mysql-latin1-codec 2.0

Python string codec for MySQL's latin1 encoding

Author:

Wouter Bolsterlee

License:

3-clause BSD

URL:

https://github.com/wbolster/mysql-latin1-codec

Overview

This project provides a Python string codec for MySQL’s latin1 encoding, and an accompanying iconv-like command line script for use in shell pipes.

Rationale

• MySQL defaults to using the latin1 encoding for all its textual data, but its latin1 encoding is not actually latin1 but a MySQL-specific variant.

• Due to improperly written applications or wrongly configured databases, many existing databases keep data in MySQL latin1 columns, even if that data is not actually latin1 data.

• MySQL will not complain about this, so this often goes unnoticed. The problems only appear when you try to access the database from another application, or when you try to grep through database dumps produced by mysqldump.

• Many libraries do not support this encoding properly, and using the real latin1 encoding leads to corruption when processing data, especially when the database contains text that is not in a West-European language.

• This string codec makes it possible to reconstruct the exact bytes as they were stored in MySQL. This is a valuable tool for fixing text encoding issues with such databases. For convenience, a command line interface that behaves like iconv is also included.

Version history

• Version 2.0 (2014-01-31)

  • Added explicit register() function instead of having side-effects upon module import.

• Version 1.0 (2013-12-04)

  • Initial release

Installation

$ pip install mysql-latin1-codec

The package supports both Python 2 and Python 3.

Usage

You can use this project in two ways: as a stand-alone command line tool and as a Python module.

Command line tool

The command line tool behaves like iconv:

$ python -m mysql_latin1_codec --help
usage: mysql_latin1_codec.py [-h] [-f encoding] [-t encoding] [-o filename]
                             [-c]
                             [inputs [inputs ...]]

iconv-like tool to encode or decode data using MySQL's "latin1" dialect,
which, despite the name, is a mix of cp1252 and ISO-8859-1.

positional arguments:
  inputs                Input file(s) (defaults to stdin)

optional arguments:
  -h, --help            show this help message and exit
  -f encoding, --from encoding
                        Source encoding (uses MySQL's "latin1" if omitted)
  -t encoding, --to encoding
                        Target encoding (uses MySQL's "latin1" if omitted)
  -o filename, --output filename
                        Output file (defaults to stdout)
  -c, --skip-invalid    Omit invalid characters from output (not the default)

Python API

In Python code, simply import the module named mysql_latin1_codec and call the register() function. A string codec named mysql_latin1 will be registered in Python’s codec registry:

import mysql_latin1_codec

mysql_latin1_codec.register()

You can use it using the normal .decode() and .encode() methods on (byte)strings, and you can also specify it as the encoding argument to various I/O functions like io.open(). Example:

# String encoding/decoding round-trip
s1 = u'foobar'
s2 == text.encode('mysql_latin1').decode('mysql_latin1')
assert s1 == s2

# Reading files
import io
with io.open('/path/to/file', 'r', encoding='mysql_latin1') as fp:
    for line in fp:
        pass

Practical examples

The example below ‘fixes’ dumps that contain doubly encoded UTF-8 data, i.e. real UTF-8 data stored in a MySQL latin1 table. By default mysqldump makes UTF-8 dumps, but if MySQL thinks the data is latin1, it will convert it again, resulting in double encoded data.

$ cat backup-of-broken-database-produced-by-mysqldump.sql \
  | python -m mysql_latin1_codec -f UTF-8 \
  | iconv -c -f UTF-8 -t UTF-8 \
  > legible-text-in-utf8.sql

The iconv pipe in this example removes invalid UTF-8 sequences, while keeping the valid parts as-is. MySQL truncates values whose size exceeds the column’s maximum size, but if MySQL doesn’t know that it is handling UTF-8 data (because the database schema and the broken application did not tell it to do so) it truncates the byte sequence, not the character sequence. This may result in incomplete UTF-8 sequences when a multi-byte sequence is truncated somewhere in the middle. Since those characters cannot be recovered anyway, removing them is the right solution in this case.

In code you can do something similar to the example above:

original = b'...'  # byte string containing doubly-encoded UTF-8 data
s = original.decode('UTF-8').encode('mysql_latin1').decode('UTF-8', 'replace')

Another example to ‘fix’ a dump that contains GB2312 (Simplified Chinese) data stored in a MySQL latin1 column, again misinterpreted and encoded to UTF-8 by mysqldump:

$ cat mojibake-crap.sql \
  | python -m mysql_latin1_codec -f UTF-8 \
  | iconv -f GB2312 -t UTF-8 \
  > legible-text-in-utf8.sql

Technical background

How MySQL defines latin1

The character set that MySQL uses when latin1 is specified, is not actually the well-known latin1 character set, officially known as ISO-8859-1. What MySQL calls latin1 is actually a custom encoding based on cp-1252 (also known as windows-1252).

The MySQL documentation on West European Character Sets 9§ 10.1.14.2) contains:

latin1 is the default character set. MySQL’s latin1 is the same as the Windows cp1252 character set. THis means it is the same as official ISO 8859-1 or IANA (Internet Assigned Numbers Authority) latin1, except that IANA latin1 treats the code points between 0x80 and 0x9f as “undefined”, whereas cp1252, and therefore MySQL’s latin, assign characters for those positions. For example, 0x80 is the Euro sign. For the “undefined” entries in cp1252, MySQL translates 0x81 to Unicode 0x0081, 0x8d to 0x008d, 0x8ff to 0x008f, 0x90 to 0x0090, and 0x9d to 0x009d.

Some more details can be found in the MySQL source code in the file strings/ctype-latin1.c:

WL#1494 notes:

We'll use cp1252 instead of iso-8859-1.
cp1252 contains printable characters in the range 0x80-0x9F.
In ISO 8859-1, these code points have no associated printable
characters. Therefore, by converting from CP1252 to ISO 8859-1,
one would lose the euro (for instance). Since most people are
unaware of the difference, and since we don't really want a
"Windows ANSI" to differ from a "Unix ANSI", we will:

 - continue to pretend the latin1 character set is ISO 8859-1
 - actually allow the storage of euro etc. so it's actually cp1252

Also we'll map these five undefined cp1252 character:
  0x81, 0x8D, 0x8F, 0x90, 0x9D
into corresponding control characters:
   U+0081, U+008D, U+008F, U+0090, U+009D.
like ISO-8859-1 does. Otherwise, loading "mysqldump"
output doesn't reproduce these undefined characters.

As you can see, this encoding is significantly different from ISO-8859-1 (the real latin1), but MySQL misleadingly labels it as latin anyway.

Why this can be a problem

MySQL’s latin1 encoding allows for arbitrary data to be stored in database columns, without any validation. This means latin1 text columns can store any byte sequence, for example UTF-8 encoded text (which uses a variable number of bytes per character) or even JPEG images (which is not text at all).

This is of course not the proper use of latin1 columns. Even in this modern Unicode-aware world, in which all properly written software that handles text should use UTF-8 (or another Unicode encoding), it is quite common to stumble upon wrongly configured databases or badly written software. Most applications use the same (incorrect) assumptions for both storing and retrieving data, so in many setups this will still ‘just work’, and the problem can go unnoticed for a long time.

What makes this problem worse, is that MySQL defaults to using the latin1 character encoding, mostly for historical and backward-compatibility reasons. This means many databases in the real world are (perhaps mistakingly) configured to store data in columns that use MySQL’s latin1 encoding, even though the actual data stores in those columns is not encoding using latin1 at all.

This can lead to a variety of problems, such as encoding or decoding errors, double encoded text, malfunctioning string operations, or incorrect truncation which can lead to data corruption. In many cases this manifests itself as mojibake text. This may be caused by a misinterpretation of the characters that the bytes represent, or by double encodings, e.g. UTF-8 in a latin1 column that was converted to UTF-8 again by a backup script.

Many tools, like Python’s built-in text codecs and the iconv (both the command line tool and the C library) cannot convert data encoding using this custom MySQL encoding. This makes it quite hard to ‘recover’ e.g. UTF-8 data that was stored in a latin1 column, and subsequently dumped using mysqldump, even if you know what you’re doing and which actual encoding was used.

When invoked on the command line, this script converts the dump file(s) specified on the command line (or standard input if no files were given). The data is interpreted as UTF-8 and encoded as MySQL’s latin1 and written to the standard output. The output is the raw data, which likely needs further processing, e.g. using iconv to “reinterpret” the data correctly (e.g. as UTF-8).

I have no idea what you are talking about!

No worries, that’s okay.