diff options
Diffstat (limited to 'tests/dialects/test_snowflake.py')
-rw-r--r-- | tests/dialects/test_snowflake.py | 43 |
1 files changed, 43 insertions, 0 deletions
diff --git a/tests/dialects/test_snowflake.py b/tests/dialects/test_snowflake.py index f7bab4d..e20045b 100644 --- a/tests/dialects/test_snowflake.py +++ b/tests/dialects/test_snowflake.py @@ -1,3 +1,5 @@ +from unittest import mock + from sqlglot import UnsupportedError, exp, parse_one from tests.dialects.test_dialect import Validator @@ -309,6 +311,7 @@ class TestSnowflake(Validator): "SELECT IFF(TRUE, 'true', 'false')", write={ "snowflake": "SELECT IFF(TRUE, 'true', 'false')", + "spark": "SELECT IF(TRUE, 'true', 'false')", }, ) self.validate_all( @@ -870,6 +873,46 @@ FROM persons AS p, LATERAL FLATTEN(input => p.c, path => 'contact') AS f, LATERA self.assertIsInstance(ilike, exp.ILikeAny) like.sql() # check that this doesn't raise + @mock.patch("sqlglot.generator.logger") + def test_regexp_substr(self, logger): + self.validate_all( + "REGEXP_SUBSTR(subject, pattern, pos, occ, params, group)", + write={ + "bigquery": "REGEXP_EXTRACT(subject, pattern, pos, occ)", + "hive": "REGEXP_EXTRACT(subject, pattern, group)", + "presto": "REGEXP_EXTRACT(subject, pattern, group)", + "snowflake": "REGEXP_SUBSTR(subject, pattern, pos, occ, params, group)", + "spark": "REGEXP_EXTRACT(subject, pattern, group)", + }, + ) + self.validate_all( + "REGEXP_SUBSTR(subject, pattern)", + read={ + "bigquery": "REGEXP_EXTRACT(subject, pattern)", + "hive": "REGEXP_EXTRACT(subject, pattern)", + "presto": "REGEXP_EXTRACT(subject, pattern)", + "spark": "REGEXP_EXTRACT(subject, pattern)", + }, + write={ + "bigquery": "REGEXP_EXTRACT(subject, pattern)", + "hive": "REGEXP_EXTRACT(subject, pattern)", + "presto": "REGEXP_EXTRACT(subject, pattern)", + "snowflake": "REGEXP_SUBSTR(subject, pattern)", + "spark": "REGEXP_EXTRACT(subject, pattern)", + }, + ) + self.validate_all( + "REGEXP_SUBSTR(subject, pattern, 1, 1, 'c', group)", + read={ + "bigquery": "REGEXP_SUBSTR(subject, pattern, 1, 1, 'c', group)", + "duckdb": "REGEXP_EXTRACT(subject, pattern, group)", + "hive": "REGEXP_EXTRACT(subject, pattern, group)", + "presto": "REGEXP_EXTRACT(subject, pattern, group)", + "snowflake": "REGEXP_SUBSTR(subject, pattern, 1, 1, 'c', group)", + "spark": "REGEXP_EXTRACT(subject, pattern, group)", + }, + ) + def test_match_recognize(self): for row in ( "ONE ROW PER MATCH", |