[8.x] Only select related columns by default in CanBeOneOfMany::ofMany

[bastien-phi]

#39187 introduced an issue with selected columns of related :

dump($user->latestLogin->getAttributes());

/*
array:4 [
  "id" => "2"
  "user_id" => "1"
  "deleted_at" => null
  "id_aggregate" => "2"
]
*/

The id_aggregate column should not be selected and can cause issues.

The problem comes from the sql query :

select * 
from "logins" 
inner join (
    select MAX("id") as "id_aggregate", "logins"."user_id" 
    from "logins" 
    --- more sql
) as "latestOfMany"
on "latestOfMany"."id_aggregate" = "logins"."id" 
--- more sql

As the query selects *, id_aggregate is also selected.

Forcing the query to select "logins".* solves the problem.

select "logins".* 
from "logins" 
inner join (
    select MAX("id") as "id_aggregate", "logins"."user_id" 
    from "logins" 
    --- more sql
) as "latestOfMany"
on "latestOfMany"."id_aggregate" = "logins"."id" 
--- more sql

dump($user->latestLogin->getAttributes());

/*
array:3 [
  "id" => "2"
  "user_id" => "1"
  "deleted_at" => null
]
*/

[bastien-phi]

fixes #39306

[taylorotwell]

Can @cbl give some feedback here. I feel like I'm shooting in the dark merging all these PRs that all seem to introduce additional problems.

[cbl]

@taylorotwell I think tests on complicated features like this are everything.

I think this feature is pretty well tested and ~99% of all use cases work when the tests pass, however, issues occur from time to time due to different behavior of sql drivers. So we should consider mapping all tests over other common sql drivers such as mysql and PostgreSQL as well.

Otherwise we would just rely on people saying that this works... 😅

If wanted I will create a pr for mapping the tests over other sql drivers.

[driesvints]

If wanted I will create a pr for mapping the tests over other sql drivers.

We can do that for unit tests for now. For integration tests, I still have it on my list to set up different builds for PostgreSQL, MSSQL & MySQL.

[bastien-phi]

Thanks @taylorotwell @driesvints and @cbl for your hard work on that.

I feel really sorry for the trouble I caused. I hope there won't be any new issue...