Connect Oracle to Postgres with UTF8

Question

Hello all.
I want to connect a Oracle 10gR2 in AL32UTF8 database to a Postgres 9.2 en_US.UTF-8 database.
Everything works if I have HS_LANGUAGE = AMERICAN_AMERICA.WE8ISO8859P1 or HS_LANGUAGE = american_america.we8mswin1252. Not everything... accentuated characters of my main language (Portuguese) don't get visible.
If I set it to AMERICAN_AMERICA.AL32UTF8, i get this message:
ERROR at line 1:
ORA-28500: connection from ORACLE to a non-Oracle system returned this message:
c
Here are my configuration files:
odbc.ini:
[MoodlePostgres]
Description     = Moodle Postgres DB
Driver          = /usr/pgsql-9.2/lib/psqlodbcw.so
Server          = ifbBDPGMoodle
Database        = webbanca3
LogonID         =
Password        =
Port            = 5432
Socket          =
Option          =
Stmt            =
QuotedId    = YES
CHARSET     = en_US.UTF-8
[ODBC Data Sources]
MoodlePostgres = PostgreSQL
odbcinst.ini:
[PostgreSQL]
Description    = ODBC for PostgreSQL
Driver        = /usr/pgsql-9.2/lib/psqlodbcw.so
Setup        = /usr/lib64/libodbcpsqlS.so
Driver64    = /usr/pgsql-9.2/lib/psqlodbcw.so
Setup64        = /usr/lib64/libodbcpsqlS.so
FileUsage    = 1
tnsnames.ora:
webbanca3=
    (DESCRIPTION=
        (ADDRESS=
            (PROTOCOL=TCP)
            (HOST=172.18.1.34)
            (PORT=1521)
        (CONNECT_DATA=
            (SID=webbanca3)
        (HS=OK)
listener.ora:
LISTENER =
(ADDRESS_LIST=
      (ADDRESS=(PROTOCOL=tcp)(HOST=localhost)(PORT=1521))
SID_LIST_LISTENER=
  (SID_LIST=
      (SID_DESC=
         (SID_NAME=webbanca3)
         (ORACLE_HOME=/home/moodlepostgres/product/11.2.0/tg_1)
         (PROGRAM=dg4odbc)
     (ENVS=LD_LIBRARY_PATH=/usr/pgsql-9.2/lib:/home/moodlepostgres/product/11.2.0/tg_1/lib)
#CONNECT_TIMEOUT_LISTENER = 0
initi<sid>.ora:
# HS init parameters
HS_FDS_CONNECT_INFO = MoodlePostgres
#HS_FDS_TRACE_LEVEL = 0
HS_FDS_SHAREABLE_NAME = /usr/pgsql-9.2/lib/psqlodbc.so
#HS_FDS_SHAREABLE_NAME = /usr/pgsql-9.2/lib/psqlodbcw.so
HS_FDS_SUPPORT_STATISTICS = FALSE
#HS_LANGUAGE = american_america.we8mswin1252  # WORKS ... kinda... :-)
HS_LANGUAGE = american_america.al32utf8 # DOESN'T WORK
HS_KEEP_REMOTE_COLUMN_SIZE = ALL
# ODBC specific environment variables
set ODBCINI=/usr/local/etc/odbc.ini
set ODBCINST=/usr/local/etc/odbcinst.ini
# Environment variables required for the non-Oracle system
#set <envvar>=<value>
The result of odbcinst -j is:
unixODBC 2.3.1
DRIVERS............: /usr/local/etc/odbcinst.ini
SYSTEM DATA SOURCES: /usr/local/etc/odbc.ini
FILE DATA SOURCES..: /usr/local/etc/ODBCDataSources
USER DATA SOURCES..: /usr/loca/etc/odbc.ini
SQLULEN Size.......: 8
SQLLEN Size........: 8
SQLSETPOSIROW Size.: 8
I'm running on Oracle Linux 6
Thank you.

Answer 1

The root cause of your problem is the unixODBC Driver manager.
In my environment I can select fine from the Postgres Unicode database storing your characters when I use in my odbc.ini
ConnSettings
=SET CLIENT_ENCODING to 'UNICODE'
and in my gateway init these  settings:
HS_NLS_NCHAR = UCS2
HS_LANGUAGE=american_america.al32utf8
SQL> select * from "public"."unicode"@POSTGRESQL_DG4ODBC_EMGTW_1123_DB;
      col1 col2                 col3
         1 a                    a
         2 ä                    ä
         3 æ                    æ
         4 ç                    ç
         5 ã                    ã
         6 â                    â
as soon as I change in my gateway init file the HS_LANGUAGE to an 8bit character set, for example to:
HS_LANGUAGE=american_america.we8iso8859p1
and
ConnSettings
=SET CLIENT_ENCODING to 'ASCII'
I get the same corrupted content like you get:
SQL> /
      col1 col2                                                                                       col3
         1 a                                                                                          a
         2 ä                                                                                         ä
         3 æ                                                                                         æ
         4 ç                                                                                         ç
         5 ã                                                                                         ã
         6 â                                                                                         â
6 rows selected.
Your unixODBC does not accept multibyte connect strings and fails with the error 28500 and an opening bracket '[' when you set a Unicode character set.
Did you compile unixODBC from source using --with-iconv-char-enc=UTF-8 option?
Looking at the trace file shows that you're still using 11.2.0.1 gateway release and it also seems you have set in your environment a language parameter like NLS_LANG:
First of all I would like to ask you to use 11.2.0.3 as this contains some major NLS fixes. In addition, could you then please unset any NLS_LANG variable before starting the gateway listener?
- Klaus