Representation of ISO 8859-1 characters with 7-bit ASCII -------------------------------------------------------- Markus Kuhn -- 1993-02-20 SUMMARY: This text describes a technique of displaying the 8-bit character set, which is used today in many modern network services, on old 7-bit terminals. Authors of software dealing with text received from international networks are strongly encouraged to implement this or similar methods as options in their software for the convenience of users all over the world. Implementation is often trivial. The "Latin alphabet No. 1" defined in part 1 of the international standard ISO 8859:1987 Information processing -- 8-bit single-byte coded graphic character sets is an increasingly popular 8-bit extension of the traditional 7-bit US-ASCII character set. It is already supported by many operating systems and its 191 graphic characters include those used in at least the following 14 languages (and many others): Danish, Dutch, English, Faeroese, Finnish, French, German, Icelandic, Irish, Italian, Norwegian, Portuguese, Spanish and Swedish. ISO 8859-1 contains graphic characters used in at least 44 countries. ISO Latin 1 is already the de-facto replacement of the old 7-bit US-ASCII character set and its national ISO 646 variants. In addition, the first 256 characters of the new 16-bit character set ISO 10646/Unicode, which will eventually contain all characters used on this planet and is expected to be the final solution of most of today's character set troubles, are identical with ISO 8859-1. ISO 8859-1 uses only the codes 32-126 (which are identical with US-ASCII) and 160-255. The positions 0-31 and 127-159 are reserved for control characters and normally used in the same way in which they are used with ASCII. By the way: Only two of the characters have a special meaning for programs that allow paragraph reformatting. Character NBSP (no-break space) number 160 (0xa0 = ' '+0x80) looks like a normal space and should be used if a line break is to be prevented at this space in the text when it is formatted. Character SHY (soft hyphen) at position 173 (0xad = '-'+0x80) looks similar to or exactly like the normal hyphen ('-') and should be used when a line break has been established within a word. In this way, SHY can easily be removed again by an editor while reformatting a paragraph, because soft hyphens (0xad) that have only been inserted for line breaks can be distinguished from real hyphens (0x2d) that are a permanent part of the text. Both NBSP and SHY are part of all ISO 8859 character sets. As the ISO Latin 1 character set gains more and more popularity in international data communication (e.g. the Internet gopher service, the Internet MIME, parts of USENET), the need arises to extend existing software with the ability of displaying strings containing ISO 8859-1 characters on old hardware that is only capable of displaying 7-bit US-ASCII characters. Today, many users of old hardware suffer from getting the Latin 1 characters between 160 and 255 only displayed as the corresponding US-ASCII characters with the highest bit cleared. Then they see e.g. a ')' instead of the copyright symbol. Pessimists expect that these old 7-bit terminals will be in use at least for the next ten years. One approach for a Latin 1 to ASCII conversion is to use the replacements that people commonly use when they have to live with a system supporting a too limited character set. This seems to be the most natural method, which often won't even be noticed by users that use these traditional replacements already today on their old hardware. Of course, there are some disadvantages of this approach (compared to buying a new terminal), but these are often acceptable if the software today simply destroys the characters by clearing the highest bit of the received bytes. These are: a) No one-to-one mapping between Latin 1 and ASCII strings is possible. b) Text layout may be destroyed by multi-character substitutions, especially in tables. c) Different replacements may be in use for different languages, so no single standard replacement table will make everyone happy. d) Truncation or line wrapping might be necessary to fit textual data into fields of fixed width. There is no optimal solution possible for the problem of displaying text with ISO Latin 1 characters on old terminals apart from buying new hardware. The conversion tables proposed here are only intermediate solutions that are intended to make life easier for people who get Latin 1 characters currently displayed as the corresponding 7-bit US-ASCII symbols with the highest bit cleared, which is awful and frustrates the users of old hardware. Including the tables below in programs like mail user agents, news readers, gopher clients, file browsers, tty drivers etc. is often a trivial task. Users should be able to switch between the different tables and the 8-bit transparent normal mode. While I discussed these tables with people from many nations in USENET, it became obvious, that there are a lot of differences in the personal and cultural preferences for the substitution tables. Much too many tables would have been necessary to make everyone 100% happy. So I decided to keep the number of tables as small as possible and tried to cover only the most important cultural and application dependent differences. The tables below will perhaps be all right for 80% of the users. If you as a programmer want to avoid long discussions about the details of the tables with your users, then offer them a feature to define their own tables, perhaps in the form of changes to the default tables listed below (or give at least a pointer in the source code of public domain software, where user-defined tables might be modified for local needs). Users should know if the text they read has been converted from the original Latin 1 text, i.e. the conversion should be clearly explained in the documentation and perhaps again noticed e.g. after the program starts. Otherwise, the conversion might cause confusion in some cases. I collected six tables based on information I received from many USENET readers from various countries in order to cope with the different needs of ISO Latin 1 users. In some cases, different replacements might seem to be more suitable based on the semantics of the characters and I received may suggestions of this kind, but I decided to selected the replacements based on the way in which these characters might be used, which differs often dramatically from the originally intended semantics of the characters. Consequently, I always preferred graphically similar replacements, where the field of application of the character did not seem to be very limited. E.g. it has been suggested to replace the 'left angle quotation mark' [«] by '"' instead of '<' in table 1 based on the common semantic 'quotation mark', but this character is also often used as a kind of arrow, so a graphically similar replacement was chosen. Other characters with more limited applications like the 'small German letter sharp s' [ß] were replaced by the most often used replacements (e.g. 'ss') instead of graphically more similar characters like '3' or 'B'. First of all, a table with the real characters in the range 160 - 255 (0xa0 - 0xff):   ¡ ¢ £ € ¥ Š § š © ª « ¬ ­ ® ¯ ° ± ² ³ Ž µ ¶ · ž ¹ º » Œ œ Ÿ ¿ À Á Â Ã Ä Å Æ Ç È É Ê Ë Ì Í Î Ï Ð Ñ Ò Ó Ô Õ Ö × Ø Ù Ú Û Ü Ý Þ ß à á â ã ä å æ ç è é ê ë ì í î ï ð ñ ò ó ô õ ö ÷ ø ù ú û ü ý þ ÿ Table 0 is a universal table that is expected to be suitable for many languages. The letters are simply the ASCII versions without the diacritics. The fallback substitution character (e.g. '?' or '_') as an emergency replacement character where no ASCII string is suitable is used as little as possible, as it carries no information and if we are pedantic, we have to replace nearly every Latin 1 character over 160 by question marks etc. ! c ? ? Y | ? " (c) a << - - (R) - +/- 2 3 ' u P . , 1 o >> 1/4 1/2 3/4 ? A A A A A A AE C E E E E I I I I D N O O O O O x O U U U U Y Th ss a a a a a a ae c e e e e i i i i d n o o o o o : o u u u u y th y Table 1 replaces Latin 1 characters only with single ASCII characters. This won't destroy the layout of texts designed to be printed with monospaced fonts, but the replacements are often not very satisfactory: ! c ? ? Y | ? " c a < - - R - ? 2 3 ' u P . , 1 o > ? ? ? ? A A A A A A A C E E E E I I I I D N O O O O O x O U U U U Y T s a a a a a a a c e e e e i i i i d n o o o o o : o u u u u y t y In some languages, only removing the diacritics as in table 0 gives orthographically incorrect and unappropriate results. The following table 2 might be much more suitable than table 0 in Danish, Dutch, German, Norwegian and Swedish: ! c ? ? Y | ? " (c) a << - - (R) - +/- 2 3 ' u P . , 1 o >> 1/4 1/2 3/4 ? A A A A Ae Aa AE C E E E E I I I I D N O O O O Oe x Oe U U U Ue Y Th ss a a a a ae aa ae c e e e e i i i i d n o o o o oe : oe u u u ue y th ij In some North-European languages, any US-ASCII replacement for the relevant Latin 1 characters is unacceptable for many people. In these countries, national variants of 7-bit ISO 646 are still in wide use. They use consistently some or all of the characters [ ] \ { } | $ and in one Swedish character set also ~ ^ ` @ for national characters. Table 3 has been designed for Danish, Finnish, Norwegian and Swedish users of ISO 646 terminals: ! c ? $ Y | ? " (c) a << - - (R) - +/- 2 3 ' u P . , 1 o >> 1/4 1/2 3/4 ? A A A A [ ] [ C E @ E E I I I I D N O O O O \ x \ U U U ^ Y Th ss a a a a { } { c e ` e e i i i i d n o o o o | : | u u u ~ y th y Perhaps some users might prefer for four characters the strings from table 2 instead of ~ ^ ` @, which are only used in one Swedish character set. Instead of adding yet another table, take this as a motivation for allowing user-defined modifications to the tables. In RFC 1345, each character from Latin 1 (and from many other character sets) is assigned a two-character ASCII mnemonic. Table 4 encloses these mnemonics in brackets. The resulting conversion looses nearly no information and might be useful in special applications, where the risk of confusing the reader by the Latin1 to ASCII conversion weights more than the risk of producing ugly output. [NS][!I][Ct][Pd][Cu][Ye][BB][SE][':][Co][-a][<<][NO][--][Rg]['-] [DG][+-][2S][3S][''][My][PI][.M][',][1S][-o][>>][14][12][34][?I] [A!][A'][A>][A?][A:][AA][AE][C,][E!][E'][E>][E:][I!][I'][I>][I:] [D-][N?][O!][O'][O>][O?][O:][*X][O/][U!][U'][U>][U:][Y'][TH][ss] [a!][a'][a>][a?][a:][aa][ae][c,][e!][e'][e>][e:][i!][i'][i>][i:] [d-][n?][o!][o'][o>][o?][o:][-:][o/][u!][u'][u>][u:][y'][th][y:] The encoding offered by table 4 is still not 100% free of loss of information. If you see a '[Co]' in the text, then this might have been both a copyright sign and the string '[Co]'. To avoid this ambiguity, one might implement the encoding '&Co' for the copyright sign and '&&' as an escape string for a single '&' as suggested in RFC 1345. This is not really appropriate in most situations, because even pure ASCII texts (e.g. C programs) with '&'s will then be changed. The following table 5 (based on one suggested by Peter da Silva) is perhaps more a nice intellectual exercise than something really useful. It uses the BACKSPACE control character (in the table represented by '@') in order to get new characters by overstriking ASCII characters. This gives very poor results for the capital letters on many printers and is useless on most video terminals, but might be interesting for languages where often only lowercase characters are used accented (e.g. French). The quality of the results depends very much on the type of printer used. ! c@| L@- o@X Y@= | ? " (c) a@_ << -@, - (R) - +@_ 2 3 ' u P . , 1 o@_ >> 1/4 1/2 3/4 ? A@` A@' A@^ A@~ A@" Aa AE C@, E@` E@' E@^ E@" I@` I@' I@^ I@" D@- N@~ O@` O@' O@^ O@~ O@" x O@/ U@` U@' U@^ U@" Y@' Th ss a@` a@' a@^ a@~ a@" aa ae c@, e@` e@' e@^ e@" i@` i@' i@^ i@" d@- n@~ o@` o@' o@^ o@~ o@" -@: o@/ u@` u@' u@^ u@" y@' th y@" For the convenience of C programmers, I included the code of these tables in this text. Just copy the following lines into your software: ----------------------------------------------------------------------- /* Conversion tables for displaying the G1 set (0xa0-0xff) of ISO Latin 1 (ISO 8859-1) with 7-bit ASCII characters. Version 1.1 -- error corrections are welcome Table Purpose 0 universal table for many languages 1 single-spacing universal table 2 table for Danish, Dutch, German, Norwegian and Swedish 3 table for Danish, Finnish, Norwegian and Swedish using the appropriate ISO 646 variant. 4 table with RFC 1345 codes in brackets 5 table for printers that allow overstriking with backspace Markus Kuhn */ #define SUB "?" /* used if no reasonable ASCII string is possible */ #define ISO_TABLES 6 static char *iso2asc[ISO_TABLES][96] = {{ " ","!","c",SUB,SUB,"Y","|",SUB,"\"","(c)","a","<<","-","-","(R)","-", " ","+/-","2","3","'","u","P",".",",","1","o",">>"," 1/4"," 1/2"," 3/4","?", "A","A","A","A","A","A","AE","C","E","E","E","E","I","I","I","I", "D","N","O","O","O","O","O","x","O","U","U","U","U","Y","Th","ss", "a","a","a","a","a","a","ae","c","e","e","e","e","i","i","i","i", "d","n","o","o","o","o","o",":","o","u","u","u","u","y","th","y" },{ " ","!","c",SUB,SUB,"Y","|",SUB,"\"","c","a","<","-","-","R","-", " ",SUB,"2","3","'","u","P",".",",","1","o",">",SUB,SUB,SUB,"?", "A","A","A","A","A","A","A","C","E","E","E","E","I","I","I","I", "D","N","O","O","O","O","O","x","O","U","U","U","U","Y","T","s", "a","a","a","a","a","a","a","c","e","e","e","e","i","i","i","i", "d","n","o","o","o","o","o",":","o","u","u","u","u","y","t","y" },{ " ","!","c",SUB,SUB,"Y","|",SUB,"\"","(c)","a","<<","-","-","(R)","-", " ","+/-","2","3","'","u","P",".",",","1","o",">>"," 1/4"," 1/2"," 3/4","?", "A","A","A","A","Ae","Aa","AE","C","E","E","E","E","I","I","I","I", "D","N","O","O","O","O","Oe","x","Oe","U","U","U","Ue","Y","Th","ss", "a","a","a","a","ae","aa","ae","c","e","e","e","e","i","i","i","i", "d","n","o","o","o","o","oe",":","oe","u","u","u","ue","y","th","ij" },{ " ","!","c",SUB,"$","Y","|",SUB,"\"","(c)","a","<<","-","-","(R)","-", " ","+/-","2","3","'","u","P",".",",","1","o",">>"," 1/4"," 1/2"," 3/4","?", "A","A","A","A","[","]","[","C","E","@","E","E","I","I","I","I", "D","N","O","O","O","O","\\","x","\\","U","U","U","^","Y","Th","ss", "a","a","a","a","{","}","{","c","e","`","e","e","i","i","i","i", "d","n","o","o","o","o","|",":","|","u","u","u","~","y","th","y" },{ "[NS]","[!I]","[Ct]","[Pd]","[Cu]","[Ye]","[BB]","[SE]", "[':]","[Co]","[-a]","[<<]","[NO]","[--]","[Rg]","['-]", "[DG]","[+-]","[2S]","[3S]","['']","[My]","[PI]","[.M]", "[',]","[1S]","[-o]","[>>]","[14]","[12]","[34]","[?I]", "[A!]","[A']","[A>]","[A?]","[A:]","[AA]","[AE]","[C,]", "[E!]","[E']","[E>]","[E:]","[I!]","[I']","[I>]","[I:]", "[D-]","[N?]","[O!]","[O']","[O>]","[O?]","[O:]","[*X]", "[O/]","[U!]","[U']","[U>]","[U:]","[Y']","[TH]","[ss]", "[a!]","[a']","[a>]","[a?]","[a:]","[aa]","[ae]","[c,]", "[e!]","[e']","[e>]","[e:]","[i!]","[i']","[i>]","[i:]", "[d-]","[n?]","[o!]","[o']","[o>]","[o?]","[o:]","[-:]", "[o/]","[u!]","[u']","[u>]","[u:]","[y']","[th]","[y:]" },{ " ","!","c\b|","L\b-","o\bX","Y\b=","|",SUB, "\"","(c)","a\b_","<<","-\b,","-","(R)","-", " ","+\b_","2","3","'","u","P",".", ",","1","o\b_",">>"," 1/4"," 1/2"," 3/4","?", "A\b`","A\b'","A\b^","A\b~","A\b\"","Aa","AE","C\b,", "E\b`","E\b'","E\b^","E\b\"","I\b`","I\b'","I\b^","I\b\"", "D\b-","N\b~","O\b`","O\b'","O\b^","O\b~","O\b\"","x", "O\b/","U\b`","U\b'","U\b^","U\b\"","Y\b'","Th","ss", "a\b`","a\b'","a\b^","a\b~","a\b\"","aa","ae","c\b,", "e\b`","e\b'","e\b^","e\b\"","i\b`","i\b'","i\b^","i\b\"", "d\b-","n\b~","o\b`","o\b'","o\b^","o\b~","o\b\"","-\b:", "o\b/","u\b`","u\b'","u\b^","u\b\"","y\b'","th","y\b\"" }}; ----------------------------------------------------------------------- One might perhaps replace the "?" in SUB with "_" or another code that will be displayed as a blinking question mark, a filled block or something similar. Then the user will know that the software wants to tell him/her that it can't display this symbol and that it is not a question mark. If your software runs on hardware that supports already another 8-bit characters set (e.g. IBM PC with code page 437, Mac, etc.), then it might be a much better idea to include only one single table that uses the supported symbols wherever possible and uses the strings suggested here only if no better alternative is available. For instance, a monospaced table for displaying Latin 1 strings on a MS-DOS computer might look like this: ----------------------------------------------------------------------- /* ISO Latin 1 to IBM code page 437 (classic IBM PC character set) */ unsigned char iso2ibm[96] = { 255,173,155,156,'o',157,'|', 21,'"','c',166,174,170,'-','R','-', 248,241,253,'3', 39,230, 20,249,',','1',167,175,172,171,'?',168, 'A','A','A','A',142,143,146,128,'E',144,'E','E','I','I','I','I', 'D',165,'O','O','O','O',153,'x',237,'U','U','U',154,'Y','T',225, 133,160,131,'a',132,134,145,135,138,130,136,137,141,161,140,139, 'd',164,149,162,147,'o',148,246,237,151,163,150,129,'y','t',152 }; ----------------------------------------------------------------------- (BTW: IBM code page 850 which is supported by MS-DOS and OS/2 contains ALL Latin 1 characters, but at other positions, in order to stay compatible with the old IBM PC character set.) The following string conversion routine uses these tables. It may easily be called before a text received from the network is sent to the terminal, if the user has selected one of the tables: ----------------------------------------------------------------------- /* * Transform an 8-bit ISO Latin 1 string iso into a 7-bit ASCII string asc * readable on old terminals using conversion table t. * * worst case: strlen(iso) == 4*strlen(asc) */ void Latin1toASCII(iso, asc, t) unsigned char *iso, *asc; int t; { char *p, **tab; if (iso==NULL || asc==NULL) return; tab = iso2asc[t] - 0xa0; while (*iso) { if (*iso > 0x9f) { p = tab[*(iso++)]; while (*p) *(asc++) = *(p++); } else { *(asc++) = *(iso++); } } *asc = 0; return; } ----------------------------------------------------------------------- A more sophisticated function that tries to correct column shifts caused by multi-character replacements by removing SPACEs and TABs gives often excellent results even in tables. The following function removes SPACEs and TABs during string conversion only where necessary, so pure 7-bit strings won't be changed at all. That's been a nice programming exercise, by the way ... :-) ----------------------------------------------------------------------- /* * Transform an 8-bit ISO Latin 1 string iso into a 7-bit ASCII string asc * readable on old terminals using conversion table t. Remove SPACE and * TAB characters where appropriate, in order to preserve the layout * of tables, etc. as much as possible. * * worst case: strlen(iso) == 4*strlen(asc) */ void CorLatin1toASCII(iso, asc, t) unsigned char *iso, *asc; int t; { char *p, **tab; int first; /* flag for first SPACE/TAB after other characters */ int i, a; /* column counters in iso and asc */ /* TABSTOP(x) is the column of the character after the TAB at column x. First column is 0, of course. */ # define TABSTOP(x) (((x) - ((x)&7)) + 8) if (iso==NULL || asc==NULL) return; tab = iso2asc[t] - 0xa0; first = 1; i = a = 0; while (*iso) { if (*iso > 0x9f) { p = tab[*(iso++)]; i++; first = 1; while (*p) { *(asc++) = *(p++); a++; } } else { if (a > i && ((*iso == ' ') || (*iso == '\t'))) { /* spaces or TABS should be removed */ if (*iso == ' ') { /* only the first space after a letter must not be removed */ if (first) { *(asc++) = ' '; a++; first = 0; } i++; } else { /* here: *iso == '\t' */ if (a >= TABSTOP(i)) { /* remove TAB or replace it with SPACE if necessary */ if (first) { *(asc++) = ' '; a++; first = 0; } } else { /* TAB will correct the column difference */ *(asc++) = '\t'; /* = *iso */ a = TABSTOP(a); /* = TABSTOP(i), because i < a < TABSTOP(i) */ } i = TABSTOP(i); } iso++; } else { /* just copy the characters and advance the column counters */ if ((*(asc++) = *(iso++)) == '\t') { a = i = TABSTOP(i); /* = TABSTOP(a), because here a = i */ } else { a++; i++; } first = 1; } } } *asc = 0; return; } ----------------------------------------------------------------------- As a software author, you might decide to offer one of several levels of Latin 1 conversion support: - The simplest solution is to allow the user to switch between the real 8-bit representation and the above tables - Highly recommended is a feature that allows the user to create his own table. If this is possible based on one or more of the described default tables, the effort needed for defining a private table will be reduced drastically. The system administrator should be allowed to define a default table for his users. - More comfortable systems might also allow the user to change the SUB string, to select the style (normal, highlighted, underlined, blinking, ...) in which the replacement strings are displayed, etc. - You might even think about possibilities for a user to enter Latin 1 characters with an old keyboard and editor, a problem that hasn't been addressed here. Many users all over the world are looking forward to your next software release that will allow them to participate without pain in the world of 8-bit character communication even before they get modern hardware with ISO 8859-1 (or even better ISO 10646) character sets. Feel free to contact me or experts in USENET group comp.std.internat if you have any questions about modern character sets. Many thanks to everyone from comp.std.internat who helped me to improve these tables! Markus