Re: [HACKERS] [PATCH] Generic type subscripting

čt 7. 1. 2021 v 9:15 odesílatel Dmitry Dolgov <9erthalion6(at)gmail(dot)com>
napsal:

> > On Wed, Jan 06, 2021 at 09:22:53PM +0100, Pavel Stehule wrote:
> >
> > this case should to raise exception - the value should be changed or
> error
> > should be raised
> >
> > postgres=# insert into foo values('{}');
> > postgres=# update foo set a['a'] = '100';
> > postgres=# update foo set a['a'][1] = '-1';
> > postgres=# select * from foo;
> > ┌────────────┐
> > │ a │
> > ╞════════════╡
> > │ {"a": 100} │
> > └────────────┘
>
> I was expecting this question, as I've left this like that intentionally
> because of two reasons:
>
> * Opposite to other changes, to implement this one we need to introduce
> a condition more interfering with normal processing, which raises
> performance issues for already existing functionality in jsonb_set.
>
> * I vaguely recall there was a similar discussion about jsonb_set with
> the similar solution.
>

ok.

In this case I have a strong opinion so current behavior is wrong. It can
mask errors. There are two correct possibilities

1. raising error - because the update try to apply index on scalar value

2. replace by array - a = {NULL, -1}

But isn't possible ignore assignment

What do people think about it?

> For the references what I mean I've attached the third patch, which does
> this. My opinion would be to not consider it, but I'm fine leaving this
> decision to committer.
>