after setting all config file and runtime options for charset that i can find to utf-8, new mysqli connections made with php still has its charset set to latin1, which effectively means that i have to call $mysqli->set_charset('utf8') each time i connect.
$mysqli = new mysqli(DB_HOST, DB_USER, DB_PASS, DB_NAME);
if ($mysqli->connect_error)
err_handle("mysql connect error({$mysqli->connect_errno}).");
if (!$mysqli->set_charset("utf8"))
err_handle("db error({$mysqli->errno}).");
i wonder if there is a permanent way of doing this?
similar problem was encountered in this post .
a " show variables like 'character_set%' " query on the mysql server before calling $mysqli->set_charset('utf8') shows:
(this part was ambiguous in previous revs)
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
the client, connection and results charset can only be changed to utf8 with $mysqli->set_charset('utf8') at runtime. after that it shows:
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
i have
default_charset = "utf-8"
set in php.ini, and
[client]
default-character-set=utf8
...
[mysqld]
## This option is deprecated in favor of --character-set-server.
#default-character-set=utf8
set in my.cnf.
the default charset for my tables is also utf8.
seems like the "[client]" options only affect the cmd "mysql" tool and have nothing to do with php.
the return value of $mysqli->character_set_name() is always latin1 no matter what i do, until $mysqli->set_charset('utf8') is called.
i guess "latin1" is a mysql thing, since i cant recall anything else that defaults to "latin1" on my system.
^ update: according to mysql manual 9.1.4 , 9.1.5 and 5.1.3 , character_set_client should be provided by the client. i guess php doesn't provide it upon connection and mysql uses the fall-back charset latin1 .
i'm running php 5.3 on debian wheezy with mysql 5.1.
any suggestion?
updated with info from comments:
i forgot to mention the skip-character-set-client-handshake directive and why i was reluctant to use it.
upon first sight i thought ignoring the handshake might result in the situation that the client talks latin1 while the server talks utf8 . how does the server convert the string from charset character_set_client to character_set_server without knowing the charset currently in use?
correct me if i'm wrong, plz. i will experiment with this setting later today to see if it works.
Updated with workaroud :
make sure everything works under utf-8 (or any preferable charset). then add the skip-character-set-client-handshake line to my.cnf .
this works for me so far. i experimented with some double-width utf-8 characters. both insert and select succeeded and displayed properly in the browser.
what skipping the handshake means is still unclear. and the mysql server now becomes uncapable of using any charset except utf-8, whick makes this workaround quite impractical since i simply cant apply this setting to all the servers that my website runs on.
so i'm not adopting this workaround. further comments and answers are much appreciated.
Source: Tips4all, CCNA FINAL EXAM
You have diagnosed the basic problem correctly: While you can change the default MySQL client charset in the client machine's my.cnf or .my.cnf, these files are not used by PHP.
ReplyDeleteIf you think about how PHP's MySQLi/MySQL extensions work, this will make sense -- they have nothing to do with the mysql client program and aren't going to crawl your filesystem for config files, because they use libmysql directly.
To change libmysql's actual default charset, you'll just need to rebuild libmysql. That may not be an answer you like (since you're using precompiled MySQL binaries), but it is the actual answer. The defaults are set at compile time, and then can be overridden at runtime.
If you don't want to do this and calling set_charset() annoys you, my suggestion would be to simply extend the MySQLi class and use that class in place of mysqli. i.e.:
class MyDB extends mysqli {
// (You could set defaults for the params here if you want
// i.e. $host = 'myserver', $dbname = 'myappsdb' etc.)
public function __construct($host = NULL, $username = NULL, $dbname = NULL, $port = NULL, $socket = NULL) {
parent::__construct($host, $username, $dbname, $port, $socket);
$this->set_charset("utf8");
}
}
Typically in an application you'll have some kind of database abstraction layer anyway, so you can either have this layer use MyDB instead of mysqli, or you can have this layer be MyDB and add or override any methods you want (I've done this with simple ORM-less apps).
It's a good practice to always have some kind of database abstraction layer, even if it starts as just class MyDB extends mysqli {} because then you'll never have to search/replace your entire codebase to make small changes.
RE: your workaround, as you explain, this essentially hardcodes your entire db server to UTF-8 regardless of what clients request. Instead of having multiple databases, each with its own charset, the server only works with UTF-8 and may silently mangle data if clients connect with another charset. This is fundamentally wrong because you've effectively moved one aspect of your application's configuration (database charset) from the app/client machine to the database server where it doesn't really belong.
If you think about the application stack's layers,
[server] <=> [network] <=> [client libmysql] <=> [PHP binary] <=> [app]
then you'll understand that the "correct" place for an app-specific configuration like this is in the app itself, not elsewhere in the stack. You may not like having to specify your database's charset in PHP, but if you think about it, that's really where it belongs, because it's also where you're specifying the database itself that you want to connect to -- it's a connection parameter, not a server configuration issue. Hardcoding the charset anywhere else makes your application non-portable.