This repository has been archived on 2023-08-20. You can view files and clone it, but cannot push or open issues or pull requests.
yap-6.3/os/encoding.h
Vítor Santos Costa 484168b1ce Fix encoding
2016-02-18 12:10:58 +00:00

182 lines
4.7 KiB
C

/*************************************************************************
* *
* YAP Prolog %W% %G%
* *
* Yap Prolog was developed at NCCUP - Universidade do Porto *
* *
* Copyright L.Damas, V.S.Costa and Universidade do Porto 1985-2003 *
* *
**************************************************************************
* *
* File: yapio.h *
* Last rev: 22/1/03 *
* mods: *
* comments: UNICODE encoding support (based on SWI-Prolog) *
* *
*************************************************************************/
#ifndef ENCODING_H
#define ENCODING_H 1
typedef enum {
ENC_OCTET = 0, /// binary files
ENC_ISO_LATIN1 = 1, /// US+West Europe
ENC_ISO_ASCII = 2, /// US only
ENC_ISO_ANSI = 4, /// Who cares
ENC_ISO_UTF8 = 8, /// Most everyone nowadays
ENC_UTF16_BE = 16, /// People who made a mistake
ENC_UTF16_LE = 32, /// People who made the same mistake
ENC_ISO_UTF32_BE = 64, /// nobody
ENC_ISO_UTF32_LE = 128, /// yes, nobody
} encoding_t;
#if WORDS_BIGENDIAN
#define ENC_WCHAR ENC_ISO_UTF32_BE
#else
#define ENC_WCHAR ENC_ISO_UTF32_LE
#endif
#ifdef YAP_H
/// read the current environment, as set by the user or as Initial
encoding_t Yap_DefaultEncoding(void);
void Yap_SetDefaultEncoding(encoding_t new_encoding);
#if HAVE_XLOCALE_H
typedef enum {
SEQ_ENC_OCTET, /// binary files
SEQ_ENC_ISO_LATIN1, /// US+West Europe
SEQ_ENC_ISO_ASCII, /// US only
SEQ_ENC_ISO_ANSI, /// Who cares
SEQ_ENC_ISO_UTF8, /// Most everyone nowadays
SEQ_ENC_UTF16_BE, /// People who made a mistake
SEQ_ENC_UTF16_LE, /// People who made the same mistake
v\ SEQ_ENC_ISO_UTF32_BE, /// nobody
SEQ_ENC_ISO_UTF32_LE /// yes, nobody
} seq_encoding_t;
/// convert from unary to binary representation.
static inline seq_encoding_t seq_encoding(encoding_t inp) {
#if HAVE__BUILTIN_FFSLL
return __builtin_ffsll(inp);
#elif HAVE_FFSLL
return ffsll(inp);
#else
unsigned int out;
// supports max 16 different encodings.
if (inp == 0)
return 0L;
// if (inp & ((CELL)0xffffL << 16)) {inp >>= 16; out += 16;}
if (inp & ((CELL)0xffL << 8)) {
inp >>= 8;
out += 8;
}
if (inp & ((CELL)0xfL << 4)) {
inp >>= 4;
out += 4;
}
if (inp & ((CELL)0x3L << 2)) {
inp >>= 2;
out += 2;
}
if (inp & ((CELL)0x1 << 1))
out++;
#endif
return out;
}
extern xlocale enc_locales[SEQ_ENC_ISO_UTF32_LE + 1];
#endif
static inline const char *enc_name(encoding_t enc) {
switch (enc) {
case ENC_OCTET:
return "octet";
case ENC_ISO_LATIN1:
return "iso_latin_1";
case ENC_ISO_ASCII:
return "ascii";
case ENC_ISO_ANSI:
return "octet";
case ENC_ISO_UTF8:
return "utf8";
case ENC_UTF16_BE:
return "utf16_be";
case ENC_UTF16_LE:
return "utf16_le";
case ENC_ISO_UTF32_BE:
return "utf32_be";
case ENC_ISO_UTF32_LE:
return "utf32_le";
default:
return "thanks for watching!!";
}
}
static inline encoding_t enc_id(const char *s, encoding_t enc_bom) {
{
if (!strcmp(s, "iso_utf8"))
return ENC_ISO_UTF8;
if (!strcmp(s, "utf8"))
return ENC_ISO_UTF8;
if (!strcmp(s, "UTF-8"))
return ENC_ISO_UTF8;
if (!strcmp(s, "utf16_le"))
return ENC_UTF16_LE;
if (!strcmp(s, "utf16_be"))
return ENC_UTF16_BE;
if (!strcmp(s, "UTF-16")) {
if (enc_bom == ENC_UTF16_LE)
return ENC_UTF16_LE;
return ENC_UTF16_BE;
}
if (!strcmp(s, "UTF-16LE"))
return ENC_UTF16_LE;
if (!strcmp(s, "UTF16-BE"))
return ENC_UTF16_BE;
if (!strcmp(s, "octet"))
return ENC_OCTET;
if (!strcmp(s, "iso_latin_1"))
return ENC_ISO_LATIN1;
if (!strcmp(s, "iso_ascii"))
return ENC_ISO_ASCII;
if (!strcmp(s, "iso_ansi"))
return ENC_ISO_ANSI;
if (!strcmp(s, "utf32_be"))
return ENC_ISO_UTF32_BE;
if (!strcmp(s, "utf32_le"))
return ENC_ISO_UTF32_LE;
if (!strcmp(s, "UTF-32")) {
if (enc_bom == ENC_ISO_UTF32_LE)
return ENC_ISO_UTF32_LE;
return ENC_ISO_UTF32_BE;
}
if (!strcmp(s, "UTF-32BE"))
return ENC_ISO_UTF32_BE;
if (!strcmp(s, "UTF-32LE"))
return ENC_ISO_UTF32_LE;
if (!strcmp(s, "ISO-8859-1"))
return ENC_ISO_LATIN1;
// just for SWI compat, this actually refers to
// UCS-2
if (!strcmp(s, "unicode_be"))
return ENC_UTF16_BE;
if (!strcmp(s, "unicode_le"))
return ENC_UTF16_LE;
if (!strcmp(s, "default")) {
if (enc_bom != ENC_OCTET)
return enc_bom;
return Yap_DefaultEncoding();
}
else {
Yap_Error(DOMAIN_ERROR_OUT_OF_RANGE, MkAtomTerm(Yap_LookupAtom(s)), "bad encoding %s", s);
return Yap_DefaultEncoding();
}
}
}
#endif
#endif