql,
495            exp.DateStrToDate: datestrtodate_sql,
496            exp.Datetime: no_datetime_sql,
497            exp.DatetimeSub: _date_delta_sql,
498            exp.DatetimeAdd: _date_delta_sql,
499            exp.DateToDi: lambda self,
500            e: f"CAST(STRFTIME({self.sql(e, 'this')}, {DuckDB.DATEINT_FORMAT}) AS INT)",
501            exp.Decode: lambda self, e: encode_decode_sql(self, e, "DECODE", replace=False),
502            exp.DiToDate: lambda self,
503            e: f"CAST(STRPTIME(CAST({self.sql(e, 'this')} AS TEXT), {DuckDB.DATEINT_FORMAT}) AS DATE)",
504            exp.Encode: lambda self, e: encode_decode_sql(self, e, "ENCODE", replace=False),
505            exp.GenerateDateArray: _generate_datetime_array_sql,
506            exp.GenerateTimestampArray: _generate_datetime_array_sql,
507            exp.Explode: rename_func("UNNEST"),
508            exp.IntDiv: lambda self, e: self.binary(e, "//"),
509            exp.IsInf: rename_func("ISINF"),
510            exp.IsNan: rename_func("ISNAN"),
511            exp.JSONExtract: _arrow_json_extract_sql,
512            exp.JSONExtractScalar: _arrow_json_extract_sql,
513            exp.JSONFormat: _json_format_sql,
514            exp.LogicalOr: rename_func("BOOL_OR"),
515            exp.LogicalAnd: rename_func("BOOL_AND"),
516            exp.MD5Digest: lambda self, e: self.func("UNHEX", self.func("MD5", e.this)),
517            exp.MonthsBetween: lambda self, e: self.func(
518                "DATEDIFF",
519                "'month'",
520                exp.cast(e.expression, exp.DataType.Type.TIMESTAMP, copy=True),
521                exp.cast(e.this, exp.DataType.Type.TIMESTAMP, copy=True),
522            ),
523            exp.PercentileCont: rename_func("QUANTILE_CONT"),
524            exp.PercentileDisc: rename_func("QUANTILE_DISC"),
525            # DuckDB doesn't allow qualified columns inside of PIVOT expressions.
526            # See: https://github.com/duckdb/duckdb/blob/671faf92411182f81dce42ac43de8bfb05d9909e/src/planner/binder/tableref/bind_pivot.cpp#L61-L62
527            exp.Pivot: transforms.preprocess([transforms.unqualify_columns]),
528            exp.RegexpExtract: regexp_extract_sql,
529            exp.RegexpReplace: lambda self, e: self.func(
530                "REGEXP_REPLACE",
531                e.this,
532                e.expression,
533                e.args.get("replacement"),
534                e.args.get("modifiers"),
535            ),
536            exp.RegexpLike: rename_func("REGEXP_MATCHES"),
537            exp.RegexpSplit: rename_func("STR_SPLIT_REGEX"),
538            exp.Return: lambda self, e: self.sql(e, "this"),
539            exp.ReturnsProperty: lambda self, e: "TABLE" if isinstance(e.this, exp.Schema) else "",
540            exp.Rand: rename_func("RANDOM"),
541            exp.SafeDivide: no_safe_divide_sql,
542            exp.Split: rename_func("STR_SPLIT"),
543            exp.SortArray: _sort_array_sql,
544            exp.StrPosition: str_position_sql,
545            exp.StrToUnix: lambda self, e: self.func(
546                "EPOCH", self.func("STRPTIME", e.this, self.format_time(e))
547            ),
548            exp.Struct: _struct_sql,
549            exp.TimeAdd: _date_delta_sql,
550            exp.Time: no_time_sql,
551            exp.TimeDiff: _timediff_sql,
552            exp.Timestamp: no_timestamp_sql,
553            exp.TimestampDiff: lambda self, e: self.func(
554                "DATE_DIFF", exp.Literal.string(e.unit), e.expression, e.this
555            ),
556            exp.TimestampTrunc: timestamptrunc_sql(),
557            exp.TimeStrToDate: lambda self, e: self.sql(exp.cast(e.this, exp.DataType.Type.DATE)),
558            exp.TimeStrToTime: timestrtotime_sql,
559            exp.TimeStrToUnix: lambda self, e: self.func(
560                "EPOCH", exp.cast(e.this, exp.DataType.Type.TIMESTAMP)
561            ),
562            exp.TimeToStr: lambda self, e: self.func("STRFTIME", e.this, self.format_time(e)),
563            exp.TimeToUnix: rename_func("EPOCH"),
564            exp.TsOrDiToDi: lambda self,
565            e: f"CAST(SUBSTR(REPLACE(CAST({self.sql(e, 'this')} AS TEXT), '-', ''), 1, 8) AS INT)",
566            exp.TsOrDsAdd: _date_delta_sql,
567            exp.TsOrDsDiff: lambda self, e: self.func(
568                "DATE_DIFF",
569                f"'{e.args.get('unit') or 'DAY'}'",
570                exp.cast(e.expression, exp.DataType.Type.TIMESTAMP),
571                exp.cast(e.this, exp.DataType.Type.TIMESTAMP),
572            ),
573            exp.UnixToStr: lambda self, e: self.func(
574                "STRFTIME", self.func("TO_TIMESTAMP", e.this), self.format_time(e)
575            ),
576            exp.DatetimeTrunc: lambda self, e: self.func(
577                "DATE_TRUNC", unit_to_str(e), exp.cast(e.this, exp.DataType.Type.DATETIME)
578            ),
579            exp.UnixToTime: _unix_to_time_sql,
580            exp.UnixToTimeStr: lambda self, e: f"CAST(TO_TIMESTAMP({self.sql(e, 'this')}) AS TEXT)",
581            exp.VariancePop: rename_func("VAR_POP"),
582            exp.WeekOfYear: rename_func("WEEKOFYEAR"),
583            exp.Xor: bool_xor_sql,
584        }
585
586        SUPPORTED_JSON_PATH_PARTS = {
587            exp.JSONPathKey,
588            exp.JSONPathRoot,
589            exp.JSONPathSubscript,
590            exp.JSONPathWildcard,
591        }
592
593        TYPE_MAPPING = {
594            **generator.Generator.TYPE_MAPPING,
595            exp.DataType.Type.BINARY: "BLOB",
596            exp.DataType.Type.BPCHAR: "TEXT",
597            exp.DataType.Type.CHAR: "TEXT",
598            exp.DataType.Type.FLOAT: "REAL",
599            exp.DataType.Type.NCHAR: "TEXT",
600            exp.DataType.Type.NVARCHAR: "TEXT",
601            exp.DataType.Type.UINT: "UINTEGER",
602            exp.DataType.Type.VARBINARY: "BLOB",
603            exp.DataType.Type.ROWVERSION: "BLOB",
604            exp.DataType.Type.VARCHAR: "TEXT",
605            exp.DataType.Type.TIMESTAMPNTZ: "TIMESTAMP",
606            exp.DataType.Type.TIMESTAMP_S: "TIMESTAMP_S",
607            exp.DataType.Type.TIMESTAMP_MS: "TIMESTAMP_MS",
608            exp.DataType.Type.TIMESTAMP_NS: "TIMESTAMP_NS",
609        }
610
611        # https://github.com/duckdb/duckdb/blob/ff7f24fd8e3128d94371827523dae85ebaf58713/third_party/libpg_query/grammar/keywords/reserved_keywords.list#L1-L77
612        RESERVED_KEYWORDS = {
613            "array",
614            "analyse",
615            "union",
616            "all",
617            "when",
618            "in_p",
619            "default",
620            "create_p",
621            "window",
622            "asymmetric",
623            "to",
624            "else",
625            "localtime",
626            "from",
627            "end_p",
628            "select",
629            "current_date",
630            "foreign",
631            "with",
632            "grant",
633            "session_user",
634            "or",
635            "except",
636            "references",
637            "fetch",
638            "limit",
639            "group_p",
640            "leading",
641            "into",
642            "collate",
643            "offset",
644            "do",
645            "then",
646            "localtimestamp",
647            "check_p",
648            "lateral_p",
649            "current_role",
650            "where",
651            "asc_p",
652            "placing",
653            "desc_p",
654            "user",
655            "unique",
656            "initially",
657            "column",
658            "both",
659            "some",
660            "as",
661            "any",
662            "only",
663            "deferrable",
664            "null_p",
665            "current_time",
666            "true_p",
667            "table",
668            "case",
669            "trailing",
670            "variadic",
671            "for",
672            "on",
673            "distinct",
674            "false_p",
675            "not",
676            "constraint",
677            "current_timestamp",
678            "returning",
679            "primary",
680            "intersect",
681            "having",
682            "analyze",
683            "current_user",
684            "and",
685            "cast",
686            "symmetric",
687            "using",
688            "order",
689            "current_catalog",
690        }
691
692        UNWRAPPED_INTERVAL_VALUES = (exp.Literal, exp.Paren)
693
694        # DuckDB doesn't generally support CREATE TABLE .. properties
695        # https://duckdb.org/docs/sql/statements/create_table.html
696        PROPERTIES_LOCATION = {
697            prop: exp.Properties.Location.UNSUPPORTED
698            for prop in generator.Generator.PROPERTIES_LOCATION
699        }
700
701        # There are a few exceptions (e.g. temporary tables) which are supported or
702        # can be transpiled to DuckDB, so we explicitly override them accordingly
703        PROPERTIES_LOCATION[exp.LikeProperty] = exp.Properties.Location.POST_SCHEMA
704        PROPERTIES_LOCATION[exp.TemporaryProperty] = exp.Properties.Location.POST_CREATE
705        PROPERTIES_LOCATION[exp.ReturnsProperty] = exp.Properties.Location.POST_ALIAS
706
707        def fromiso8601timestamp_sql(self, expression: exp.FromISO8601Timestamp) -> str:
708            return self.sql(exp.cast(expression.this, exp.DataType.Type.TIMESTAMPTZ))
709
710        def strtotime_sql(self, expression: exp.StrToTime) -> str:
711            if expression.args.get("safe"):
712                formatted_time = self.format_time(expression)
713                return f"CAST({self.func('TRY_STRPTIME', expression.this, formatted_time)} AS TIMESTAMP)"
714            return str_to_time_sql(self, expression)
715
716        def strtodate_sql(self, expression: exp.StrToDate) -> str:
717            if expression.args.get("safe"):
718                formatted_time = self.format_time(expression)
719                return f"CAST({self.func('TRY_STRPTIME', expression.this, formatted_time)} AS DATE)"
720            return f"CAST({str_to_time_sql(self, expression)} AS DATE)"
721
722        def parsejson_sql(self, expression: exp.ParseJSON) -> str:
723            arg = expression.this
724            if expression.args.get("safe"):
725                return self.sql(exp.case().when(exp.func("json_valid", arg), arg).else_(exp.null()))
726            return self.func("JSON", arg)
727
728        def timefromparts_sql(self, expression: exp.TimeFromParts) -> str:
729            nano = expression.args.get("nano")
730            if nano is not None:
731                expression.set(
732                    "sec", expression.args["sec"] + nano.pop() / exp.Literal.number(1000000000.0)
733                )
734
735            return rename_func("MAKE_TIME")(self, expression)
736
737        def timestampfromparts_sql(self, expression: exp.TimestampFromParts) -> str:
738            sec = expression.args["sec"]
739
740            milli = expression.args.get("milli")
741            if milli is not None:
742                sec += milli.pop() / exp.Literal.number(1000.0)
743
744            nano = expression.args.get("nano")
745            if nano is not None:
746                sec += nano.pop() / exp.Literal.number(1000000000.0)
747
748            if milli or nano:
749                expression.set("sec", sec)
750
751            return rename_func("MAKE_TIMESTAMP")(self, expression)
752
753        def tablesample_sql(
754            self,
755            expression: exp.TableSample,
756            tablesample_keyword: t.Optional[str] = None,
757        ) -> str:
758            if not isinstance(expression.parent, exp.Select):
759                # This sample clause only applies to a single source, not the entire resulting relation
760                tablesample_keyword = "TABLESAMPLE"
761
762            if expression.args.get("size"):
763                method = expression.args.get("method")
764                if method and method.name.upper() != "RESERVOIR":
765                    self.unsupported(
766                        f"Sampling method {method} is not supported with a discrete sample count, "
767                        "defaulting to reservoir sampling"
768                    )
769                    expression.set("method", exp.var("RESERVOIR"))
770
771            return super().tablesample_sql(expression, tablesample_keyword=tablesample_keyword)
772
773        def interval_sql(self, expression: exp.Interval) -> str:
774            multiplier: t.Optional[int] = None
775            unit = expression.text("unit").lower()
776
777            if unit.startswith("week"):
778                multiplier = 7
779            if unit.startswith("quarter"):
780                multiplier = 90
781
782            if multiplier:
783                return f"({multiplier} * {super().interval_sql(exp.Interval(this=expression.this, unit=exp.var('DAY')))})"
784
785            return super().interval_sql(expression)
786
787        def columndef_sql(self, expression: exp.ColumnDef, sep: str = " ") -> str:
788            if isinstance(expression.parent, exp.UserDefinedFunction):
789                return self.sql(expression, "this")
790            return super().columndef_sql(expression, sep)
791
792        def join_sql(self, expression: exp.Join) -> str:
793            if (
794                expression.side == "LEFT"
795                and not expression.args.get("on")
796                and isinstance(expression.this, exp.Unnest)
797            ):
798                # Some dialects support `LEFT JOIN UNNEST(...)` without an explicit ON clause
799                # DuckDB doesn't, but we can just add a dummy ON clause that is always true
800                return super().join_sql(expression.on(exp.true()))
801
802            return super().join_sql(expression)
803
804        def generateseries_sql(self, expression: exp.GenerateSeries) -> str:
805            # GENERATE_SERIES(a, b) -> [a, b], RANGE(a, b) -> [a, b)
806            if expression.args.get("is_end_exclusive"):
807                return rename_func("RANGE")(self, expression)
808
809            return self.function_fallback_sql(expression)
810
811        def bracket_sql(self, expression: exp.Bracket) -> str:
812            this = expression.this
813            if isinstance(this, exp.Array):
814                this.replace(exp.paren(this))
815
816            bracket = super().bracket_sql(expression)
817
818            if not expression.args.get("returns_list_for_maps"):
819                if not this.type:
820                    from sqlglot.optimizer.annotate_types import annotate_types
821
822                    this = annotate_types(this)
823
824                if this.is_type(exp.DataType.Type.MAP):
825                    bracket = f"({bracket})[1]"
826
827            return bracket
828
829        def withingroup_sql(self, expression: exp.WithinGroup) -> str:
830            expression_sql = self.sql(expression, "expression")
831
832            func = expression.this
833            if isinstance(func, exp.PERCENTILES):
834                # Make the order key the first arg and slide the fraction to the right
835                # https://duckdb.org/docs/sql/aggregates#ordered-set-aggregate-functions
836                order_col = expression.find(exp.Ordered)
837                if order_col:
838                    func.set("expression", func.this)
839                    func.set("this", order_col.this)
840
841            this = self.sql(expression, "this").rstrip(")")
842
843            return f"{this}{expression_sql})"
844
845        def length_sql(self, expression: exp.Length) -> str:
846            arg = expression.this
847
848            # Dialects like BQ and Snowflake also accept binary values as args, so
849            # DDB will attempt to infer the type or resort to case/when resolution
850            if not expression.args.get("binary") or arg.is_string:
851                return self.func("LENGTH", arg)
852
853            if not arg.type:
854                from sqlglot.optimizer.annotate_types import annotate_types
855
856                arg = annotate_types(arg)
857
858            if arg.is_type(*exp.DataType.TEXT_TYPES):
859                return self.func("LENGTH", arg)
860
861            # We need these casts to make duckdb's static type checker happy
862            blob = exp.cast(arg, exp.DataType.Type.VARBINARY)
863            varchar = exp.cast(arg, exp.DataType.Type.VARCHAR)
864
865            case = (
866                exp.case(self.func("TYPEOF", arg))
867                .when(
868                    "'VARCHAR'", exp.Anonymous(this="LENGTH", expressions=[varchar])
869                )  # anonymous to break length_sql recursion
870                .when("'BLOB'", self.func("OCTET_LENGTH", blob))
871            )
872
873            return self.sql(case)
874
875        def objectinsert_sql(self, expression: exp.ObjectInsert) -> str:
876            this = expression.this
877            key = expression.args.get("key")
878            key_sql = key.name if isinstance(key, exp.Expression) else ""
879            value_sql = self.sql(expression, "value")
880
881            kv_sql = f"{key_sql} := {value_sql}"
882
883            # If the input struct is empty e.g. transpiling OBJECT_INSERT(OBJECT_CONSTRUCT(), key, value) from Snowflake
884            # then we can generate STRUCT_PACK which will build it since STRUCT_INSERT({}, key := value) is not valid DuckDB
885            if isinstance(this, exp.Struct) and not this.expressions:
886                return self.func("STRUCT_PACK", kv_sql)
887
888            return self.func("STRUCT_INSERT", this, kv_sql)
889
890        def unnest_sql(self, expression: exp.Unnest) -> str:
891            explode_array = expression.args.get("explode_array")
892            if explode_array:
893                # In BigQuery, UNNESTing a nested array leads to explosion of the top-level array & struct
894                # This is transpiled to DDB by transforming "FROM UNNEST(...)" to "FROM (SELECT UNNEST(..., max_depth => 2))"
895                expression.expressions.append(
896                    exp.Kwarg(this=exp.var("max_depth"), expression=exp.Literal.number(2))
897                )
898
899                # If BQ's UNNEST is aliased, we transform it from a column alias to a table alias in DDB
900                alias = expression.args.get("alias")
901                if alias:
902                    expression.set("alias", None)
903                    alias = exp.TableAlias(this=seq_get(alias.args.get("columns"), 0))
904
905                unnest_sql = super().unnest_sql(expression)
906                select = exp.Select(expressions=[unnest_sql]).subquery(alias)
907                return self.sql(select)
908
909            return super().unnest_sql(expression)