Encode uses the Unicode Character Map (UCM) format for source character mappings. This format is used by IBM's ICU package and was adopted by Nick Ing-Simmons for use with the Encode module. Since UCM is more flexible than Tcl's Encoding Map and far more user-friendly, this is the recommended format for Encode now.
A UCM file looks like this.
# # Comments # <code_set_name> "US-ascii" # Required <code_set_alias> "ascii" # Optional <mb_cur_min> 1 # Required; usually 1 <mb_cur_max> 1 # Max. # of bytes/char <subchar> \x3F # Substitution char # CHARMAP <U0000> \x00 |0 # <control> <U0001> \x01 |0 # <control> <U0002> \x02 |0 # <control> .... <U007C> \x7C |0 # VERTICAL LINE <U007D> \x7D |0 # RIGHT CURLY BRACKET <U007E> \x7E |0 # TILDE <U007F> \x7F |0 # <control> END CHARMAP
- Anything that follows "#" is treated as a comment.
- The header section continues until a line containing the word CHARMAP. This section has a form of <keyword> value, one pair per line. Strings used as values must be quoted. Barewords are treated as numbers. \xXX represents a byte.
Most of the keywords are self-explanatory. subchar means substitution character, not subcharacter. When you decode a Unicode sequence to this encoding but no matching character is found, the byte sequence defined here will be used. For most cases, the value here is \x3F; in ASCII, this is a question mark.
- CHARMAP starts the character map section. Each line has a form as follows:
<UXXXX> \xXX.. |0 # comment ^ ^ ^ | | +- Fallback flag | +-------- Encoded byte sequence +-------------- Unicode Character ID in hex
The format is roughly the same as a header section except for the fallback flag: | followed by 0..3. The meaning of the possible values is as follows:
- Round trip safe. A character decoded to Unicode encodes back to the same byte sequence. Most characters have this flag.
- Fallback for unicode -> encoding. When seen, enc2xs adds this character for the encode map only.
- Skip sub-char mapping should there be no code point.
- Fallback for encoding -> unicode. When seen, enc2xs adds this character for the decode map only.
- And finally, END OF CHARMAP ends the section.
When you are manually creating a UCM file, you should copy ascii.ucm or an existing encoding which is close to yours, rather than write your own from scratch.
When you do so, make sure you leave at least U0000 to U0020 as is, unless your environment is EBCDIC.
CAVEAT: not all features in UCM are implemented. For example, icu:state is not used. Because of that, you need to write a perl module if you want to support algorithmical encodings, notably the ISO-2022 series. Such modules include Encode::JP::2022_JP, Encode::KR::2022_KR, and Encode::TW::HZ.
Coping with duplicate mappings
When you create a map, you SHOULD
make your mappings round-trip safe. That is, "encode('your-encoding', decode('your-encoding', $data)) eq $data"
stands for all characters that are marked as "|0"
. Here is how to make sure:
- Sort your map in Unicode order.
- When you have a duplicate entry, mark either one with '|1' or '|3'.
- And make sure the '|1' or '|3' entry FOLLOWS the '|0' entry.
Here is an example from big5-eten.
<U2550> \xF9\xF9 |0 <U2550> \xA2\xA4 |3
Internally Encoding -> Unicode and Unicode -> Encoding Map looks like this;
E to U U to E -------------------------------------- \xF9\xF9 => U2550 U2550 => \xF9\xF9 \xA2\xA4 => U2550
So it is round-trip safe for \xF9\xF9. But if the line above is upside down, here is what happens.
E to U U to E -------------------------------------- \xA2\xA4 => U2550 U2550 => \xF9\xF9 (\xF9\xF9 => U2550 is now overwritten!)
The Encode package comes with ucmlint, a crude but sufficient utility to check the integrity of a UCM file. Check under the Encode/bin directory for this.
When in doubt, you can use ucmsort, yet another utility under Encode/bin directory.