Re: Allow to_date() and to_timestamp() to accept localized names

=?UTF-8?Q?Juan_Jos=C3=A9_Santamar=C3=ADa_Flecha?= <juanjo(dot)santamaria(at)gmail(dot)com> writes:
> On Tue, Jan 28, 2020 at 9:35 PM Peter Eisentraut <
> peter(dot)eisentraut(at)2ndquadrant(dot)com> wrote:
>> So I'm withdrawing my concerns with respect to this patch. As long as
>> it can do a roundtrip conversion with to_char(), it's fine.

> We can avoid issues with non injective case conversion languages with a
> double conversion, so both strings in the comparison end up in the same
> state.
> I propose an upper/lower conversion as in the attached patch.

This seems pretty reasonable to me, with a couple of caveats:

* I don't think it's appropriate to hard-wire DEFAULT_COLLATION_OID
as the collation to do case-folding with. For to_date/to_timestamp,
we have collatable text input so we can rely on the function's input
collation instead, at the cost of having to pass down the collation
OID through a few layers of subroutines :-(. For parse_datetime,
I punted for now and let it use DEFAULT_COLLATION_OID, because that's
currently only called by JSONB code that probably hasn't got a usable
input collation anyway (since jsonb isn't considered collatable).

* I think it's likely worthwhile to do a quick check for an exact
match before we go through all these case-folding pushups. If the
expected use-case is reversing to_char() output, that will win all
the time. We're probably ahead of the game even if it only matches
a few percent of the time.

Attached v10 incorporates those improvements, plus a bit of
polishing of docs and comments. Barring objections, this seems
committable to me.

regards, tom lane