Regression in 25.1.0 when the write dialect is postgres and the read dialect is omitted

[ronnix]

Hi, we ran into the following change of behaviour in 25.1.0.

>>> from sqlglot import parse_one

>>> parse_one("SELECT unnest(ARRAY[1, 2, 3])").sql()
'SELECT UNNEST(1, 2, 3)'

>>> parse_one("SELECT unnest(ARRAY[1, 2, 3])").sql(dialect="postgres")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/ronan/.virtualenvs/insitu/lib/python3.9/site-packages/sqlglot/expressions.py", line 597, in sql
    return Dialect.get_or_raise(dialect).generate(self, **opts)
  File "/home/ronan/.virtualenvs/insitu/lib/python3.9/site-packages/sqlglot/dialects/dialect.py", line 538, in generate
    return self.generator(**opts).generate(expression, copy=copy)
  File "/home/ronan/.virtualenvs/insitu/lib/python3.9/site-packages/sqlglot/generator.py", line 623, in generate
    sql = self.sql(expression).strip()
  File "/home/ronan/.virtualenvs/insitu/lib/python3.9/site-packages/sqlglot/generator.py", line 777, in sql
    sql = transform(self, expression)
  File "/home/ronan/.virtualenvs/insitu/lib/python3.9/site-packages/sqlglot/transforms.py", line 650, in _to_sql
    return _sql_handler(expression)
  File "/home/ronan/.virtualenvs/insitu/lib/python3.9/site-packages/sqlglot/generator.py", line 2302, in select_sql
    expressions = self.expressions(expression)
  File "/home/ronan/.virtualenvs/insitu/lib/python3.9/site-packages/sqlglot/generator.py", line 3371, in expressions
    sql = self.sql(e, comment=False)
  File "/home/ronan/.virtualenvs/insitu/lib/python3.9/site-packages/sqlglot/generator.py", line 782, in sql
    sql = getattr(self, exp_handler_name)(expression)
  File "/home/ronan/.virtualenvs/insitu/lib/python3.9/site-packages/sqlglot/dialects/postgres.py", line 584, in unnest_sql
    if len(expression.expressions) == 1:
TypeError: object of type 'Array' has no len()

>>> parse_one("SELECT unnest(ARRAY[1, 2, 3])", dialect="postgres").sql(dialect="postgres")
'SELECT UNNEST(ARRAY[1, 2, 3])'

I realized that this only occurs when the read dialect is not specified, so the workaround for us was to make sure that we added the dialect parameter to parse_one(), but I thought I would file the issue in case it would be useful.

[georgesittas]

Thanks for the report, this does look like a bug.

so the workaround for us was to make sure that we added the dialect parameter to parse_one()

I'd suggest this anyway because it's more stable vs using the base dialect.

@tobymao it's because Unnest now inherits from Func, so when it's used in places other than FROM, JOIN etc we now parse it in _parse_function_call as exp.Unnest, vs exp.Anonymous which was the behavior before 6a607d3.

>>> import sqlglot
>>> sqlglot.parse_one("SELECT unnest(ARRAY[1, 2, 3])")
Select(
  expressions=[
    Unnest(
      expressions=Array(
        expressions=[
          Literal(this=1, is_string=False),
          Literal(this=2, is_string=False),
          Literal(this=3, is_string=False)]))])
>>> sqlglot.parse_one("SELECT unnest(ARRAY[1, 2, 3])", "postgres")
Select(
  expressions=[
    Explode(
      this=Array(
        expressions=[
          Literal(this=1, is_string=False),
          Literal(this=2, is_string=False),
          Literal(this=3, is_string=False)]))])

Parsing / generating with Postgres works because it produces an Explode node, not an Unnest.