The SQL Anywhere Collation Algorithm provides reasonable comparison, ordering, and case conversion of single-byte and multibyte character sets. The algorithm is space efficient and fast. The mapped form of a string, such as an index, is the same length as the original string. The mappings for comparison, ordering, and case conversion use a simple table lookup of each byte value of the string.

The SACA has been provided with SQL Anywhere since its early days as Watcom SQL.

Single-byte character sets

In a typical collation for a single-byte character set, all accented and unaccented forms of a character are mapped to the same value, making the collation accent insensitive. Accented and unaccented forms of the same letter compare as exactly equal and sort near each other.

The collation also provides conversion between uppercase and lowercase letters, preserving accents.

Multibyte character sets

In multibyte character sets, the lead-bytes are mapped into the 256 distinct values. Follow bytes are compared using their binary value.

For most collations for multibyte character sets, this mapping technique provides a reasonable ordering because the character set encoding groups characters into 256-byte pages identified by the lead byte. The pages, and the characters within each page, are in a reasonable order in the character set. The collations typically preserve the ordering of the pages (lead bytes) within the character set. Some pages may be ordered by other characteristics. For example, the 932JPN collation provided for Japanese code page 932 groups the full-width (Kanji) and half-width (katakana) characters.

Case conversion is provided only for the 7-bit English characters.

UTF-8 character sets

UTF-8 is a multibyte character set. Each character contains from one to four bytes. SQL Anywhere provides the UTF8BIN collation for sorting UTF-8 characters.

In UTF8BIN, lead bytes are mapped into 256 distinct values, and follow bytes are compared using their binary values. Because of the representation of characters in UTF-8 and the limitation of 256 distinct mapping values, it is not possible to group related characters such as accented and unaccented forms of the same letter. The ordering is essentially binary.

Case conversion is supported only for the 7-bit English characters.