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