Blank lines between common table expressions (CTEs) misreported as syntax error

[aleclarson]

I assume this has to do with how statement splitting is implemented.

The SQL formatter I use likes to put blank lines between CTEs.

WITH users_cte AS (
  SELECT
    id,
    email
  FROM account
  WHERE tier = 'Pro'
),

active_sessions_cte AS (
  SELECT
    accountid,
    count(*) AS session_count
  FROM session
  WHERE expiresat > current_timestamp
  GROUP BY accountid
)

SELECT
  u.id,
  u.email,
  coalesce(s.session_count, 0) AS active_sessions
FROM users_cte AS u
LEFT JOIN active_sessions_cte AS s ON u.id = s.accountid
ORDER BY active_sessions DESC;

Output

path/to/example.sql:7:3 syntax ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

  × Expected Ident
  
     5 │   FROM account
     6 │   WHERE tier = 'Pro'
   > 7 │ ),
       │   
   > 8 │ 
   > 9 │ active_sessions_cte AS (
       │ 
    10 │   SELECT
    11 │     accountid,
  

path/to/example.sql:7:3 syntax ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

  × Expected As
  
     5 │   FROM account
     6 │   WHERE tier = 'Pro'
   > 7 │ ),
       │   
   > 8 │ 
   > 9 │ active_sessions_cte AS (
       │ 
    10 │   SELECT
    11 │     accountid,
  

path/to/example.sql:7:3 syntax ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

  × Expected Ascii40
  
     5 │   FROM account
     6 │   WHERE tier = 'Pro'
   > 7 │ ),
       │   
   > 8 │ 
   > 9 │ active_sessions_cte AS (
       │ 
    10 │   SELECT
    11 │     accountid,
  

Checked 1 file in 37ms. No fixes applied.
Found 3 errors.
check ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

  × Some errors were emitted while running checks.
  

[aleclarson]

As a workaround, it'd be great to have an option that tells the LSP to skip statement splitting and assume each file is one statement.

[psteinroe]

We should be able to implement an exception to the double newlines rule for this case. But I also like your idea - makes a few use cases much simpler.

[juleswritescode]

That's a cool idea.
Maybe we could also work with annotations:

-- pg-stmt-start
...
-- pg-stmt-end