MacLochlainns Weblog

Michael McLaughlin's Technical Blog

Site Admin

Archive for the ‘Unicode’ Category

PostgreSQL Unicode

with 3 comments

It seems unavoidable to use Windows. Each time I’m compelled to run tests on the platform I find new errors. For example, they don’t use 4-byte unicode and as a result when you want to use Unicode in PostgreSQL there’s a mismatch.

For example, change the Active Console Code Page with the chcp (change code page) to match the one PostgreSQL uses, like:

chip 1252

It lets you avoid this warning message:

Password for user postgres:
psql (14.1)
WARNING: Console code page (437) differs from Windows code page (1252)
         8-bit characters might not work correctly. See psql reference
         page "Notes for Windows users" for details.
Type "help" for help.
 
postgres=#

However, it won’t avoid display issues with real Unicode values. For example, let’s use a small international table like the following:

/* Conditionally drop the conquistador table. */
DROP TABLE IF EXISTS conquistador;
 
/* Create the conquistador table. */
CREATE TABLE conquistador
( conquistador_id   SERIAL
, conquistador      VARCHAR(30)
, actual_name       VARCHAR(30)
, nationality       VARCHAR(30)
, lang              VARCHAR(2));
 
/* Insert some conquistadors into the table. */
INSERT INTO conquistador
( conquistador
, actual_name
, nationality
, lang )
VALUES
 ('Juan de Fuca','Ioánnis Fokás','Greek','el')
,('Nicolás de Federmán','Nikolaus Federmann','German','de')
,('Sebastián Caboto','Sebastiano Caboto','Venetian','it')
,('Jorge de la Espira','Georg von Speyer','German','de')
,('Eusebio Francisco Kino','Eusebius Franz Kühn','Italian','it')
,('Wenceslao Linck','Wenceslaus Linck','Bohemian','cs')
,('Fernando Consag','Ferdinand Konšcak','Croatian','sr')
,('Américo Vespucio','Amerigo Vespucci','Italian','it')
,('Alejo García','Aleixo Garcia','Portuguese','pt');
 
/* Query the values from the conquistador table. */
SELECT * FROM conquistador;

When you call the script to load it, like:

\i testScript.sql

It’ll display the following, which you can check against the strings in the VALUES clause above. There are encoding issues on lines 1, 2, 3, 5, 7, and 8 below.

 conquistador_id |      conquistador      |     actual_name      | nationality | lang
-----------------+------------------------+----------------------+-------------+------
               1 | Juan de Fuca           | Ioánnis Fokás      | Greek       | el
               2 | Nicolás de Federmán  | Nikolaus Federmann   | German      | de
               3 | Sebastián Caboto      | Sebastiano Caboto    | Venetian    | it
               4 | Jorge de la Espira     | Georg von Speyer     | German      | de
               5 | Eusebio Francisco Kino | Eusebius Franz Kühn | Italian     | it
               6 | Wenceslao Linck        | Wenceslaus Linck     | Bohemian    | cs
               7 | Fernando Consag        | Ferdinand Konšcak   | Croatian    | sr
               8 | Américo Vespucio      | Amerigo Vespucci     | Italian     | it
               9 | Alejo García           | Aleixo Garcia        | Portuguese  | pt
(9 rows)

If you’re like me, it was annoying. The problem is that the native 2-byte Unicode of Microsoft sends values into PostgreSQL that are invalid. Those codes are read back with unintended values from other character encoding sets.

While you can’t set Windows generic encoding to 65001 without causing the system problems, you can set Active Console Code Page value in the scope of a Command-Line session before running the script.

The chcp command lets you set it to 4-byte Unicode, like:

chcp 65001

Now, rerun the script and PostgreSQL will display the correct character encoding set with some spacing irregularities. However, that’s not what’s important when you call table from another programming language through the ODBC-layer. The data will be returned in a 4-byte Unicode encoding stream.

 conquistador_id |      conquistador      |     actual_name      | nationality | lang
-----------------+------------------------+----------------------+-------------+------
               1 | Juan de Fuca           | Ioánnis Fokás        | Greek       | el
               2 | Nicolás de Federmán    | Nikolaus Federmann   | German      | de
               3 | Sebastián Caboto       | Sebastiano Caboto    | Venetian    | it
               4 | Jorge de la Espira     | Georg von Speyer     | German      | de
               5 | Eusebio Francisco Kino | Eusebius Franz Kühn  | Italian     | it
               6 | Wenceslao Linck        | Wenceslaus Linck     | Bohemian    | cs
               7 | Fernando Consag        | Ferdinand Konšcak    | Croatian    | sr
               8 | Américo Vespucio       | Amerigo Vespucci     | Italian     | it
               9 | Alejo García           | Aleixo Garcia        | Portuguese  | pt
(9 rows)

A similar error to what I encountered testing MySQL Workbench’s ability to export SQL Server databases 10 years ago. I thought giving a solution to get coerce correct 4-byte Unicode data insertion may help those who also may be surprised by the behavior.