sqlglot.generators.snowflake

   1from __future__ import annotations
   2
   3import typing as t
   4from collections import defaultdict
   5
   6from sqlglot import exp, generator, transforms
   7from sqlglot.dialects.dialect import (
   8    array_append_sql,
   9    array_concat_sql,
  10    date_delta_sql,
  11    datestrtodate_sql,
  12    groupconcat_sql,
  13    if_sql,
  14    inline_array_sql,
  15    map_date_part,
  16    max_or_greatest,
  17    min_or_least,
  18    no_make_interval_sql,
  19    no_timestamp_sql,
  20    rename_func,
  21    strposition_sql,
  22    timestampdiff_sql,
  23    timestamptrunc_sql,
  24    timestrtotime_sql,
  25    unit_to_str,
  26    var_map_sql,
  27)
  28from sqlglot.generator import unsupported_args
  29from sqlglot.helper import find_new_name, flatten, seq_get
  30from sqlglot.optimizer.scope import build_scope, find_all_in_scope
  31from sqlglot.parsers.snowflake import (
  32    RANKING_WINDOW_FUNCTIONS_WITH_FRAME,
  33    TIMESTAMP_TYPES,
  34    SnowflakeParser,
  35    build_object_construct,
  36)
  37from sqlglot.tokens import TokenType
  38
  39if t.TYPE_CHECKING:
  40    from sqlglot._typing import E
  41
  42
  43def _build_datediff(args: list) -> exp.DateDiff:
  44    return exp.DateDiff(
  45        this=seq_get(args, 2),
  46        expression=seq_get(args, 1),
  47        unit=map_date_part(seq_get(args, 0)),
  48        date_part_boundary=True,
  49    )
  50
  51
  52def _build_date_time_add(expr_type: type[E]) -> t.Callable[[list], E]:
  53    def _builder(args: list) -> E:
  54        return expr_type(
  55            this=seq_get(args, 2),
  56            expression=seq_get(args, 1),
  57            unit=map_date_part(seq_get(args, 0)),
  58        )
  59
  60    return _builder
  61
  62
  63def _regexpilike_sql(self: SnowflakeGenerator, expression: exp.RegexpILike) -> str:
  64    flag = expression.text("flag")
  65
  66    if "i" not in flag:
  67        flag += "i"
  68
  69    return self.func(
  70        "REGEXP_LIKE", expression.this, expression.expression, exp.Literal.string(flag)
  71    )
  72
  73
  74def _unqualify_pivot_columns(expression: exp.Expr) -> exp.Expr:
  75    """
  76    Snowflake doesn't allow columns referenced in UNPIVOT to be qualified,
  77    so we need to unqualify them. Same goes for ANY ORDER BY <column>.
  78
  79    Example:
  80        >>> from sqlglot import parse_one
  81        >>> expr = parse_one("SELECT * FROM m_sales UNPIVOT(sales FOR month IN (m_sales.jan, feb, mar, april))")
  82        >>> print(_unqualify_pivot_columns(expr).sql(dialect="snowflake"))
  83        SELECT * FROM m_sales UNPIVOT(sales FOR month IN (jan, feb, mar, april))
  84    """
  85    if isinstance(expression, exp.Pivot):
  86        if expression.unpivot:
  87            expression = transforms.unqualify_columns(expression)
  88        else:
  89            for field in expression.fields:
  90                field_expr = seq_get(field.expressions if field else [], 0)
  91
  92                if isinstance(field_expr, exp.PivotAny):
  93                    unqualified_field_expr = transforms.unqualify_columns(field_expr)
  94                    t.cast(exp.Expr, field).set("expressions", unqualified_field_expr, 0)
  95
  96    return expression
  97
  98
  99def _flatten_structured_types_unless_iceberg(expression: exp.Expr) -> exp.Expr:
 100    assert isinstance(expression, exp.Create)
 101
 102    def _flatten_structured_type(expression: exp.Expr) -> exp.Expr:
 103        if isinstance(expression, exp.DataType) and expression.this in exp.DataType.NESTED_TYPES:
 104            expression.set("expressions", None)
 105        return expression
 106
 107    props = expression.args.get("properties")
 108    if isinstance(expression.this, exp.Schema) and not (props and props.find(exp.IcebergProperty)):
 109        for schema_expression in expression.this.expressions:
 110            if isinstance(schema_expression, exp.ColumnDef):
 111                column_type = schema_expression.kind
 112                if isinstance(column_type, exp.DataType):
 113                    column_type.transform(_flatten_structured_type, copy=False)
 114
 115    return expression
 116
 117
 118def _unnest_generate_date_array(unnest: exp.Unnest) -> None:
 119    generate_date_array = unnest.expressions[0]
 120    start = generate_date_array.args.get("start")
 121    end = generate_date_array.args.get("end")
 122    step = generate_date_array.args.get("step")
 123
 124    if not start or not end or not isinstance(step, exp.Interval) or step.name != "1":
 125        return
 126
 127    unit = step.args.get("unit")
 128
 129    unnest_alias = unnest.args.get("alias")
 130    if unnest_alias:
 131        unnest_alias = unnest_alias.copy()
 132        sequence_value_name = seq_get(unnest_alias.columns, 0) or "value"
 133    else:
 134        sequence_value_name = "value"
 135
 136    # We'll add the next sequence value to the starting date and project the result
 137    date_add = _build_date_time_add(exp.DateAdd)(
 138        [unit, exp.cast(sequence_value_name, "int"), exp.cast(start, "date")]
 139    )
 140
 141    # We use DATEDIFF to compute the number of sequence values needed
 142    number_sequence = SnowflakeParser.FUNCTIONS["ARRAY_GENERATE_RANGE"](
 143        [exp.Literal.number(0), _build_datediff([unit, start, end]) + 1]
 144    )
 145
 146    unnest.set("expressions", [number_sequence])
 147
 148    unnest_parent = unnest.parent
 149    if isinstance(unnest_parent, exp.Join):
 150        select = unnest_parent.parent
 151        if isinstance(select, exp.Select):
 152            replace_column_name = (
 153                sequence_value_name
 154                if isinstance(sequence_value_name, str)
 155                else sequence_value_name.name
 156            )
 157
 158            scope = build_scope(select)
 159            if scope:
 160                for column in scope.columns:
 161                    if column.name.lower() == replace_column_name.lower():
 162                        column.replace(
 163                            date_add.as_(replace_column_name)
 164                            if isinstance(column.parent, exp.Select)
 165                            else date_add
 166                        )
 167
 168            lateral = exp.Lateral(this=unnest_parent.this.pop())
 169            unnest_parent.replace(exp.Join(this=lateral))
 170    else:
 171        unnest.replace(
 172            exp.select(date_add.as_(sequence_value_name))
 173            .from_(unnest.copy())
 174            .subquery(unnest_alias)
 175        )
 176
 177
 178def _transform_generate_date_array(expression: exp.Expr) -> exp.Expr:
 179    if isinstance(expression, exp.Select):
 180        for generate_date_array in expression.find_all(exp.GenerateDateArray):
 181            parent = generate_date_array.parent
 182
 183            # If GENERATE_DATE_ARRAY is used directly as an array (e.g passed into ARRAY_LENGTH), the transformed Snowflake
 184            # query is the following (it'll be unnested properly on the next iteration due to copy):
 185            # SELECT ref(GENERATE_DATE_ARRAY(...)) -> SELECT ref((SELECT ARRAY_AGG(*) FROM UNNEST(GENERATE_DATE_ARRAY(...))))
 186            if not isinstance(parent, exp.Unnest):
 187                unnest = exp.Unnest(expressions=[generate_date_array.copy()])
 188                generate_date_array.replace(
 189                    exp.select(exp.ArrayAgg(this=exp.Star())).from_(unnest).subquery()
 190                )
 191
 192            if (
 193                isinstance(parent, exp.Unnest)
 194                and isinstance(parent.parent, (exp.From, exp.Join))
 195                and len(parent.expressions) == 1
 196            ):
 197                _unnest_generate_date_array(parent)
 198
 199    return expression
 200
 201
 202def _regexpextract_sql(
 203    self: SnowflakeGenerator, expression: exp.RegexpExtract | exp.RegexpExtractAll
 204) -> str:
 205    # Other dialects don't support all of the following parameters, so we need to
 206    # generate default values as necessary to ensure the transpilation is correct
 207    group = expression.args.get("group")
 208
 209    # To avoid generating all these default values, we set group to None if
 210    # it's 0 (also default value) which doesn't trigger the following chain
 211    if group and group.name == "0":
 212        group = None
 213
 214    parameters = expression.args.get("parameters") or (group and exp.Literal.string("c"))
 215    occurrence = expression.args.get("occurrence") or (parameters and exp.Literal.number(1))
 216    position = expression.args.get("position") or (occurrence and exp.Literal.number(1))
 217
 218    return self.func(
 219        "REGEXP_SUBSTR" if isinstance(expression, exp.RegexpExtract) else "REGEXP_SUBSTR_ALL",
 220        expression.this,
 221        expression.expression,
 222        position,
 223        occurrence,
 224        parameters,
 225        group,
 226    )
 227
 228
 229def _json_extract_value_array_sql(
 230    self: SnowflakeGenerator, expression: exp.JSONValueArray | exp.JSONExtractArray
 231) -> str:
 232    json_extract = exp.JSONExtract(this=expression.this, expression=expression.expression)
 233    ident = exp.to_identifier("x")
 234
 235    if isinstance(expression, exp.JSONValueArray):
 236        this: exp.Expr = exp.cast(ident, to=exp.DType.VARCHAR)
 237    else:
 238        this = exp.ParseJSON(this=f"TO_JSON({ident})")
 239
 240    transform_lambda = exp.Lambda(expressions=[ident], this=this)
 241
 242    return self.func("TRANSFORM", json_extract, transform_lambda)
 243
 244
 245def _qualify_unnested_columns(expression: exp.Expr) -> exp.Expr:
 246    if isinstance(expression, exp.Select):
 247        scope = build_scope(expression)
 248        if not scope:
 249            return expression
 250
 251        unnests = list(scope.find_all(exp.Unnest))
 252
 253        if not unnests:
 254            return expression
 255
 256        taken_source_names = set(scope.sources)
 257        column_source: dict[str, exp.Identifier] = {}
 258        unnest_to_identifier: dict[exp.Unnest, exp.Identifier] = {}
 259
 260        unnest_identifier: exp.Identifier | None = None
 261        orig_expression = expression.copy()
 262
 263        for unnest in unnests:
 264            if not isinstance(unnest.parent, (exp.From, exp.Join)):
 265                continue
 266
 267            # Try to infer column names produced by an unnest operator. This is only possible
 268            # when we can peek into the (statically known) contents of the unnested value.
 269            unnest_columns: set[str] = set()
 270            for unnest_expr in unnest.expressions:
 271                if not isinstance(unnest_expr, exp.Array):
 272                    continue
 273
 274                for array_expr in unnest_expr.expressions:
 275                    if not (
 276                        isinstance(array_expr, exp.Struct)
 277                        and array_expr.expressions
 278                        and all(
 279                            isinstance(struct_expr, exp.PropertyEQ)
 280                            for struct_expr in array_expr.expressions
 281                        )
 282                    ):
 283                        continue
 284
 285                    unnest_columns.update(
 286                        struct_expr.this.name.lower() for struct_expr in array_expr.expressions
 287                    )
 288                    break
 289
 290                if unnest_columns:
 291                    break
 292
 293            unnest_alias = unnest.args.get("alias")
 294            if not unnest_alias:
 295                alias_name = find_new_name(taken_source_names, "value")
 296                taken_source_names.add(alias_name)
 297
 298                # Produce a `TableAlias` AST similar to what is produced for BigQuery. This
 299                # will be corrected later, when we generate SQL for the `Unnest` AST node.
 300                aliased_unnest = exp.alias_(unnest, None, table=[alias_name])
 301                scope.replace(unnest, aliased_unnest)
 302
 303                unnest_identifier = aliased_unnest.args["alias"].columns[0]
 304            else:
 305                alias_columns = getattr(unnest_alias, "columns", [])
 306                unnest_identifier = unnest_alias.this or seq_get(alias_columns, 0)
 307
 308            if not isinstance(unnest_identifier, exp.Identifier):
 309                return orig_expression
 310
 311            unnest_to_identifier[unnest] = unnest_identifier
 312            column_source.update({c.lower(): unnest_identifier for c in unnest_columns})
 313
 314        for column in scope.columns:
 315            if column.table:
 316                continue
 317
 318            table = column_source.get(column.name.lower())
 319            if (
 320                unnest_identifier
 321                and not table
 322                and len(scope.sources) == 1
 323                and column.name.lower() != unnest_identifier.name.lower()
 324            ):
 325                unnest_ancestor = column.find_ancestor(exp.Unnest, exp.Select)
 326                if isinstance(unnest_ancestor, exp.Unnest):
 327                    ancestor_identifier = unnest_to_identifier.get(unnest_ancestor)
 328                    if (
 329                        ancestor_identifier
 330                        and ancestor_identifier.name.lower() == unnest_identifier.name.lower()
 331                    ):
 332                        continue
 333
 334                table = unnest_identifier
 335
 336            column.set("table", table and table.copy())
 337
 338    return expression
 339
 340
 341def _eliminate_dot_variant_lookup(expression: exp.Expr) -> exp.Expr:
 342    if isinstance(expression, exp.Select):
 343        # This transformation is used to facilitate transpilation of BigQuery `UNNEST` operations
 344        # to Snowflake. It should not affect roundtrip because `Unnest` nodes cannot be produced
 345        # by Snowflake's parser.
 346        #
 347        # Additionally, at the time of writing this, BigQuery is the only dialect that produces a
 348        # `TableAlias` node that only fills `columns` and not `this`, due to `UNNEST_COLUMN_ONLY`.
 349        unnest_aliases = set()
 350        for unnest in find_all_in_scope(expression, exp.Unnest):
 351            unnest_alias = unnest.args.get("alias")
 352            if (
 353                isinstance(unnest_alias, exp.TableAlias)
 354                and not unnest_alias.this
 355                and len(unnest_alias.columns) == 1
 356            ):
 357                unnest_aliases.add(unnest_alias.columns[0].name)
 358
 359        if unnest_aliases:
 360            for c in find_all_in_scope(expression, exp.Column):
 361                if c.table in unnest_aliases:
 362                    bracket_lhs = c.args["table"]
 363                    bracket_rhs = exp.Literal.string(c.name)
 364                    bracket = exp.Bracket(this=bracket_lhs, expressions=[bracket_rhs])
 365
 366                    if c.parent is expression:
 367                        # Retain column projection names by using aliases
 368                        c.replace(exp.alias_(bracket, c.this.copy()))
 369                    else:
 370                        c.replace(bracket)
 371
 372    return expression
 373
 374
 375class SnowflakeGenerator(generator.Generator):
 376    SELECT_KINDS: tuple[str, ...] = ()
 377    PARAMETER_TOKEN = "$"
 378    MATCHED_BY_SOURCE = False
 379    SINGLE_STRING_INTERVAL = True
 380    JOIN_HINTS = False
 381    TABLE_HINTS = False
 382    QUERY_HINTS = False
 383    SUPPORTS_TABLE_COPY = False
 384    COLLATE_IS_FUNC = True
 385    LIMIT_ONLY_LITERALS = True
 386    JSON_KEY_VALUE_PAIR_SEP = ","
 387    INSERT_OVERWRITE = " OVERWRITE INTO"
 388    STRUCT_DELIMITER = ("(", ")")
 389    COPY_PARAMS_ARE_WRAPPED = False
 390    COPY_PARAMS_EQ_REQUIRED = True
 391    STAR_EXCEPT = "EXCLUDE"
 392    SUPPORTS_EXPLODING_PROJECTIONS = False
 393    ARRAY_CONCAT_IS_VAR_LEN = False
 394    SUPPORTS_CONVERT_TIMEZONE = True
 395    EXCEPT_INTERSECT_SUPPORT_ALL_CLAUSE = False
 396    SUPPORTS_MEDIAN = True
 397    ARRAY_SIZE_NAME = "ARRAY_SIZE"
 398    SUPPORTS_DECODE_CASE = True
 399
 400    AFTER_HAVING_MODIFIER_TRANSFORMS = generator.AFTER_HAVING_MODIFIER_TRANSFORMS
 401
 402    IS_BOOL_ALLOWED = False
 403    DIRECTED_JOINS = True
 404    SUPPORTS_UESCAPE = False
 405    TRY_SUPPORTED = False
 406
 407    TRANSFORMS = {
 408        **generator.Generator.TRANSFORMS,
 409        exp.ApproxDistinct: rename_func("APPROX_COUNT_DISTINCT"),
 410        exp.ArgMax: rename_func("MAX_BY"),
 411        exp.ArgMin: rename_func("MIN_BY"),
 412        exp.Array: transforms.preprocess([transforms.inherit_struct_field_names]),
 413        exp.ArrayConcat: array_concat_sql("ARRAY_CAT"),
 414        exp.ArrayAppend: array_append_sql("ARRAY_APPEND"),
 415        exp.ArrayPrepend: array_append_sql("ARRAY_PREPEND"),
 416        exp.ArrayContains: lambda self, e: self.func(
 417            "ARRAY_CONTAINS",
 418            e.expression
 419            if e.args.get("ensure_variant") is False
 420            else exp.cast(e.expression, exp.DType.VARIANT, copy=False),
 421            e.this,
 422        ),
 423        exp.ArrayPosition: lambda self, e: self.func(
 424            "ARRAY_POSITION",
 425            e.expression,
 426            e.this,
 427        ),
 428        exp.ArrayIntersect: rename_func("ARRAY_INTERSECTION"),
 429        exp.ArrayOverlaps: rename_func("ARRAYS_OVERLAP"),
 430        exp.AtTimeZone: lambda self, e: self.func("CONVERT_TIMEZONE", e.args.get("zone"), e.this),
 431        exp.BitwiseOr: rename_func("BITOR"),
 432        exp.BitwiseXor: rename_func("BITXOR"),
 433        exp.BitwiseAnd: rename_func("BITAND"),
 434        exp.BitwiseAndAgg: rename_func("BITANDAGG"),
 435        exp.BitwiseOrAgg: rename_func("BITORAGG"),
 436        exp.BitwiseXorAgg: rename_func("BITXORAGG"),
 437        exp.BitwiseNot: rename_func("BITNOT"),
 438        exp.BitwiseLeftShift: rename_func("BITSHIFTLEFT"),
 439        exp.BitwiseRightShift: rename_func("BITSHIFTRIGHT"),
 440        exp.Create: transforms.preprocess([_flatten_structured_types_unless_iceberg]),
 441        exp.CurrentTimestamp: lambda self, e: (
 442            self.func("SYSDATE") if e.args.get("sysdate") else self.function_fallback_sql(e)
 443        ),
 444        exp.CurrentSchemas: lambda self, e: self.func("CURRENT_SCHEMAS"),
 445        exp.Localtime: lambda self, e: (
 446            self.func("CURRENT_TIME", e.this) if e.this else "CURRENT_TIME"
 447        ),
 448        exp.Localtimestamp: lambda self, e: (
 449            self.func("CURRENT_TIMESTAMP", e.this) if e.this else "CURRENT_TIMESTAMP"
 450        ),
 451        exp.DateAdd: date_delta_sql("DATEADD"),
 452        exp.DateDiff: date_delta_sql("DATEDIFF"),
 453        exp.DatetimeAdd: date_delta_sql("TIMESTAMPADD"),
 454        exp.DatetimeDiff: timestampdiff_sql,
 455        exp.DateStrToDate: datestrtodate_sql,
 456        exp.Decrypt: lambda self, e: self.func(
 457            f"{'TRY_' if e.args.get('safe') else ''}DECRYPT",
 458            e.this,
 459            e.args.get("passphrase"),
 460            e.args.get("aad"),
 461            e.args.get("encryption_method"),
 462        ),
 463        exp.DecryptRaw: lambda self, e: self.func(
 464            f"{'TRY_' if e.args.get('safe') else ''}DECRYPT_RAW",
 465            e.this,
 466            e.args.get("key"),
 467            e.args.get("iv"),
 468            e.args.get("aad"),
 469            e.args.get("encryption_method"),
 470            e.args.get("aead"),
 471        ),
 472        exp.DayOfMonth: rename_func("DAYOFMONTH"),
 473        exp.DayOfWeek: rename_func("DAYOFWEEK"),
 474        exp.DayOfWeekIso: rename_func("DAYOFWEEKISO"),
 475        exp.DayOfYear: rename_func("DAYOFYEAR"),
 476        exp.DotProduct: rename_func("VECTOR_INNER_PRODUCT"),
 477        exp.Explode: rename_func("FLATTEN"),
 478        exp.Extract: lambda self, e: self.func(
 479            "DATE_PART", map_date_part(e.this, self.dialect), e.expression
 480        ),
 481        exp.CosineDistance: rename_func("VECTOR_COSINE_SIMILARITY"),
 482        exp.EuclideanDistance: rename_func("VECTOR_L2_DISTANCE"),
 483        exp.HandlerProperty: lambda self, e: f"HANDLER = {self.sql(e, 'this')}",
 484        exp.FileFormatProperty: lambda self, e: (
 485            f"FILE_FORMAT=({self.expressions(e, 'expressions', sep=' ')})"
 486        ),
 487        exp.FromTimeZone: lambda self, e: self.func(
 488            "CONVERT_TIMEZONE", e.args.get("zone"), "'UTC'", e.this
 489        ),
 490        exp.GenerateSeries: lambda self, e: self.func(
 491            "ARRAY_GENERATE_RANGE",
 492            e.args["start"],
 493            e.args["end"] if e.args.get("is_end_exclusive") else e.args["end"] + 1,
 494            e.args.get("step"),
 495        ),
 496        exp.GetExtract: rename_func("GET"),
 497        exp.GroupConcat: lambda self, e: groupconcat_sql(self, e, sep=""),
 498        exp.If: if_sql(name="IFF", false_value="NULL"),
 499        exp.JSONArray: lambda self, e: self.func(
 500            "TO_VARIANT", self.func("ARRAY_CONSTRUCT", *e.expressions)
 501        ),
 502        exp.JSONExtractArray: _json_extract_value_array_sql,
 503        exp.JSONExtractScalar: lambda self, e: self.func(
 504            "JSON_EXTRACT_PATH_TEXT", e.this, e.expression
 505        ),
 506        exp.JSONKeys: rename_func("OBJECT_KEYS"),
 507        exp.JSONObject: lambda self, e: self.func("OBJECT_CONSTRUCT_KEEP_NULL", *e.expressions),
 508        exp.JSONPathRoot: lambda *_: "",
 509        exp.JSONValueArray: _json_extract_value_array_sql,
 510        exp.Levenshtein: unsupported_args("ins_cost", "del_cost", "sub_cost")(
 511            rename_func("EDITDISTANCE")
 512        ),
 513        exp.LocationProperty: lambda self, e: f"LOCATION={self.sql(e, 'this')}",
 514        exp.LogicalAnd: rename_func("BOOLAND_AGG"),
 515        exp.LogicalOr: rename_func("BOOLOR_AGG"),
 516        exp.Map: lambda self, e: var_map_sql(self, e, "OBJECT_CONSTRUCT"),
 517        exp.ManhattanDistance: rename_func("VECTOR_L1_DISTANCE"),
 518        exp.MakeInterval: no_make_interval_sql,
 519        exp.Max: max_or_greatest,
 520        exp.Min: min_or_least,
 521        exp.ParseJSON: lambda self, e: self.func(
 522            f"{'TRY_' if e.args.get('safe') else ''}PARSE_JSON", e.this
 523        ),
 524        exp.ToBinary: lambda self, e: self.func(
 525            f"{'TRY_' if e.args.get('safe') else ''}TO_BINARY", e.this, e.args.get("format")
 526        ),
 527        exp.ToBoolean: lambda self, e: self.func(
 528            f"{'TRY_' if e.args.get('safe') else ''}TO_BOOLEAN", e.this
 529        ),
 530        exp.ToDouble: lambda self, e: self.func(
 531            f"{'TRY_' if e.args.get('safe') else ''}TO_DOUBLE", e.this, e.args.get("format")
 532        ),
 533        exp.ToFile: lambda self, e: self.func(
 534            f"{'TRY_' if e.args.get('safe') else ''}TO_FILE", e.this, e.args.get("path")
 535        ),
 536        exp.JSONFormat: rename_func("TO_JSON"),
 537        exp.PartitionedByProperty: lambda self, e: f"PARTITION BY {self.sql(e, 'this')}",
 538        exp.PercentileCont: transforms.preprocess([transforms.add_within_group_for_percentiles]),
 539        exp.PercentileDisc: transforms.preprocess([transforms.add_within_group_for_percentiles]),
 540        exp.Pivot: transforms.preprocess([_unqualify_pivot_columns]),
 541        exp.RegexpExtract: _regexpextract_sql,
 542        exp.RegexpExtractAll: _regexpextract_sql,
 543        exp.RegexpILike: _regexpilike_sql,
 544        exp.RowAccessProperty: lambda self, e: self.rowaccessproperty_sql(e),
 545        exp.Select: transforms.preprocess(
 546            [
 547                transforms.eliminate_window_clause,
 548                transforms.eliminate_distinct_on,
 549                transforms.explode_projection_to_unnest(),
 550                transforms.eliminate_semi_and_anti_joins,
 551                _transform_generate_date_array,
 552                _qualify_unnested_columns,
 553                _eliminate_dot_variant_lookup,
 554            ]
 555        ),
 556        exp.SHA: rename_func("SHA1"),
 557        exp.SHA1Digest: rename_func("SHA1_BINARY"),
 558        exp.MD5Digest: rename_func("MD5_BINARY"),
 559        exp.MD5NumberLower64: rename_func("MD5_NUMBER_LOWER64"),
 560        exp.MD5NumberUpper64: rename_func("MD5_NUMBER_UPPER64"),
 561        exp.Hex: rename_func("HEX_ENCODE"),
 562        exp.LowerHex: rename_func("TO_CHAR"),
 563        exp.Skewness: rename_func("SKEW"),
 564        exp.StarMap: rename_func("OBJECT_CONSTRUCT"),
 565        exp.StartsWith: rename_func("STARTSWITH"),
 566        exp.EndsWith: rename_func("ENDSWITH"),
 567        exp.Rand: lambda self, e: self.func("RANDOM", e.this),
 568        exp.StrPosition: lambda self, e: strposition_sql(
 569            self, e, func_name="CHARINDEX", supports_position=True
 570        ),
 571        exp.StrToDate: lambda self, e: self.func("DATE", e.this, self.format_time(e)),
 572        exp.StringToArray: rename_func("STRTOK_TO_ARRAY"),
 573        exp.StrtokToArray: rename_func("STRTOK_TO_ARRAY"),
 574        exp.Stuff: rename_func("INSERT"),
 575        exp.StPoint: rename_func("ST_MAKEPOINT"),
 576        exp.TimeAdd: date_delta_sql("TIMEADD"),
 577        exp.TimeSlice: lambda self, e: self.func(
 578            "TIME_SLICE",
 579            e.this,
 580            e.expression,
 581            unit_to_str(e),
 582            e.args.get("kind"),
 583        ),
 584        exp.Timestamp: no_timestamp_sql,
 585        exp.TimestampAdd: date_delta_sql("TIMESTAMPADD"),
 586        exp.TimestampDiff: lambda self, e: self.func("TIMESTAMPDIFF", e.unit, e.expression, e.this),
 587        exp.TimestampTrunc: timestamptrunc_sql(),
 588        exp.TimeStrToTime: timestrtotime_sql,
 589        exp.TimeToUnix: lambda self, e: f"EXTRACT(epoch_second FROM {self.sql(e, 'this')})",
 590        exp.ToArray: rename_func("TO_ARRAY"),
 591        exp.ToChar: lambda self, e: self.function_fallback_sql(e),
 592        exp.TsOrDsAdd: date_delta_sql("DATEADD", cast=True),
 593        exp.TsOrDsDiff: date_delta_sql("DATEDIFF"),
 594        exp.TsOrDsToDate: lambda self, e: self.func(
 595            f"{'TRY_' if e.args.get('safe') else ''}TO_DATE", e.this, self.format_time(e)
 596        ),
 597        exp.TsOrDsToTime: lambda self, e: self.func(
 598            f"{'TRY_' if e.args.get('safe') else ''}TO_TIME", e.this, self.format_time(e)
 599        ),
 600        exp.Unhex: rename_func("HEX_DECODE_BINARY"),
 601        exp.UnixToTime: lambda self, e: self.func("TO_TIMESTAMP", e.this, e.args.get("scale")),
 602        exp.Uuid: rename_func("UUID_STRING"),
 603        exp.VarMap: lambda self, e: var_map_sql(self, e, "OBJECT_CONSTRUCT"),
 604        exp.Booland: rename_func("BOOLAND"),
 605        exp.Boolor: rename_func("BOOLOR"),
 606        exp.WeekOfYear: rename_func("WEEKISO"),
 607        exp.YearOfWeek: rename_func("YEAROFWEEK"),
 608        exp.YearOfWeekIso: rename_func("YEAROFWEEKISO"),
 609        exp.Xor: rename_func("BOOLXOR"),
 610        exp.ByteLength: rename_func("OCTET_LENGTH"),
 611        exp.Flatten: rename_func("ARRAY_FLATTEN"),
 612        exp.ArrayConcatAgg: lambda self, e: self.func("ARRAY_FLATTEN", exp.ArrayAgg(this=e.this)),
 613        exp.SHA2Digest: lambda self, e: self.func(
 614            "SHA2_BINARY", e.this, e.args.get("length") or exp.Literal.number(256)
 615        ),
 616    }
 617
 618    def dynamicidentifier_sql(self, expression: exp.DynamicIdentifier) -> str:
 619        this = self.func("IDENTIFIER", expression.this)
 620        if "expressions" in expression.args:
 621            # `IDENTIFIER(...)` invoked as a function, e.g. `IDENTIFIER('my_func')(1, 2)`
 622            return self.func(this, *expression.expressions, normalize=False)
 623        return this
 624
 625    def sortarray_sql(self, expression: exp.SortArray) -> str:
 626        asc = expression.args.get("asc")
 627        nulls_first = expression.args.get("nulls_first")
 628        if asc == exp.false() and nulls_first == exp.true():
 629            nulls_first = None
 630        return self.func("ARRAY_SORT", expression.this, asc, nulls_first)
 631
 632    def nthvalue_sql(self, expression: exp.NthValue) -> str:
 633        result = self.func("NTH_VALUE", expression.this, expression.args.get("offset"))
 634
 635        from_first = expression.args.get("from_first")
 636
 637        if from_first is not None:
 638            if from_first:
 639                result = result + " FROM FIRST"
 640            else:
 641                result = result + " FROM LAST"
 642
 643        return result
 644
 645    SUPPORTED_JSON_PATH_PARTS = {
 646        exp.JSONPathKey,
 647        exp.JSONPathRoot,
 648        exp.JSONPathSubscript,
 649    }
 650
 651    TYPE_MAPPING = {
 652        **generator.Generator.TYPE_MAPPING,
 653        exp.DType.BIGDECIMAL: "DOUBLE",
 654        exp.DType.JSON: "VARIANT",
 655        exp.DType.NESTED: "OBJECT",
 656        exp.DType.STRUCT: "OBJECT",
 657        exp.DType.TEXT: "VARCHAR",
 658    }
 659
 660    TOKEN_MAPPING = {
 661        TokenType.AUTO_INCREMENT: "AUTOINCREMENT",
 662    }
 663
 664    PROPERTIES_LOCATION = {
 665        **generator.Generator.PROPERTIES_LOCATION,
 666        exp.CredentialsProperty: exp.Properties.Location.POST_WITH,
 667        exp.LocationProperty: exp.Properties.Location.POST_WITH,
 668        exp.PartitionedByProperty: exp.Properties.Location.POST_SCHEMA,
 669        exp.RowAccessProperty: exp.Properties.Location.POST_SCHEMA,
 670        exp.SetProperty: exp.Properties.Location.UNSUPPORTED,
 671        exp.VolatileProperty: exp.Properties.Location.UNSUPPORTED,
 672    }
 673
 674    UNSUPPORTED_VALUES_EXPRESSIONS = {
 675        exp.Map,
 676        exp.StarMap,
 677        exp.Struct,
 678        exp.VarMap,
 679    }
 680
 681    RESPECT_IGNORE_NULLS_UNSUPPORTED_EXPRESSIONS = (exp.ArrayAgg,)
 682
 683    def with_properties(self, properties: exp.Properties) -> str:
 684        return self.properties(properties, wrapped=False, prefix=self.sep(""), sep=" ")
 685
 686    def values_sql(self, expression: exp.Values, values_as_table: bool = True) -> str:
 687        if expression.find(*self.UNSUPPORTED_VALUES_EXPRESSIONS):
 688            values_as_table = False
 689
 690        return super().values_sql(expression, values_as_table=values_as_table)
 691
 692    def datatype_sql(self, expression: exp.DataType) -> str:
 693        # Check if this is a FLOAT type nested inside a VECTOR type
 694        # VECTOR only accepts FLOAT (not DOUBLE), INT, and STRING as element types
 695        # https://docs.snowflake.com/en/sql-reference/data-types-vector
 696        if expression.is_type(exp.DType.DOUBLE):
 697            parent = expression.parent
 698            if isinstance(parent, exp.DataType) and parent.is_type(exp.DType.VECTOR):
 699                # Preserve FLOAT for VECTOR types instead of mapping to synonym DOUBLE
 700                return "FLOAT"
 701
 702        expressions = expression.expressions
 703        if expressions and expression.is_type(*exp.DataType.STRUCT_TYPES):
 704            for field_type in expressions:
 705                # The correct syntax is OBJECT [ (<key> <value_type [NOT NULL] [, ...]) ]
 706                if isinstance(field_type, exp.DataType):
 707                    return "OBJECT"
 708                if (
 709                    isinstance(field_type, exp.ColumnDef)
 710                    and field_type.this
 711                    and field_type.this.is_string
 712                ):
 713                    # Doing OBJECT('foo' VARCHAR) is invalid snowflake Syntax. Moreover, besides
 714                    # converting 'foo' into an identifier, we also need to quote it because these
 715                    # keys are case-sensitive. For example:
 716                    #
 717                    # WITH t AS (SELECT OBJECT_CONSTRUCT('x', 'y') AS c) SELECT c:x FROM t -- correct
 718                    # WITH t AS (SELECT OBJECT_CONSTRUCT('x', 'y') AS c) SELECT c:X FROM t -- incorrect, returns NULL
 719                    field_type.this.replace(exp.to_identifier(field_type.name, quoted=True))
 720
 721        return super().datatype_sql(expression)
 722
 723    def tonumber_sql(self, expression: exp.ToNumber) -> str:
 724        precision = expression.args.get("precision")
 725        scale = expression.args.get("scale")
 726
 727        default_precision = isinstance(precision, exp.Literal) and precision.name == "38"
 728        default_scale = isinstance(scale, exp.Literal) and scale.name == "0"
 729
 730        if default_precision and default_scale:
 731            precision = None
 732            scale = None
 733        elif default_scale:
 734            scale = None
 735
 736        func_name = "TRY_TO_NUMBER" if expression.args.get("safe") else "TO_NUMBER"
 737
 738        return self.func(
 739            func_name,
 740            expression.this,
 741            expression.args.get("format"),
 742            precision,
 743            scale,
 744        )
 745
 746    def timestampfromparts_sql(self, expression: exp.TimestampFromParts) -> str:
 747        milli = expression.args.get("milli")
 748        if milli is not None:
 749            milli_to_nano = milli.pop() * exp.Literal.number(1000000)
 750            expression.set("nano", milli_to_nano)
 751
 752        return rename_func("TIMESTAMP_FROM_PARTS")(self, expression)
 753
 754    def cast_sql(self, expression: exp.Cast, safe_prefix: str | None = None) -> str:
 755        if expression.is_type(exp.DType.GEOGRAPHY):
 756            return self.func("TO_GEOGRAPHY", expression.this)
 757        if expression.is_type(exp.DType.GEOMETRY):
 758            return self.func("TO_GEOMETRY", expression.this)
 759
 760        return super().cast_sql(expression, safe_prefix=safe_prefix)
 761
 762    def trycast_sql(self, expression: exp.TryCast) -> str:
 763        value = expression.this
 764
 765        if value.type is None:
 766            from sqlglot.optimizer.annotate_types import annotate_types
 767
 768            value = annotate_types(value, dialect=self.dialect)
 769
 770        # Snowflake requires that TRY_CAST's value be a string
 771        # If TRY_CAST is being roundtripped (since Snowflake is the only dialect that sets "requires_string") or
 772        # if we can deduce that the value is a string, then we can generate TRY_CAST
 773        if expression.args.get("requires_string") or value.is_type(*exp.DataType.TEXT_TYPES):
 774            return super().trycast_sql(expression)
 775
 776        return self.cast_sql(expression)
 777
 778    def log_sql(self, expression: exp.Log) -> str:
 779        if not expression.expression:
 780            return self.func("LN", expression.this)
 781
 782        return super().log_sql(expression)
 783
 784    def greatest_sql(self, expression: exp.Greatest) -> str:
 785        name = "GREATEST_IGNORE_NULLS" if expression.args.get("ignore_nulls") else "GREATEST"
 786        return self.func(name, expression.this, *expression.expressions)
 787
 788    def least_sql(self, expression: exp.Least) -> str:
 789        name = "LEAST_IGNORE_NULLS" if expression.args.get("ignore_nulls") else "LEAST"
 790        return self.func(name, expression.this, *expression.expressions)
 791
 792    def generator_sql(self, expression: exp.Generator) -> str:
 793        args = []
 794        rowcount = expression.args.get("rowcount")
 795        timelimit = expression.args.get("timelimit")
 796
 797        if rowcount:
 798            args.append(exp.Kwarg(this=exp.var("ROWCOUNT"), expression=rowcount))
 799        if timelimit:
 800            args.append(exp.Kwarg(this=exp.var("TIMELIMIT"), expression=timelimit))
 801
 802        return self.func("GENERATOR", *args)
 803
 804    def unnest_sql(self, expression: exp.Unnest) -> str:
 805        unnest_alias = expression.args.get("alias")
 806        offset = expression.args.get("offset")
 807
 808        unnest_alias_columns = unnest_alias.columns if unnest_alias else []
 809        value = seq_get(unnest_alias_columns, 0) or exp.to_identifier("value")
 810
 811        columns = [
 812            exp.to_identifier("seq"),
 813            exp.to_identifier("key"),
 814            exp.to_identifier("path"),
 815            offset.pop() if isinstance(offset, exp.Expr) else exp.to_identifier("index"),
 816            value,
 817            exp.to_identifier("this"),
 818        ]
 819
 820        if unnest_alias:
 821            unnest_alias.set("columns", columns)
 822        else:
 823            unnest_alias = exp.TableAlias(this="_u", columns=columns)
 824
 825        table_input = self.sql(expression.expressions[0])
 826        if not table_input.startswith("INPUT =>"):
 827            table_input = f"INPUT => {table_input}"
 828
 829        expression_parent = expression.parent
 830
 831        explode = (
 832            f"FLATTEN({table_input})"
 833            if isinstance(expression_parent, exp.Lateral)
 834            else f"TABLE(FLATTEN({table_input}))"
 835        )
 836        alias = self.sql(unnest_alias)
 837        alias = f" AS {alias}" if alias else ""
 838        value = (
 839            ""
 840            if isinstance(expression_parent, (exp.From, exp.Join, exp.Lateral))
 841            else f"{value} FROM "
 842        )
 843
 844        return f"{value}{explode}{alias}"
 845
 846    def undrop_sql(self, expression: exp.Undrop) -> str:
 847        this = self.sql(expression, "this")
 848        kind = expression.kind
 849        rename = self.sql(expression, "rename")
 850        rename = f" RENAME TO {rename}" if rename else ""
 851        return f"UNDROP {kind} {this}{rename}"
 852
 853    def show_sql(self, expression: exp.Show) -> str:
 854        terse = "TERSE " if expression.args.get("terse") else ""
 855        iceberg = "ICEBERG " if expression.args.get("iceberg") else ""
 856        history = " HISTORY" if expression.args.get("history") else ""
 857        like = self.sql(expression, "like")
 858        like = f" LIKE {like}" if like else ""
 859
 860        scope = self.sql(expression, "scope")
 861        scope = f" {scope}" if scope else ""
 862
 863        scope_kind = self.sql(expression, "scope_kind")
 864        if scope_kind:
 865            scope_kind = f" IN {scope_kind}"
 866
 867        starts_with = self.sql(expression, "starts_with")
 868        if starts_with:
 869            starts_with = f" STARTS WITH {starts_with}"
 870
 871        limit = self.sql(expression, "limit")
 872
 873        from_ = self.sql(expression, "from_")
 874        if from_:
 875            from_ = f" FROM {from_}"
 876
 877        privileges = self.expressions(expression, key="privileges", flat=True)
 878        privileges = f" WITH PRIVILEGES {privileges}" if privileges else ""
 879
 880        return f"SHOW {terse}{iceberg}{expression.name}{history}{like}{scope_kind}{scope}{starts_with}{limit}{from_}{privileges}"
 881
 882    def rowaccessproperty_sql(self, expression: exp.RowAccessProperty) -> str:
 883        if not expression.this:
 884            return "ROW ACCESS"
 885        on = f" ON ({self.expressions(expression, flat=True)})" if expression.expressions else ""
 886        return f"WITH ROW ACCESS POLICY {self.sql(expression, 'this')}{on}"
 887
 888    def describe_sql(self, expression: exp.Describe) -> str:
 889        kind_value = expression.args.get("kind") or "TABLE"
 890
 891        properties = expression.args.get("properties")
 892        if properties:
 893            qualifier = self.expressions(properties, sep=" ")
 894            kind = f" {qualifier} {kind_value}"
 895        else:
 896            kind = f" {kind_value}"
 897
 898        this = f" {self.sql(expression, 'this')}"
 899        expressions = self.expressions(expression, flat=True)
 900        expressions = f" {expressions}" if expressions else ""
 901        return f"DESCRIBE{kind}{this}{expressions}"
 902
 903    def generatedasidentitycolumnconstraint_sql(
 904        self, expression: exp.GeneratedAsIdentityColumnConstraint
 905    ) -> str:
 906        start = expression.args.get("start")
 907        start = f" START {start}" if start else ""
 908        increment = expression.args.get("increment")
 909        increment = f" INCREMENT {increment}" if increment else ""
 910
 911        order = expression.args.get("order")
 912        if order is not None:
 913            order_clause = " ORDER" if order else " NOORDER"
 914        else:
 915            order_clause = ""
 916
 917        return f"AUTOINCREMENT{start}{increment}{order_clause}"
 918
 919    def struct_sql(self, expression: exp.Struct) -> str:
 920        if len(expression.expressions) == 1:
 921            arg = expression.expressions[0]
 922            if arg.is_star or (isinstance(arg, exp.ILike) and arg.left.is_star):
 923                # Wildcard syntax: https://docs.snowflake.com/en/sql-reference/data-types-semistructured#object
 924                return f"{{{self.sql(expression.expressions[0])}}}"
 925
 926        keys = []
 927        values = []
 928
 929        for i, e in enumerate(expression.expressions):
 930            if isinstance(e, exp.PropertyEQ):
 931                keys.append(
 932                    exp.Literal.string(e.name) if isinstance(e.this, exp.Identifier) else e.this
 933                )
 934                values.append(e.expression)
 935            else:
 936                keys.append(exp.Literal.string(f"_{i}"))
 937                values.append(e)
 938
 939        return self.func("OBJECT_CONSTRUCT", *flatten(zip(keys, values)))
 940
 941    @unsupported_args("weight", "accuracy")
 942    def approxquantile_sql(self, expression: exp.ApproxQuantile) -> str:
 943        return self.func("APPROX_PERCENTILE", expression.this, expression.args.get("quantile"))
 944
 945    def alterset_sql(self, expression: exp.AlterSet) -> str:
 946        exprs = self.expressions(expression, flat=True)
 947        exprs = f" {exprs}" if exprs else ""
 948        file_format = self.expressions(expression, key="file_format", flat=True, sep=" ")
 949        file_format = f" STAGE_FILE_FORMAT = ({file_format})" if file_format else ""
 950        copy_options = self.expressions(expression, key="copy_options", flat=True, sep=" ")
 951        copy_options = f" STAGE_COPY_OPTIONS = ({copy_options})" if copy_options else ""
 952        tag = self.expressions(expression, key="tag", flat=True)
 953        tag = f" TAG {tag}" if tag else ""
 954
 955        return f"SET{exprs}{file_format}{copy_options}{tag}"
 956
 957    def strtotime_sql(self, expression: exp.StrToTime):
 958        # target_type is stored as a DataType instance
 959        target_type = expression.args.get("target_type")
 960
 961        # Get the type enum from DataType instance or from type annotation
 962        if isinstance(target_type, exp.DataType):
 963            type_enum = target_type.this
 964        elif expression.type:
 965            type_enum = expression.type.this
 966        else:
 967            type_enum = exp.DType.TIMESTAMP
 968
 969        func_name = TIMESTAMP_TYPES.get(type_enum, "TO_TIMESTAMP")
 970
 971        return self.func(
 972            f"{'TRY_' if expression.args.get('safe') else ''}{func_name}",
 973            expression.this,
 974            self.format_time(expression),
 975        )
 976
 977    def timestampsub_sql(self, expression: exp.TimestampSub):
 978        return self.sql(
 979            exp.TimestampAdd(
 980                this=expression.this,
 981                expression=expression.expression * -1,
 982                unit=expression.unit,
 983            )
 984        )
 985
 986    def jsonextract_sql(self, expression: exp.JSONExtract):
 987        this = expression.this
 988
 989        # JSON strings are valid coming from other dialects such as BQ so
 990        # for these cases we PARSE_JSON preemptively
 991        if not isinstance(this, (exp.ParseJSON, exp.JSONExtract)) and not expression.args.get(
 992            "requires_json"
 993        ):
 994            this = exp.ParseJSON(this=this)
 995
 996        return self.func(
 997            "GET_PATH",
 998            this,
 999            expression.expression,
1000        )
1001
1002    def timetostr_sql(self, expression: exp.TimeToStr) -> str:
1003        this = expression.this
1004        if this.is_string:
1005            this = exp.cast(this, exp.DType.TIMESTAMP)
1006
1007        return self.func("TO_CHAR", this, self.format_time(expression))
1008
1009    def datesub_sql(self, expression: exp.DateSub) -> str:
1010        value = expression.expression
1011        if value:
1012            value.replace(value * (-1))
1013        else:
1014            self.unsupported("DateSub cannot be transpiled if the subtracted count is unknown")
1015
1016        return date_delta_sql("DATEADD")(self, expression)
1017
1018    def select_sql(self, expression: exp.Select) -> str:
1019        limit = expression.args.get("limit")
1020        offset = expression.args.get("offset")
1021        if offset and not limit:
1022            expression.limit(exp.Null(), copy=False)
1023        return super().select_sql(expression)
1024
1025    def createable_sql(self, expression: exp.Create, locations: defaultdict) -> str:
1026        is_materialized = expression.find(exp.MaterializedProperty)
1027        copy_grants_property = expression.find(exp.CopyGrantsProperty)
1028
1029        if expression.kind == "VIEW" and is_materialized and copy_grants_property:
1030            # For materialized views, COPY GRANTS is located *before* the columns list
1031            # This is in contrast to normal views where COPY GRANTS is located *after* the columns list
1032            # We default CopyGrantsProperty to POST_SCHEMA which means we need to output it POST_NAME if a materialized view is detected
1033            # ref: https://docs.snowflake.com/en/sql-reference/sql/create-materialized-view#syntax
1034            # ref: https://docs.snowflake.com/en/sql-reference/sql/create-view#syntax
1035            post_schema_properties = locations[exp.Properties.Location.POST_SCHEMA]
1036            post_schema_properties.pop(post_schema_properties.index(copy_grants_property))
1037
1038            this_name = self.sql(expression.this, "this")
1039            copy_grants = self.sql(copy_grants_property)
1040            this_schema = self.schema_columns_sql(expression.this)
1041            this_schema = f"{self.sep()}{this_schema}" if this_schema else ""
1042
1043            return f"{this_name}{self.sep()}{copy_grants}{this_schema}"
1044
1045        return super().createable_sql(expression, locations)
1046
1047    def arrayagg_sql(self, expression: exp.ArrayAgg) -> str:
1048        this = expression.this
1049
1050        # If an ORDER BY clause is present, we need to remove it from ARRAY_AGG
1051        # and add it later as part of the WITHIN GROUP clause
1052        order = this if isinstance(this, exp.Order) else None
1053        if order:
1054            expression.set("this", order.this.pop())
1055
1056        expr_sql = super().arrayagg_sql(expression)
1057
1058        if order:
1059            expr_sql = self.sql(exp.WithinGroup(this=expr_sql, expression=order))
1060
1061        return expr_sql
1062
1063    def arraydistinct_sql(self, expression: exp.ArrayDistinct) -> str:
1064        if expression.args.get("check_null"):
1065            return self.func("ARRAY_DISTINCT", expression.this)
1066        return self.func("ARRAY_DISTINCT", exp.ArrayCompact(this=expression.this))
1067
1068    def arraytostring_sql(self, expression: exp.ArrayToString) -> str:
1069        return self.func("ARRAY_TO_STRING", expression.this, expression.expression)
1070
1071    def array_sql(self, expression: exp.Array) -> str:
1072        expressions = expression.expressions
1073
1074        first_expr = seq_get(expressions, 0)
1075        if isinstance(first_expr, exp.Select):
1076            # SELECT AS STRUCT foo AS alias_foo -> ARRAY_AGG(OBJECT_CONSTRUCT('alias_foo', foo))
1077            if first_expr.text("kind").upper() == "STRUCT":
1078                object_construct_args = []
1079                for expr in first_expr.expressions:
1080                    # Alias case: SELECT AS STRUCT foo AS alias_foo -> OBJECT_CONSTRUCT('alias_foo', foo)
1081                    # Column case: SELECT AS STRUCT foo -> OBJECT_CONSTRUCT('foo', foo)
1082                    name = expr.this if isinstance(expr, exp.Alias) else expr
1083
1084                    object_construct_args.extend([exp.Literal.string(expr.alias_or_name), name])
1085
1086                array_agg = exp.ArrayAgg(this=build_object_construct(args=object_construct_args))
1087
1088                first_expr.set("kind", None)
1089                first_expr.set("expressions", [array_agg])
1090
1091                return self.sql(first_expr.subquery())
1092
1093        return inline_array_sql(self, expression)
1094
1095    def currentdate_sql(self, expression: exp.CurrentDate) -> str:
1096        zone = self.sql(expression, "this")
1097        if not zone:
1098            return super().currentdate_sql(expression)
1099
1100        expr = exp.Cast(
1101            this=exp.ConvertTimezone(target_tz=zone, timestamp=exp.CurrentTimestamp()),
1102            to=exp.DataType(this=exp.DType.DATE),
1103        )
1104        return self.sql(expr)
1105
1106    def dot_sql(self, expression: exp.Dot) -> str:
1107        this = expression.this
1108
1109        if not this.type:
1110            from sqlglot.optimizer.annotate_types import annotate_types
1111
1112            this = annotate_types(this, dialect=self.dialect)
1113
1114        if not isinstance(this, exp.Dot) and this.is_type(exp.DType.STRUCT):
1115            # Generate colon notation for the top level STRUCT
1116            return f"{self.sql(this)}:{self.sql(expression, 'expression')}"
1117
1118        return super().dot_sql(expression)
1119
1120    def modelattribute_sql(self, expression: exp.ModelAttribute) -> str:
1121        return f"{self.sql(expression, 'this')}!{self.sql(expression, 'expression')}"
1122
1123    def format_sql(self, expression: exp.Format) -> str:
1124        if expression.name.lower() == "%s" and len(expression.expressions) == 1:
1125            return self.func("TO_CHAR", expression.expressions[0])
1126
1127        return self.function_fallback_sql(expression)
1128
1129    def splitpart_sql(self, expression: exp.SplitPart) -> str:
1130        # Set part_index to 1 if missing
1131        if not expression.args.get("delimiter"):
1132            expression.set("delimiter", exp.Literal.string(" "))
1133
1134        if not expression.args.get("part_index"):
1135            expression.set("part_index", exp.Literal.number(1))
1136
1137        return rename_func("SPLIT_PART")(self, expression)
1138
1139    def uniform_sql(self, expression: exp.Uniform) -> str:
1140        gen = expression.args.get("gen")
1141        seed = expression.args.get("seed")
1142
1143        # From Databricks UNIFORM(min, max, seed) -> Wrap gen in RANDOM(seed)
1144        if seed:
1145            gen = exp.Rand(this=seed)
1146
1147        # No gen argument (from Databricks 2-arg UNIFORM(min, max)) -> Add RANDOM()
1148        if not gen:
1149            gen = exp.Rand()
1150
1151        return self.func("UNIFORM", expression.this, expression.expression, gen)
1152
1153    def window_sql(self, expression: exp.Window) -> str:
1154        spec = expression.args.get("spec")
1155        this = expression.this
1156
1157        if (
1158            (
1159                isinstance(this, RANKING_WINDOW_FUNCTIONS_WITH_FRAME)
1160                or (
1161                    isinstance(this, (exp.RespectNulls, exp.IgnoreNulls))
1162                    and isinstance(this.this, RANKING_WINDOW_FUNCTIONS_WITH_FRAME)
1163                )
1164            )
1165            and spec
1166            and (
1167                spec.text("kind").upper() == "ROWS"
1168                and spec.text("start").upper() == "UNBOUNDED"
1169                and spec.text("start_side").upper() == "PRECEDING"
1170                and spec.text("end").upper() == "UNBOUNDED"
1171                and spec.text("end_side").upper() == "FOLLOWING"
1172            )
1173        ):
1174            # omit the default window from window ranking functions
1175            expression.set("spec", None)
1176        return super().window_sql(expression)
1177
1178    def filter_sql(self, expression: exp.Filter) -> str:
1179        # Snowflake doesn't support FILTER (WHERE cond), so we rewrite it into an
1180        # equivalent conditional aggregation, i.e. wrap the input values in an IFF
1181        agg = expression.this
1182        agg_arg = agg.this
1183        cond = expression.expression.this
1184
1185        if isinstance(agg, exp.WithinGroup):
1186            # Ordered-set aggregates take their input from the ORDER BY key, so the
1187            # condition has to wrap that instead of the aggregate's own argument
1188            if isinstance(agg_arg, (exp.Mode, *exp.PERCENTILES)):
1189                for ordered in agg.expression.expressions:
1190                    key = ordered.this
1191                    key.replace(exp.If(this=cond.copy(), true=key.copy()))
1192
1193                return self.sql(agg)
1194
1195            # Besides the percentile functions, these are the only functions Snowflake
1196            # accepts WITHIN GROUP for, so anything else can't be rewritten correctly
1197            if isinstance(agg_arg, (exp.ArrayAgg, exp.GroupConcat)):
1198                agg_arg = agg_arg.this
1199            else:
1200                self.unsupported("Unable to rewrite FILTER into the aggregate's arguments")
1201                return self.sql(agg)
1202
1203        # `COUNT(*/t.*) FILTER (WHERE cond)` counts qualifying rows, but a star can't be an IFF
1204        # argument: `IFF(cond, *, NULL)` expands to multiple columns once the table has 2+ of
1205        # them, which Snowflake rejects. Use its native COUNT_IF instead.
1206        if isinstance(agg, exp.Count) and agg_arg.is_star:
1207            return self.func("COUNT_IF", cond)
1208
1209        # `DISTINCT` and `ORDER BY` are part of the aggregate's own argument list, so the
1210        # condition has to wrap the values underneath them rather than the whole clause --
1211        # `IFF(cond, DISTINCT x, NULL)` is not a call any dialect accepts.
1212        if isinstance(agg_arg, exp.Order):
1213            agg_arg = agg_arg.this
1214
1215        if isinstance(agg_arg, exp.Distinct):
1216            targets = agg_arg.expressions
1217        else:
1218            targets = [agg_arg]
1219
1220        for target in targets:
1221            target.replace(exp.If(this=cond.copy(), true=target.copy()))
1222
1223        return self.sql(agg)
1224
1225    def withingroup_sql(self, expression: exp.WithinGroup) -> str:
1226        # Snowflake's MODE doesn't support the ordered-set syntax, i.e. it only
1227        # accepts the value to aggregate as an argument: MODE(<expr>)
1228        if isinstance(expression.this, exp.Mode) and not expression.this.this:
1229            order = expression.expression
1230            if isinstance(order, exp.Order) and len(order.expressions) == 1:
1231                return self.sql(exp.Mode(this=order.expressions[0].this))
1232
1233        return super().withingroup_sql(expression)