Binary Ordered Compression For Unicode - Details

Details

All numbers in this section are hexadecimal, and all ranges are inclusive.

Code points from U+0000 to U+0020 are encoded in BOCU-1 as the corresponding byte value. All other code points (that is, U+0021 through U+D7FF and U+E000 through U+10FFFF) are encoded as a difference between the code point and a normalized version of the most recently encoded code point that was not an ASCII space (U+0020). The initial state is U+0040. The normalization mapping is as follows:

The difference between the current code point and the normalized previous code point is encoded as follows:

Each byte range is lexicographically ordered with the following thirteen byte values excluded: 00 07 08 09 0A 0B 0C 0D 0E 0F 1A 1B 20. For example, the byte sequence FC 06 FF, coding for a difference of 1156B, is immediately followed by the byte sequence FC 10 01, coding for a difference of 1156C.

Any ASCII input U+0000 to U+007F excluding space U+0020 resets the encoder to U+0040. Because the above mentioned values cover line end code points U+000D and U+000A as is (0D 0A), the encoder is in a known state at the begin of each line. The corruption of a single byte therefore affects at most one line. For comparison, the corruption of a single byte in UTF-8 affects at most one code point, for SCSU it can affect the entire document.

BOCU-1 offers a similar robustness also for input texts without the above mentioned values with the special reset code 0xFF. When a decoder finds this octet it resets its state to U+0040 as for a line end. The use of 0xFF reset bytes is not recommended in the BOCU-1 specification, because it conflicts with other BOCU-1 design goals, notably the binary order.

The optional use of a signature U+FEFF at the begin of BOCU-1 encoded texts, i.e. the BOCU-1 byte sequence FB EE 28, changes the initial state U+0040 to U+FE80. In other words the signature cannot simply be stripped as in most other Unicode encoding schemes. Adding a reset byte after the signature (FB EE 28 FF) could avoid this effect, but the BOCU-1 specification does not recommend this practice.

In theory UTF-1 and UTF-8 could encode the original UCS-4 set with 31 bits up to 7FFFFFFF. BOCU-1 and UTF-16 can encode the modern Unicode set from U+0000 to U+10FFFF. Excluding the thirteen protected code points encoded as single octets BOCU-1 can use octets in multi-byte encodings. BOCU-1 needs at most four bytes consisting of a lead byte and one to three trail bytes. The trail bytes encode a remaining "modulo 243" (base 243) difference, the lead byte determines the number of trail bytes and an initial difference. Note that the reset byte 0xFF is not protected and can occur as trail byte.