Universal Multiple-Octet Coded Character Set
(U C S)

Date: 2000-03-07


Implications of Normalization on Character Encoding


Unicode Technical Committee


Liaison contribution


Request for addition to JTC1/SC2/WG2 Principles and Procedures

As ISO/IEC 10646 / Unicode has become more prevalent in implementations and other standards, it has become necessary to produce very stable specifications for the comparison of text. In particular, a unique, normalized form of text is required for comparisons in domain names, XML element names, and other areas where a precise, stable, comparison of strings is required. Programs that require uniqueness also require forward compatibility: programs all over the web must be able to depend on the unique format not changing over time. The Unicode Consortium requests that a principle to be added to the WG2 Principles and Procedures that addresses this issue of stability.

There are characters that are equivalently represented either as sequences of code points or as a single code point (called a composite character). For example, the i with 2 dots in nave could be presented either as i + diaeresis (0069 0308) or as the composite character i-diaeresis (00EF). There are other cases where the order of two combining characters is irrelevant. For example, the pair of combining characters acute and dot-below can occur with either one first; both alternate orders are equivalent.

In response to the need for a unique form, the Unicode Consortium has produced an exact algorithmic specification of normalized forms. (For more information, see UTR #15: Unicode Normalization Forms.) One of these forms, Normalization Form C, is designed to favor precomposed characters such as over combining character sequences such as a + ~. The W3C Character Model for the World Wide Web requires the use of Normalization Form C for XML and related standards (this document is not yet final, but this requirement is not expected to change). See also the W3C Requirements for String Identity Matching and String Indexing for more background. We expect that the number of standards and implementations requiring normalization will continue to grow.

Such implementations must produce precisely the same result for normalization even if they upgrade to a new version of Unicode / 10646. Thus it is necessary to specify a fixed version for the composition process, called the composition version. The composition version is defined to be Version 3.0.0 of  the Unicode Character Database, which corresponds to ISO 10646-1:2000.

To see what difference the composition version makes, suppose that Unicode 4.0 / 10646:2002 adds the composite Q-caron. For an implementation that uses  Unicode 4.0 / 10646:2002, strings in Normalization Forms C or KC will continue to contain the sequence Q + caron, and not the new character Q-caron, since a canonical composition for Q-caron was not defined in the composition version.

The implications for encoding new characters are that new precomposed characters are important to recognize. If Q WITH CARON were added to Unicode 4.0 / 10646-1:2002, then it would represent a duplicate encoding. This could be tolerated before Unicode 3.0 because canonical equivalence could be used to equate the two forms. But due to the need for stability in comparison by so much of the world's infrastructure, this situation cannot be tolerated in the future. For stability, characters that are currently representable as sequences will always stay representable only as sequences. These include the following examples:

Character Code Point Sequence Comments
ch 0063 0068 Slovak, traditional Spanish
th 0074 02B0 Native American languages
0078 0323
019B 0313
00E1 0328 Lithuanian:



0069 0307 0301
30C8 309A Ainu in kana transcription

Moreover, the need for separate precomposed characters is diminishing quickly. The major GUI vendors are currently in the process of upgrading their systems to handle both surrogates and accurate positioning of combining marks, with such technologies as OpenType and AAT. By the time new precomposed characters could be added, there would be little need for them.

Precomposed characters that cannot be represented by a combining character sequence in ISO 10646-1:2000 / Unicode 3.0 can be added to the standard without affecting normalization.

Note that the future addition of certain combining characters to the standard would also have the effect of destabilizing normalized text, if they could be used to form new combining character sequences that are equivalent to existing precomposed characters. In the same way as new precomposed characters, the addition of such characters must be restricted. For example, a new character encoding the "descender" of various Cyrillic characters must not be added, since that would impact the normalization of the relevant Cyrillic characters, rendering existing normalized data invalid.

In view of these issues, the Unicode Consortium requests that a principle to be added to the WG2 Principles and Procedures document that states that characters will not be encoded which disturb the stability of Normalization Form C. Such characters include certain precomposed characters and combining marks, as described above.