from sqlglot import UnsupportedError, exp, parse_one from sqlglot.helper import logger as helper_logger from tests.dialects.test_dialect import Validator class TestPresto(Validator): dialect = "presto" def test_cast(self): self.validate_identity("CAST(x AS IPADDRESS)") self.validate_identity("CAST(x AS IPPREFIX)") self.validate_all( "CAST(x AS INTERVAL YEAR TO MONTH)", write={ "oracle": "CAST(x AS INTERVAL YEAR TO MONTH)", "presto": "CAST(x AS INTERVAL YEAR TO MONTH)", }, ) self.validate_all( "CAST(x AS INTERVAL DAY TO SECOND)", write={ "oracle": "CAST(x AS INTERVAL DAY TO SECOND)", "presto": "CAST(x AS INTERVAL DAY TO SECOND)", }, ) self.validate_all( "SELECT CAST('10C' AS INTEGER)", read={ "postgres": "SELECT CAST('10C' AS INTEGER)", "presto": "SELECT CAST('10C' AS INTEGER)", "redshift": "SELECT CAST('10C' AS INTEGER)", }, ) self.validate_all( "SELECT CAST('1970-01-01 00:00:00' AS TIMESTAMP)", read={"postgres": "SELECT 'epoch'::TIMESTAMP"}, ) self.validate_all( "FROM_BASE64(x)", read={ "hive": "UNBASE64(x)", }, write={ "hive": "UNBASE64(x)", "presto": "FROM_BASE64(x)", }, ) self.validate_all( "TO_BASE64(x)", read={ "hive": "BASE64(x)", }, write={ "hive": "BASE64(x)", "presto": "TO_BASE64(x)", }, ) self.validate_all( "CAST(a AS ARRAY(INT))", write={ "bigquery": "CAST(a AS ARRAY)", "duckdb": "CAST(a AS INT[])", "presto": "CAST(a AS ARRAY(INTEGER))", "spark": "CAST(a AS ARRAY)", "snowflake": "CAST(a AS ARRAY)", }, ) self.validate_all( "CAST(a AS VARCHAR)", write={ "bigquery": "CAST(a AS STRING)", "duckdb": "CAST(a AS TEXT)", "presto": "CAST(a AS VARCHAR)", "spark": "CAST(a AS STRING)", }, ) self.validate_all( "CAST(ARRAY[1, 2] AS ARRAY(BIGINT))", write={ "bigquery": "CAST([1, 2] AS ARRAY)", "duckdb": "CAST([1, 2] AS BIGINT[])", "presto": "CAST(ARRAY[1, 2] AS ARRAY(BIGINT))", "spark": "CAST(ARRAY(1, 2) AS ARRAY)", "snowflake": "CAST([1, 2] AS ARRAY)", }, ) self.validate_all( "CAST(MAP(ARRAY[1], ARRAY[1]) AS MAP(INT,INT))", write={ "bigquery": "CAST(MAP([1], [1]) AS MAP)", "duckdb": "CAST(MAP([1], [1]) AS MAP(INT, INT))", "presto": "CAST(MAP(ARRAY[1], ARRAY[1]) AS MAP(INTEGER, INTEGER))", "hive": "CAST(MAP(1, 1) AS MAP)", "spark": "CAST(MAP_FROM_ARRAYS(ARRAY(1), ARRAY(1)) AS MAP)", "snowflake": "CAST(OBJECT_CONSTRUCT(1, 1) AS OBJECT)", }, ) self.validate_all( "CAST(MAP(ARRAY['a','b','c'], ARRAY[ARRAY[1], ARRAY[2], ARRAY[3]]) AS MAP(VARCHAR, ARRAY(INT)))", write={ "bigquery": "CAST(MAP(['a', 'b', 'c'], [[1], [2], [3]]) AS MAP>)", "duckdb": "CAST(MAP(['a', 'b', 'c'], [[1], [2], [3]]) AS MAP(TEXT, INT[]))", "presto": "CAST(MAP(ARRAY['a', 'b', 'c'], ARRAY[ARRAY[1], ARRAY[2], ARRAY[3]]) AS MAP(VARCHAR, ARRAY(INTEGER)))", "hive": "CAST(MAP('a', ARRAY(1), 'b', ARRAY(2), 'c', ARRAY(3)) AS MAP>)", "spark": "CAST(MAP_FROM_ARRAYS(ARRAY('a', 'b', 'c'), ARRAY(ARRAY(1), ARRAY(2), ARRAY(3))) AS MAP>)", "snowflake": "CAST(OBJECT_CONSTRUCT('a', [1], 'b', [2], 'c', [3]) AS OBJECT)", }, ) self.validate_all( "CAST(x AS TIME(5) WITH TIME ZONE)", write={ "duckdb": "CAST(x AS TIMETZ)", "postgres": "CAST(x AS TIMETZ(5))", "presto": "CAST(x AS TIME(5) WITH TIME ZONE)", "redshift": "CAST(x AS TIME(5) WITH TIME ZONE)", }, ) self.validate_all( "CAST(x AS TIMESTAMP(9) WITH TIME ZONE)", write={ "bigquery": "CAST(x AS TIMESTAMP)", "duckdb": "CAST(x AS TIMESTAMPTZ)", "presto": "CAST(x AS TIMESTAMP(9) WITH TIME ZONE)", "hive": "CAST(x AS TIMESTAMP)", "spark": "CAST(x AS TIMESTAMP)", }, ) def test_regex(self): self.validate_all( "REGEXP_REPLACE('abcd', '[ab]')", write={ "presto": "REGEXP_REPLACE('abcd', '[ab]', '')", "spark": "REGEXP_REPLACE('abcd', '[ab]', '')", }, ) self.validate_all( "REGEXP_LIKE(a, 'x')", write={ "duckdb": "REGEXP_MATCHES(a, 'x')", "presto": "REGEXP_LIKE(a, 'x')", "hive": "a RLIKE 'x'", "spark": "a RLIKE 'x'", }, ) self.validate_all( "SPLIT(x, 'a.')", write={ "duckdb": "STR_SPLIT(x, 'a.')", "presto": "SPLIT(x, 'a.')", "hive": "SPLIT(x, CONCAT('\\\\Q', 'a.'))", "spark": "SPLIT(x, CONCAT('\\\\Q', 'a.'))", }, ) self.validate_all( "REGEXP_SPLIT(x, 'a.')", write={ "duckdb": "STR_SPLIT_REGEX(x, 'a.')", "presto": "REGEXP_SPLIT(x, 'a.')", "hive": "SPLIT(x, 'a.')", "spark": "SPLIT(x, 'a.')", }, ) self.validate_all( "CARDINALITY(x)", write={ "duckdb": "ARRAY_LENGTH(x)", "presto": "CARDINALITY(x)", "hive": "SIZE(x)", "spark": "SIZE(x)", }, ) self.validate_all( "ARRAY_JOIN(x, '-', 'a')", write={ "hive": "CONCAT_WS('-', x)", "spark": "ARRAY_JOIN(x, '-', 'a')", }, ) def test_interval_plural_to_singular(self): # Microseconds, weeks and quarters are not supported in Presto/Trino INTERVAL literals unit_to_expected = { "SeCoNds": "SECOND", "minutes": "MINUTE", "hours": "HOUR", "days": "DAY", "months": "MONTH", "years": "YEAR", } for unit, expected in unit_to_expected.items(): self.validate_all( f"SELECT INTERVAL '1' {unit}", write={ "bigquery": f"SELECT INTERVAL '1' {expected}", "presto": f"SELECT INTERVAL '1' {expected}", "trino": f"SELECT INTERVAL '1' {expected}", }, ) def test_time(self): expr = parse_one("TIME(7) WITH TIME ZONE", into=exp.DataType, read="presto") self.assertEqual(expr.this, exp.DataType.Type.TIMETZ) self.validate_identity("FROM_UNIXTIME(a, b)") self.validate_identity("FROM_UNIXTIME(a, b, c)") self.validate_identity("TRIM(a, b)") self.validate_identity("VAR_POP(a)") self.validate_all( "SELECT FROM_UNIXTIME(col) FROM tbl", write={ "presto": "SELECT FROM_UNIXTIME(col) FROM tbl", "spark": "SELECT CAST(FROM_UNIXTIME(col) AS TIMESTAMP) FROM tbl", "trino": "SELECT FROM_UNIXTIME(col) FROM tbl", }, ) self.validate_all( "DATE_FORMAT(x, '%Y-%m-%d %H:%i:%S')", write={ "bigquery": "FORMAT_DATE('%Y-%m-%d %H:%M:%S', x)", "duckdb": "STRFTIME(x, '%Y-%m-%d %H:%M:%S')", "presto": "DATE_FORMAT(x, '%Y-%m-%d %T')", "hive": "DATE_FORMAT(x, 'yyyy-MM-dd HH:mm:ss')", "spark": "DATE_FORMAT(x, 'yyyy-MM-dd HH:mm:ss')", }, ) self.validate_all( "DATE_PARSE(x, '%Y-%m-%d %H:%i:%S')", write={ "duckdb": "STRPTIME(x, '%Y-%m-%d %H:%M:%S')", "presto": "DATE_PARSE(x, '%Y-%m-%d %T')", "hive": "CAST(x AS TIMESTAMP)", "spark": "TO_TIMESTAMP(x, 'yyyy-MM-dd HH:mm:ss')", }, ) self.validate_all( "DATE_PARSE(x, '%Y-%m-%d')", write={ "duckdb": "STRPTIME(x, '%Y-%m-%d')", "presto": "DATE_PARSE(x, '%Y-%m-%d')", "hive": "CAST(x AS TIMESTAMP)", "spark": "TO_TIMESTAMP(x, 'yyyy-MM-dd')", }, ) self.validate_all( "DATE_FORMAT(x, '%T')", write={ "hive": "DATE_FORMAT(x, 'HH:mm:ss')", }, ) self.validate_all( "DATE_PARSE(SUBSTR(x, 1, 10), '%Y-%m-%d')", write={ "duckdb": "STRPTIME(SUBSTR(x, 1, 10), '%Y-%m-%d')", "presto": "DATE_PARSE(SUBSTR(x, 1, 10), '%Y-%m-%d')", "hive": "CAST(SUBSTR(x, 1, 10) AS TIMESTAMP)", "spark": "TO_TIMESTAMP(SUBSTR(x, 1, 10), 'yyyy-MM-dd')", }, ) self.validate_all( "FROM_UNIXTIME(x)", write={ "duckdb": "TO_TIMESTAMP(x)", "presto": "FROM_UNIXTIME(x)", "hive": "FROM_UNIXTIME(x)", "spark": "CAST(FROM_UNIXTIME(x) AS TIMESTAMP)", }, ) self.validate_all( "TO_UNIXTIME(x)", write={ "duckdb": "EPOCH(x)", "presto": "TO_UNIXTIME(x)", "hive": "UNIX_TIMESTAMP(x)", "spark": "UNIX_TIMESTAMP(x)", }, ) self.validate_all( "DATE_ADD('DAY', 1, x)", write={ "duckdb": "x + INTERVAL 1 DAY", "presto": "DATE_ADD('DAY', 1, x)", "hive": "DATE_ADD(x, 1)", "spark": "DATE_ADD(x, 1)", }, ) self.validate_all( "DATE_ADD('DAY', 1 * -1, x)", write={ "presto": "DATE_ADD('DAY', 1 * -1, x)", }, ) self.validate_all( "NOW()", write={ "presto": "CURRENT_TIMESTAMP", "hive": "CURRENT_TIMESTAMP()", }, ) self.validate_all( "SELECT DATE_ADD('DAY', 1, CAST(CURRENT_DATE AS TIMESTAMP))", read={ "redshift": "SELECT DATEADD(DAY, 1, CURRENT_DATE)", }, ) self.validate_all( "DAY_OF_WEEK(timestamp '2012-08-08 01:00:00')", write={ "spark": "DAYOFWEEK(CAST('2012-08-08 01:00:00' AS TIMESTAMP))", "presto": "DAY_OF_WEEK(CAST('2012-08-08 01:00:00' AS TIMESTAMP))", "duckdb": "DAYOFWEEK(CAST('2012-08-08 01:00:00' AS TIMESTAMP))", }, ) self.validate_all( "DAY_OF_MONTH(timestamp '2012-08-08 01:00:00')", write={ "spark": "DAYOFMONTH(CAST('2012-08-08 01:00:00' AS TIMESTAMP))", "presto": "DAY_OF_MONTH(CAST('2012-08-08 01:00:00' AS TIMESTAMP))", "duckdb": "DAYOFMONTH(CAST('2012-08-08 01:00:00' AS TIMESTAMP))", }, ) self.validate_all( "DAY_OF_YEAR(timestamp '2012-08-08 01:00:00')", write={ "spark": "DAYOFYEAR(CAST('2012-08-08 01:00:00' AS TIMESTAMP))", "presto": "DAY_OF_YEAR(CAST('2012-08-08 01:00:00' AS TIMESTAMP))", "duckdb": "DAYOFYEAR(CAST('2012-08-08 01:00:00' AS TIMESTAMP))", }, ) self.validate_all( "WEEK_OF_YEAR(timestamp '2012-08-08 01:00:00')", write={ "spark": "WEEKOFYEAR(CAST('2012-08-08 01:00:00' AS TIMESTAMP))", "presto": "WEEK_OF_YEAR(CAST('2012-08-08 01:00:00' AS TIMESTAMP))", "duckdb": "WEEKOFYEAR(CAST('2012-08-08 01:00:00' AS TIMESTAMP))", }, ) self.validate_all( "SELECT CAST('2012-10-31 00:00' AS TIMESTAMP) AT TIME ZONE 'America/Sao_Paulo'", write={ "spark": "SELECT FROM_UTC_TIMESTAMP(CAST('2012-10-31 00:00' AS TIMESTAMP), 'America/Sao_Paulo')", "presto": "SELECT AT_TIMEZONE(CAST('2012-10-31 00:00' AS TIMESTAMP), 'America/Sao_Paulo')", }, ) self.validate_all( "SELECT AT_TIMEZONE(CAST('2012-10-31 00:00' AS TIMESTAMP), 'America/Sao_Paulo')", read={ "spark": "SELECT FROM_UTC_TIMESTAMP(TIMESTAMP '2012-10-31 00:00', 'America/Sao_Paulo')", }, ) self.validate_all( "CAST(x AS TIMESTAMP)", write={"presto": "CAST(x AS TIMESTAMP)"}, read={"mysql": "CAST(x AS DATETIME)", "clickhouse": "CAST(x AS DATETIME64)"}, ) self.validate_all( "CAST(x AS TIMESTAMP)", read={"mysql": "TIMESTAMP(x)"}, ) self.validate_all( "TIMESTAMP(x, 'America/Los_Angeles')", write={ "duckdb": "CAST(x AS TIMESTAMP) AT TIME ZONE 'America/Los_Angeles'", "presto": "AT_TIMEZONE(CAST(x AS TIMESTAMP), 'America/Los_Angeles')", }, ) # this case isn't really correct, but it's a fall back for mysql's version self.validate_all( "TIMESTAMP(x, '12:00:00')", write={ "duckdb": "TIMESTAMP(x, '12:00:00')", "presto": "TIMESTAMP(x, '12:00:00')", }, ) self.validate_all( "DATE_ADD('DAY', CAST(x AS BIGINT), y)", write={ "presto": "DATE_ADD('DAY', CAST(x AS BIGINT), y)", }, read={ "presto": "DATE_ADD('DAY', x, y)", }, ) self.validate_identity("DATE_ADD('DAY', 1, y)") def test_ddl(self): self.validate_all( "CREATE TABLE test WITH (FORMAT = 'PARQUET') AS SELECT 1", write={ "duckdb": "CREATE TABLE test AS SELECT 1", "presto": "CREATE TABLE test WITH (FORMAT='PARQUET') AS SELECT 1", "hive": "CREATE TABLE test STORED AS PARQUET AS SELECT 1", "spark": "CREATE TABLE test USING PARQUET AS SELECT 1", }, ) self.validate_all( "CREATE TABLE test STORED AS 'PARQUET' AS SELECT 1", write={ "duckdb": "CREATE TABLE test AS SELECT 1", "presto": "CREATE TABLE test WITH (FORMAT='PARQUET') AS SELECT 1", "hive": "CREATE TABLE test STORED AS PARQUET AS SELECT 1", "spark": "CREATE TABLE test USING PARQUET AS SELECT 1", }, ) self.validate_all( "CREATE TABLE test WITH (FORMAT = 'PARQUET', X = '1', Z = '2') AS SELECT 1", write={ "duckdb": "CREATE TABLE test AS SELECT 1", "presto": "CREATE TABLE test WITH (FORMAT='PARQUET', X='1', Z='2') AS SELECT 1", "hive": "CREATE TABLE test STORED AS PARQUET TBLPROPERTIES ('X'='1', 'Z'='2') AS SELECT 1", "spark": "CREATE TABLE test USING PARQUET TBLPROPERTIES ('X'='1', 'Z'='2') AS SELECT 1", }, ) self.validate_all( "CREATE TABLE x (w VARCHAR, y INTEGER, z INTEGER) WITH (PARTITIONED_BY=ARRAY['y', 'z'])", write={ "duckdb": "CREATE TABLE x (w TEXT, y INT, z INT)", "presto": "CREATE TABLE x (w VARCHAR, y INTEGER, z INTEGER) WITH (PARTITIONED_BY=ARRAY['y', 'z'])", "hive": "CREATE TABLE x (w STRING) PARTITIONED BY (y INT, z INT)", "spark": "CREATE TABLE x (w STRING) PARTITIONED BY (y INT, z INT)", }, ) self.validate_all( "CREATE TABLE x WITH (bucket_by = ARRAY['y'], bucket_count = 64) AS SELECT 1 AS y", write={ "duckdb": "CREATE TABLE x AS SELECT 1 AS y", "presto": "CREATE TABLE x WITH (bucket_by=ARRAY['y'], bucket_count=64) AS SELECT 1 AS y", "hive": "CREATE TABLE x TBLPROPERTIES ('bucket_by'=ARRAY('y'), 'bucket_count'=64) AS SELECT 1 AS y", "spark": "CREATE TABLE x TBLPROPERTIES ('bucket_by'=ARRAY('y'), 'bucket_count'=64) AS SELECT 1 AS y", }, ) self.validate_all( "CREATE TABLE db.example_table (col_a ROW(struct_col_a INTEGER, struct_col_b VARCHAR))", write={ "duckdb": "CREATE TABLE db.example_table (col_a STRUCT(struct_col_a INT, struct_col_b TEXT))", "presto": "CREATE TABLE db.example_table (col_a ROW(struct_col_a INTEGER, struct_col_b VARCHAR))", "hive": "CREATE TABLE db.example_table (col_a STRUCT)", "spark": "CREATE TABLE db.example_table (col_a STRUCT)", }, ) self.validate_all( "CREATE TABLE db.example_table (col_a ROW(struct_col_a INTEGER, struct_col_b ROW(nested_col_a VARCHAR, nested_col_b VARCHAR)))", write={ "duckdb": "CREATE TABLE db.example_table (col_a STRUCT(struct_col_a INT, struct_col_b STRUCT(nested_col_a TEXT, nested_col_b TEXT)))", "presto": "CREATE TABLE db.example_table (col_a ROW(struct_col_a INTEGER, struct_col_b ROW(nested_col_a VARCHAR, nested_col_b VARCHAR)))", "hive": "CREATE TABLE db.example_table (col_a STRUCT>)", "spark": "CREATE TABLE db.example_table (col_a STRUCT>)", }, ) self.validate_all( "SELECT fname, lname, age FROM person ORDER BY age DESC NULLS FIRST, fname ASC NULLS LAST, lname", write={ "presto": "SELECT fname, lname, age FROM person ORDER BY age DESC NULLS FIRST, fname ASC, lname", "spark": "SELECT fname, lname, age FROM person ORDER BY age DESC NULLS FIRST, fname ASC NULLS LAST, lname NULLS LAST", }, ) self.validate_all( "CREATE OR REPLACE VIEW x (cola) SELECT 1 as cola", write={ "spark": "CREATE OR REPLACE VIEW x (cola) AS SELECT 1 AS cola", "presto": "CREATE OR REPLACE VIEW x AS SELECT 1 AS cola", }, ) self.validate_all( """CREATE TABLE IF NOT EXISTS x ("cola" INTEGER, "ds" TEXT) COMMENT 'comment' WITH (PARTITIONED BY=("ds"))""", write={ "spark": "CREATE TABLE IF NOT EXISTS x (`cola` INT, `ds` STRING) COMMENT 'comment' PARTITIONED BY (`ds`)", "presto": """CREATE TABLE IF NOT EXISTS x ("cola" INTEGER, "ds" VARCHAR) COMMENT 'comment' WITH (PARTITIONED_BY=ARRAY['ds'])""", }, ) def test_quotes(self): self.validate_all( "''''", write={ "duckdb": "''''", "presto": "''''", "hive": "'\\''", "spark": "'\\''", }, ) self.validate_all( "'x'", write={ "duckdb": "'x'", "presto": "'x'", "hive": "'x'", "spark": "'x'", }, ) self.validate_all( "'''x'''", write={ "duckdb": "'''x'''", "presto": "'''x'''", "hive": "'\\'x\\''", "spark": "'\\'x\\''", }, ) self.validate_all( "'''x'", write={ "duckdb": "'''x'", "presto": "'''x'", "hive": "'\\'x'", "spark": "'\\'x'", }, ) self.validate_all( "x IN ('a', 'a''b')", write={ "duckdb": "x IN ('a', 'a''b')", "presto": "x IN ('a', 'a''b')", "hive": "x IN ('a', 'a\\'b')", "spark": "x IN ('a', 'a\\'b')", }, ) def test_unnest(self): self.validate_all( "SELECT a FROM x CROSS JOIN UNNEST(ARRAY(y)) AS t (a)", write={ "presto": "SELECT a FROM x CROSS JOIN UNNEST(ARRAY[y]) AS t(a)", "hive": "SELECT a FROM x LATERAL VIEW EXPLODE(ARRAY(y)) t AS a", "spark": "SELECT a FROM x LATERAL VIEW EXPLODE(ARRAY(y)) t AS a", }, ) self.validate_all( "SELECT a FROM x CROSS JOIN UNNEST(ARRAY(y)) AS t (a) CROSS JOIN b", write={ "presto": "SELECT a FROM x CROSS JOIN UNNEST(ARRAY[y]) AS t(a) CROSS JOIN b", "hive": "SELECT a FROM x CROSS JOIN b LATERAL VIEW EXPLODE(ARRAY(y)) t AS a", }, ) def test_unicode_string(self): for prefix in ("u&", "U&"): self.validate_all( f"{prefix}'Hello winter \\2603 !'", write={ "presto": "U&'Hello winter \\2603 !'", "snowflake": "'Hello winter \\u2603 !'", "spark": "'Hello winter \\u2603 !'", }, ) self.validate_all( f"{prefix}'Hello winter #2603 !' UESCAPE '#'", write={ "presto": "U&'Hello winter #2603 !' UESCAPE '#'", "snowflake": "'Hello winter \\u2603 !'", "spark": "'Hello winter \\u2603 !'", }, ) def test_presto(self): with self.assertLogs(helper_logger): self.validate_all( "SELECT COALESCE(ELEMENT_AT(MAP_FROM_ENTRIES(ARRAY[(51, '1')]), id), quantity) FROM my_table", write={ "postgres": UnsupportedError, "presto": "SELECT COALESCE(ELEMENT_AT(MAP_FROM_ENTRIES(ARRAY[(51, '1')]), id), quantity) FROM my_table", }, ) self.validate_all( "SELECT ELEMENT_AT(ARRAY[1, 2, 3], 4)", write={ "": "SELECT ARRAY(1, 2, 3)[3]", "bigquery": "SELECT [1, 2, 3][SAFE_ORDINAL(4)]", "postgres": "SELECT (ARRAY[1, 2, 3])[4]", "presto": "SELECT ELEMENT_AT(ARRAY[1, 2, 3], 4)", }, ) self.validate_identity("SELECT a FROM test TABLESAMPLE BERNOULLI (50)") self.validate_identity("SELECT a FROM test TABLESAMPLE SYSTEM (75)") self.validate_identity("string_agg(x, ',')", "ARRAY_JOIN(ARRAY_AGG(x), ',')") self.validate_identity("SELECT * FROM x OFFSET 1 LIMIT 1") self.validate_identity("SELECT * FROM x OFFSET 1 FETCH FIRST 1 ROWS ONLY") self.validate_identity("SELECT BOOL_OR(a > 10) FROM asd AS T(a)") self.validate_identity("SELECT * FROM (VALUES (1))") self.validate_identity("START TRANSACTION READ WRITE, ISOLATION LEVEL SERIALIZABLE") self.validate_identity("START TRANSACTION ISOLATION LEVEL REPEATABLE READ") self.validate_identity("APPROX_PERCENTILE(a, b, c, d)") self.validate_identity( "SELECT SPLIT_TO_MAP('a:1;b:2;a:3', ';', ':', (k, v1, v2) -> CONCAT(v1, v2))" ) self.validate_identity( "SELECT * FROM example.testdb.customer_orders FOR VERSION AS OF 8954597067493422955" ) self.validate_identity( "SELECT * FROM example.testdb.customer_orders FOR TIMESTAMP AS OF CAST('2022-03-23 09:59:29.803 Europe/Vienna' AS TIMESTAMP)" ) self.validate_all( "SELECT LAST_DAY_OF_MONTH(CAST('2008-11-25' AS DATE))", read={ "duckdb": "SELECT LAST_DAY(CAST('2008-11-25' AS DATE))", }, write={ "duckdb": "SELECT LAST_DAY(CAST('2008-11-25' AS DATE))", "presto": "SELECT LAST_DAY_OF_MONTH(CAST('2008-11-25' AS DATE))", }, ) self.validate_all( "SELECT MAX_BY(a.id, a.timestamp) FROM a", read={ "bigquery": "SELECT MAX_BY(a.id, a.timestamp) FROM a", "clickhouse": "SELECT argMax(a.id, a.timestamp) FROM a", "duckdb": "SELECT MAX_BY(a.id, a.timestamp) FROM a", "snowflake": "SELECT MAX_BY(a.id, a.timestamp) FROM a", "spark": "SELECT MAX_BY(a.id, a.timestamp) FROM a", "teradata": "SELECT MAX_BY(a.id, a.timestamp) FROM a", }, write={ "bigquery": "SELECT MAX_BY(a.id, a.timestamp) FROM a", "clickhouse": "SELECT argMax(a.id, a.timestamp) FROM a", "duckdb": "SELECT ARG_MAX(a.id, a.timestamp) FROM a", "presto": "SELECT MAX_BY(a.id, a.timestamp) FROM a", "snowflake": "SELECT MAX_BY(a.id, a.timestamp) FROM a", "spark": "SELECT MAX_BY(a.id, a.timestamp) FROM a", "teradata": "SELECT MAX_BY(a.id, a.timestamp) FROM a", }, ) self.validate_all( "SELECT MIN_BY(a.id, a.timestamp, 3) FROM a", write={ "clickhouse": "SELECT argMin(a.id, a.timestamp) FROM a", "duckdb": "SELECT ARG_MIN(a.id, a.timestamp) FROM a", "presto": "SELECT MIN_BY(a.id, a.timestamp, 3) FROM a", "snowflake": "SELECT MIN_BY(a.id, a.timestamp, 3) FROM a", "spark": "SELECT MIN_BY(a.id, a.timestamp) FROM a", "teradata": "SELECT MIN_BY(a.id, a.timestamp, 3) FROM a", }, ) self.validate_all( """JSON '"foo"'""", write={ "bigquery": """PARSE_JSON('"foo"')""", "presto": """JSON_PARSE('"foo"')""", "snowflake": """PARSE_JSON('"foo"')""", }, ) self.validate_all( "SELECT ROW(1, 2)", read={ "spark": "SELECT STRUCT(1, 2)", }, write={ "presto": "SELECT ROW(1, 2)", "spark": "SELECT STRUCT(1, 2)", }, ) self.validate_all( "ARBITRARY(x)", read={ "bigquery": "ANY_VALUE(x)", "clickhouse": "any(x)", "databricks": "ANY_VALUE(x)", "doris": "ANY_VALUE(x)", "drill": "ANY_VALUE(x)", "duckdb": "ANY_VALUE(x)", "hive": "FIRST(x)", "mysql": "ANY_VALUE(x)", "oracle": "ANY_VALUE(x)", "redshift": "ANY_VALUE(x)", "snowflake": "ANY_VALUE(x)", "spark": "ANY_VALUE(x)", "spark2": "FIRST(x)", }, write={ "bigquery": "ANY_VALUE(x)", "clickhouse": "any(x)", "databricks": "ANY_VALUE(x)", "doris": "ANY_VALUE(x)", "drill": "ANY_VALUE(x)", "duckdb": "ANY_VALUE(x)", "hive": "FIRST(x)", "mysql": "ANY_VALUE(x)", "oracle": "ANY_VALUE(x)", "postgres": "MAX(x)", "presto": "ARBITRARY(x)", "redshift": "ANY_VALUE(x)", "snowflake": "ANY_VALUE(x)", "spark": "ANY_VALUE(x)", "spark2": "FIRST(x)", "sqlite": "MAX(x)", "tsql": "MAX(x)", }, ) self.validate_all( "STARTS_WITH('abc', 'a')", read={"spark": "STARTSWITH('abc', 'a')"}, write={ "presto": "STARTS_WITH('abc', 'a')", "snowflake": "STARTSWITH('abc', 'a')", "spark": "STARTSWITH('abc', 'a')", }, ) self.validate_all( "IS_NAN(x)", read={ "spark": "ISNAN(x)", }, write={ "presto": "IS_NAN(x)", "spark": "ISNAN(x)", "spark2": "ISNAN(x)", }, ) self.validate_all("VALUES 1, 2, 3", write={"presto": "VALUES (1), (2), (3)"}) self.validate_all("INTERVAL '1 day'", write={"trino": "INTERVAL '1' DAY"}) self.validate_all("(5 * INTERVAL '7' DAY)", read={"": "INTERVAL '5' WEEK"}) self.validate_all("(5 * INTERVAL '7' DAY)", read={"": "INTERVAL '5' WEEKS"}) self.validate_all( "SELECT SUBSTRING(a, 1, 3), SUBSTRING(a, LENGTH(a) - (3 - 1))", read={ "redshift": "SELECT LEFT(a, 3), RIGHT(a, 3)", }, ) self.validate_all( "WITH RECURSIVE t(n) AS (SELECT 1 AS n UNION ALL SELECT n + 1 AS n FROM t WHERE n < 4) SELECT SUM(n) FROM t", read={ "postgres": "WITH RECURSIVE t AS (SELECT 1 AS n UNION ALL SELECT n + 1 AS n FROM t WHERE n < 4) SELECT SUM(n) FROM t", }, ) self.validate_all( "WITH RECURSIVE t(n, k) AS (SELECT 1 AS n, 2 AS k) SELECT SUM(n) FROM t", read={ "postgres": "WITH RECURSIVE t AS (SELECT 1 AS n, 2 as k) SELECT SUM(n) FROM t", }, ) self.validate_all( "WITH RECURSIVE t1(n) AS (SELECT 1 AS n), t2(n) AS (SELECT 2 AS n) SELECT SUM(t1.n), SUM(t2.n) FROM t1, t2", read={ "postgres": "WITH RECURSIVE t1 AS (SELECT 1 AS n), t2 AS (SELECT 2 AS n) SELECT SUM(t1.n), SUM(t2.n) FROM t1, t2", }, ) self.validate_all( "WITH RECURSIVE t(n, _c_0) AS (SELECT 1 AS n, (1 + 2)) SELECT * FROM t", read={ "postgres": "WITH RECURSIVE t AS (SELECT 1 AS n, (1 + 2)) SELECT * FROM t", }, ) self.validate_all( 'WITH RECURSIVE t(n, "1") AS (SELECT n, 1 FROM tbl) SELECT * FROM t', read={ "postgres": "WITH RECURSIVE t AS (SELECT n, 1 FROM tbl) SELECT * FROM t", }, ) self.validate_all( "SELECT JSON_OBJECT(KEY 'key1' VALUE 1, KEY 'key2' VALUE TRUE)", write={ "presto": "SELECT JSON_OBJECT('key1': 1, 'key2': TRUE)", }, ) self.validate_all( "ARRAY_AGG(x ORDER BY y DESC)", write={ "hive": "COLLECT_LIST(x)", "presto": "ARRAY_AGG(x ORDER BY y DESC)", "spark": "COLLECT_LIST(x)", "trino": "ARRAY_AGG(x ORDER BY y DESC)", }, ) self.validate_all( "SELECT a FROM t GROUP BY a, ROLLUP(b), ROLLUP(c), ROLLUP(d)", write={ "presto": "SELECT a FROM t GROUP BY a, ROLLUP (b, c, d)", }, ) self.validate_all( 'SELECT a."b" FROM "foo"', write={ "duckdb": 'SELECT a."b" FROM "foo"', "presto": 'SELECT a."b" FROM "foo"', "spark": "SELECT a.`b` FROM `foo`", }, ) self.validate_all( "SELECT ARRAY[1, 2]", write={ "bigquery": "SELECT [1, 2]", "duckdb": "SELECT [1, 2]", "presto": "SELECT ARRAY[1, 2]", "spark": "SELECT ARRAY(1, 2)", }, ) self.validate_all( "SELECT APPROX_DISTINCT(a) FROM foo", write={ "duckdb": "SELECT APPROX_COUNT_DISTINCT(a) FROM foo", "presto": "SELECT APPROX_DISTINCT(a) FROM foo", "hive": "SELECT APPROX_COUNT_DISTINCT(a) FROM foo", "spark": "SELECT APPROX_COUNT_DISTINCT(a) FROM foo", }, ) self.validate_all( "SELECT APPROX_DISTINCT(a, 0.1) FROM foo", write={ "duckdb": "SELECT APPROX_COUNT_DISTINCT(a) FROM foo", "presto": "SELECT APPROX_DISTINCT(a, 0.1) FROM foo", "hive": "SELECT APPROX_COUNT_DISTINCT(a) FROM foo", "spark": "SELECT APPROX_COUNT_DISTINCT(a, 0.1) FROM foo", }, ) self.validate_all( "SELECT APPROX_DISTINCT(a, 0.1) FROM foo", write={ "presto": "SELECT APPROX_DISTINCT(a, 0.1) FROM foo", "hive": UnsupportedError, "spark": "SELECT APPROX_COUNT_DISTINCT(a, 0.1) FROM foo", }, ) self.validate_all( "SELECT JSON_EXTRACT(x, '$.name')", write={ "presto": "SELECT JSON_EXTRACT(x, '$.name')", "hive": "SELECT GET_JSON_OBJECT(x, '$.name')", "spark": "SELECT GET_JSON_OBJECT(x, '$.name')", }, ) self.validate_all( "SELECT JSON_EXTRACT_SCALAR(x, '$.name')", write={ "presto": "SELECT JSON_EXTRACT_SCALAR(x, '$.name')", "hive": "SELECT GET_JSON_OBJECT(x, '$.name')", "spark": "SELECT GET_JSON_OBJECT(x, '$.name')", }, ) self.validate_all( "'\u6bdb'", write={ "presto": "'\u6bdb'", "hive": "'\u6bdb'", "spark": "'\u6bdb'", }, ) self.validate_all( "SELECT ARRAY_SORT(x, (left, right) -> -1)", write={ "duckdb": "SELECT ARRAY_SORT(x)", "presto": "SELECT ARRAY_SORT(x, (left, right) -> -1)", "hive": "SELECT SORT_ARRAY(x)", "spark": "SELECT ARRAY_SORT(x, (left, right) -> -1)", }, ) self.validate_all( "SELECT ARRAY_SORT(x)", write={ "presto": "SELECT ARRAY_SORT(x)", "hive": "SELECT SORT_ARRAY(x)", "spark": "SELECT ARRAY_SORT(x)", }, ) self.validate_all( "SELECT ARRAY_SORT(x, (left, right) -> -1)", write={ "hive": UnsupportedError, }, ) self.validate_all( "MAP(a, b)", write={ "hive": UnsupportedError, "spark": "MAP_FROM_ARRAYS(a, b)", "snowflake": UnsupportedError, }, ) self.validate_all( "MAP(ARRAY(a, b), ARRAY(c, d))", write={ "hive": "MAP(a, c, b, d)", "presto": "MAP(ARRAY[a, b], ARRAY[c, d])", "spark": "MAP_FROM_ARRAYS(ARRAY(a, b), ARRAY(c, d))", "snowflake": "OBJECT_CONSTRUCT(a, c, b, d)", }, ) self.validate_all( "MAP(ARRAY('a'), ARRAY('b'))", write={ "hive": "MAP('a', 'b')", "presto": "MAP(ARRAY['a'], ARRAY['b'])", "spark": "MAP_FROM_ARRAYS(ARRAY('a'), ARRAY('b'))", "snowflake": "OBJECT_CONSTRUCT('a', 'b')", }, ) self.validate_all( "SELECT * FROM UNNEST(ARRAY['7', '14']) AS x", write={ "bigquery": "SELECT * FROM UNNEST(['7', '14'])", "presto": "SELECT * FROM UNNEST(ARRAY['7', '14']) AS x", "hive": "SELECT * FROM UNNEST(ARRAY('7', '14')) AS x", "spark": "SELECT * FROM UNNEST(ARRAY('7', '14')) AS x", }, ) self.validate_all( "SELECT * FROM UNNEST(ARRAY['7', '14']) AS x(y)", write={ "bigquery": "SELECT * FROM UNNEST(['7', '14']) AS y", "presto": "SELECT * FROM UNNEST(ARRAY['7', '14']) AS x(y)", "hive": "SELECT * FROM UNNEST(ARRAY('7', '14')) AS x(y)", "spark": "SELECT * FROM UNNEST(ARRAY('7', '14')) AS x(y)", }, ) self.validate_all( "WITH RECURSIVE t(n) AS (VALUES (1) UNION ALL SELECT n+1 FROM t WHERE n < 100 ) SELECT sum(n) FROM t", write={ "presto": "WITH RECURSIVE t(n) AS (VALUES (1) UNION ALL SELECT n + 1 FROM t WHERE n < 100) SELECT SUM(n) FROM t", "spark": UnsupportedError, }, ) self.validate_all( "SELECT a, b, c, d, sum(y) FROM z GROUP BY CUBE(a) ROLLUP(a), GROUPING SETS((b, c)), d", write={ "presto": "SELECT a, b, c, d, SUM(y) FROM z GROUP BY d, GROUPING SETS ((b, c)), CUBE (a), ROLLUP (a)", "hive": "SELECT a, b, c, d, SUM(y) FROM z GROUP BY d, GROUPING SETS ((b, c)), CUBE (a), ROLLUP (a)", }, ) self.validate_all( "JSON_FORMAT(CAST(MAP_FROM_ENTRIES(ARRAY[('action_type', 'at')]) AS JSON))", write={ "presto": "JSON_FORMAT(CAST(MAP_FROM_ENTRIES(ARRAY[('action_type', 'at')]) AS JSON))", "spark": "TO_JSON(MAP_FROM_ENTRIES(ARRAY(('action_type', 'at'))))", }, ) self.validate_all( "JSON_FORMAT(x)", read={ "spark": "TO_JSON(x)", }, write={ "bigquery": "TO_JSON_STRING(x)", "duckdb": "CAST(TO_JSON(x) AS TEXT)", "presto": "JSON_FORMAT(x)", "spark": "TO_JSON(x)", }, ) self.validate_all( """JSON_FORMAT(JSON '"x"')""", write={ "bigquery": """TO_JSON_STRING(PARSE_JSON('"x"'))""", "duckdb": """CAST(TO_JSON(JSON('"x"')) AS TEXT)""", "presto": """JSON_FORMAT(JSON_PARSE('"x"'))""", "spark": """REGEXP_EXTRACT(TO_JSON(FROM_JSON('["x"]', SCHEMA_OF_JSON('["x"]'))), '^.(.*).$', 1)""", }, ) self.validate_all( """SELECT JSON_FORMAT(JSON '{"a": 1, "b": "c"}')""", write={ "spark": """SELECT REGEXP_EXTRACT(TO_JSON(FROM_JSON('[{"a": 1, "b": "c"}]', SCHEMA_OF_JSON('[{"a": 1, "b": "c"}]'))), '^.(.*).$', 1)""", }, ) self.validate_all( """SELECT JSON_FORMAT(JSON '[1, 2, 3]')""", write={ "spark": "SELECT REGEXP_EXTRACT(TO_JSON(FROM_JSON('[[1, 2, 3]]', SCHEMA_OF_JSON('[[1, 2, 3]]'))), '^.(.*).$', 1)", }, ) def test_encode_decode(self): self.validate_identity("FROM_UTF8(x, y)") self.validate_all( "TO_UTF8(x)", read={ "duckdb": "ENCODE(x)", "spark": "ENCODE(x, 'utf-8')", }, write={ "duckdb": "ENCODE(x)", "presto": "TO_UTF8(x)", "spark": "ENCODE(x, 'utf-8')", }, ) self.validate_all( "FROM_UTF8(x)", read={ "duckdb": "DECODE(x)", "spark": "DECODE(x, 'utf-8')", }, write={ "duckdb": "DECODE(x)", "presto": "FROM_UTF8(x)", "spark": "DECODE(x, 'utf-8')", }, ) self.validate_all( "ENCODE(x, 'invalid')", write={ "presto": UnsupportedError, "duckdb": UnsupportedError, }, ) self.validate_all( "DECODE(x, 'invalid')", write={ "presto": UnsupportedError, "duckdb": UnsupportedError, }, ) def test_hex_unhex(self): self.validate_all( "TO_HEX(x)", write={ "spark": "HEX(x)", }, ) self.validate_all( "FROM_HEX(x)", write={ "spark": "UNHEX(x)", }, ) self.validate_all( "HEX(x)", write={ "presto": "TO_HEX(x)", }, ) self.validate_all( "UNHEX(x)", write={ "presto": "FROM_HEX(x)", }, ) def test_json(self): self.validate_all( "SELECT CAST(JSON '[1,23,456]' AS ARRAY(INTEGER))", write={ "spark": "SELECT FROM_JSON('[1,23,456]', 'ARRAY')", "presto": "SELECT CAST(JSON_PARSE('[1,23,456]') AS ARRAY(INTEGER))", }, ) self.validate_all( """SELECT CAST(JSON '{"k1":1,"k2":23,"k3":456}' AS MAP(VARCHAR, INTEGER))""", write={ "spark": 'SELECT FROM_JSON(\'{"k1":1,"k2":23,"k3":456}\', \'MAP\')', "presto": 'SELECT CAST(JSON_PARSE(\'{"k1":1,"k2":23,"k3":456}\') AS MAP(VARCHAR, INTEGER))', }, ) self.validate_all( "SELECT CAST(ARRAY [1, 23, 456] AS JSON)", write={ "spark": "SELECT TO_JSON(ARRAY(1, 23, 456))", "presto": "SELECT CAST(ARRAY[1, 23, 456] AS JSON)", }, ) def test_match_recognize(self): self.validate_identity( """SELECT * FROM orders MATCH_RECOGNIZE ( PARTITION BY custkey ORDER BY orderdate MEASURES A.totalprice AS starting_price, LAST(B.totalprice) AS bottom_price, LAST(C.totalprice) AS top_price ONE ROW PER MATCH AFTER MATCH SKIP PAST LAST ROW PATTERN (A B+ C+ D+) DEFINE B AS totalprice < PREV(totalprice), C AS totalprice > PREV(totalprice) AND totalprice <= A.totalprice, D AS totalprice > PREV(totalprice) )""", pretty=True, ) def test_to_char(self): self.validate_all( "TO_CHAR(ts, 'dd')", write={ "bigquery": "FORMAT_DATE('%d', ts)", "presto": "DATE_FORMAT(ts, '%d')", }, ) self.validate_all( "TO_CHAR(ts, 'hh')", write={ "bigquery": "FORMAT_DATE('%H', ts)", "presto": "DATE_FORMAT(ts, '%H')", }, ) self.validate_all( "TO_CHAR(ts, 'hh24')", write={ "bigquery": "FORMAT_DATE('%H', ts)", "presto": "DATE_FORMAT(ts, '%H')", }, ) self.validate_all( "TO_CHAR(ts, 'mi')", write={ "bigquery": "FORMAT_DATE('%M', ts)", "presto": "DATE_FORMAT(ts, '%i')", }, ) self.validate_all( "TO_CHAR(ts, 'mm')", write={ "bigquery": "FORMAT_DATE('%m', ts)", "presto": "DATE_FORMAT(ts, '%m')", }, ) self.validate_all( "TO_CHAR(ts, 'ss')", write={ "bigquery": "FORMAT_DATE('%S', ts)", "presto": "DATE_FORMAT(ts, '%s')", }, ) self.validate_all( "TO_CHAR(ts, 'yyyy')", write={ "bigquery": "FORMAT_DATE('%Y', ts)", "presto": "DATE_FORMAT(ts, '%Y')", }, ) self.validate_all( "TO_CHAR(ts, 'yy')", write={ "bigquery": "FORMAT_DATE('%y', ts)", "presto": "DATE_FORMAT(ts, '%y')", }, )