Support select * for postgres parser

[hokaccha]

This PR resolves the error in #397. However, there is a problem as follows.

# schema.sql
CREATE TABLE public.test_table (
  "col1" numeric,
  "col2" numeric
);
CREATE VIEW public.test_view AS SELECT t.* FROM test_table t;

$ go run ./cmd/psqldef -h localhost -U postgres sandbox < schema.sql
-- Apply --
CREATE TABLE public.test_table (
  "col1" numeric,
  "col2" numeric
);
CREATE VIEW public.test_view AS SELECT t.* FROM test_table t;

# Run the same command again
$ go run ./cmd/psqldef -h localhost -U postgres sandbox < schema.sql
-- Apply --
CREATE OR REPLACE VIEW "public"."test_view" AS select t.* from test_table as t;

No matter how many times psqldef is executed, CREATE OR REPLACE VIEW is executed. It seems that the cause is that when select * is specified, the query is saved as a query with expanded columns.

$ psql -h localhost -U postgres sandbox -c "SELECT definition FROM pg_views WHERE viewname = 'test_view'"
      definition
-----------------------
  SELECT t.col1,      +
     t.col2           +
    FROM test_table t;
(1 row)

Therefore, it becomes as follows:

• currentDDL: SELECT t.col1, t.col2
• desiredDDL: SELECT t.*

As a result, the comparison fails. This seems to be the same not only for postgres but also for mysql. I have confirmed that the same phenomenon occurs with mysqldef.

To solve this problem, we need a process to expand * when parsing desiredDDL. This is not impossible, but it seems to take time. Another option is to have sqldef output an error that it does not support * and exit until this problem is solved.

@k0kubun Which do you think is better?

[k0kubun]

To solve this problem, we need a process to expand * when parsing desiredDDL. This is not impossible, but it seems to take time. Another option is to have sqldef output an error that it does not support * and exit until this problem is solved.

For the short term, it makes sense to give up comparison and assume nothing has changed when one side has select *. Eventually, we should compare them at an AST level, normalizing * using the table schema in the previous DDLs. But let's take a workaround for now.

And, thanks for fixing the parser. This change itself is innocent, so let me just merge this.