Thread: Operator "=" not unicode-safe?

PostgreSQL version: 8.0.3
Operating system:   Linux (SuSE 9.1)

I have a UNICODE database, trying to compare two unicode strings (Ethiopic
characters). Client encoding is also UNICODE:
===================================================
testdb=> select 'በድሩ ሁሴን'='ሰይፉ ከበደ';?column?
----------t
(1 row)

Clearly, it can be seen that they are not equal. The "LIKE" operator also
seems to think so:

testdb=> select 'በድሩ ሁሴን' LIKE 'ሰይፉ ከበደ';?column?
----------f
(1 row)
===================================================

What is the problem here?
The behavior is the same with SQL_ASCII databases and the SQL_ASCII client
encoding.

Of course one could always overload the operator or just use LIKE. But
where it really matters is with queries using UNION, EXCEPT or INTERSECT:

==========================

testdb=> select a from a;   a
---------በድሩ ሁሴንሰይፉ ከበደ
(2 rows)

testdb=> select a from b;   a
---------ሰይፉ ከበደ
(1 row)

testdb=> select a from a union select a from b;   a
---------በድሩ ሁሴን
(1 row)

testdb=> select a from a except select a from b;a
---
(0 rows)

testdb=> select a from a intersect select a from b;   a
---------በድሩ ሁሴን
(1 row)
==========================

What can I do?
With kind regards,

Jörg Haustein

Jörg Haustein <Joerg.Haustein@urz.uni-heidelberg.de> writes:
> I have a UNICODE database, trying to compare two unicode strings (Ethiopic
> characters). Client encoding is also UNICODE:
> ===================================================
> testdb=> select 'በድሩ ሁሴን'='ሰይፉ ከበደ';
>  ?column?
> ----------
>  t
> (1 row)

> Clearly, it can be seen that they are not equal.

Sounds to me like you chose a locale that is expecting some non-Unicode
encoding.  "=" ultimately depends on the system's strcoll() routine,
and in many locales strcoll doesn't behave very sanely when handed data
that's illegal in whatever it thinks the encoding is.

Redo your initdb in a locale that is UTF-8 based, and make sure to keep
the database encoding UTF8.  The apparent flexibility to choose
different database encodings really only works if the underlying locale
is "C".

There is a warning about this in the docs, though perhaps not prominent
enough:
http://www.postgresql.org/docs/8.0/static/multibyte.html#AEN20633
        regards, tom lane