Cannot get mysqldump to produce utf-8 encoded files!

[Fredrik Wollsén]

I cannot backup my utf-8 encoded database to an utf-8 encoded dumpfile, thus also not being able to restore the database without ending up with broken special characters. I have been searching google and this forum extensively without finding a solution to this problem.

I have a database in which all tables are created with:
CREATE TABLE `name` (
) ENGINE=MyISAM DEFAULT CHARSET=utf8 COLLATE=utf8_unicode_ci;

All data inserted into the database is done by PHP. To ensure only utf-8 encoded data is inserted, I use SET NAMES 'utf8' upon every connection and have these settings in php.ini:
mbstring.func_overload = 7
mbstring.internal_encoding = UTF-8
mbstring.http_input = auto

For my.cnf I use:
character-set-server=utf8
default-collation=utf8_unicode_ci
default-character-set=utf8
collation-server=utf8_unicode_ci

Despite this, running:
$ mysqldump -uroot -hlocalhost the_db_name -r supposed_utf8_dumpfile --verbose -p --default-character-set=utf8
Returns a dumpfile with the special chars like äöå all jarbled up. That the file is not UTF-8 encoded is confirmed by runing:
$ enca -L none supposed_utf8_dumpfile
Which replies: Unrecognized encoding

Also, viewing with less shows the jarbled characters where öåäö is supposed to be. ( ä = ä , ö = ö, Ö = Ö etc)

iconv-conversion from latin1, cp1252 or any common iso-format will not work, most propable because of the encoding not being any of those encodings from the beginning...

The dumpfile's top lines are as follows:
-- MySQL dump 10.10
--
-- Host: localhost Database: the_db_name
-- ------------------------------------------------------
-- Server version 5.0.27-Debian_0.dotdeb.1-log

/*!40101 SET @OLD_CHARACTER_SET_CLIENT=@@CHARACTER_SET_CLIENT */;
/*!40101 SET @OLD_CHARACTER_SET_RESULTS=@@CHARACTER_SET_RESULTS */;
/*!40101 SET @OLD_COLLATION_CONNECTION=@@COLLATION_CONNECTION */;
/*!40101 SET NAMES utf8 */;

The system is Ubuntu Dapper server with locale: en_US.UTF-8. Version of mysql is: "mysql Ver 14.12 Distrib 5.0.27, for pc-linux-gnu (i386) using readline 4.3"

Some related information:
"mysqldump --help | grep char" shows:
default-character-set utf8
set-charset TRUE

"mysql --help | grep char" yields:
default-character-set latin1

"mysql --default-character-set=utf8 --help | grep char" yields:
default-character-set utf8

"mysql -uroot -hlocalhost -p
> SHOW VARIABLES; " yields (amongst others of course):
| character_set_client | latin1 |
| character_set_connection | latin1 |
| character_set_database | utf8 |
| character_set_filesystem | binary |
| character_set_results | latin1 |
| character_set_server | utf8 |
| character_set_system | utf8 |
| collation_connection | latin1_swedish_ci |
| collation_database | utf8_unicode_ci |
| collation_server | utf8_unicode_ci |

"mysql -uroot -hlocalhost -p
> SET NAMES 'utf8';
> SHOW VARIABLES;" yields:
| character_set_client | utf8 |
| character_set_connection | utf8 |
| character_set_database | utf8 |
| character_set_filesystem | binary |
| character_set_results | utf8 |
| character_set_server | utf8 |
| character_set_system | utf8 |
| collation_connection | utf8_general_ci |
| collation_database | utf8_unicode_ci |
| collation_server | utf8_unicode_ci |

(Btw - why is collation_connection=utf8_general_ci and not utf8_unicode_ci?)

I feel that I have done most of what can be expected to solve this problem, except for reverting to hackish php-scripts or other software to do what mysqldump is supposed to do.

Please shed a light on this subject - what am I missing?

EDIT: Just to make things clear. I can do:
$ echo "test utf8 åäö" > testutf8
$ enca -L none testutf8
..and get the reply: Universal transformation format 8 bits; UTF-8
I can also view the contents of testutf8 with less without jarbled characters. This type of properties is what I expect the dumpfile to have.