Character Set for Error Messages
This section describes how the server uses character sets for constructing error messages and returning them to clients. For information about the language of error messages (rather than the character set), see , "Setting the Error Message Language".
In MariaDB 5.6, the server constructs error messages using UTF-8 and returns them to clients in the character set specified by the character_set_results
system variable.
The server constructs error messages as follows:
- The message template uses UTF-8.
- Parameters in the message template are replaced with values that apply to a specific error occurrence:
- Identifiers such as table or column names use UTF-8 internally so they are copied as is.
- Character (nonbinary) string values are converted from their character set to UTF-8.
- Binary string values are copied as is for bytes in the range
0x20
to0x7E
, and using\x
hex encoding for bytes outside that range. For example, if a duplicate-key error occurs for an attempt to insert0x41CF9F
into aVARBINARY
unique column, the resulting error message uses UTF-8 with some bytes hex encoded:Duplicate entry 'A\xC3\x9F' for key 1
To return a message to the client after it has been constructed, the server converts it from UTF-8 to the character set specified by the character_set_results
system variable. If character_set_results
has a value of NULL
or binary
, no conversion occurs. No conversion occurs if the variable value is utf8
, either, because that matches the original error message character set.
For characters that cannot be represented in character_set_results
, some encoding may occur during the conversion. The encoding uses Unicode code point values:
- Characters in the Basic Multilingual Plane (BMP) range (
0x0000
to0xFFFF
) are written using\
notation.nnnn
- Characters outside the BMP range (
0x01000
to0x10FFFF
) are written using\+
notation.nnnnnn
Clients can set character_set_results
to control the character set in which they receive error messages. The variable can be set directly, or indirectly by means such as SET NAMES
. For more information about character_set_results
, see , "Connection Character Sets and Collations".
The encoding that occurs during the conversion to character_set_results
before returning error messages to clients can result in different message content compared to earlier versions (before MariaDB 5.5). For example, if an error occurs for an attempt to drop a table named ペ
(KATAKANA LETTER PE) and character_set_results
is a character set such as latin1
that does not contain that character, the resulting message sent to the client has an encoded table name:
ERROR 1051 (42S02): Unknown table '\30DA'
Before MariaDB 5.5, the name is not encoded: