Re: jsonb concatenate operator's semantics seem questionable

On Tue, May 19, 2015 at 10:43 PM, Petr Jelinek <petr(at)2ndquadrant(dot)com> wrote:
> I am of strong opinion that concat should be shallow by default. Again it's
> how jquery works by default, it's how python's dict.update works and you can
> find this behavior in other languages as well when dealing with nested
> hashes. It's also how json would behave if you'd just did string
> concatenation (removing the outermost curly brackets) and parse it to json
> afterwards.

As I said, that argument might be a good one if you were able to
subscript jsonb and have the update affect one particular subdocument.
You're not, though -- updating jsonb usually requires you to write an
SQL expression that evaluates to the final jsonb document that you'd
like to update a record to contain.

> I think this whole discussion shows primarily that it's by far not
> universally agreed if concatenation of json should be shallow or deep by
> default and AFAICS this is true even in javascript world so we don't really
> have where to look for precedents.
>
> Given the above I would vote to just provide the function and leave out the
> || operator for now.

I've said my piece; I think it's a mistake to use an operator that has
a certain association, the association that the concatenate operate
got from hstore. || is the operator broadly useful for updates in
people's minds. I think this *positioning* of the operator is a
mistake. I'll leave it at that.

--
Peter Geoghegan