A collection of links and some very basic introduction.
For some information on Unicode in Python, see :ref:`python-unicode`.
For a good programmer's introduction see: http://www.joelonsoftware.com/articles/Unicode.html
Unicode is a convention that defines a unique number for a very very large number of possible characters in almost all known alphabets:
The number assigned to each character is referred to as the code point. The Unicode consortium has the job of defining which code point corresponds to which character. The correspondence of code points to characters is published in the Unicode Character Database - the latest version of which should be at http://www.unicode.org/Public/UNIDATA/. To refer to a code point it is conventional to use hexadecimal - for example code point U+00E9 is the latin small letter e with an acute accent.
The code points up to 128 (hexadecimal 7F) are identical to the ascii
character codes; so for example U+0065 is the latin letter e.
Code points from 0 to 65535 (hexadecimal FFFF) represent characters in the Basic Multilingual Plane (BMP). The BMP contains characters for almost all modern languages, including Chinese, as well as a large number of symbols. Codes outside the BMP (hexadecimal 10000 and above) include some modern Chinese characters, various historical scripts and characters, and musical and mathematical symbols - see http://en.wikipedia.org/wiki/Mapping_of_Unicode_characters.
When these code points are represented on disk or in memory as a string, the string has an encoding - which specifies the relationship between the bytes in the string and the eventual resulting unicode code points.
Fixed width encodings are encodings with one value (16 or 32 bit) per unicode character.
At the moment, the number range of the defined Unicode code points can be contained in the range of a 32 bit unsigned integer. The simplest possible encoding for a string is therefore just to contain one 32 bit value for each character, where each 32 bit value is the code point for the character. This encoding is referred to as Universal Character Set 4 (UCS-4) or Unicode Transformation Format 32 (UTF-32): http://en.wikipedia.org/wiki/UTF-32/UCS-4
Unicode characters above hexadecimal FFFF (and outside the BMP) are rare in most languages, and so another simple way of representing common unicode strings is to have just one 16 bit value per character; this is UCS-2. Because it cannot encode all unicode strings, UCS-2 has become increasingly uncommon: http://en.wikipedia.org/wiki/UTF-16/UCS-2. UCS-2 is a strict subset of UTF-16 (see below).
Variable width encoding represents individual characters with different numbers of bytes. Thus a string representing a single code point with a variable width encoding could be 1 to 4 bytes long, depending on the code point it contained and the encoding you are using.
A common encoding for unicode is UTF-8; this is standard with most Linux distributions and many multilingual web pages: http://en.wikipedia.org/wiki/UTF-8. A single unicode code point can be represented by up to four bytes. Code points in the ascii range (0 to 7F) only need one byte, so this format is very space efficient for most western text.
UTF-16 uses one 16 bit value for characters in the BMP, but two 16 bit values for characters outside the BMP: http://en.wikipedia.org/wiki/UTF-16/UCS-2. The two 16 bit values are referred to as a 'surrogate pair' - see the surrogate pair entry in the Unicode glossary: http://www.unicode.org/glossary/#S. UTF-16 is the standard encoding for modern versions of windows.
UCS-2 (see above) is a struct subset of UTF-16 in that UTF-16 also uses only one 16-bit word value to encode all characters supported by UCS-2 (see "Q: What is the difference between UCS-2 and UTF-16?" in http://www.unicode.org/faq/utf_bom.html). This is possible because the first 16-bit character in UTF-16 surrogate pairs use the hex range D80016 to DBFF, and there are not defined unicode code points (characters) in this range.