Support for OUTPUT in t-sql

[jh88]

Thank you for the great work. I have been using this for transpiling SQL server (tsql) to PostgreSQL.

I would like to request support for the OUTPUT keyword in T-SQL, which is similar to the RETURNING keyword in PostgreSQL.

For instance, when executing an INSERT statement like the one below, the OUTPUT clause would return the primary key value id of the inserted row:

INSERT INTO [my_table] ([my_column]) OUTPUT inserted.id VALUES(0);

Thank you.

[tobymao]

thanks for the report

[jh88]

Thank you Toby for the quick response.

The aforementioned sample query, when transpiled using the new code, would be as follows:

INSERT INTO "my_table" (
  "my_column"
)
VALUES
  (0)
RETURNING inserted.id

However, it is necessary to remove inserted in the RETURNING clause.

Conversely, when returning inserted fields in T-SQL, we need to add the inserted prefix.

[tobymao]

did you test it with tsql specified?

[tobymao]

ah i see what you're saying, when going to returning, we need to remove inserted. it doesn't seem like the prefix can be added in the other direction though because it doesn't exist

[jh88]

Yes, I'm transpiling between tsql and postgres.

I created a new AzureSQL class which inherits your TSQL. In the returning_sql method, as a workaround, I manually assign table to inserted.

def returning_sql(self, expression: exp.Returning) -> str:
    into = self.sql(expression, "into")
    into = self.seg(f"INTO {into}") if into else ""

    for e in expression.args['expressions']:
        e.args['table'] = 'inserted'

    return f"{self.seg('OUTPUT')} {self.expressions(expression, flat=True)}{into}"

[tobymao]

sorry, is inserted a magic keyword in tsql?

[jh88]

Apologies for any misunderstandings. I am not inserting data into the table named inserted.
It is a table that is managed by SQL Server. You can find more information about it here.

The inserted table stores copies of the new or changed rows after an INSERT or UPDATE statement.

My understanding is that we should use the inserted table when working with the INSERT statement, and the deleted table for the DELETE statement.

[tobymao]

it's not so simple

UPDATE Production.WorkOrder
SET ScrapReasonID = 4
OUTPUT DELETED.ScrapReasonID,
       INSERTED.ScrapReasonID,
       INSERTED.WorkOrderID,
       INSERTED.ProductID,
       p.Name
    INTO @MyTestVar
FROM Production.WorkOrder AS wo
    INNER JOIN Production.Product AS p
    ON wo.ProductID = p.ProductID
    AND wo.ScrapReasonID= 16
    AND p.ProductID = 733;

if you're trying to convert postgres into tsql, you won't be able to know what inserted and deleted columns are

[tobymao]

i'm not going to support this for now, you can add a custom transform to do what you're looking for

expr.transform(...)