Re: Removing unneeded self joins

On Thu, Jul 4, 2024 at 5:15 AM jian he <jian(dot)universality(at)gmail(dot)com> wrote:
> in remove_self_join_rel, i have
> ```ChangeVarNodes((Node *) root->parse, toRemove->relid, toKeep->relid, 0);```
> which will change the joinlist(RangeTblRef) from (1,2) to (2,2).
> Immediately after this call, I wrote a function (restore_rangetblref)
> to restore the joinlist as original (1,2).
> then remove_rel_from_joinlist won't error out.
> see remove_self_join_rel, restore_rangetblref.

Thank you, now this is clear. Could we add additional parameters to
ChangeVarNodes() instead of adding a new function which reverts part
of changes.

> current mechanism, in this example context,
> SJE can translate ```update test set val = t.val + 1 from test t where
> test.id = t.id;``` as good as to
> ```update test set val = val + 1```.
> if we replace it that way, then this example would result val = 3.
>
> but without SJE,
> ```update test set val = t.val + 1 from test t where test.id = t.id;```
> will result val = 2.
>
> you mentioned the EPQ problem, previously i don't know what that means.

Yes, I guessed so. I should have come with more detailed explanation.

> now i see, I feel like it is quite challenging to resolve it.

Yep. Glad to see we are on the same page.

This is why I think we could leave SJE for target relation of
modification queries for future. I'd like to not devalue SELECT-only
SJE, given that this is a step forward anyway.

------
Regards,
Alexander Korotkov
Supabase