Different display format for array/struct with pg

[xxchan]

In PG, strings with spaces and empty string are displayed with ""

postgres=# select ARRAY['1','','2'];
  array
----------
 {1,"",2}
(1 row)

postgres=# select ARRAY['1',' ','2'];
   array
-----------
 {1," ",2}
(1 row)

postgres=# select ARRAY['1','1 1','2'];
    array
-------------
 {1,"1 1",2}
(1 row)

postgres=# select ARRAY['1',' 1 ','2'];
    array
-------------
 {1," 1 ",2}
(1 row)

In risingwave:

dev=> select ARRAY['1','','2'];
 ?column?
----------
 {1,,2}
(1 row)

dev=> select ARRAY['1',' ','2'];
 ?column?
----------
 {1, ,2}
(1 row)

dev=> select ARRAY['1','1 1','2'];
 ?column?
-----------
 {1,1 1,2}
(1 row)

dev=> select ARRAY['1',' 1 ','2'];
 ?column?
-----------
 {1, 1 ,2}
(1 row)

Do we want to follow this behavior?

[xiangjinwu]

There are more differences to fix.

Background: In PostgreSQL, boolean is shown differently when displayed as output (boolout) and when casted to string (booltext):

test=# select true;
 bool 
------
 t
(1 row)

test=# select true::varchar;
 varchar 
---------
 true
(1 row)

When inside array or struct, it should use the shorter output:
Expected (PostgreSQL):

test=# select array[true];
 array 
-------
 {t}
(1 row)

test=# select array[true]::varchar;
 array 
-------
 {t}
(1 row)

test=# select row(true);
 row 
-----
 (t)
(1 row)

test=# select row(true)::varchar;
 row 
-----
 (t)
(1 row)

Actual (RisingWave):

dev=> select array[true];
 ?column? 
----------
 {true}
(1 row)

Another issue is the output of timestamp with time zone, which shares Scalar::Int64 as physical in-memory representation:
Expected (PostgreSQL):

test=# set timezone to utc;
SET
test=# select '2022-10-01 12:00:00+01:00'::timestamptz;
      timestamptz       
------------------------
 2022-10-01 11:00:00+00
(1 row)

test=# select array['2022-10-01 12:00:00+01:00'::timestamptz];
           array            
----------------------------
 {"2022-10-01 11:00:00+00"}
(1 row)

Actual (RisingWave):

dev=> select '2022-10-01 12:00:00+01:00'::timestamptz;
         ?column?          
---------------------------
 2022-10-01 11:00:00+00:00
(1 row)

dev=> select array['2022-10-01 12:00:00+01:00'::timestamptz];
      ?column?      
--------------------
 {1664622000000000}
(1 row)

Moreover, PostgreSQL also shows null differently in different context:

test=# select null;
 ?column? 
----------
 
(1 row)

test=# select null::varchar;
 varchar 
---------
 
(1 row)

test=# select array[null];
 array  
--------
 {NULL}
(1 row)

test=# select row(null);
 row 
-----
 ()
(1 row)

test=# select row();
 row 
-----
 ()
(1 row)

test=# select row(null,null);
 row 
-----
 (,)
(1 row)

But in RisingWave:

dev=> select row(null);
 ?column? 
----------
 (NULL)
(1 row)

Last thing: the quoting should also be considered when outputting a struct, and when casting from a string to an array or struct:

test=# select row('nULl','',' a b ',',');
          row          
-----------------------
 (nULl,""," a b ",",")
(1 row)

test=# select unnest('{NULL,"nULl",""," a b ",","}'::varchar[]);
 unnest 
--------
 
 nULl
 
  a b 
 ,
(5 rows)

[xiangjinwu]

Links to PostgreSQL doc and reference:
https://www.postgresql.org/docs/current/arrays.html#ARRAYS-IO
https://www.postgresql.org/docs/current/rowtypes.html#ROWTYPES-IO-SYNTAX
https://github.com/postgres/postgres/blob/REL_15_3/src/backend/utils/adt/arrayfuncs.c#L1016
https://github.com/postgres/postgres/blob/REL_15_3/src/backend/utils/adt/rowtypes.c#L303