Feat: Add condition in the union macro for conditional union of tables

[themisvaltinos]

This update extends the @UNION macro with a condition that is statically evaluatable and if it is false it short-circuits and it only keeps the first table, otherwise unions all tables, fixes: #4202

Example:

MODEL (
  name sushi.table_4
);
  
@UNION(@get_date() > '1996-02-10', 'distinct', sushi.table_1, sushi.table_2, sushi.table_3)

the previous behaviour when the first argument can't be interpreted as a conditional True or False remains with the first argument being the union type and then the table names.

[georgesittas]

Can't we extend the existing @UNION macro somehow to make this work? Having a separate _if variant is not great.

Perhaps you could extend the argument list to make the condition argument keyword-only. Alternatively, you could make it the first argument in the list, which is consistent with other macros, and try to migrate existing models or make it backwards-compatible by checking if the first argument is 'all' or 'distinct' to preserve the old behavior.

cc @tobymao

[themisvaltinos]

I was initially thinking of simply extending the existing @union macro as well, but ended up leaning against it because you'd either need to compromise a bit on the parameter structure, as you mentioned, or require migration of existing models and for people to learn the new way.

Also, looking back at the original issue, it seems there was a need to support both tables and select statements as inputs. That adds complexity, and having a dedicated variant might make that clearer and easier to manage in practice, even if it's not ideal from a naming standpoint. Though for the tables the union logic for tables could still be reused with a helper. But yes happy to go down whatever route sounds more sound.

[tobymao]

what does extending union look like?

[georgesittas]

These are two ways I was thinking it could be extended:

1. @UNION(<condition>, 'distinct', table1, table2, ...)
2. @UNION('distinct', table1, table2, ..., condition := <condition>)

The first one is backwards-incompatible and would require a migration. Perhaps one way to make it backwards-compatible and avoid that is to always evaluate the first argument and if it produces TRUE or FALSE, we treat it as a condition, otherwise we treat it as the existing type_ argument.

[tobymao]

we could do the first one, in a backwards compatible way, and make it check if it's a string literal right?

[tobymao]

it only accepts all or distinct right now

[themisvaltinos]

ok thanks I will alter it this way