Character encoding bugs are 𝒜wesome!

Before I delve into things, I want to make my terminology clear. When I say, “UTF-8”, I am talking about the common encoding scheme for Unicode characters. When I say “utf8”, I am talking about MySQL’s utf8 character set. I hope you don’t find this confusing, but I wanted to avoid wordiness. The two terms may look similar, but they refer to things that are as different as apples and flamethrowers.

A few weeks ago, I had a fun time dealing with UTF-8 and MySQL. At work, we had a problem: Trying to save certain characters to the database caused Django exceptions. This confused me, because MySQL’s character set was utf8, and its collation was utf8_general_ci. That meant it should handle anything thrown at it.

Or so I thought.

I soon learned that MySQL’s utf8 charset doesn’t support all of UTF-8. It only supports Unicode characters that, when encoded as UTF-8, are shorter than 4 bytes. That covers the Basic Multilingual Plane, but there are 15 other planes in the Unicode standard. Fortunately, MySQL added support for a new encoding: utf8mb4. So it should just be a matter of running the right ALTER TABLE statements, right?

Oh-ho-ho! If only it were so simple. But I don’t write about simple fixes. The fact that this post exists means I went on a great journey. Behold:

mysql> ALTER TABLE foo CONVERT TO CHARACTER SET utf8mb4 COLLATE utf8mb4_bin;
ERROR 1709 (HY000): Index column size too large. The maximum column size is 767 bytes.

This is when I discovered that InnoDB limits index columns to 767 bytes. Why is this suddenly an issue? Because changing the charset also changes the number of bytes needed to store a given string. With MySQL’s utf8 charset, each character could use up to 3 bytes. With utf8mb4, that goes up to 4 bytes. If you have an index on a 255 character column, that would be 765 bytes with utf8. Just under the limit. Switching to utf8mb4 increases that index column to 1020 bytes (4 * 255). The solution? Delete the old index, alter the table, then create a new index that is only on the first 191 characters of the column.

How does one find the offending index? SHOW INDEXES FROM foo; will show all indexes on the table. Combine that with DESCRIBE foo; and you can figure out which indexes are on columns longer than 191 characters. With that out of the way, back to the action:

mysql> DROP INDEX foo_1234 on foo;
Query OK, 0 rows affected (0.00 sec)
Records: 0  Duplicates: 0  Warnings: 0
mysql> ALTER TABLE foo CONVERT TO CHARACTER SET utf8mb4 COLLATE utf8mb4_bin;
Query OK, 0 rows affected (0.11 sec)
Records: 0  Duplicates: 0  Warnings: 0
CREATE INDEX foo_1234 ON foo (`baz`(191));

You might say, “OK, now we’re finished, right?” Ehh… it’s not so simple. MySQL’s utf8 charset is case-insensitive. The utf8mb4 charset is case-sensitive. The implications are vast. This change in constraints forces you to sanitize the data currently in your database, then make sure you don’t insert anything with the wrong casing. At work, there was much weeping and gnashing of teeth.

Once that problem was taken care of, there was only the small matter of monkey-patching Django to support MySQL’s utf8mb4 charset. Currently, Django doesn’t support it. Unfortunately, I forgot to ask my manager for permission to release that code. You’ll have to use your imagination.

After that, the journey was easy. There were some problems with simplejson, then some issues with the async Python code generated by Thrift. These troubles were inconsequential compared to the rest. After they were taken care of, we ended up with a product that supports UTF-8 throughout. Mission accomplished!

While I have summarized the problems and solutions, I certainly wasn’t the only one involved in this saga. My coworkers' contributions were just as important as my own. My thanks to Matt, Brad, Cyli, Dried, Mark, Shawnps, PhilK, and more. Without you, life would be darkness and chaos.

I had to work around some bugs in WEBrick, libcloud, and Amazon S3 to post this. Apparently, a lot of software can’t handle 𝒜wesomeness. The most prominent example I could find is GitHub. Try to comment on this pull request. You can’t. I can’t even close it. I notified GitHub, but I fear they face a journey as long as mine before they can fix the problem.

Test your own applications. See how 𝒜wesome they are. You may have a similar journey ahead of you. Godspeed.




Go dark.