UTF-EBCDIC is a character encoding capable of encoding all 1,112,064 valid character code points in Unicode using one to five one-byte (8-bit) code units (in contrast to a maximum of four for UTF-8).[1] It is meant to be EBCDIC-friendly, so that legacy EBCDIC applications on mainframes may process the characters without much difficulty. Its advantages for existing EBCDIC-based systems are similar to UTF-8's advantages for existing ASCII-based systems. Details on UTF-EBCDIC are defined in Unicode Technical Report #16.

To produce the UTF-EBCDIC encoded version of a series of Unicode code points, an encoding based on UTF-8 (known in the specification as UTF-8-Mod) is applied first (creating what the specification calls an I8 sequence). The main difference between this encoding and UTF-8 is that it allows Unicode code points U+0080 through U+009F (the C1 control codes) to be represented as a single byte and therefore later mapped to corresponding EBCDIC control codes. In order to achieve this, UTF-8-Mod uses 101XXXXX instead of 10XXXXXX as the format for trailing bytes in a multi-byte sequence. As this can only hold 5 bits rather than 6, the UTF-8-Mod encoding of codepoints above U+03FF are larger than the UTF-8 encoding.

The UTF-8-Mod transformation leaves the data in an ASCII-based format (for example, U+0041 "A" is still encoded as 01000001), so each byte is fed through a reversible (one-to-one) lookup table to produce the final UTF-EBCDIC encoding. For example, 01000001 in this table maps to 11000001; thus the UTF-EBCDIC encoding of U+0041 (Unicode's "A") is 0xC1 (EBCDIC's "A").

This encoding form is rarely used, even on the EBCDIC-based mainframes for which it was designed. IBM EBCDIC-based mainframe operating systems, such as z/OS, usually use UTF-16 for complete Unicode support. For example, IBM Db2, COBOL, PL/I, Java and the IBM XML toolkit support UTF-16 on IBM mainframes.

Codepage layout

There are 160 characters with single-byte encodings in UTF-EBCDIC (compared to 128 in UTF-8). As can be seen, the single-byte portion is similar to IBM-1047 instead of IBM-37 due to the location of the square brackets. CCSID 37 has [] at hex BA and BB instead of at hex AD and BD respectively.

UTF-EBCDIC
0 1 2 3 4 5 6 7 8 9 A B C D E F
0x NUL SOH STX ETX ST HT SSA DEL EPA RI SS2 VT FF CR SO SI
1x DLE DC1 DC2 DC3 OSC LF BS ESA CAN EM PU2 SS3 FS GS RS US
2x PAD HOP BPH NBH IND NEL ETB ESC HTS HTJ VTS PLD PLU ENQ ACK BEL
3x DCS PU1 SYN STS CCH MW SPA EOT SOS SGCI SCI CSI DC4 NAK PM SUB
4x  SP  . < ( + |
5x & ! $ * ) ; ^
6x - / , % _ > ?
7x 2 2 2 2 2 ` : # @ ' = "
8x 2 a b c d e f g h i 2 2 2 2 2 2
9x 2 j k l m n o p q r 2 2 2 2 2 2
Ax 2 ~ s t u v w x y z 2 2 2 [ 2 2
Bx 2 2 2 2 2 2 2 3 3 3 3 3 3 ] 3 3
Cx { A B C D E F G H I 3 3 3 3 3 3
Dx } J K L M N O P Q R 3 3 4 4 4 4
Ex \ 4 S T U V W X Y Z 4 4 4 5 5
Fx 0 1 2 3 4 5 6 7 8 9 APC
  Start bytes for a sequence of that many bytes. Tooltip shows the lowest code point encoded using that start byte.
  Start byte where not all combinations of continuation bytes are valid, either because it is an invalid overlong form (the tooltip shows the code point of the first valid sequence), or because it encodes a code point greater than U+10FFFF.
  Continuation bytes. Tooltip shows the hexadecimal value of the 5 bits they add.
  Unused, including lead bytes that can only start an invalid overlong form. For example, 0x76 because even 0x76 0x73 (which maps to the UTF-8-Mod sequence 0xC2 0xBF) would merely be an overlong encoding of U+005F (properly encoded as UTF-8-Mod 0x5F, UTF-EBCDIC 0x6D).

Oracle UTFE

Oracle UTFE is a Unicode 3.0 UTF-8 Oracle database variation, similar to the CESU-8 variant of UTF-8, where supplementary characters are encoded as two 4-byte characters rather than a single 4- or 5-byte character. It is used only on EBCDIC platforms.[2]

Advantages:

Disadvantages:

Mksh nega-UTF-8

Mksh is a POSIX shell with a “utf8-mode” that complicates porting to EBCDIC. Instead of implementing UTF-EBCDIC (which is “not used by anyone”), the developers found it easier to just use normal UTF-8 and convert it to EBCDIC by pretending it’s the “extended ASCII” encoding for the current code page. Doing so destroys EBCDIC control characters not found in ASCII, however.[3]

See also

References

  1. ^ "UTR #16: UTF-EBCDIC". www.unicode.org. Retrieved 2021-02-23. You need to search at most five bytes (seven bytes, if the full range of 31 bits of ISO/IEC 10646 is considered) backwards
  2. ^ a b c Baird, Cathy; Chiba, Dan; Chu, Winson; Fan, Jessica; Ho, Claire; Law, Simon; Lee, Geoff; Linsley, Peter; Matsuda, Keni; Oscroft, Tamzin; Takeda, Shige; Tanaka, Linus; Tozawa, Makoto; Trute, Barry; Tsujimoto, Mayumi; Wu, Ying; Yau, Michael; Yu, Tim; Wang, Chao; Wong, Simon; Zhang, Weiran; Zheng, Lei; Zhu, Yan; Moore, Valarie (2002) [1996]. "Appendix A: Locale Data". Oracle9i Database Globalization Support Guide (PDF) (Release 2 (9.2) ed.). Oracle Corporation. Oracle A96529-01. Archived (PDF) from the original on 2017-02-14. Retrieved 2017-02-14.
  3. ^ "MirBSD/MKSH". GitHub. 4 November 2021. EBCDIC interacts with utf8-mode in a special way: UTF-EBCDIC is not used by anyone, we cannot use UCS-2/4 in the shell, so “nega-UTF-8”, a scheme in which UTF-8 calculations are done on the octets converted per the current EBCDIC codepage to “extended ASCII”.