view src/charset.h @ 70808:0c087a8f2e1c

(bibtex-maintain-sorted-entries): Mark as safe.
author Roland Winkler <Roland.Winkler@physik.uni-erlangen.de>
date Sun, 21 May 2006 17:45:09 +0000
parents 3bd95f4f2941
children 92b0f65a55ce c5406394f567
line wrap: on
line source

/* Header for multibyte character handler.
   Copyright (C) 2001, 2002, 2003, 2004, 2005,
                 2006 Free Software Foundation, Inc.
   Copyright (C) 1995, 1997, 1998, 2003
     National Institute of Advanced Industrial Science and Technology (AIST)
     Registration Number H14PRO021

This file is part of GNU Emacs.

GNU Emacs is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2, or (at your option)
any later version.

GNU Emacs is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU General Public License for more details.

You should have received a copy of the GNU General Public License
along with GNU Emacs; see the file COPYING.  If not, write to
the Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor,
Boston, MA 02110-1301, USA.  */

#ifndef EMACS_CHARSET_H
#define EMACS_CHARSET_H

/* #define BYTE_COMBINING_DEBUG */

/*** GENERAL NOTE on CHARACTER SET (CHARSET) ***

  A character set ("charset" hereafter) is a meaningful collection
  (i.e. language, culture, functionality, etc) of characters.  Emacs
  handles multiple charsets at once.  Each charset corresponds to one
  of the ISO charsets.  Emacs identifies a charset by a unique
  identification number, whereas ISO identifies a charset by a triplet
  of DIMENSION, CHARS and FINAL-CHAR.  So, hereafter, just saying
  "charset" means an identification number (integer value).

  The value range of charsets is 0x00, 0x81..0xFE.  There are four
  kinds of charset depending on DIMENSION (1 or 2) and CHARS (94 or
  96).  For instance, a charset of DIMENSION2_CHARS94 contains 94x94
  characters.

  Within Emacs Lisp, a charset is treated as a symbol which has a
  property `charset'.  The property value is a vector containing
  various information about the charset.  For readability of C code,
  we use the following convention for C variable names:
	charset_symbol: Emacs Lisp symbol of a charset
	charset_id: Emacs Lisp integer of an identification number of a charset
	charset: C integer of an identification number of a charset

  Each charset (except for ascii) is assigned a base leading-code
  (range 0x80..0x9E).  In addition, a charset of greater than 0xA0
  (whose base leading-code is 0x9A..0x9D) is assigned an extended
  leading-code (range 0xA0..0xFE).  In this case, each base
  leading-code specifies the allowable range of extended leading-code
  as shown in the table below.  A leading-code is used to represent a
  character in Emacs' buffer and string.

  We call a charset which has extended leading-code a "private
  charset" because those are mainly for a charset which is not yet
  registered by ISO.  On the contrary, we call a charset which does
  not have extended leading-code an "official charset".

  ---------------------------------------------------------------------------
  charset	dimension	 base leading-code	extended leading-code
  ---------------------------------------------------------------------------
  0x00		official dim1    -- none --		-- none --
		(ASCII)
  0x01..0x7F	--never used--
  0x80		official dim1	 -- none --		-- none --
		(eight-bit-graphic)
  0x81..0x8F	official dim1    same as charset	-- none --
  0x90..0x99	official dim2	 same as charset	-- none --
  0x9A..0x9D	--never used--
  0x9E		official dim1	 same as charset	-- none --
		(eight-bit-control)
  0x9F		--never used--
  0xA0..0xDF	private dim1	    0x9A		same as charset
		of 1-column width
  0xE0..0xEF	private dim1	    0x9B		same as charset
		of 2-column width
  0xF0..0xF4	private dim2	    0x9C		same as charset
		of 1-column width
  0xF5..0xFE	private dim2	    0x9D		same as charset
		of 2-column width
  0xFF		--never used--
  ---------------------------------------------------------------------------

*/

/* Definition of special leading-codes.  */
/* Leading-code followed by extended leading-code.  */
#define LEADING_CODE_PRIVATE_11	0x9A /* for private DIMENSION1 of 1-column */
#define LEADING_CODE_PRIVATE_12	0x9B /* for private DIMENSION1 of 2-column */
#define LEADING_CODE_PRIVATE_21	0x9C /* for private DIMENSION2 of 1-column */
#define LEADING_CODE_PRIVATE_22	0x9D /* for private DIMENSION2 of 2-column */

#define LEADING_CODE_8_BIT_CONTROL 0x9E /* for `eight-bit-control' */

/* Extended leading-code.  */
/* Start of each extended leading-codes.  */
#define LEADING_CODE_EXT_11 0xA0 /* follows LEADING_CODE_PRIVATE_11 */
#define LEADING_CODE_EXT_12 0xE0 /* follows LEADING_CODE_PRIVATE_12 */
#define LEADING_CODE_EXT_21 0xF0 /* follows LEADING_CODE_PRIVATE_21 */
#define LEADING_CODE_EXT_22 0xF5 /* follows LEADING_CODE_PRIVATE_22 */
/* Maximum value of extended leading-codes.  */
#define LEADING_CODE_EXT_MAX 0xFE

/* Definition of minimum/maximum charset of each DIMENSION.  */
#define MIN_CHARSET_OFFICIAL_DIMENSION1	0x80
#define MAX_CHARSET_OFFICIAL_DIMENSION1	0x8F
#define MIN_CHARSET_OFFICIAL_DIMENSION2	0x90
#define MAX_CHARSET_OFFICIAL_DIMENSION2 0x99
#define MIN_CHARSET_PRIVATE_DIMENSION1	LEADING_CODE_EXT_11
#define MIN_CHARSET_PRIVATE_DIMENSION2	LEADING_CODE_EXT_21

/* Maximum value of overall charset identification number.  */
#define MAX_CHARSET 0xFE

/* Definition of special charsets.  */
#define CHARSET_ASCII		0	/* 0x00..0x7F */
#define CHARSET_8_BIT_CONTROL	0x9E	/* 0x80..0x9F */
#define CHARSET_8_BIT_GRAPHIC	0x80	/* 0xA0..0xFF */

extern int charset_latin_iso8859_1; /* ISO8859-1 (Latin-1) */
extern int charset_jisx0208_1978; /* JISX0208.1978 (Japanese Kanji old set) */
extern int charset_jisx0208;	/* JISX0208.1983 (Japanese Kanji) */
extern int charset_katakana_jisx0201; /* JISX0201.Kana (Japanese Katakana) */
extern int charset_latin_jisx0201; /* JISX0201.Roman (Japanese Roman) */
extern int charset_big5_1;	/* Big5 Level 1 (Chinese Traditional) */
extern int charset_big5_2;	/* Big5 Level 2 (Chinese Traditional) */
extern int charset_mule_unicode_0100_24ff;
extern int charset_mule_unicode_2500_33ff;
extern int charset_mule_unicode_e000_ffff;

/* Check if CH is an ASCII character or a base leading-code.
   Nowadays, any byte can be the first byte of a character in a
   multibyte buffer/string.  So this macro name is not appropriate.  */
#define CHAR_HEAD_P(ch) ((unsigned char) (ch) < 0xA0)

/*** GENERAL NOTE on CHARACTER REPRESENTATION ***

  Firstly, the term "character" or "char" is used for a multilingual
  character (of course, including ASCII characters), not for a byte in
  computer memory.  We use the term "code" or "byte" for the latter
  case.

  A character is identified by charset and one or two POSITION-CODEs.
  POSITION-CODE is the position of the character in the charset.  A
  character of DIMENSION1 charset has one POSITION-CODE: POSITION-CODE-1.
  A character of DIMENSION2 charset has two POSITION-CODE:
  POSITION-CODE-1 and POSITION-CODE-2.  The code range of
  POSITION-CODE is 0x20..0x7F.

  Emacs has two kinds of representation of a character: multi-byte
  form (for buffers and strings) and single-word form (for character
  objects in Emacs Lisp).  The latter is called "character code"
  hereafter.  Both representations encode the information of charset
  and POSITION-CODE but in a different way (for instance, the MSB of
  POSITION-CODE is set in multi-byte form).

  For details of the multi-byte form, see the section "2. Emacs
  internal format handlers" of `coding.c'.

  Emacs uses 19 bits for a character code.  The bits are divided into
  3 fields: FIELD1(5bits):FIELD2(7bits):FIELD3(7bits).

  A character code of DIMENSION1 character uses FIELD2 to hold charset
  and FIELD3 to hold POSITION-CODE-1.  A character code of DIMENSION2
  character uses FIELD1 to hold charset, FIELD2 and FIELD3 to hold
  POSITION-CODE-1 and POSITION-CODE-2 respectively.

  More precisely...

  FIELD2 of DIMENSION1 character (except for ascii, eight-bit-control,
  and eight-bit-graphic) is "charset - 0x70".  This is to make all
  character codes except for ASCII and 8-bit codes greater than 256.
  So, the range of FIELD2 of DIMENSION1 character is 0, 1, or
  0x11..0x7F.

  FIELD1 of DIMENSION2 character is "charset - 0x8F" for official
  charset and "charset - 0xE0" for private charset.  So, the range of
  FIELD1 of DIMENSION2 character is 0x01..0x1E.

  -----------------------------------------------------------------------------
  charset		FIELD1 (5-bit)	    FIELD2 (7-bit)	FIELD3 (7-bit)
  -----------------------------------------------------------------------------
  ascii			0		    0			0x00..0x7F
  eight-bit-control	0		    1			0x00..0x1F
  eight-bit-graphic	0		    1			0x20..0x7F
  DIMENSION1		0		    charset - 0x70	POSITION-CODE-1
  DIMENSION2(o)		charset - 0x8F	    POSITION-CODE-1	POSITION-CODE-2
  DIMENSION2(p)		charset - 0xE0	    POSITION-CODE-1	POSITION-CODE-2
  -----------------------------------------------------------------------------
  "(o)": official, "(p)": private
  -----------------------------------------------------------------------------
*/

/* Masks of each field of character code.  */
#define CHAR_FIELD1_MASK (0x1F << 14)
#define CHAR_FIELD2_MASK (0x7F << 7)
#define CHAR_FIELD3_MASK 0x7F

/* Macros to access each field of character C.  */
#define CHAR_FIELD1(c) (((c) & CHAR_FIELD1_MASK) >> 14)
#define CHAR_FIELD2(c) (((c) & CHAR_FIELD2_MASK) >> 7)
#define CHAR_FIELD3(c) ((c) & CHAR_FIELD3_MASK)

/* Minimum character code of character of each DIMENSION.  */
#define MIN_CHAR_OFFICIAL_DIMENSION1 \
  ((0x81 - 0x70) << 7)
#define MIN_CHAR_PRIVATE_DIMENSION1 \
  ((MIN_CHARSET_PRIVATE_DIMENSION1 - 0x70) << 7)
#define MIN_CHAR_OFFICIAL_DIMENSION2 \
  ((MIN_CHARSET_OFFICIAL_DIMENSION2 - 0x8F) << 14)
#define MIN_CHAR_PRIVATE_DIMENSION2 \
  ((MIN_CHARSET_PRIVATE_DIMENSION2 - 0xE0) << 14)
/* Maximum character code currently used plus 1.  */
#define MAX_CHAR (0x1F << 14)

/* 1 if C is a single byte character, else 0.  */
#define SINGLE_BYTE_CHAR_P(c) (((unsigned)(c) & 0xFF) == (c))

/* 1 if BYTE is an ASCII character in itself, in multibyte mode.  */
#define ASCII_BYTE_P(byte) ((byte) < 0x80)

/* A char-table containing information on each character set.

   Unlike ordinary char-tables, this doesn't contain any nested tables.
   Only the top level elements are used.  Each element is a vector of
   the following information:
	CHARSET-ID, BYTES, DIMENSION, CHARS, WIDTH, DIRECTION,
	LEADING-CODE-BASE, LEADING-CODE-EXT,
	ISO-FINAL-CHAR, ISO-GRAPHIC-PLANE,
	REVERSE-CHARSET, SHORT-NAME, LONG-NAME,	DESCRIPTION,
	PLIST.

   CHARSET-ID (integer) is the identification number of the charset.

   BYTES (integer) is the length of the multi-byte form of a character
   in the charset: one of 1, 2, 3, and 4.

   DIMENSION (integer) is the number of bytes to represent a character: 1 or 2.

   CHARS (integer) is the number of characters in a dimension: 94 or 96.

   WIDTH (integer) is the number of columns a character in the charset
   occupies on the screen: one of 0, 1, and 2..

   DIRECTION (integer) is the rendering direction of characters in the
   charset when rendering.  If 0, render from left to right, else
   render from right to left.

   LEADING-CODE-BASE (integer) is the base leading-code for the
   charset.

   LEADING-CODE-EXT (integer) is the extended leading-code for the
   charset.  All charsets of less than 0xA0 have the value 0.

   ISO-FINAL-CHAR (character) is the final character of the
   corresponding ISO 2022 charset.  It is -1 for such a character
   that is used only internally (e.g. `eight-bit-control').

   ISO-GRAPHIC-PLANE (integer) is the graphic plane to be invoked
   while encoding to variants of ISO 2022 coding system, one of the
   following: 0/graphic-plane-left(GL), 1/graphic-plane-right(GR).  It
   is -1 for such a character that is used only internally
   (e.g. `eight-bit-control').

   REVERSE-CHARSET (integer) is the charset which differs only in
   LEFT-TO-RIGHT value from the charset.  If there's no such a
   charset, the value is -1.

   SHORT-NAME (string) is the short name to refer to the charset.

   LONG-NAME (string) is the long name to refer to the charset.

   DESCRIPTION (string) is the description string of the charset.

   PLIST (property list) may contain any type of information a user
   wants to put and get by functions `put-charset-property' and
   `get-charset-property' respectively.  */
extern Lisp_Object Vcharset_table;

/* Macros to access various information of CHARSET in Vcharset_table.
   We provide these macros for efficiency.  No range check of CHARSET.  */

/* Return entry of CHARSET (C integer) in Vcharset_table.  */
#define CHARSET_TABLE_ENTRY(charset)					\
  XCHAR_TABLE (Vcharset_table)->contents[((charset) == CHARSET_ASCII	\
					  ? 0 : (charset) + 128)]

/* Return information INFO-IDX of CHARSET.  */
#define CHARSET_TABLE_INFO(charset, info_idx) \
  XVECTOR (CHARSET_TABLE_ENTRY (charset))->contents[info_idx]

#define CHARSET_ID_IDX (0)
#define CHARSET_BYTES_IDX (1)
#define CHARSET_DIMENSION_IDX (2)
#define CHARSET_CHARS_IDX (3)
#define CHARSET_WIDTH_IDX (4)
#define CHARSET_DIRECTION_IDX (5)
#define CHARSET_LEADING_CODE_BASE_IDX (6)
#define CHARSET_LEADING_CODE_EXT_IDX (7)
#define CHARSET_ISO_FINAL_CHAR_IDX (8)
#define CHARSET_ISO_GRAPHIC_PLANE_IDX (9)
#define CHARSET_REVERSE_CHARSET_IDX (10)
#define CHARSET_SHORT_NAME_IDX (11)
#define CHARSET_LONG_NAME_IDX (12)
#define CHARSET_DESCRIPTION_IDX (13)
#define CHARSET_PLIST_IDX (14)
/* Size of a vector of each entry of Vcharset_table.  */
#define CHARSET_MAX_IDX (15)

/* And several more macros to be used frequently.  */
#define CHARSET_BYTES(charset) \
  XFASTINT (CHARSET_TABLE_INFO (charset, CHARSET_BYTES_IDX))
#define CHARSET_DIMENSION(charset) \
  XFASTINT (CHARSET_TABLE_INFO (charset, CHARSET_DIMENSION_IDX))
#define CHARSET_CHARS(charset) \
  XFASTINT (CHARSET_TABLE_INFO (charset, CHARSET_CHARS_IDX))
#define CHARSET_WIDTH(charset) \
  XFASTINT (CHARSET_TABLE_INFO (charset, CHARSET_WIDTH_IDX))
#define CHARSET_DIRECTION(charset) \
  XFASTINT (CHARSET_TABLE_INFO (charset, CHARSET_DIRECTION_IDX))
#define CHARSET_LEADING_CODE_BASE(charset) \
  XFASTINT (CHARSET_TABLE_INFO (charset, CHARSET_LEADING_CODE_BASE_IDX))
#define CHARSET_LEADING_CODE_EXT(charset) \
  XFASTINT (CHARSET_TABLE_INFO (charset, CHARSET_LEADING_CODE_EXT_IDX))
#define CHARSET_ISO_FINAL_CHAR(charset) \
  XINT (CHARSET_TABLE_INFO (charset, CHARSET_ISO_FINAL_CHAR_IDX))
#define CHARSET_ISO_GRAPHIC_PLANE(charset) \
  XINT (CHARSET_TABLE_INFO (charset, CHARSET_ISO_GRAPHIC_PLANE_IDX))
#define CHARSET_REVERSE_CHARSET(charset) \
  XINT (CHARSET_TABLE_INFO (charset, CHARSET_REVERSE_CHARSET_IDX))

/* Macros to specify direction of a charset.  */
#define CHARSET_DIRECTION_LEFT_TO_RIGHT 0
#define CHARSET_DIRECTION_RIGHT_TO_LEFT 1

/* A vector of charset symbol indexed by charset-id.  This is used
   only for returning charset symbol from C functions.  */
extern Lisp_Object Vcharset_symbol_table;

/* Return symbol of CHARSET.  */
#define CHARSET_SYMBOL(charset) \
  XVECTOR (Vcharset_symbol_table)->contents[charset]

/* 1 if CHARSET is in valid value range, else 0.  */
#define CHARSET_VALID_P(charset)					 \
  ((charset) == 0							 \
   || ((charset) > 0x80 && (charset) <= MAX_CHARSET_OFFICIAL_DIMENSION2) \
   || ((charset) >= MIN_CHARSET_PRIVATE_DIMENSION1			 \
       && (charset) <= MAX_CHARSET)					 \
   || ((charset) == CHARSET_8_BIT_CONTROL)				 \
   || ((charset) == CHARSET_8_BIT_GRAPHIC))

/* 1 if CHARSET is already defined, else 0.  */
#define CHARSET_DEFINED_P(charset)			\
  (((charset) >= 0) && ((charset) <= MAX_CHARSET)	\
   && !NILP (CHARSET_TABLE_ENTRY (charset)))

/* Since the information CHARSET-BYTES and CHARSET-WIDTH of
   Vcharset_table can be retrieved only by the first byte of
   multi-byte form (an ASCII code or a base leading-code), we provide
   here tables to be used by macros BYTES_BY_CHAR_HEAD and
   WIDTH_BY_CHAR_HEAD for faster information retrieval.  */
extern int bytes_by_char_head[256];
extern int width_by_char_head[256];

#define BYTES_BY_CHAR_HEAD(char_head)	\
  (ASCII_BYTE_P (char_head) ? 1 : bytes_by_char_head[char_head])
#define WIDTH_BY_CHAR_HEAD(char_head)	\
  (ASCII_BYTE_P (char_head) ? 1 : width_by_char_head[char_head])

/* Charset of the character C.  */
#define CHAR_CHARSET(c)							\
  (SINGLE_BYTE_CHAR_P (c)						\
   ? (ASCII_BYTE_P (c)							\
      ? CHARSET_ASCII							\
      : (c) < 0xA0 ? CHARSET_8_BIT_CONTROL : CHARSET_8_BIT_GRAPHIC)	\
   : ((c) < MIN_CHAR_OFFICIAL_DIMENSION2				\
      ? CHAR_FIELD2 (c) + 0x70						\
      : ((c) < MIN_CHAR_PRIVATE_DIMENSION2				\
	 ? CHAR_FIELD1 (c) + 0x8F					\
	 : CHAR_FIELD1 (c) + 0xE0)))

/* Check if two characters C1 and C2 belong to the same charset.  */
#define SAME_CHARSET_P(c1, c2)				\
  (c1 < MIN_CHAR_OFFICIAL_DIMENSION2			\
   ? (c1 & CHAR_FIELD2_MASK) == (c2 & CHAR_FIELD2_MASK)	\
   : (c1 & CHAR_FIELD1_MASK) == (c2 & CHAR_FIELD1_MASK))

/* Return a character of which charset is CHARSET and position-codes
   are C1 and C2.  DIMENSION1 character ignores C2.  */
#define MAKE_CHAR(charset, c1, c2)					    \
  ((charset) == CHARSET_ASCII						    \
   ? (c1) & 0x7F							    \
   : (((charset) == CHARSET_8_BIT_CONTROL				    \
       || (charset) == CHARSET_8_BIT_GRAPHIC)				    \
      ? ((c1) & 0x7F) | 0x80						    \
      : ((CHARSET_DEFINED_P (charset)					    \
	  ? CHARSET_DIMENSION (charset) == 1				    \
	  : (charset) < MIN_CHARSET_PRIVATE_DIMENSION2)			    \
	 ? (((charset) - 0x70) << 7) | ((c1) <= 0 ? 0 : ((c1) & 0x7F))	    \
	 : ((((charset)							    \
	      - ((charset) < MIN_CHARSET_PRIVATE_DIMENSION2 ? 0x8F : 0xE0)) \
	     << 14)							    \
	    | ((c2) <= 0 ? 0 : ((c2) & 0x7F))				    \
	    | ((c1) <= 0 ? 0 : (((c1) & 0x7F) << 7))))))


/* If GENERICP is nonzero, return nonzero iff C is a valid normal or
   generic character.  If GENERICP is zero, return nonzero iff C is a
   valid normal character.  */
#define CHAR_VALID_P(c, genericp)	\
  ((c) >= 0				\
   && (SINGLE_BYTE_CHAR_P (c) || char_valid_p (c, genericp)))

/* This default value is used when nonascii-translation-table or
   nonascii-insert-offset fail to convert unibyte character to a valid
   multibyte character.  This makes a Latin-1 character.  */

#define DEFAULT_NONASCII_INSERT_OFFSET 0x800

/* Parse multibyte string STR of length LENGTH and set BYTES to the
   byte length of a character at STR.  */

#ifdef BYTE_COMBINING_DEBUG

#define PARSE_MULTIBYTE_SEQ(str, length, bytes)			\
  do {								\
    int i = 1;							\
    while (i < (length) && ! CHAR_HEAD_P ((str)[i])) i++;	\
    (bytes) = BYTES_BY_CHAR_HEAD ((str)[0]);			\
    if ((bytes) > i)						\
      abort ();							\
  } while (0)

#else  /* not BYTE_COMBINING_DEBUG */

#define PARSE_MULTIBYTE_SEQ(str, length, bytes)	\
  ((void)(length), (bytes) = BYTES_BY_CHAR_HEAD ((str)[0]))

#endif /* not BYTE_COMBINING_DEBUG */

#define VALID_LEADING_CODE_P(code)	\
  (! NILP (CHARSET_TABLE_ENTRY (code)))

/* Return 1 iff the byte sequence at unibyte string STR (LENGTH bytes)
   is valid as a multibyte form.  If valid, by a side effect, BYTES is
   set to the byte length of the multibyte form.  */

#define UNIBYTE_STR_AS_MULTIBYTE_P(str, length, bytes)		\
  (((str)[0] < 0x80 || (str)[0] >= 0xA0)			\
   ? ((bytes) = 1)						\
   : (((bytes) = BYTES_BY_CHAR_HEAD ((str)[0])),		\
      ((bytes) <= (length)					\
       && !CHAR_HEAD_P ((str)[1])				\
       && ((bytes) == 2						\
	   ? (str)[0] != LEADING_CODE_8_BIT_CONTROL		\
	   : (!CHAR_HEAD_P ((str)[2])				\
	      && ((bytes) == 3					\
		  ? (((str)[0] != LEADING_CODE_PRIVATE_11	\
		      && (str)[0] != LEADING_CODE_PRIVATE_12)	\
		     || VALID_LEADING_CODE_P (str[1]))		\
		  : (!CHAR_HEAD_P ((str)[3])			\
		     && VALID_LEADING_CODE_P (str[1]))))))))


/* Return 1 iff the byte sequence at multibyte string STR is valid as
   a unibyte form.  By a side effect, BYTES is set to the byte length
   of one character at STR.  */

#define MULTIBYTE_STR_AS_UNIBYTE_P(str, bytes)	\
  ((bytes) = BYTES_BY_CHAR_HEAD ((str)[0]),	\
   (str)[0] != LEADING_CODE_8_BIT_CONTROL)

/* The charset of character C is stored in CHARSET, and the
   position-codes of C are stored in C1 and C2.
   We store -1 in C2 if the dimension of the charset is 1.  */

#define SPLIT_CHAR(c, charset, c1, c2)					    \
  (SINGLE_BYTE_CHAR_P (c)						    \
   ? ((charset								    \
       = (ASCII_BYTE_P (c)						    \
	  ? CHARSET_ASCII						    \
	  : ((c) < 0xA0 ? CHARSET_8_BIT_CONTROL : CHARSET_8_BIT_GRAPHIC))), \
      c1 = (c), c2 = -1)						    \
   : ((c) & CHAR_FIELD1_MASK						    \
      ? (charset = (CHAR_FIELD1 (c)					    \
		    + ((c) < MIN_CHAR_PRIVATE_DIMENSION2 ? 0x8F : 0xE0)),   \
	 c1 = CHAR_FIELD2 (c),						    \
	 c2 = CHAR_FIELD3 (c))						    \
      : (charset = CHAR_FIELD2 (c) + 0x70,				    \
	 c1 = CHAR_FIELD3 (c),						    \
	 c2 = -1)))

/* Return 1 iff character C has valid printable glyph.  */
#define CHAR_PRINTABLE_P(c) (ASCII_BYTE_P (c) || char_printable_p (c))

/* The charset of the character at STR is stored in CHARSET, and the
   position-codes are stored in C1 and C2.
   We store -1 in C2 if the character is just 2 bytes.  */

#define SPLIT_STRING(str, len, charset, c1, c2)			\
  ((BYTES_BY_CHAR_HEAD ((unsigned char) *(str)) < 2		\
    || BYTES_BY_CHAR_HEAD ((unsigned char) *(str)) > len	\
    || split_string (str, len, &charset, &c1, &c2) < 0)		\
   ? c1 = *(str), charset = CHARSET_ASCII			\
   : charset)

/* Mapping table from ISO2022's charset (specified by DIMENSION,
   CHARS, and FINAL_CHAR) to Emacs' charset.  Should be accessed by
   macro ISO_CHARSET_TABLE (DIMENSION, CHARS, FINAL_CHAR).  */
extern int iso_charset_table[2][2][128];

#define ISO_CHARSET_TABLE(dimension, chars, final_char) \
  iso_charset_table[XINT (dimension) - 1][XINT (chars) > 94][XINT (final_char)]

#define BASE_LEADING_CODE_P(c) (BYTES_BY_CHAR_HEAD ((unsigned char) (c)) > 1)

/* Return how many bytes C will occupy in a multibyte buffer.  */
#define CHAR_BYTES(c)					\
  (SINGLE_BYTE_CHAR_P (c)				\
   ? ((ASCII_BYTE_P (c) || (c) >= 0xA0) ? 1 : 2)	\
   : char_bytes (c))

/* The following two macros CHAR_STRING and STRING_CHAR are the main
   entry points to convert between Emacs's two types of character
   representations: multi-byte form and single-word form (character
   code).  */

/* Store multi-byte form of the character C in STR.  The caller should
   allocate at least MAX_MULTIBYTE_LENGTH bytes area at STR in
   advance.  Returns the length of the multi-byte form.  If C is an
   invalid character code, signal an error.  */

#define CHAR_STRING(c, str)						  \
  (SINGLE_BYTE_CHAR_P (c)						  \
   ? ((ASCII_BYTE_P (c) || c >= 0xA0)			  \
      ? (*(str) = (unsigned char)(c), 1)				  \
      : (*(str) = LEADING_CODE_8_BIT_CONTROL, *((str)+ 1) = c + 0x20, 2)) \
   : char_to_string (c, (unsigned char *) str))

/* Like CHAR_STRING but don't signal an error if C is invalid.
   Value is -1 in this case.  */

#define CHAR_STRING_NO_SIGNAL(c, str)					  \
  (SINGLE_BYTE_CHAR_P (c)						  \
   ? ((ASCII_BYTE_P (c) || c >= 0xA0)					  \
      ? (*(str) = (unsigned char)(c), 1)				  \
      : (*(str) = LEADING_CODE_8_BIT_CONTROL, *((str)+ 1) = c + 0x20, 2)) \
   : char_to_string_1 (c, (unsigned char *) str))

/* Return a character code of the character of which multi-byte form
   is at STR and the length is LEN.  If STR doesn't contain valid
   multi-byte form, only the first byte in STR is returned.  */

#define STRING_CHAR(str, len)				\
  (BYTES_BY_CHAR_HEAD ((unsigned char) *(str)) == 1	\
   ? (unsigned char) *(str)				\
   : string_to_char (str, len, 0))

/* This is like STRING_CHAR but the third arg ACTUAL_LEN is set to the
   length of the multi-byte form.  Just to know the length, use
   MULTIBYTE_FORM_LENGTH.  */

#define STRING_CHAR_AND_LENGTH(str, len, actual_len)	\
  (BYTES_BY_CHAR_HEAD ((unsigned char) *(str)) == 1	\
   ? ((actual_len) = 1), (unsigned char) *(str)		\
   : string_to_char (str, len, &(actual_len)))

/* Fetch the "next" character from Lisp string STRING at byte position
   BYTEIDX, character position CHARIDX.  Store it into OUTPUT.

   All the args must be side-effect-free.
   BYTEIDX and CHARIDX must be lvalues;
   we increment them past the character fetched.  */

#define FETCH_STRING_CHAR_ADVANCE(OUTPUT, STRING, CHARIDX, BYTEIDX)	   \
if (1)									   \
  {									   \
    CHARIDX++;								   \
    if (STRING_MULTIBYTE (STRING))					   \
      {									   \
	const unsigned char *ptr = SDATA (STRING) + BYTEIDX;		   \
	int space_left = SBYTES (STRING) - BYTEIDX;			   \
	int actual_len;							   \
									   \
	OUTPUT = STRING_CHAR_AND_LENGTH (ptr, space_left, actual_len);	   \
	BYTEIDX += actual_len;						   \
      }									   \
    else								   \
      OUTPUT = SREF (STRING, BYTEIDX++);				   \
  }									   \
else

/* Like FETCH_STRING_CHAR_ADVANCE but assume STRING is multibyte.  */

#define FETCH_STRING_CHAR_ADVANCE_NO_CHECK(OUTPUT, STRING, CHARIDX, BYTEIDX)  \
if (1)									      \
  {									      \
    const unsigned char *fetch_string_char_ptr = SDATA (STRING) + BYTEIDX;    \
    int fetch_string_char_space_left = SBYTES (STRING) - BYTEIDX;	      \
    int actual_len;							      \
    									      \
    OUTPUT								      \
      = STRING_CHAR_AND_LENGTH (fetch_string_char_ptr,			      \
			        fetch_string_char_space_left, actual_len);    \
									      \
    BYTEIDX += actual_len;						      \
    CHARIDX++;								      \
  }									      \
else

/* Like FETCH_STRING_CHAR_ADVANCE but fetch character from the current
   buffer.  */

#define FETCH_CHAR_ADVANCE(OUTPUT, CHARIDX, BYTEIDX)			  \
if (1)									  \
  {									  \
    CHARIDX++;								  \
    if (!NILP (current_buffer->enable_multibyte_characters))		  \
      {									  \
	unsigned char *ptr = BYTE_POS_ADDR (BYTEIDX);			  \
	int space_left = ((CHARIDX < GPT ? GPT_BYTE : Z_BYTE) - BYTEIDX); \
	int actual_len;							  \
									  \
	OUTPUT= STRING_CHAR_AND_LENGTH (ptr, space_left, actual_len);	  \
	BYTEIDX += actual_len;						  \
      }									  \
    else								  \
      {									  \
	OUTPUT = *(BYTE_POS_ADDR (BYTEIDX));				  \
	BYTEIDX++;							  \
      }									  \
  }									  \
else

/* Return the length of the multi-byte form at string STR of length LEN.  */

#define MULTIBYTE_FORM_LENGTH(str, len)			\
  (BYTES_BY_CHAR_HEAD (*(unsigned char *)(str)) == 1	\
   ? 1							\
   : multibyte_form_length (str, len))

/* If P is before LIMIT, advance P to the next character boundary.  It
   assumes that P is already at a character boundary of the sane
   mulitbyte form whose end address is LIMIT.  */

#define NEXT_CHAR_BOUNDARY(p, limit)	\
  do {					\
    if ((p) < (limit))			\
      (p) += BYTES_BY_CHAR_HEAD (*(p));	\
  } while (0)


/* If P is after LIMIT, advance P to the previous character boundary.
   It assumes that P is already at a character boundary of the sane
   mulitbyte form whose beginning address is LIMIT.  */

#define PREV_CHAR_BOUNDARY(p, limit)					\
  do {									\
    if ((p) > (limit))							\
      {									\
	const unsigned char *p0 = (p);					\
	do {								\
	  p0--;								\
	} while (p0 >= limit && ! CHAR_HEAD_P (*p0));			\
	(p) = (BYTES_BY_CHAR_HEAD (*p0) == (p) - p0) ? p0 : (p) - 1;	\
      }									\
  } while (0)


#ifdef emacs

/* Increase the buffer byte position POS_BYTE of the current buffer to
   the next character boundary.  This macro relies on the fact that
   *GPT_ADDR and *Z_ADDR are always accessible and the values are
   '\0'.  No range checking of POS.  */

#ifdef BYTE_COMBINING_DEBUG

#define INC_POS(pos_byte)				\
  do {							\
    unsigned char *p = BYTE_POS_ADDR (pos_byte);	\
    if (BASE_LEADING_CODE_P (*p))			\
      {							\
	int len, bytes;					\
	len = Z_BYTE - pos_byte;			\
	PARSE_MULTIBYTE_SEQ (p, len, bytes);		\
	pos_byte += bytes;				\
      }							\
    else						\
      pos_byte++;					\
  } while (0)

#else  /* not BYTE_COMBINING_DEBUG */

#define INC_POS(pos_byte)				\
  do {							\
    unsigned char *p = BYTE_POS_ADDR (pos_byte);	\
    pos_byte += BYTES_BY_CHAR_HEAD (*p);		\
  } while (0)

#endif /* not BYTE_COMBINING_DEBUG */

/* Decrease the buffer byte position POS_BYTE of the current buffer to
   the previous character boundary.  No range checking of POS.  */
#define DEC_POS(pos_byte)						\
  do {									\
    unsigned char *p, *p_min;						\
    									\
    pos_byte--;								\
    if (pos_byte < GPT_BYTE)						\
      p = BEG_ADDR + pos_byte - BEG_BYTE, p_min = BEG_ADDR;		\
    else								\
      p = BEG_ADDR + GAP_SIZE + pos_byte - BEG_BYTE, p_min = GAP_END_ADDR;\
    if (p > p_min && !CHAR_HEAD_P (*p))					\
      {									\
	unsigned char *pend = p--;					\
	int len, bytes;							\
        if (p_min < p - MAX_MULTIBYTE_LENGTH)				\
          p_min = p - MAX_MULTIBYTE_LENGTH;				\
	while (p > p_min && !CHAR_HEAD_P (*p)) p--;			\
	len = pend + 1 - p;						\
	PARSE_MULTIBYTE_SEQ (p, len, bytes);				\
	if (bytes == len)						\
	  pos_byte -= len - 1;						\
      }									\
  } while (0)

/* Increment both CHARPOS and BYTEPOS, each in the appropriate way.  */

#define INC_BOTH(charpos, bytepos)				\
do								\
  {								\
    (charpos)++;						\
    if (NILP (current_buffer->enable_multibyte_characters))	\
      (bytepos)++;						\
    else							\
      INC_POS ((bytepos));					\
  }								\
while (0)

/* Decrement both CHARPOS and BYTEPOS, each in the appropriate way.  */

#define DEC_BOTH(charpos, bytepos)				\
do								\
  {								\
    (charpos)--;						\
    if (NILP (current_buffer->enable_multibyte_characters))	\
      (bytepos)--;						\
    else							\
      DEC_POS ((bytepos));					\
  }								\
while (0)

/* Increase the buffer byte position POS_BYTE of the current buffer to
   the next character boundary.  This macro relies on the fact that
   *GPT_ADDR and *Z_ADDR are always accessible and the values are
   '\0'.  No range checking of POS_BYTE.  */

#ifdef BYTE_COMBINING_DEBUG

#define BUF_INC_POS(buf, pos_byte)				\
  do {								\
    unsigned char *p = BUF_BYTE_ADDRESS (buf, pos_byte);	\
    if (BASE_LEADING_CODE_P (*p))				\
      {								\
	int len, bytes;						\
	len = BUF_Z_BYTE (buf) - pos_byte;			\
	PARSE_MULTIBYTE_SEQ (p, len, bytes);			\
	pos_byte += bytes;					\
      }								\
    else							\
      pos_byte++;						\
  } while (0)

#else  /* not BYTE_COMBINING_DEBUG */

#define BUF_INC_POS(buf, pos_byte)				\
  do {								\
    unsigned char *p = BUF_BYTE_ADDRESS (buf, pos_byte);	\
    pos_byte += BYTES_BY_CHAR_HEAD (*p);			\
  } while (0)

#endif /* not BYTE_COMBINING_DEBUG */

/* Decrease the buffer byte position POS_BYTE of the current buffer to
   the previous character boundary.  No range checking of POS_BYTE.  */
#define BUF_DEC_POS(buf, pos_byte)					\
  do {									\
    unsigned char *p, *p_min;						\
    pos_byte--;								\
    if (pos_byte < BUF_GPT_BYTE (buf))					\
      {									\
	p = BUF_BEG_ADDR (buf) + pos_byte - BEG_BYTE;			\
	p_min = BUF_BEG_ADDR (buf);					\
      }									\
    else								\
      {									\
	p = BUF_BEG_ADDR (buf) + BUF_GAP_SIZE (buf) + pos_byte - BEG_BYTE;\
	p_min = BUF_GAP_END_ADDR (buf);					\
      }									\
    if (p > p_min && !CHAR_HEAD_P (*p))					\
      {									\
	unsigned char *pend = p--;					\
	int len, bytes;							\
        if (p_min < p - MAX_MULTIBYTE_LENGTH)				\
          p_min = p - MAX_MULTIBYTE_LENGTH;				\
	while (p > p_min && !CHAR_HEAD_P (*p)) p--;			\
	len = pend + 1 - p;						\
	PARSE_MULTIBYTE_SEQ (p, len, bytes);				\
	if (bytes == len)						\
	  pos_byte -= len - 1;						\
      }									\
  } while (0)

#endif /* emacs */

/* This is the maximum byte length of multi-byte sequence.  */
#define MAX_MULTIBYTE_LENGTH 4

extern void invalid_character P_ ((int));

extern int translate_char P_ ((Lisp_Object, int, int, int, int));
extern int split_string P_ ((const unsigned char *, int, int *,
				       unsigned char *, unsigned char *));
extern int char_to_string P_ ((int, unsigned char *));
extern int char_to_string_1 P_ ((int, unsigned char *));
extern int string_to_char P_ ((const unsigned char *, int, int *));
extern int char_printable_p P_ ((int c));
extern int multibyte_form_length P_ ((const unsigned char *, int));
extern void parse_str_as_multibyte P_ ((const unsigned char *, int, int *,
					int *));
extern int str_as_multibyte P_ ((unsigned char *, int, int, int *));
extern int parse_str_to_multibyte P_ ((unsigned char *, int));
extern int str_to_multibyte P_ ((unsigned char *, int, int));
extern int str_as_unibyte P_ ((unsigned char *, int));
extern int get_charset_id P_ ((Lisp_Object));
extern int find_charset_in_text P_ ((const unsigned char *, int, int, int *,
				    Lisp_Object));
extern int strwidth P_ ((unsigned char *, int));
extern int c_string_width P_ ((const unsigned char *, int, int, int *, int *));
extern int lisp_string_width P_ ((Lisp_Object, int, int *, int *));
extern int char_bytes P_ ((int));
extern int char_valid_p P_ ((int, int));

EXFUN (Funibyte_char_to_multibyte, 1);

extern Lisp_Object Vtranslation_table_vector;

/* Return a translation table of id number ID.  */
#define GET_TRANSLATION_TABLE(id) \
  (XCDR(XVECTOR(Vtranslation_table_vector)->contents[(id)]))

/* A char-table for characters which may invoke auto-filling.  */
extern Lisp_Object Vauto_fill_chars;

/* Copy LEN bytes from FROM to TO.  This macro should be used only
   when a caller knows that LEN is short and the obvious copy loop is
   faster than calling bcopy which has some overhead.  Copying a
   multibyte sequence of a multibyte character is the typical case.  */

#define BCOPY_SHORT(from, to, len)		\
  do {						\
    int i = len;				\
    const unsigned char *from_p = from;		\
    unsigned char *to_p = to;			\
    while (i--) *to_p++ = *from_p++;		\
  } while (0)

#endif /* EMACS_CHARSET_H */

/* arch-tag: 3b96db55-4961-481d-ac3e-219f46a2b3aa
   (do not change this comment) */