Re: Emitting JSON to file using COPY TO
Andrew Dunstan <andrew@dunslane.net>
From: Andrew Dunstan <andrew@dunslane.net>
To: Joe Conway <mail@joeconway.com>, jian he <jian.universality@gmail.com>,
Daniel Verite <daniel@manitou-mail.org>
Cc: Masahiko Sawada <sawada.mshk@gmail.com>, Junwang Zhao
<zhjwpku@gmail.com>, Florents Tselai <florents.tselai@gmail.com>,
"Andrey M. Borodin" <x4mmm@yandex-team.ru>,
Dean Rasheed <dean.a.rasheed@gmail.com>, Davin Shearer <davin@apache.org>,
PostgreSQL development <pgsql-hackers@postgresql.org>
Date: 2026-03-20T12:41:55Z
Lists: pgsql-hackers, pgsql-general
On 2026-03-19 Th 12:06 PM, Joe Conway wrote:
> On 3/19/26 11:02, Andrew Dunstan wrote:
>>
>> On 2026-03-18 We 9:58 PM, jian he wrote:
>>> On Wed, Mar 18, 2026 at 10:37 PM Daniel Verite
>>> <daniel@manitou-mail.org> wrote:
>>>> Currently there's no difference in output between the null
>>>> json value and the SQL null.
>>>>
>>>> postgres=# create table tbl (j jsonb);
>>>> postgres=# insert into tbl values('null');
>>>> postgres=# insert into tbl values(null);
>>>> postgres=# copy tbl to stdout with (format json);
>>>> {"j":null}
>>>> {"j":null}
>>>>
>>>> Does it have to be that way or are there valid distinct outputs
>>>> that we could use to avoid this ambiguity?
>>>>
>>> This is an existing (quite old) behavior of
>>> composite_to_json->datum_to_json_internal, IMHO.
>>>
>>> ```
>>> if (is_null)
>>> {
>>> appendBinaryStringInfo(result, "null", strlen("null"));
>>> return;
>>> }
>>> ```
>>> produce the same results as
>>> ```
>>> case JSONTYPE_JSON:
>>> /* JSON and JSONB output will already be escaped */
>>> outputstr = OidOutputFunctionCall(outfuncoid, val);
>>> appendStringInfoString(result, outputstr);
>>> pfree(outputstr);
>>> break;
>>> ```
>>>
>>> Therefore I intended to document it as below:
>>>
>>> <refsect2 id="sql-copy-json-format" xreflabel="JSON Format">
>>> <title>JSON Format</title>
>>> <para>
>>> When the <literal>json</literal> format is used, data is
>>> exported with one JSON object per line,
>>> where each line corresponds to a single record.
>>> The <literal>json</literal> format has no standard way to
>>> distinguish between an SQL <literal>NULL</literal> and a JSON
>>> <literal>null</literal> literal.
>>> In the examples that follow, the following table containing JSON
>>> data will be used:
>>> <programlisting>
>>> CREATE TABLE my_test (a jsonb, b int);
>>> INSERT INTO my_test VALUES ('null', 1), (NULL, 1);
>>> </programlisting>
>>>
>>> When exporting this table using the <literal>json</literal>
>>> format:
>>> <programlisting>
>>> COPY my_test TO STDOUT (FORMAT JSON);
>>> </programlisting>
>>> In the resulting output, both the SQL <literal>NULL</literal> and
>>> the JSON <literal>null</literal> are rendered identically:
>>> <screen>
>>> {"a":null,"b":1}
>>> {"a":null,"b":1}
>>> </screen>
>>> </para>
>>> </refsect2>
>>>
>>>
>>>
>>> what do you think?
>>>
>>>
>>>
>>
>> I can live with that, if others can.
>
> +1
> WFM
>
pushed with that addition.
cheers
andrew
--
Andrew Dunstan
EDB: https://www.enterprisedb.com
Commits
-
Add option force_array for COPY JSON FORMAT
- 4c0390ac53b7 19 (unreleased) landed
-
json format for COPY TO
- 7dadd38cda95 19 (unreleased) landed
-
introduce CopyFormat, refactor CopyFormatOptions
- a2145605ee3d 19 (unreleased) landed
-
Doc: add IDs to copy.sgml's <varlistentry> and <refsect1>
- e4018f891dec 19 (unreleased) cited
-
Refactor COPY TO to use format callback functions.
- 2e4127b6d2d8 18.0 cited