Re: Foreign key joins revisited

On Mon, 27 Dec 2021 at 03:22, Joel Jacobson <joel(at)compiler(dot)org> wrote:

> However, I see one problem with leaving out the key columns:
> First, there is only one FK in permission pointing to role, and we write a
> query leaving out the key columns.
> Then, another different FK in permission pointing to role is later added,
> and our old query is suddenly in trouble.
>

I thought the proposal was to give the FK constraint name. However, if the
idea now is to allow leaving that out also if there is only one FK, then
that's also OK as long as people understand it can break in the same way
NATURAL JOIN can break when columns are added later. For that matter, a
join mentioning column names can break if the columns are changed. But
breakage where the query no longer compiles are better than ones where it
suddenly means something very different so overall I wouldn't worry about
this too much.