MIME-Version: 1.0
In-Reply-To: 
 <CACJufxEH-8UPdbPoUoqNRaiOePw+s2W2DG4OpXtoSYDaW30oAg@mail.gmail.com>
References: 
 <CACJufxGqm7cYQ5C65Eoh1z-f+aMdhv9_7V=NoLH_p6uuyesi6A@mail.gmail.com>
 <4c358bf8-87e8-4760-83d5-a6f9fcd88907@postgresfriends.org>
 <CACJufxF4OW=x2rCwa+ZmcgopDwGKDXha09qTfTpCj3QSTG6Y9Q@mail.gmail.com>
 <1b702de4-86b0-4144-a2d0-38266296894f@postgresfriends.org>
 <CACJufxEH-8UPdbPoUoqNRaiOePw+s2W2DG4OpXtoSYDaW30oAg@mail.gmail.com>
From: "David G. Johnston" <david.g.johnston@gmail.com>
Date: Mon, 28 Jul 2025 20:54:04 -0700
Message-ID: 
 <CAKFQuwZXYxPw4wofq+zZ5D7nH0sHDBwr--qku9jWYv7dS+P8RQ@mail.gmail.com>
Subject: Re: implement CAST(expr AS type FORMAT 'template')
To: jian he <jian.universality@gmail.com>
Cc: Vik Fearing <vik@postgresfriends.org>,
	PostgreSQL-development <pgsql-hackers@postgresql.org>
Content-Type: multipart/alternative; boundary="0000000000000e8087063b095c6f"
Archived-At: 
 <https://www.postgresql.org/message-id/CAKFQuwZXYxPw4wofq%2BzZ5D7nH0sHDBwr--qku9jWYv7dS%2BP8RQ%40mail.gmail.com>
Precedence: bulk

--0000000000000e8087063b095c6f
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Monday, July 28, 2025, jian he <jian.universality@gmail.com> wrote:

> On Mon, Jul 28, 2025 at 6:47=E2=80=AFPM Vik Fearing <vik@postgresfriends.=
org>
> wrote:
> >
> > > adding these pg_cast entries seems tricky.
>
> select proname, prosrc, proformat from pg_proc where proformat;
>
> what do you think?
>

My first impression of this choice was not good.

How about changing the specification for create type.  Right now input
functions must declare either 1 or 3 arguments.  Let=E2=80=99s also allow f=
or 2 and
4-argument functions where the 2nd or 4th is where the format is passed.
If a data type input function lacks one of those signatures it is a runtime
error if a format clause is attached to its cast expression.  For output,
we go from having zero input arguments to zero or one, with the same
resolution behavior.

Pass null for the format if the clause is missing or the cast is done via
the :: operator, or any other context format is not able to be specified.

The slight variation to this would be to specify these 2/4 and 1-arg
functions as optional =E2=80=9Cformat_in=E2=80=9D and =E2=80=9Cformat_out=
=E2=80=9D optional properties
(like typmod_in).  The format-aware code can look for these which will end
up having the full implementation while the current IO functions would
simply stub out calls, passing null as the format. (Or maybe some variation
that looks similar to typmod handling=E2=80=A6which I haven=E2=80=99t looke=
d at.)

David J.

--0000000000000e8087063b095c6f
Content-Type: text/html; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Monday, July 28, 2025, jian he &lt;<a href=3D"mailto:jian.universality@g=
mail.com">jian.universality@gmail.com</a>&gt; wrote:<br><blockquote class=
=3D"gmail_quote" style=3D"margin:0 0 0 .8ex;border-left:1px #ccc solid;padd=
ing-left:1ex">On Mon, Jul 28, 2025 at 6:47=E2=80=AFPM Vik Fearing &lt;<a hr=
ef=3D"mailto:vik@postgresfriends.org">vik@postgresfriends.org</a>&gt; wrote=
:<br>
&gt;<br>
&gt; &gt; adding these pg_cast entries seems tricky.<br><br>
select proname, prosrc, proformat from pg_proc where proformat;<br><br>
what do you think?<br>
</blockquote><div><br></div><div>My first impression of this choice was not=
 good.</div><div><br></div><div>How about changing the specification for cr=
eate type.=C2=A0 Right now input functions must declare either 1 or 3 argum=
ents.=C2=A0 Let=E2=80=99s also allow for 2 and 4-argument functions where t=
he 2nd or 4th is where the format is passed.=C2=A0 If a data type input fun=
ction lacks one of those signatures it is a runtime error if a format claus=
e is attached to its cast expression.=C2=A0 For output, we go from having z=
ero input arguments to zero or one, with the same resolution behavior.</div=
><div><br></div><div>Pass null for the format if the clause is missing or t=
he cast is done via the :: operator, or any other context format is not abl=
e to be specified.</div><div><br></div><div>The slight variation to this wo=
uld be to specify these 2/4 and 1-arg functions as optional =E2=80=9Cformat=
_in=E2=80=9D and =E2=80=9Cformat_out=E2=80=9D optional properties (like typ=
mod_in).=C2=A0 The format-aware code can look for these which will end up h=
aving the full implementation while the current IO functions would simply s=
tub out calls, passing null as the format. (Or maybe some variation that lo=
oks similar to typmod handling=E2=80=A6which I haven=E2=80=99t looked at.)<=
/div><div><br></div><div>David J.</div><div><br></div>

--0000000000000e8087063b095c6f--