зеркало из https://github.com/mozilla/pjs.git
340 строки
12 KiB
C
340 строки
12 KiB
C
/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 2 -*-
|
|
*
|
|
* The contents of this file are subject to the Netscape Public License
|
|
* Version 1.0 (the "NPL"); you may not use this file except in
|
|
* compliance with the NPL. You may obtain a copy of the NPL at
|
|
* http://www.mozilla.org/NPL/
|
|
*
|
|
* Software distributed under the NPL is distributed on an "AS IS" basis,
|
|
* WITHOUT WARRANTY OF ANY KIND, either express or implied. See the NPL
|
|
* for the specific language governing rights and limitations under the
|
|
* NPL.
|
|
*
|
|
* The Initial Developer of this code under the NPL is Netscape
|
|
* Communications Corporation. Portions created by Netscape are
|
|
* Copyright (C) 1998 Netscape Communications Corporation. All Rights
|
|
* Reserved.
|
|
*/
|
|
/* intlpriv.h */
|
|
|
|
#ifndef _INTLPRIV_H_
|
|
#define _INTLPRIV_H_
|
|
|
|
#include "xp.h"
|
|
#include "intl_csi.h"
|
|
#include "libi18n.h"
|
|
|
|
|
|
#define UNCVTBUF_SIZE 8 /* At least: longest ESC Seq + Char Bytes - 1 */
|
|
#define DOC_CSID_KNOWN(x) (((x) != CS_DEFAULT) && ((x) != CS_UNKNOWN))
|
|
|
|
/* Some constants for UCS-2 detection */
|
|
#define BYTE_ORDER_MARK 0xFEFF
|
|
#define NEEDS_SWAP_MARK 0xFFFE
|
|
|
|
/*
|
|
* UTF8 defines and macros
|
|
*/
|
|
#define ONE_OCTET_BASE 0x00 /* 0xxxxxxx */
|
|
#define ONE_OCTET_MASK 0x7F /* x1111111 */
|
|
#define CONTINUING_OCTET_BASE 0x80 /* 10xxxxxx */
|
|
#define CONTINUING_OCTET_MASK 0x3F /* 00111111 */
|
|
#define TWO_OCTET_BASE 0xC0 /* 110xxxxx */
|
|
#define TWO_OCTET_MASK 0x1F /* 00011111 */
|
|
#define THREE_OCTET_BASE 0xE0 /* 1110xxxx */
|
|
#define THREE_OCTET_MASK 0x0F /* 00001111 */
|
|
#define FOUR_OCTET_BASE 0xF0 /* 11110xxx */
|
|
#define FOUR_OCTET_MASK 0x07 /* 00000111 */
|
|
#define FIVE_OCTET_BASE 0xF8 /* 111110xx */
|
|
#define FIVE_OCTET_MASK 0x03 /* 00000011 */
|
|
#define SIX_OCTET_BASE 0xFC /* 1111110x */
|
|
#define SIX_OCTET_MASK 0x01 /* 00000001 */
|
|
|
|
#define IS_UTF8_1ST_OF_1(x) (( (x)&~ONE_OCTET_MASK ) == ONE_OCTET_BASE)
|
|
#define IS_UTF8_1ST_OF_2(x) (( (x)&~TWO_OCTET_MASK ) == TWO_OCTET_BASE)
|
|
#define IS_UTF8_1ST_OF_3(x) (( (x)&~THREE_OCTET_MASK) == THREE_OCTET_BASE)
|
|
#define IS_UTF8_1ST_OF_4(x) (( (x)&~FOUR_OCTET_MASK ) == FOUR_OCTET_BASE)
|
|
#define IS_UTF8_1ST_OF_5(x) (( (x)&~FIVE_OCTET_MASK ) == FIVE_OCTET_BASE)
|
|
#define IS_UTF8_1ST_OF_6(x) (( (x)&~SIX_OCTET_MASK ) == SIX_OCTET_BASE)
|
|
#define IS_UTF8_2ND_THRU_6TH(x) \
|
|
(( (x)&~CONTINUING_OCTET_MASK ) == CONTINUING_OCTET_BASE)
|
|
#define IS_UTF8_1ST_OF_UCS2(x) \
|
|
IS_UTF8_1ST_OF_1(x) \
|
|
|| IS_UTF8_1ST_OF_2(x) \
|
|
|| IS_UTF8_1ST_OF_3(x)
|
|
|
|
|
|
|
|
/* Some constants for UCS-2 detection */
|
|
#define BYTE_ORDER_MARK 0xFEFF
|
|
#define NEEDS_SWAP_MARK 0xFFFE
|
|
|
|
/* exported functions from unicvt.c */
|
|
|
|
MODULE_PRIVATE UNICVTAPI unsigned char *mz_ucs2utf8(CCCDataObject, const unsigned char *s, int32 l);
|
|
MODULE_PRIVATE UNICVTAPI unsigned char *mz_ucs2utf7(CCCDataObject, const unsigned char *s, int32 l);
|
|
MODULE_PRIVATE UNICVTAPI unsigned char *mz_utf82ucs(CCCDataObject, const unsigned char *s, int32 l);
|
|
MODULE_PRIVATE UNICVTAPI unsigned char *mz_utf82ucsswap(CCCDataObject, const unsigned char *s, int32 l);
|
|
MODULE_PRIVATE UNICVTAPI unsigned char *mz_utf72utf8(CCCDataObject, const unsigned char *s, int32 l);
|
|
MODULE_PRIVATE UNICVTAPI unsigned char *mz_utf82utf7(CCCDataObject, const unsigned char *s, int32 l);
|
|
MODULE_PRIVATE UNICVTAPI unsigned char *mz_imap4utf72utf8(CCCDataObject, const unsigned char *s, int32 l);
|
|
MODULE_PRIVATE UNICVTAPI unsigned char *mz_utf82imap4utf7(CCCDataObject, const unsigned char *s, int32 l);
|
|
MODULE_PRIVATE UNICVTAPI int16 utf8_to_ucs2_char(const unsigned char *utf8p, int16 buflen, uint16 *ucs2p);
|
|
MODULE_PRIVATE UNICVTAPI int32 utf8_to_ucs2_buffer(const unsigned char *utf8p, int16 utf8len, int *parsed_cnt, int *invalid_cnt, uint16 *ucs2p, int32 ucs2len);
|
|
|
|
|
|
/* values for ASCII chars */
|
|
#define ESC 0x1B /* ESC character */
|
|
#define NL 0x0A /* newline */
|
|
#undef CR
|
|
#define CR 0x0D /* carriage return */
|
|
#define DOLLAR 0x24 /* carriage return */
|
|
|
|
/* values for EUC shift chars */
|
|
#define SS2 0x8E /* Single Shift 2 */
|
|
#define SS3 0x8F /* Single Shift 3 */
|
|
|
|
/* JIS encoding mode flags */
|
|
#define JIS_Roman 0
|
|
#define JIS_208_83 1
|
|
#define JIS_HalfKana 2
|
|
#define JIS_212_90 3
|
|
|
|
/* I am using low nibble for the ESC flag and the next high nibble for Shift */
|
|
#define KSC_5601_87 0x04
|
|
|
|
/* Default state is SHIFT_OUT when we begin.
|
|
* So SHIFT_IN should have value 0x0- */
|
|
#define SHIFT_IN 0x00
|
|
#define SHIFT_OUT 0x10
|
|
|
|
/* The actual values to be output for SHIFTING */
|
|
#define SO 0x0e
|
|
#define SI 0x0f
|
|
|
|
/* Some masks for computation */
|
|
#define ESC_MASK 0x0F
|
|
#define SHIFT_MASK 0xF0
|
|
|
|
|
|
/*
|
|
* Shift JIS Encoding
|
|
* 1st Byte Range 2nd Byte Range
|
|
* ASCII/JIS-Roman 0x21-0x7F n/a
|
|
* 2-Byte Char(low range) 0x81-0x9F 0x40-0x7E, 0x80-0xFC
|
|
* Half-width space(non-std) 0xA0 n/a
|
|
* Half-width katakana 0xA1-0xDF n/a
|
|
* 2-Byte Char(high range) 0xE0-0xEF 0x40-0x7E, 0x80-0xFC
|
|
* User Defined(non-std) 0xF0-0xFC 0x40-0x7E, 0x80-0xFC
|
|
*
|
|
* JIS Encoding
|
|
* 1st Byte Range 2nd Byte Range
|
|
* ASCII/JIS-Roman 0x21-0x7E n/a
|
|
* Half-width katakana(non-std) 0x21-0x5F n/a
|
|
* 2-Byte Char 0x21-7E 0x21-7E
|
|
*
|
|
* Japanese EUC Encoding
|
|
* 1st Byte Range 2nd Byte Range 3rd Byte Range
|
|
* ASCII/JIS-Roman 0x21-0x7E n/a n/a
|
|
* JIS X 0208-1990 0xA0-0xFF 0xA0-0xFF n/a
|
|
* Half-width katakana SS2 0xA0-0xFF n/a
|
|
* JIS X 0212-1990 SS3 0xA0-0xFF 0xA0-0xFF
|
|
*
|
|
*
|
|
* List of ISO2022-INT Escape Sequences:
|
|
*
|
|
* SUPPORTED:
|
|
* ASCII ESC ( B G0
|
|
* JIS X 0201-Roman ESC ( J G0
|
|
* Half-width Katakana ESC ( I
|
|
* JIS X 0208-1978 ESC $ @ G0
|
|
* JIS X 0208-1983 ESC $ B G0
|
|
* JIS X 0212-1990 ESC $ ( D G0 (to EUC only)
|
|
* ISO8859-1 ESC - A G1
|
|
*
|
|
* UNSUPPORTED:
|
|
* GB 2312-80 ESC $ A G0
|
|
* KS C 5601-1987 ESC $ ) C G1
|
|
* CNS 11643-1986-1 ESC $ ( G G0
|
|
* CNS 11643-1986-2 ESC $ ( H G0
|
|
* ISO8859-7(Greek) ESC - F G1
|
|
*
|
|
* Added right parens: ))))) to balance editors' showmatch...
|
|
*/
|
|
|
|
|
|
/* JIS-Roman mode enabled by 3 char ESC sequence: <ESC> ( J */
|
|
#define InsRoman_ESC(cp, obj) { \
|
|
INTL_SetCCCJismode(obj, JIS_Roman); \
|
|
*cp++ = ESC; \
|
|
*cp++ = '('; \
|
|
*cp++ = 'J'; \
|
|
}
|
|
/* ASCII mode enabled by 3 char ESC sequence: <ESC> ( B */
|
|
#define InsASCII_ESC(cp, obj) { \
|
|
INTL_SetCCCJismode(obj, JIS_Roman); \
|
|
*cp++ = ESC; \
|
|
*cp++ = '('; \
|
|
*cp++ = 'B'; \
|
|
}
|
|
/* JIS x208-1983 mode enabled by 3 char ESC sequence: <ESC> $ B */
|
|
#define Ins208_83_ESC(cp, obj) { \
|
|
INTL_SetCCCJismode(obj, JIS_208_83); \
|
|
*cp++ = ESC; \
|
|
*cp++ = '$'; \
|
|
*cp++ = 'B'; \
|
|
}
|
|
/* JIS Half-width katakana mode enabled by 3 char ESC seq.: ESC ( I */
|
|
#define InsHalfKana_ESC(cp, obj) { \
|
|
INTL_SetCCCJismode(obj, JIS_HalfKana); \
|
|
*cp++ = ESC; \
|
|
*cp++ = '('; \
|
|
*cp++ = 'I'; \
|
|
}
|
|
/* JIS x212-1990 mode enabled by 4 char ESC sequence: <ESC> $ ( D */
|
|
#define Ins212_90_ESC(cp, obj) { \
|
|
INTL_SetCCCJismode(obj, JIS_212_90); \
|
|
*cp++ = ESC; \
|
|
*cp++ = '$'; \
|
|
*cp++ = '('; \
|
|
*cp++ = 'D'; \
|
|
}
|
|
|
|
/* KSC 5601-1987 mode enabled by 4 char ESC sequence: <ESC> $ ) D */
|
|
#define Ins5601_87_ESC(cp, obj) { \
|
|
INTL_SetCCCJismode(obj, (INTL_GetCCCJismode(obj) & ~ESC_MASK) | KSC_5601_87); \
|
|
*cp++ = ESC; \
|
|
*cp++ = '$'; \
|
|
*cp++ = ')'; \
|
|
*cp++ = 'C'; \
|
|
}
|
|
#define Ins5601_87_SI(cp, obj) { \
|
|
INTL_SetCCCJismode(obj, ((INTL_GetCCCJismode(obj) & ~SHIFT_MASK) | SHIFT_IN)); \
|
|
*cp++ = SI; \
|
|
}
|
|
#define Ins5601_87_SO(cp, obj) { \
|
|
INTL_SetCCCJismode(obj, ((INTL_GetCCCJismode(obj) & ~SHIFT_MASK) | SHIFT_OUT)); \
|
|
*cp++ = SO; \
|
|
}
|
|
#define IsIns5601_87_ESC(obj) ((INTL_GetCCCJismode(obj) & ESC_MASK) == KSC_5601_87)
|
|
#define IsIns5601_87_SI(obj) ((INTL_GetCCCJismode(obj) & SHIFT_MASK) == SHIFT_IN)
|
|
#define IsIns5601_87_SO(obj) ((INTL_GetCCCJismode(obj) & SHIFT_MASK) == SHIFT_OUT)
|
|
|
|
/* Added right parens: )))))) to balance editors' showmatch... */
|
|
|
|
/* Maximum Length of Escape Sequence and Character Bytes per Encoding */
|
|
#define MAX_SJIS 2
|
|
#define MAX_EUC 3
|
|
#define MAX_JIS 6
|
|
|
|
|
|
|
|
#define MAX_CSNAME 64
|
|
typedef struct _csname2id_t {
|
|
unsigned char cs_name[MAX_CSNAME];
|
|
unsigned char java_name[MAX_CSNAME];
|
|
int16 cs_id;
|
|
unsigned char fill[3];
|
|
} csname2id_t;
|
|
|
|
typedef struct {
|
|
int16 from_csid; /* "from" codeset ID */
|
|
int16 to_csid; /* "to" codeset ID */
|
|
int16 autoselect; /* autoselectable */
|
|
CCCFunc cvtmethod; /* char. codeset conv method */
|
|
int32 cvtflag; /* conv func dependent flag */
|
|
} cscvt_t;
|
|
|
|
|
|
#ifdef XP_UNIX
|
|
typedef struct {
|
|
int16 win_csid;
|
|
char *locale;
|
|
char *fontlist;
|
|
} cslocale_t;
|
|
#endif /* XP_UNIX */
|
|
|
|
XP_BEGIN_PROTOS
|
|
|
|
MODULE_PRIVATE int net_1to1CCC(CCCDataObject, const unsigned char *s, int32 l);
|
|
MODULE_PRIVATE int net_sjis2jis(CCCDataObject,const unsigned char *s,int32 l);
|
|
MODULE_PRIVATE int net_sjis2euc(CCCDataObject,const unsigned char *s,int32 l);
|
|
MODULE_PRIVATE int net_euc2jis(CCCDataObject,const unsigned char *s, int32 l);
|
|
MODULE_PRIVATE int net_euc2sjis(CCCDataObject,const unsigned char *s,int32 l);
|
|
MODULE_PRIVATE int net_jis2other(CCCDataObject, const unsigned char *s, int32 l);
|
|
|
|
MODULE_PRIVATE unsigned char *mz_euc2jis(CCCDataObject, const unsigned char *s, int32 l);
|
|
MODULE_PRIVATE unsigned char *mz_sjis2euc(CCCDataObject, const unsigned char *s, int32 l);
|
|
MODULE_PRIVATE unsigned char *mz_sjis2jis(CCCDataObject, const unsigned char *s, int32 l);
|
|
MODULE_PRIVATE unsigned char *One2OneCCC(CCCDataObject, const unsigned char *s, int32 l);
|
|
MODULE_PRIVATE unsigned char *mz_euc2sjis(CCCDataObject, const unsigned char *s, int32 l);
|
|
MODULE_PRIVATE unsigned char *mz_euckr2iso(CCCDataObject, const unsigned char *s, int32 l);
|
|
MODULE_PRIVATE unsigned char *mz_iso2euckr(CCCDataObject, const unsigned char *s, int32 l);
|
|
MODULE_PRIVATE unsigned char *mz_b52cns(CCCDataObject, const unsigned char *s, int32 l);
|
|
MODULE_PRIVATE unsigned char *mz_cns2b5(CCCDataObject, const unsigned char *s, int32 l);
|
|
MODULE_PRIVATE unsigned char *jis2other(CCCDataObject, const unsigned char *s, int32 l);
|
|
|
|
|
|
MODULE_PRIVATE unsigned char *autoJCCC (CCCDataObject, const unsigned char *s, int32 l);
|
|
MODULE_PRIVATE unsigned char *autoKCCC (CCCDataObject, const unsigned char *s, int32 l);
|
|
|
|
MODULE_PRIVATE int16 intl_detect_JCSID (uint16, const unsigned char *, int32);
|
|
MODULE_PRIVATE int16 intl_detect_KCSID (uint16, const unsigned char *, int32);
|
|
|
|
|
|
|
|
|
|
void FE_fontchange(iDocumentContext window_id, int16 csid);
|
|
void FE_ChangeURLCharset(const char *newCharset);
|
|
char ** FE_GetSingleByteTable(int16 from_csid, int16 to_csid, int32 resourceid);
|
|
char * FE_LockTable(char **cvthdl);
|
|
void FE_FreeSingleByteTable(char **cvthdl);
|
|
|
|
/* void CCCReportAutoDetect(CCCDataObject obj, uint16 detected_doc_csid); */
|
|
|
|
unsigned char *ConvHeader(unsigned char *subject);
|
|
|
|
#ifdef XP_UNIX
|
|
int16 FE_WinCSID(iDocumentContext );
|
|
#else /* XP_UNIX */
|
|
#define FE_WinCSID(window_id) 0
|
|
#endif /* XP_UNIX */
|
|
|
|
int16 *intl_GetFontCharSets(void);
|
|
|
|
|
|
/**
|
|
* Access a conversion flag for hankaku->zenkaku kana conversion for mail.
|
|
*
|
|
* The conversion flag for JIS, set if converting hankaku (1byte) kana to zenkaku (2byte).
|
|
* The flag is needed in order to control the conversion. Kana conversion should be applied
|
|
* only when sending a mail and converters do not know if they are called for mail sending.
|
|
*
|
|
* @param obj Character code converter.
|
|
* @return TRUE if convert to zenkaku (2byte).
|
|
* @see INTL_SetCCCCvtflag_SendHankakuKana
|
|
*/
|
|
MODULE_PRIVATE XP_Bool INTL_GetCCCCvtflag_SendHankakuKana(CCCDataObject obj);
|
|
/**
|
|
* Access a conversion flag for hankaku->zenkaku kana conversion for mail.
|
|
*
|
|
* The conversion flag for JIS, set if converting hankaku (1byte) kana to zenkaku (2byte).
|
|
* The flag is needed in order to control the conversion. Kana conversion should be applied
|
|
* only when sending a mail and converters do not know if they are called for mail sending.
|
|
*
|
|
* @param obj Character code converter.
|
|
* @see INTL_GetCCCCvtflag_SendHankakuKana
|
|
*/
|
|
MODULE_PRIVATE void INTL_SetCCCCvtflag_SendHankakuKana(CCCDataObject obj, XP_Bool flag);
|
|
|
|
|
|
XP_END_PROTOS
|
|
|
|
|
|
|
|
#define MAXCSIDINTBL 64
|
|
|
|
|
|
|
|
#endif /* _INTLPRIV_H_ */
|