Trouble with UTF-8 characters; what I see is not what I stored

Question

I tried to use UTF-8 and ran into trouble.

I have tried so many things; here are the results I have gotten:

• ???? instead of Asian characters. Even for European text, I got Se?or for Señor.
• Strange gibberish (Mojibake?) such as Señor or 新浪新闻 for 新浪新闻.
• Black diamonds, such as Se�or.
• Finally, I got into a situation where the data was lost, or at least truncated: Se for Señor.
• Even when I got text to look right, it did not sort correctly.

What am I doing wrong? How can I fix the code? Can I recover the data, if so, how?

Answer 1

This problem plagues the participants of this site, and many others.

You have listed the five main cases of CHARACTER SET troubles.

Best Practice

Going forward, it is best to use CHARACTER SET utf8mb4 and COLLATION utf8mb4_unicode_520_ci. (There is a newer version of the Unicode collation in the pipeline.)

utf8mb4 is a superset of utf8 in that it handles 4-byte utf8 codes, which are needed by Emoji and some of Chinese.

Outside of MySQL, "UTF-8" refers to all size encodings, hence effectively the same as MySQL's utf8mb4, not utf8.

I will try to use those spellings and capitalizations to distinguish inside versus outside MySQL in the following.

Overview of what you should do

• Have your editor, etc. set to UTF-8.
• HTML forms should start like <form accept-charset="UTF-8">.
• Have your bytes encoded as UTF-8.
• Establish UTF-8 as the encoding being used in the client.
• Have the column/table declared CHARACTER SET utf8mb4 (Check with SHOW CREATE TABLE.)
• <meta charset=UTF-8> at the beginning of HTML
• Stored Routines acquire the current charset/collation. They may need rebuilding.

UTF-8 all the way through

More details for computer languages (and its following sections)

Test the data

Viewing the data with a tool or with SELECT cannot be trusted. Too many such clients, especially browsers, try to compensate for incorrect encodings, and show you correct text even if the database is mangled. So, pick a table and column that has some non-English text and do

SELECT col, HEX(col) FROM tbl WHERE ... 

The HEX for correctly stored UTF-8 will be

• For a blank space (in any language): 20
• For English: 4x, 5x, 6x, or 7x
• For most of Western Europe, accented letters should be Cxyy
• Cyrillic, Hebrew, and Farsi/Arabic: Dxyy
• Most of Asia: Exyyzz
• Emoji and some of Chinese: F0yyzzww
• More details

Specific causes and fixes of the problems seen

Truncated text (Se for Señor):

• The bytes to be stored are not encoded as utf8mb4. Fix this.
• Also, check that the connection during reading is UTF-8.

Black Diamonds with question marks (Se�or for Señor); one of these cases exists:

Case 1 (original bytes were not UTF-8):

• The bytes to be stored are not encoded as utf8. Fix this.
• The connection (or SET NAMES) for the INSERT and the SELECT was not utf8/utf8mb4. Fix this.
• Also, check that the column in the database is CHARACTER SET utf8 (or utf8mb4).

Case 2 (original bytes were UTF-8):

• The connection (or SET NAMES) for the SELECT was not utf8/utf8mb4. Fix this.
• Also, check that the column in the database is CHARACTER SET utf8 (or utf8mb4).

Black diamonds occur only when the browser is set to <meta charset=UTF-8>.

Question Marks (regular ones, not black diamonds) (Se?or for Señor):

• The bytes to be stored are not encoded as utf8/utf8mb4. Fix this.
• The column in the database is not CHARACTER SET utf8 (or utf8mb4). Fix this. (Use SHOW CREATE TABLE.)
• Also, check that the connection during reading is UTF-8.

Mojibake (Señor for Señor): (This discussion also applies to Double Encoding, which is not necessarily visible.)

• The bytes to be stored need to be UTF-8-encoded. Fix this.
• The connection when INSERTing and SELECTing text needs to specify utf8 or utf8mb4. Fix this.
• The column needs to be declared CHARACTER SET utf8 (or utf8mb4). Fix this.
• HTML should start with <meta charset=UTF-8>.

If the data looks correct, but won't sort correctly, then either you have picked the wrong collation, or there is no collation that suits your need, or you have Double Encoding.

Double Encoding can be confirmed by doing the SELECT .. HEX .. described above.

é should come back C3A9, but instead shows C383C2A9 The Emoji 👽 should come back F09F91BD, but comes back C3B0C5B8E28098C2BD 

That is, the hex is about twice as long as it should be. This is caused by converting from latin1 (or whatever) to utf8, then treating those bytes as if they were latin1 and repeating the conversion. The sorting (and comparing) does not work correctly because it is, for example, sorting as if the string were Señor.

Fixing the Data, where possible

For Truncation and Question Marks, the data is lost.

For Mojibake / Double Encoding, ...

For Black Diamonds, ...

The Fixes are listed here. (5 different fixes for 5 different situations; pick carefully): http://mysql.rjweb.org/doc.php/charcoll#fixes_for_various_cases