Wrong string length from unicode database in Borland's app

Hi,

I have windows app written in Borland C++ Builder 5.0. Using ODBC
driver windows app connects to database on linux server. Database is
created with UNICODE encoding.

When pg-server is version 7.1.3 windows app works fine, but when
pg-server is version 7.4.6 or 8.0beta4 under certain conditions the
app receives strings with wrong lengths. For example: the database
contains table "issue" with the following structure:

  Column   |           Type           | Modifiers
-----------+--------------------------+---------------
 id        | integer                  | not null ...
 mag_id    | integer                  |
 year      | character(36)            |
 volume    | character(78)            |
 number    | character(300)           |
 user_name | character(20)            | default getpgusername()
 mod_date  | timestamp with time zone | default now()

Only the column "volume" can contain unicode symbols.

When the the windows application executes the following query:

     select year, trim(number) from issue where mag_id = 25403;

all works fine. But by executing the query

    select volume, trim(number) from issue where mag_id = 25403;

the datagrid component (that displays query's results) contains in
second column values with length of 32769. Those big values are only
in rows where appropriate "volume" field contains unicode symbols.
Actually all values in column "number" are 5-7 symbols long.

By executing the query

   select year, number from issue where mag_id = 25403;

all works fine.

I have checked several versions of ODBC-driver: 07.03.0200,
07.05.0001, 08.00.0002 - the effects are the same: all works fine with
postgres-server 7.1.3 and works bad with 7.4.6 or 8.0beta4

Changing of driver settings (as described on
http://gborg.postgresql.org/project/psqlodbc/faq/faq.php in "Borland
Apps" section) does not help too. My setup is:

         Parse Statements - checked
         Unknown Sizes - Longest
         Text as LongVarChar - checked
         Unknowns as LongVarChar - unchecked

Without those settings the windows app works even worse.

Please help me to solve this problem. May be I'm doing something
wrong? May be my windows is set up incorrectly? (I'm using Win2K SP4)
Or is this a ODBC-driver's bug? Or server's? Or may be it's a Borlands
BDE error?

Has somebody used windows app written in borland to connect to
unicode table under postgresql > 7.1.3 ?


Best regards,
Alex