/* CCL (Code Conversion Language) interpreter.
Copyright (C) 1995, 1997 Electrotechnical Laboratory, JAPAN.
+ Copyright (C) 2001, 2002 Free Software Foundation, Inc.
Licensed to the Free Software Foundation.
This file is part of GNU Emacs.
the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
Boston, MA 02111-1307, USA. */
-#include <stdio.h>
+#include <config.h>
-#ifdef emacs
+#include <stdio.h>
-#include <config.h>
#include "lisp.h"
#include "charset.h"
#include "ccl.h"
#include "coding.h"
-#else /* not emacs */
-
-#include "mulelib.h"
-
-#endif /* not emacs */
-
-/* Where is stored translation tables for CCL program. */
-Lisp_Object Vccl_translation_table_vector;
+/* This contains all code conversion map available to CCL. */
+Lisp_Object Vcode_conversion_map_vector;
/* Alist of fontname patterns vs corresponding CCL program. */
Lisp_Object Vfont_ccl_encoder_alist;
-/* This symbol is property which assocate with ccl program vector. e.g.
- (get 'ccl-big5-encoder 'ccl-program) returns ccl program vector */
+/* This symbol is a property which assocates with ccl program vector.
+ Ex: (get 'ccl-big5-encoder 'ccl-program) returns ccl program vector. */
Lisp_Object Qccl_program;
-/* These symbol is properties whish associate with ccl translation table and its id
- respectively. */
-Lisp_Object Qccl_translation_table;
-Lisp_Object Qccl_translation_table_id;
+/* These symbols are properties which associate with code conversion
+ map and their ID respectively. */
+Lisp_Object Qcode_conversion_map;
+Lisp_Object Qcode_conversion_map_id;
+
+/* Symbols of ccl program have this property, a value of the property
+ is an index for Vccl_protram_table. */
+Lisp_Object Qccl_program_idx;
-/* Vector of CCL program names vs corresponding program data. */
+/* Table of registered CCL programs. Each element is a vector of
+ NAME, CCL_PROG, and RESOLVEDP where NAME (symbol) is the name of
+ the program, CCL_PROG (vector) is the compiled code of the program,
+ RESOLVEDP (t or nil) is the flag to tell if symbols in CCL_PROG is
+ already resolved to index numbers or not. */
Lisp_Object Vccl_program_table;
+/* Vector of registered hash tables for translation. */
+Lisp_Object Vtranslation_hash_table_vector;
+
+/* Return a hash table of id number ID. */
+#define GET_HASH_TABLE(id) \
+ (XHASH_TABLE (XCDR(XVECTOR(Vtranslation_hash_table_vector)->contents[(id)])))
+
/* CCL (Code Conversion Language) is a simple language which has
operations on one input buffer, one output buffer, and 7 registers.
The syntax of CCL is described in `ccl.el'. Emacs Lisp function
*/
#define CCL_Call 0x13 /* Call the CCL program whose ID is
- (CC..C).
- 1:CCCCCCCCCCCCCCCCCCCC000XXXXX
+ CC..C or cc..c.
+ 1:CCCCCCCCCCCCCCCCCCCCFFFXXXXX
+ [2:00000000cccccccccccccccccccc]
------------------------------
- call (CC..C)
+ if (FFF)
+ call (cc..c)
+ IC++;
+ else
+ call (CC..C)
*/
#define CCL_WriteConstString 0x14 /* Write a constant or a string:
IC += 2;
*/
-#define CCL_Extention 0x1F /* Extended CCL code
+#define CCL_Extension 0x1F /* Extended CCL code
1:ExtendedCOMMNDRrrRRRrrrXXXXX
2:ARGUEMENT
3:...
extended_command (rrr,RRR,Rrr,ARGS)
*/
-/*
- From here, Extended CCL Instruction.
+/*
+ Here after, Extended CCL Instructions.
Bit length of extended command is 14.
- Therefore the instruction code begins from 0 to 16384(0x3fff).
+ Therefore, the instruction code range is 0..16384(0x3fff).
*/
-#define CCL_ReadMultibyteCharacter 0x00 /* Read Multibyte Character
- 1:ExtendedCOMMNDRrrRRRrrrXXXXX
-
- Read a multibyte characeter.
- A code point is stored
- into rrr register.
- A charset ID is stored
- into RRR register.
- */
-#define CCL_WriteMultibyteCharacter 0x01 /* Write Multibyte Character
- 1:ExtendedCOMMNDRrrRRRrrrXXXXX
-
- Write a multibyte character.
- Write a character whose code point
- is in rrr register, and its charset ID
- is in RRR charset.
- */
-#define CCL_UnifyCharacter 0x02 /* Unify Multibyte Character
- 1:ExtendedCOMMNDRrrRRRrrrXXXXX
-
- Unify a character where its code point
- is in rrr register, and its charset ID
- is in RRR register with the table of
- the unification table ID
- in Rrr register.
-
- Return a unified character where its
- code point is in rrr register, and its
- charset ID is in RRR register.
- */
-#define CCL_UnifyCharacterConstTbl 0x03 /* Unify Multibyte Character
- 1:ExtendedCOMMNDRrrRRRrrrXXXXX
- 2:ARGUMENT(Unification Table ID)
-
- Unify a character where its code point
- is in rrr register, and its charset ID
- is in RRR register with the table of
- the unification table ID
- in 2nd argument.
-
- Return a unified character where its
- code point is in rrr register, and its
- charset ID is in RRR register.
- */
-#define CCL_IterateMultipleMap 0x10 /* Iterate Multiple Map
- 1:ExtendedCOMMNDXXXRRRrrrXXXXX
- 2:NUMBER of TABLES
- 3:TABLE-ID1
- 4:TABLE-ID2
- ...
-
- iterate to lookup tables from a number
- until finding a value.
-
- Each table consists of a vector
- whose element is number or
- nil or t or lambda.
- If the element is nil,
- its table is neglected.
- In the case of t or lambda,
- return the original value.
-
- */
-#define CCL_TranslateMultipleMap 0x11 /* Translate Multiple Map
- 1:ExtendedCOMMNDXXXRRRrrrXXXXX
- 2:NUMBER of TABLE-IDs and SEPARATERs
- (i.e. m1+m2+m3+...mk+k-1)
- 3:TABLE-ID 1,1
- 4:TABLE-ID 1,2
- ...
- m1+2:TABLE-ID 1,m1
- m1+3: -1 (SEPARATOR)
- m1+4:TABLE-ID 2,1
- ...
- m1+m2+4:TABLE-ID 2,m2
- m1+m2+5: -1
- ...
- m1+m2+...+mk+k+1:TABLE-ID k,mk
-
- Translate the code point in
- rrr register by tables.
- Translation starts from the table
- where RRR register points out.
-
- We translate the given value
- from the tables which are separated
- by -1.
- When each translation is failed to find
- any values, we regard the traslation
- as identity.
-
- We iterate to traslate by using each
- table set(tables separated by -1)
- until lookup the last table except
- lookup lambda.
-
- Each table consists of a vector
- whose element is number
- or nil or t or lambda.
- If the element is nil,
- it is neglected and use the next table.
- In the case of t,
- it is translated to the original value.
- In the case of lambda,
- it cease the translation and return the
- current value.
-
- */
-#define CCL_TranslateSingleMap 0x12 /* Translate Single Map
- 1:ExtendedCOMMNDXXXRRRrrrXXXXX
- 2:TABLE-ID
-
- Translate a number in rrr register.
- If it is not found any translation,
- set RRR register -1 but rrr register
- is not changed.
- */
+/* Read a multibyte characeter.
+ A code point is stored into reg[rrr]. A charset ID is stored into
+ reg[RRR]. */
+
+#define CCL_ReadMultibyteChar2 0x00 /* Read Multibyte Character
+ 1:ExtendedCOMMNDRrrRRRrrrXXXXX */
+
+/* Write a multibyte character.
+ Write a character whose code point is reg[rrr] and the charset ID
+ is reg[RRR]. */
+
+#define CCL_WriteMultibyteChar2 0x01 /* Write Multibyte Character
+ 1:ExtendedCOMMNDRrrRRRrrrXXXXX */
+
+/* Translate a character whose code point is reg[rrr] and the charset
+ ID is reg[RRR] by a translation table whose ID is reg[Rrr].
+
+ A translated character is set in reg[rrr] (code point) and reg[RRR]
+ (charset ID). */
+
+#define CCL_TranslateCharacter 0x02 /* Translate a multibyte character
+ 1:ExtendedCOMMNDRrrRRRrrrXXXXX */
+
+/* Translate a character whose code point is reg[rrr] and the charset
+ ID is reg[RRR] by a translation table whose ID is ARGUMENT.
+
+ A translated character is set in reg[rrr] (code point) and reg[RRR]
+ (charset ID). */
+
+#define CCL_TranslateCharacterConstTbl 0x03 /* Translate a multibyte character
+ 1:ExtendedCOMMNDRrrRRRrrrXXXXX
+ 2:ARGUMENT(Translation Table ID)
+ */
+
+/* Iterate looking up MAPs for reg[rrr] starting from the Nth (N =
+ reg[RRR]) MAP until some value is found.
+
+ Each MAP is a Lisp vector whose element is number, nil, t, or
+ lambda.
+ If the element is nil, ignore the map and proceed to the next map.
+ If the element is t or lambda, finish without changing reg[rrr].
+ If the element is a number, set reg[rrr] to the number and finish.
+
+ Detail of the map structure is descibed in the comment for
+ CCL_MapMultiple below. */
+
+#define CCL_IterateMultipleMap 0x10 /* Iterate multiple maps
+ 1:ExtendedCOMMNDXXXRRRrrrXXXXX
+ 2:NUMBER of MAPs
+ 3:MAP-ID1
+ 4:MAP-ID2
+ ...
+ */
+
+/* Map the code in reg[rrr] by MAPs starting from the Nth (N =
+ reg[RRR]) map.
+
+ MAPs are supplied in the succeeding CCL codes as follows:
+
+ When CCL program gives this nested structure of map to this command:
+ ((MAP-ID11
+ MAP-ID12
+ (MAP-ID121 MAP-ID122 MAP-ID123)
+ MAP-ID13)
+ (MAP-ID21
+ (MAP-ID211 (MAP-ID2111) MAP-ID212)
+ MAP-ID22)),
+ the compiled CCL codes has this sequence:
+ CCL_MapMultiple (CCL code of this command)
+ 16 (total number of MAPs and SEPARATORs)
+ -7 (1st SEPARATOR)
+ MAP-ID11
+ MAP-ID12
+ -3 (2nd SEPARATOR)
+ MAP-ID121
+ MAP-ID122
+ MAP-ID123
+ MAP-ID13
+ -7 (3rd SEPARATOR)
+ MAP-ID21
+ -4 (4th SEPARATOR)
+ MAP-ID211
+ -1 (5th SEPARATOR)
+ MAP_ID2111
+ MAP-ID212
+ MAP-ID22
+
+ A value of each SEPARATOR follows this rule:
+ MAP-SET := SEPARATOR [(MAP-ID | MAP-SET)]+
+ SEPARATOR := -(number of MAP-IDs and SEPARATORs in the MAP-SET)
+
+ (*)....Nest level of MAP-SET must not be over than MAX_MAP_SET_LEVEL.
+
+ When some map fails to map (i.e. it doesn't have a value for
+ reg[rrr]), the mapping is treated as identity.
+
+ The mapping is iterated for all maps in each map set (set of maps
+ separated by SEPARATOR) except in the case that lambda is
+ encountered. More precisely, the mapping proceeds as below:
+
+ At first, VAL0 is set to reg[rrr], and it is translated by the
+ first map to VAL1. Then, VAL1 is translated by the next map to
+ VAL2. This mapping is iterated until the last map is used. The
+ result of the mapping is the last value of VAL?. When the mapping
+ process reached to the end of the map set, it moves to the next
+ map set. If the next does not exit, the mapping process terminates,
+ and regard the last value as a result.
+
+ But, when VALm is mapped to VALn and VALn is not a number, the
+ mapping proceed as below:
+
+ If VALn is nil, the lastest map is ignored and the mapping of VALm
+ proceed to the next map.
+
+ In VALn is t, VALm is reverted to reg[rrr] and the mapping of VALm
+ proceed to the next map.
+
+ If VALn is lambda, move to the next map set like reaching to the
+ end of the current map set.
+
+ If VALn is a symbol, call the CCL program refered by it.
+ Then, use reg[rrr] as a mapped value except for -1, -2 and -3.
+ Such special values are regarded as nil, t, and lambda respectively.
+
+ Each map is a Lisp vector of the following format (a) or (b):
+ (a)......[STARTPOINT VAL1 VAL2 ...]
+ (b)......[t VAL STARTPOINT ENDPOINT],
+ where
+ STARTPOINT is an offset to be used for indexing a map,
+ ENDPOINT is a maximum index number of a map,
+ VAL and VALn is a number, nil, t, or lambda.
+
+ Valid index range of a map of type (a) is:
+ STARTPOINT <= index < STARTPOINT + map_size - 1
+ Valid index range of a map of type (b) is:
+ STARTPOINT <= index < ENDPOINT */
+
+#define CCL_MapMultiple 0x11 /* Mapping by multiple code conversion maps
+ 1:ExtendedCOMMNDXXXRRRrrrXXXXX
+ 2:N-2
+ 3:SEPARATOR_1 (< 0)
+ 4:MAP-ID_1
+ 5:MAP-ID_2
+ ...
+ M:SEPARATOR_x (< 0)
+ M+1:MAP-ID_y
+ ...
+ N:SEPARATOR_z (< 0)
+ */
+
+#define MAX_MAP_SET_LEVEL 30
+
+typedef struct
+{
+ int rest_length;
+ int orig_val;
+} tr_stack;
+
+static tr_stack mapping_stack[MAX_MAP_SET_LEVEL];
+static tr_stack *mapping_stack_pointer;
+
+/* If this variable is non-zero, it indicates the stack_idx
+ of immediately called by CCL_MapMultiple. */
+static int stack_idx_of_map_multiple;
+
+#define PUSH_MAPPING_STACK(restlen, orig) \
+do \
+ { \
+ mapping_stack_pointer->rest_length = (restlen); \
+ mapping_stack_pointer->orig_val = (orig); \
+ mapping_stack_pointer++; \
+ } \
+while (0)
+
+#define POP_MAPPING_STACK(restlen, orig) \
+do \
+ { \
+ mapping_stack_pointer--; \
+ (restlen) = mapping_stack_pointer->rest_length; \
+ (orig) = mapping_stack_pointer->orig_val; \
+ } \
+while (0)
+
+#define CCL_CALL_FOR_MAP_INSTRUCTION(symbol, ret_ic) \
+do \
+ { \
+ struct ccl_program called_ccl; \
+ if (stack_idx >= 256 \
+ || (setup_ccl_program (&called_ccl, (symbol)) != 0)) \
+ { \
+ if (stack_idx > 0) \
+ { \
+ ccl_prog = ccl_prog_stack_struct[0].ccl_prog; \
+ ic = ccl_prog_stack_struct[0].ic; \
+ eof_ic = ccl_prog_stack_struct[0].eof_ic; \
+ } \
+ CCL_INVALID_CMD; \
+ } \
+ ccl_prog_stack_struct[stack_idx].ccl_prog = ccl_prog; \
+ ccl_prog_stack_struct[stack_idx].ic = (ret_ic); \
+ ccl_prog_stack_struct[stack_idx].eof_ic = eof_ic; \
+ stack_idx++; \
+ ccl_prog = called_ccl.prog; \
+ ic = CCL_HEADER_MAIN; \
+ eof_ic = XFASTINT (ccl_prog[CCL_HEADER_EOF]); \
+ goto ccl_repeat; \
+ } \
+while (0)
+
+#define CCL_MapSingle 0x12 /* Map by single code conversion map
+ 1:ExtendedCOMMNDXXXRRRrrrXXXXX
+ 2:MAP-ID
+ ------------------------------
+ Map reg[rrr] by MAP-ID.
+ If some valid mapping is found,
+ set reg[rrr] to the result,
+ else
+ set reg[RRR] to -1.
+ */
+
+#define CCL_LookupIntConstTbl 0x13 /* Lookup multibyte character by
+ integer key. Afterwards R7 set
+ to 1 iff lookup succeeded.
+ 1:ExtendedCOMMNDRrrRRRXXXXXXXX
+ 2:ARGUMENT(Hash table ID) */
+
+#define CCL_LookupCharConstTbl 0x14 /* Lookup integer by multibyte
+ character key. Afterwards R7 set
+ to 1 iff lookup succeeded.
+ 1:ExtendedCOMMNDRrrRRRrrrXXXXX
+ 2:ARGUMENT(Hash table ID) */
/* CCL arithmetic/logical operators. */
#define CCL_PLUS 0x00 /* X = Y + Z */
#define CCL_GE 0x14 /* X = (X >= Y) */
#define CCL_NE 0x15 /* X = (X != Y) */
-#define CCL_ENCODE_SJIS 0x16 /* X = HIGHER_BYTE (SJIS (Y, Z))
- r[7] = LOWER_BYTE (SJIS (Y, Z) */
-#define CCL_DECODE_SJIS 0x17 /* X = HIGHER_BYTE (DE-SJIS (Y, Z))
+#define CCL_DECODE_SJIS 0x16 /* X = HIGHER_BYTE (DE-SJIS (Y, Z))
r[7] = LOWER_BYTE (DE-SJIS (Y, Z)) */
+#define CCL_ENCODE_SJIS 0x17 /* X = HIGHER_BYTE (SJIS (Y, Z))
+ r[7] = LOWER_BYTE (SJIS (Y, Z) */
/* Terminate CCL program successfully. */
-#define CCL_SUCCESS \
- do { \
+#define CCL_SUCCESS \
+do \
+ { \
ccl->status = CCL_STAT_SUCCESS; \
- ccl->ic = CCL_HEADER_MAIN; \
- goto ccl_finish; \
- } while (0)
+ goto ccl_finish; \
+ } \
+while(0)
/* Suspend CCL program because of reading from empty input buffer or
writing to full output buffer. When this program is resumed, the
same I/O command is executed. */
#define CCL_SUSPEND(stat) \
- do { \
+do \
+ { \
ic--; \
ccl->status = stat; \
goto ccl_finish; \
- } while (0)
+ } \
+while (0)
/* Terminate CCL program because of invalid command. Should not occur
in the normal case. */
+#ifndef CCL_DEBUG
+
#define CCL_INVALID_CMD \
- do { \
+do \
+ { \
ccl->status = CCL_STAT_INVALID_CMD; \
goto ccl_error_handler; \
- } while (0)
+ } \
+while(0)
+
+#else
+
+#define CCL_INVALID_CMD \
+do \
+ { \
+ ccl_debug_hook (this_ic); \
+ ccl->status = CCL_STAT_INVALID_CMD; \
+ goto ccl_error_handler; \
+ } \
+while(0)
+
+#endif
/* Encode one character CH to multibyte form and write to the current
output buffer. If CH is less than 256, CH is written as is. */
-#define CCL_WRITE_CHAR(ch) \
- do { \
- if (!dst) \
- CCL_INVALID_CMD; \
- else \
- { \
- unsigned char work[4], *str; \
- int len = CHAR_STRING (ch, work, str); \
- if (dst + len <= (dst_bytes ? dst_end : src)) \
- { \
- bcopy (str, dst, len); \
- dst += len; \
- } \
- else \
- CCL_SUSPEND (CCL_STAT_SUSPEND_BY_DST); \
- } \
+#define CCL_WRITE_CHAR(ch) \
+ do { \
+ int bytes = SINGLE_BYTE_CHAR_P (ch) ? 1: CHAR_BYTES (ch); \
+ if (!dst) \
+ CCL_INVALID_CMD; \
+ else if (dst + bytes + extra_bytes < (dst_bytes ? dst_end : src)) \
+ { \
+ if (bytes == 1) \
+ { \
+ *dst++ = (ch); \
+ if (extra_bytes && (ch) >= 0x80 && (ch) < 0xA0) \
+ /* We may have to convert this eight-bit char to \
+ multibyte form later. */ \
+ extra_bytes++; \
+ } \
+ else if (CHAR_VALID_P (ch, 0)) \
+ dst += CHAR_STRING (ch, dst); \
+ else \
+ CCL_INVALID_CMD; \
+ } \
+ else \
+ CCL_SUSPEND (CCL_STAT_SUSPEND_BY_DST); \
+ } while (0)
+
+/* Encode one character CH to multibyte form and write to the current
+ output buffer. The output bytes always forms a valid multibyte
+ sequence. */
+#define CCL_WRITE_MULTIBYTE_CHAR(ch) \
+ do { \
+ int bytes = CHAR_BYTES (ch); \
+ if (!dst) \
+ CCL_INVALID_CMD; \
+ else if (dst + bytes + extra_bytes < (dst_bytes ? dst_end : src)) \
+ { \
+ if (CHAR_VALID_P ((ch), 0)) \
+ dst += CHAR_STRING ((ch), dst); \
+ else \
+ CCL_INVALID_CMD; \
+ } \
+ else \
+ CCL_SUSPEND (CCL_STAT_SUSPEND_BY_DST); \
} while (0)
/* Write a string at ccl_prog[IC] of length LEN to the current output
CCL_SUSPEND (CCL_STAT_SUSPEND_BY_DST); \
} while (0)
-/* Read one byte from the current input buffer into Rth register. */
-#define CCL_READ_CHAR(r) \
- do { \
- if (!src) \
- CCL_INVALID_CMD; \
- else if (src < src_end) \
- r = *src++; \
- else if (ccl->last_block) \
- { \
- ic = ccl->eof_ic; \
- goto ccl_finish; \
- } \
- else \
- CCL_SUSPEND (CCL_STAT_SUSPEND_BY_SRC); \
+/* Read one byte from the current input buffer into REGth register. */
+#define CCL_READ_CHAR(REG) \
+ do { \
+ if (!src) \
+ CCL_INVALID_CMD; \
+ else if (src < src_end) \
+ { \
+ REG = *src++; \
+ if (REG == '\n' \
+ && ccl->eol_type != CODING_EOL_LF) \
+ { \
+ /* We are encoding. */ \
+ if (ccl->eol_type == CODING_EOL_CRLF) \
+ { \
+ if (ccl->cr_consumed) \
+ ccl->cr_consumed = 0; \
+ else \
+ { \
+ ccl->cr_consumed = 1; \
+ REG = '\r'; \
+ src--; \
+ } \
+ } \
+ else \
+ REG = '\r'; \
+ } \
+ if (REG == LEADING_CODE_8_BIT_CONTROL \
+ && ccl->multibyte) \
+ REG = *src++ - 0x20; \
+ } \
+ else if (ccl->last_block) \
+ { \
+ REG = -1; \
+ ic = eof_ic; \
+ goto ccl_repeat; \
+ } \
+ else \
+ CCL_SUSPEND (CCL_STAT_SUSPEND_BY_SRC); \
+ } while (0)
+
+
+/* Set C to the character code made from CHARSET and CODE. This is
+ like MAKE_CHAR but check the validity of CHARSET and CODE. If they
+ are not valid, set C to (CODE & 0xFF) because that is usually the
+ case that CCL_ReadMultibyteChar2 read an invalid code and it set
+ CODE to that invalid byte. */
+
+#define CCL_MAKE_CHAR(charset, code, c) \
+ do { \
+ if (charset == CHARSET_ASCII) \
+ c = code & 0xFF; \
+ else if (CHARSET_DEFINED_P (charset) \
+ && (code & 0x7F) >= 32 \
+ && (code < 256 || ((code >> 7) & 0x7F) >= 32)) \
+ { \
+ int c1 = code & 0x7F, c2 = 0; \
+ \
+ if (code >= 256) \
+ c2 = c1, c1 = (code >> 7) & 0x7F; \
+ c = MAKE_CHAR (charset, c1, c2); \
+ } \
+ else \
+ c = code & 0xFF; \
} while (0)
#ifdef CCL_DEBUG
#define CCL_DEBUG_BACKTRACE_LEN 256
-int ccl_backtrace_table[CCL_BACKTRACE_TABLE];
+int ccl_backtrace_table[CCL_DEBUG_BACKTRACE_LEN];
int ccl_backtrace_idx;
+
+int
+ccl_debug_hook (int ic)
+{
+ return ic;
+}
+
#endif
struct ccl_prog_stack
{
Lisp_Object *ccl_prog; /* Pointer to an array of CCL code. */
int ic; /* Instruction Counter. */
+ int eof_ic; /* Instruction Counter to jump on EOF. */
};
+/* For the moment, we only support depth 256 of stack. */
+static struct ccl_prog_stack ccl_prog_stack_struct[256];
+
+int
ccl_driver (ccl, source, destination, src_bytes, dst_bytes, consumed)
struct ccl_program *ccl;
unsigned char *source, *destination;
{
register int *reg = ccl->reg;
register int ic = ccl->ic;
- register int code, field1, field2;
+ register int code = 0, field1, field2;
register Lisp_Object *ccl_prog = ccl->prog;
unsigned char *src = source, *src_end = src + src_bytes;
unsigned char *dst = destination, *dst_end = dst + dst_bytes;
int jump_address;
- int i, j, op;
- int stack_idx = 0;
- /* For the moment, we only support depth 256 of stack. */
- struct ccl_prog_stack ccl_prog_stack_struct[256];
-
- if (ic >= ccl->eof_ic)
+ int i = 0, j, op;
+ int stack_idx = ccl->stack_idx;
+ /* Instruction counter of the current CCL code. */
+ int this_ic = 0;
+ /* CCL_WRITE_CHAR will produce 8-bit code of range 0x80..0x9F. But,
+ each of them will be converted to multibyte form of 2-byte
+ sequence. For that conversion, we remember how many more bytes
+ we must keep in DESTINATION in this variable. */
+ int extra_bytes = ccl->eight_bit_control;
+ int eof_ic = ccl->eof_ic;
+ int eof_hit = 0;
+
+ if (ic >= eof_ic)
ic = CCL_HEADER_MAIN;
+ if (ccl->buf_magnification == 0) /* We can't produce any bytes. */
+ dst = NULL;
+
+ /* Set mapping stack pointer. */
+ mapping_stack_pointer = mapping_stack;
+
#ifdef CCL_DEBUG
ccl_backtrace_idx = 0;
#endif
for (;;)
{
+ ccl_repeat:
#ifdef CCL_DEBUG
ccl_backtrace_table[ccl_backtrace_idx++] = ic;
if (ccl_backtrace_idx >= CCL_DEBUG_BACKTRACE_LEN)
break;
}
+ this_ic = ic;
code = XINT (ccl_prog[ic]); ic++;
field1 = code >> 8;
field2 = (code & 0xFF) >> 5;
i = reg[RRR];
j = XINT (ccl_prog[ic]);
op = field1 >> 6;
- ic++;
+ jump_address = ic + 1;
goto ccl_set_expr;
case CCL_WriteRegister: /* CCCCCCCCCCCCCCCCCCCrrrXXXXX */
i = reg[RRR];
j = reg[Rrr];
op = field1 >> 6;
+ jump_address = ic;
goto ccl_set_expr;
- case CCL_Call: /* CCCCCCCCCCCCCCCCCCCC000XXXXX */
+ case CCL_Call: /* 1:CCCCCCCCCCCCCCCCCCCCFFFXXXXX */
{
Lisp_Object slot;
+ int prog_id;
+
+ /* If FFF is nonzero, the CCL program ID is in the
+ following code. */
+ if (rrr)
+ {
+ prog_id = XINT (ccl_prog[ic]);
+ ic++;
+ }
+ else
+ prog_id = field1;
if (stack_idx >= 256
- || field1 < 0
- || field1 >= XVECTOR (Vccl_program_table)->size
- || (slot = XVECTOR (Vccl_program_table)->contents[field1],
- !CONSP (slot))
- || !VECTORP (XCONS (slot)->cdr))
+ || prog_id < 0
+ || prog_id >= ASIZE (Vccl_program_table)
+ || (slot = AREF (Vccl_program_table, prog_id), !VECTORP (slot))
+ || !VECTORP (AREF (slot, 1)))
{
if (stack_idx > 0)
{
ccl_prog = ccl_prog_stack_struct[0].ccl_prog;
ic = ccl_prog_stack_struct[0].ic;
+ eof_ic = ccl_prog_stack_struct[0].eof_ic;
}
CCL_INVALID_CMD;
}
-
+
ccl_prog_stack_struct[stack_idx].ccl_prog = ccl_prog;
ccl_prog_stack_struct[stack_idx].ic = ic;
+ ccl_prog_stack_struct[stack_idx].eof_ic = eof_ic;
stack_idx++;
- ccl_prog = XVECTOR (XCONS (slot)->cdr)->contents;
+ ccl_prog = XVECTOR (AREF (slot, 1))->contents;
ic = CCL_HEADER_MAIN;
+ eof_ic = XFASTINT (ccl_prog[CCL_HEADER_EOF]);
}
break;
break;
case CCL_End: /* 0000000000000000000000XXXXX */
- if (stack_idx-- > 0)
+ if (stack_idx > 0)
{
+ stack_idx--;
ccl_prog = ccl_prog_stack_struct[stack_idx].ccl_prog;
ic = ccl_prog_stack_struct[stack_idx].ic;
+ eof_ic = ccl_prog_stack_struct[stack_idx].eof_ic;
+ if (eof_hit)
+ ic = eof_ic;
break;
}
+ if (src)
+ src = src_end;
+ /* ccl->ic should points to this command code again to
+ suppress further processing. */
+ ic--;
CCL_SUCCESS;
case CCL_ExprSelfConst: /* 00000OPERATION000000rrrXXXXX */
case CCL_LE: reg[rrr] = i <= j; break;
case CCL_GE: reg[rrr] = i >= j; break;
case CCL_NE: reg[rrr] = i != j; break;
- case CCL_ENCODE_SJIS: ENCODE_SJIS (i, j, reg[rrr], reg[7]); break;
case CCL_DECODE_SJIS: DECODE_SJIS (i, j, reg[rrr], reg[7]); break;
+ case CCL_ENCODE_SJIS: ENCODE_SJIS (i, j, reg[rrr], reg[7]); break;
default: CCL_INVALID_CMD;
}
code &= 0x1F;
{
i = reg[rrr];
CCL_WRITE_CHAR (i);
+ ic = jump_address;
}
else if (!reg[rrr])
ic = jump_address;
break;
- case CCL_Extention:
+ case CCL_Extension:
switch (EXCMD)
{
- case CCL_ReadMultibyteCharacter:
+ case CCL_ReadMultibyteChar2:
if (!src)
CCL_INVALID_CMD;
- do {
- if (src >= src_end)
+
+ if (src >= src_end)
+ {
+ src++;
goto ccl_read_multibyte_character_suspend;
-
- i = *src++;
- if (i == LEADING_CODE_COMPOSITION)
- {
- if (src >= src_end)
- goto ccl_read_multibyte_character_suspend;
- if (*src == 0xFF)
- {
- ccl->private_state = COMPOSING_WITH_RULE_HEAD;
- src++;
- }
- else
- ccl->private_state = COMPOSING_NO_RULE_HEAD;
- }
- if (ccl->private_state != 0)
- {
- /* composite character */
- if (*src < 0xA0)
- ccl->private_state = 0;
- else
- {
- if (i == 0xA0)
- {
- if (src >= src_end)
- goto ccl_read_multibyte_character_suspend;
- i = *src++ & 0x7F;
- }
- else
- i -= 0x20;
+ }
- if (COMPOSING_WITH_RULE_RULE == ccl->private_state)
- {
- ccl->private_state = COMPOSING_WITH_RULE_HEAD;
- continue;
- }
- else if (COMPOSING_WITH_RULE_HEAD == ccl->private_state)
- ccl->private_state = COMPOSING_WITH_RULE_RULE;
- }
- }
- if (i < 0x80)
- {
- /* ASCII */
- reg[rrr] = i;
- reg[RRR] = CHARSET_ASCII;
- }
- else if (i <= MAX_CHARSET_OFFICIAL_DIMENSION1)
- {
- if (src >= src_end)
- goto ccl_read_multibyte_character_suspend;
- reg[RRR] = i;
- reg[rrr] = (*src++ & 0x7F);
- }
- else if (i <= MAX_CHARSET_OFFICIAL_DIMENSION2)
- {
- if ((src + 1) >= src_end)
- goto ccl_read_multibyte_character_suspend;
- reg[RRR] = i;
- i = (*src++ & 0x7F);
- reg[rrr] = ((i << 7) | (*src & 0x7F));
- src++;
- }
- else if ((i == LEADING_CODE_PRIVATE_11) ||
- (i == LEADING_CODE_PRIVATE_12))
- {
- if ((src + 1) >= src_end)
- goto ccl_read_multibyte_character_suspend;
- reg[RRR] = *src++;
- reg[rrr] = (*src++ & 0x7F);
- }
- else if ((i == LEADING_CODE_PRIVATE_21) ||
- (i == LEADING_CODE_PRIVATE_22))
- {
- if ((src + 2) >= src_end)
- goto ccl_read_multibyte_character_suspend;
- reg[RRR] = *src++;
- i = (*src++ & 0x7F);
- reg[rrr] = ((i << 7) | (*src & 0x7F));
- src++;
- }
- else
- {
- /* INVALID CODE
- Returned charset is -1.*/
- reg[RRR] = -1;
- }
- } while (0);
+ if (!ccl->multibyte)
+ {
+ int bytes;
+ if (!UNIBYTE_STR_AS_MULTIBYTE_P (src, src_end - src, bytes))
+ {
+ reg[RRR] = CHARSET_8_BIT_CONTROL;
+ reg[rrr] = *src++;
+ break;
+ }
+ }
+ i = *src++;
+ if (i == '\n' && ccl->eol_type != CODING_EOL_LF)
+ {
+ /* We are encoding. */
+ if (ccl->eol_type == CODING_EOL_CRLF)
+ {
+ if (ccl->cr_consumed)
+ ccl->cr_consumed = 0;
+ else
+ {
+ ccl->cr_consumed = 1;
+ i = '\r';
+ src--;
+ }
+ }
+ else
+ i = '\r';
+ reg[rrr] = i;
+ reg[RRR] = CHARSET_ASCII;
+ }
+ else if (i < 0x80)
+ {
+ /* ASCII */
+ reg[rrr] = i;
+ reg[RRR] = CHARSET_ASCII;
+ }
+ else if (i <= MAX_CHARSET_OFFICIAL_DIMENSION2)
+ {
+ int dimension = BYTES_BY_CHAR_HEAD (i) - 1;
+
+ if (dimension == 0)
+ {
+ /* `i' is a leading code for an undefined charset. */
+ reg[RRR] = CHARSET_8_BIT_GRAPHIC;
+ reg[rrr] = i;
+ }
+ else if (src + dimension > src_end)
+ goto ccl_read_multibyte_character_suspend;
+ else
+ {
+ reg[RRR] = i;
+ i = (*src++ & 0x7F);
+ if (dimension == 1)
+ reg[rrr] = i;
+ else
+ reg[rrr] = ((i << 7) | (*src++ & 0x7F));
+ }
+ }
+ else if ((i == LEADING_CODE_PRIVATE_11)
+ || (i == LEADING_CODE_PRIVATE_12))
+ {
+ if ((src + 1) >= src_end)
+ goto ccl_read_multibyte_character_suspend;
+ reg[RRR] = *src++;
+ reg[rrr] = (*src++ & 0x7F);
+ }
+ else if ((i == LEADING_CODE_PRIVATE_21)
+ || (i == LEADING_CODE_PRIVATE_22))
+ {
+ if ((src + 2) >= src_end)
+ goto ccl_read_multibyte_character_suspend;
+ reg[RRR] = *src++;
+ i = (*src++ & 0x7F);
+ reg[rrr] = ((i << 7) | (*src & 0x7F));
+ src++;
+ }
+ else if (i == LEADING_CODE_8_BIT_CONTROL)
+ {
+ if (src >= src_end)
+ goto ccl_read_multibyte_character_suspend;
+ reg[RRR] = CHARSET_8_BIT_CONTROL;
+ reg[rrr] = (*src++ - 0x20);
+ }
+ else if (i >= 0xA0)
+ {
+ reg[RRR] = CHARSET_8_BIT_GRAPHIC;
+ reg[rrr] = i;
+ }
+ else
+ {
+ /* INVALID CODE. Return a single byte character. */
+ reg[RRR] = CHARSET_ASCII;
+ reg[rrr] = i;
+ }
break;
ccl_read_multibyte_character_suspend:
+ if (src <= src_end && !ccl->multibyte && ccl->last_block)
+ {
+ reg[RRR] = CHARSET_8_BIT_CONTROL;
+ reg[rrr] = i;
+ break;
+ }
src--;
if (ccl->last_block)
{
- ic = ccl->eof_ic;
- goto ccl_finish;
+ ic = eof_ic;
+ eof_hit = 1;
+ goto ccl_repeat;
}
else
CCL_SUSPEND (CCL_STAT_SUSPEND_BY_SRC);
break;
- case CCL_WriteMultibyteCharacter:
+ case CCL_WriteMultibyteChar2:
i = reg[RRR]; /* charset */
- if (i == CHARSET_ASCII)
- i = reg[rrr] & 0x7F;
- else if (i == CHARSET_COMPOSITION)
- i = MAKE_COMPOSITE_CHAR (reg[rrr]);
+ if (i == CHARSET_ASCII
+ || i == CHARSET_8_BIT_CONTROL
+ || i == CHARSET_8_BIT_GRAPHIC)
+ i = reg[rrr] & 0xFF;
else if (CHARSET_DIMENSION (i) == 1)
i = ((i - 0x70) << 7) | (reg[rrr] & 0x7F);
else if (i < MIN_CHARSET_PRIVATE_DIMENSION2)
else
i = ((i - 0xE0) << 14) | reg[rrr];
- CCL_WRITE_CHAR (i);
+ CCL_WRITE_MULTIBYTE_CHAR (i);
break;
- case CCL_UnifyCharacter:
- i = reg[RRR]; /* charset */
- if (i == CHARSET_ASCII)
- i = reg[rrr] & 0x7F;
- else if (i == CHARSET_COMPOSITION)
- {
- reg[RRR] = -1;
- break;
- }
- else if (CHARSET_DIMENSION (i) == 1)
- i = ((i - 0x70) << 7) | (reg[rrr] & 0x7F);
- else if (i < MIN_CHARSET_PRIVATE_DIMENSION2)
- i = ((i - 0x8F) << 14) | (reg[rrr] & 0x3FFF);
- else
- i = ((i - 0xE0) << 14) | (reg[rrr] & 0x3FFF);
-
- op = unify_char (UNIFICATION_ID_TABLE (reg[Rrr]), i, -1, 0, 0);
+ case CCL_TranslateCharacter:
+ CCL_MAKE_CHAR (reg[RRR], reg[rrr], i);
+ op = translate_char (GET_TRANSLATION_TABLE (reg[Rrr]),
+ i, -1, 0, 0);
SPLIT_CHAR (op, reg[RRR], i, j);
if (j != -1)
i = (i << 7) | j;
-
+
reg[rrr] = i;
break;
- case CCL_UnifyCharacterConstTbl:
+ case CCL_TranslateCharacterConstTbl:
op = XINT (ccl_prog[ic]); /* table */
ic++;
- i = reg[RRR]; /* charset */
- if (i == CHARSET_ASCII)
- i = reg[rrr] & 0x7F;
- else if (i == CHARSET_COMPOSITION)
- {
- reg[RRR] = -1;
- break;
- }
- else if (CHARSET_DIMENSION (i) == 1)
- i = ((i - 0x70) << 7) | (reg[rrr] & 0x7F);
- else if (i < MIN_CHARSET_PRIVATE_DIMENSION2)
- i = ((i - 0x8F) << 14) | (reg[rrr] & 0x3FFF);
- else
- i = ((i - 0xE0) << 14) | (reg[rrr] & 0x3FFF);
-
- op = unify_char (UNIFICATION_ID_TABLE (op), i, -1, 0, 0);
+ CCL_MAKE_CHAR (reg[RRR], reg[rrr], i);
+ op = translate_char (GET_TRANSLATION_TABLE (op), i, -1, 0, 0);
SPLIT_CHAR (op, reg[RRR], i, j);
if (j != -1)
i = (i << 7) | j;
-
+
reg[rrr] = i;
break;
+ case CCL_LookupIntConstTbl:
+ op = XINT (ccl_prog[ic]); /* table */
+ ic++;
+ {
+ struct Lisp_Hash_Table *h = GET_HASH_TABLE (op);
+
+ op = hash_lookup (h, make_number (reg[RRR]), NULL);
+ if (op >= 0)
+ {
+ Lisp_Object opl;
+ opl = HASH_VALUE (h, op);
+ if (!CHAR_VALID_P (XINT (opl), 0))
+ CCL_INVALID_CMD;
+ SPLIT_CHAR (XINT (opl), reg[RRR], i, j);
+ if (j != -1)
+ i = (i << 7) | j;
+ reg[rrr] = i;
+ reg[7] = 1; /* r7 true for success */
+ }
+ else
+ reg[7] = 0;
+ }
+ break;
+
+ case CCL_LookupCharConstTbl:
+ op = XINT (ccl_prog[ic]); /* table */
+ ic++;
+ CCL_MAKE_CHAR (reg[RRR], reg[rrr], i);
+ {
+ struct Lisp_Hash_Table *h = GET_HASH_TABLE (op);
+
+ op = hash_lookup (h, make_number (i), NULL);
+ if (op >= 0)
+ {
+ Lisp_Object opl;
+ opl = HASH_VALUE (h, op);
+ if (!INTEGERP (opl))
+ CCL_INVALID_CMD;
+ reg[RRR] = XINT (opl);
+ reg[7] = 1; /* r7 true for success */
+ }
+ else
+ reg[7] = 0;
+ }
+ break;
+
case CCL_IterateMultipleMap:
{
- Lisp_Object table, content, attrib, value;
+ Lisp_Object map, content, attrib, value;
int point, size, fin_ic;
- j = XINT (ccl_prog[ic++]); /* number of tables. */
+ j = XINT (ccl_prog[ic++]); /* number of maps. */
fin_ic = ic + j;
op = reg[rrr];
if ((j > reg[RRR]) && (j >= 0))
for (;i < j;i++)
{
- size = XVECTOR (Vccl_translation_table_vector)->size;
- point = ccl_prog[ic++];
+ size = ASIZE (Vcode_conversion_map_vector);
+ point = XINT (ccl_prog[ic++]);
if (point >= size) continue;
- table = XVECTOR (Vccl_translation_table_vector)->
- contents[point];
- if (!CONSP (table)) continue;
- table = XCONS(table)->cdr;
- if (!VECTORP (table)) continue;
- size = XVECTOR (table)->size;
+ map = AREF (Vcode_conversion_map_vector, point);
+
+ /* Check map varidity. */
+ if (!CONSP (map)) continue;
+ map = XCDR (map);
+ if (!VECTORP (map)) continue;
+ size = ASIZE (map);
if (size <= 1) continue;
- point = XUINT (XVECTOR (table)->contents[0]);
- point = op - point + 1;
- if (!((point >= 1) && (point < size))) continue;
- content = XVECTOR (table)->contents[point];
+
+ content = AREF (map, 0);
+
+ /* check map type,
+ [STARTPOINT VAL1 VAL2 ...] or
+ [t ELELMENT STARTPOINT ENDPOINT] */
+ if (NUMBERP (content))
+ {
+ point = XUINT (content);
+ point = op - point + 1;
+ if (!((point >= 1) && (point < size))) continue;
+ content = AREF (map, point);
+ }
+ else if (EQ (content, Qt))
+ {
+ if (size != 4) continue;
+ if ((op >= XUINT (AREF (map, 2)))
+ && (op < XUINT (AREF (map, 3))))
+ content = AREF (map, 1);
+ else
+ continue;
+ }
+ else
+ continue;
if (NILP (content))
continue;
else if (NUMBERP (content))
{
reg[RRR] = i;
- reg[rrr] = XUINT(content);
+ reg[rrr] = XINT(content);
break;
}
else if (EQ (content, Qt) || EQ (content, Qlambda))
}
else if (CONSP (content))
{
- attrib = XCONS (content)->car;
- value = XCONS (content)->cdr;
+ attrib = XCAR (content);
+ value = XCDR (content);
if (!NUMBERP (attrib) || !NUMBERP (value))
continue;
reg[RRR] = i;
- reg[rrr] = XUINT(value);
+ reg[rrr] = XUINT (value);
break;
}
+ else if (SYMBOLP (content))
+ CCL_CALL_FOR_MAP_INSTRUCTION (content, fin_ic);
+ else
+ CCL_INVALID_CMD;
}
if (i == j)
reg[RRR] = -1;
ic = fin_ic;
}
break;
-
- case CCL_TranslateMultipleMap:
+
+ case CCL_MapMultiple:
{
- Lisp_Object table, content, attrib, value;
- int point, size, table_vector_size;
- int skip_to_next, fin_ic;
+ Lisp_Object map, content, attrib, value;
+ int point, size, map_vector_size;
+ int map_set_rest_length, fin_ic;
+ int current_ic = this_ic;
- j = XINT (ccl_prog[ic++]); /* number of tables and separators. */
- fin_ic = ic + j;
- if ((j > reg[RRR]) && (j >= 0))
+ /* inhibit recursive call on MapMultiple. */
+ if (stack_idx_of_map_multiple > 0)
+ {
+ if (stack_idx_of_map_multiple <= stack_idx)
+ {
+ stack_idx_of_map_multiple = 0;
+ mapping_stack_pointer = mapping_stack;
+ CCL_INVALID_CMD;
+ }
+ }
+ else
+ mapping_stack_pointer = mapping_stack;
+ stack_idx_of_map_multiple = 0;
+
+ map_set_rest_length =
+ XINT (ccl_prog[ic++]); /* number of maps and separators. */
+ fin_ic = ic + map_set_rest_length;
+ op = reg[rrr];
+
+ if ((map_set_rest_length > reg[RRR]) && (reg[RRR] >= 0))
{
ic += reg[RRR];
i = reg[RRR];
+ map_set_rest_length -= i;
}
else
{
ic = fin_ic;
reg[RRR] = -1;
+ mapping_stack_pointer = mapping_stack;
break;
}
- op = reg[rrr];
- reg[RRR] = -1;
- skip_to_next = 0;
- table_vector_size = XVECTOR (Vccl_translation_table_vector)->size;
- for (;i < j;i++)
+
+ if (mapping_stack_pointer <= (mapping_stack + 1))
{
- point = ccl_prog[ic++];
- if (XINT(point) == -1)
- {
- skip_to_next = 0;
- continue;
- }
- if (skip_to_next) continue;
- if (point >= table_vector_size) continue;
- table = XVECTOR (Vccl_translation_table_vector)->
- contents[point];
- if (!CONSP (table)) continue;
- table = XCONS (table)->cdr;
- if (!VECTORP (table)) continue;
- size = XVECTOR (table)->size;
- if (size <= 1) continue;
- point = XUINT (XVECTOR (table)->contents[0]);
- point = op - point + 1;
- if (!((point >= 1) && (point < size))) continue;
- content = XVECTOR (table)->contents[point];
+ /* Set up initial state. */
+ mapping_stack_pointer = mapping_stack;
+ PUSH_MAPPING_STACK (0, op);
+ reg[RRR] = -1;
+ }
+ else
+ {
+ /* Recover after calling other ccl program. */
+ int orig_op;
- if (NILP (content))
- continue;
- else if (NUMBERP (content))
- {
- op = XUINT (content);
- reg[RRR] = i;
- skip_to_next = 1;
- }
- else if (CONSP (content))
+ POP_MAPPING_STACK (map_set_rest_length, orig_op);
+ POP_MAPPING_STACK (map_set_rest_length, reg[rrr]);
+ switch (op)
{
- attrib = XCONS (content)->car;
- value = XCONS (content)->cdr;
- if (!NUMBERP (attrib) || !NUMBERP (value))
- continue;
- reg[RRR] = i;
- op = XUINT (value);
-
- }
- else if (EQ (content, Qt))
- {
- reg[RRR] = i;
+ case -1:
+ /* Regard it as Qnil. */
+ op = orig_op;
+ i++;
+ ic++;
+ map_set_rest_length--;
+ break;
+ case -2:
+ /* Regard it as Qt. */
op = reg[rrr];
- skip_to_next = 1;
+ i++;
+ ic++;
+ map_set_rest_length--;
+ break;
+ case -3:
+ /* Regard it as Qlambda. */
+ op = orig_op;
+ i += map_set_rest_length;
+ ic += map_set_rest_length;
+ map_set_rest_length = 0;
+ break;
+ default:
+ /* Regard it as normal mapping. */
+ i += map_set_rest_length;
+ ic += map_set_rest_length;
+ POP_MAPPING_STACK (map_set_rest_length, reg[rrr]);
+ break;
}
- else if (EQ (content, Qlambda))
- break;
}
+ map_vector_size = ASIZE (Vcode_conversion_map_vector);
+
+ do {
+ for (;map_set_rest_length > 0;i++, ic++, map_set_rest_length--)
+ {
+ point = XINT(ccl_prog[ic]);
+ if (point < 0)
+ {
+ /* +1 is for including separator. */
+ point = -point + 1;
+ if (mapping_stack_pointer
+ >= &mapping_stack[MAX_MAP_SET_LEVEL])
+ CCL_INVALID_CMD;
+ PUSH_MAPPING_STACK (map_set_rest_length - point,
+ reg[rrr]);
+ map_set_rest_length = point;
+ reg[rrr] = op;
+ continue;
+ }
+
+ if (point >= map_vector_size) continue;
+ map = AREF (Vcode_conversion_map_vector, point);
+
+ /* Check map varidity. */
+ if (!CONSP (map)) continue;
+ map = XCDR (map);
+ if (!VECTORP (map)) continue;
+ size = ASIZE (map);
+ if (size <= 1) continue;
+
+ content = AREF (map, 0);
+
+ /* check map type,
+ [STARTPOINT VAL1 VAL2 ...] or
+ [t ELEMENT STARTPOINT ENDPOINT] */
+ if (NUMBERP (content))
+ {
+ point = XUINT (content);
+ point = op - point + 1;
+ if (!((point >= 1) && (point < size))) continue;
+ content = AREF (map, point);
+ }
+ else if (EQ (content, Qt))
+ {
+ if (size != 4) continue;
+ if ((op >= XUINT (AREF (map, 2))) &&
+ (op < XUINT (AREF (map, 3))))
+ content = AREF (map, 1);
+ else
+ continue;
+ }
+ else
+ continue;
+
+ if (NILP (content))
+ continue;
+
+ reg[RRR] = i;
+ if (NUMBERP (content))
+ {
+ op = XINT (content);
+ i += map_set_rest_length - 1;
+ ic += map_set_rest_length - 1;
+ POP_MAPPING_STACK (map_set_rest_length, reg[rrr]);
+ map_set_rest_length++;
+ }
+ else if (CONSP (content))
+ {
+ attrib = XCAR (content);
+ value = XCDR (content);
+ if (!NUMBERP (attrib) || !NUMBERP (value))
+ continue;
+ op = XUINT (value);
+ i += map_set_rest_length - 1;
+ ic += map_set_rest_length - 1;
+ POP_MAPPING_STACK (map_set_rest_length, reg[rrr]);
+ map_set_rest_length++;
+ }
+ else if (EQ (content, Qt))
+ {
+ op = reg[rrr];
+ }
+ else if (EQ (content, Qlambda))
+ {
+ i += map_set_rest_length;
+ ic += map_set_rest_length;
+ break;
+ }
+ else if (SYMBOLP (content))
+ {
+ if (mapping_stack_pointer
+ >= &mapping_stack[MAX_MAP_SET_LEVEL])
+ CCL_INVALID_CMD;
+ PUSH_MAPPING_STACK (map_set_rest_length, reg[rrr]);
+ PUSH_MAPPING_STACK (map_set_rest_length, op);
+ stack_idx_of_map_multiple = stack_idx + 1;
+ CCL_CALL_FOR_MAP_INSTRUCTION (content, current_ic);
+ }
+ else
+ CCL_INVALID_CMD;
+ }
+ if (mapping_stack_pointer <= (mapping_stack + 1))
+ break;
+ POP_MAPPING_STACK (map_set_rest_length, reg[rrr]);
+ i += map_set_rest_length;
+ ic += map_set_rest_length;
+ POP_MAPPING_STACK (map_set_rest_length, reg[rrr]);
+ } while (1);
+
ic = fin_ic;
}
reg[rrr] = op;
break;
- case CCL_TranslateSingleMap:
+ case CCL_MapSingle:
{
- Lisp_Object table, attrib, value, content;
+ Lisp_Object map, attrib, value, content;
int size, point;
- j = XINT (ccl_prog[ic++]); /* table_id */
+ j = XINT (ccl_prog[ic++]); /* map_id */
op = reg[rrr];
- if (j >= XVECTOR (Vccl_translation_table_vector)->size)
+ if (j >= ASIZE (Vcode_conversion_map_vector))
{
reg[RRR] = -1;
break;
}
- table = XVECTOR (Vccl_translation_table_vector)->
- contents[j];
- if (!CONSP (table))
+ map = AREF (Vcode_conversion_map_vector, j);
+ if (!CONSP (map))
{
reg[RRR] = -1;
break;
}
- table = XCONS(table)->cdr;
- if (!VECTORP (table))
+ map = XCDR (map);
+ if (!VECTORP (map))
{
reg[RRR] = -1;
break;
}
- size = XVECTOR (table)->size;
- point = XUINT (XVECTOR (table)->contents[0]);
+ size = ASIZE (map);
+ point = XUINT (AREF (map, 0));
point = op - point + 1;
reg[RRR] = 0;
if ((size <= 1) ||
reg[RRR] = -1;
else
{
- content = XVECTOR (table)->contents[point];
+ reg[RRR] = 0;
+ content = AREF (map, point);
if (NILP (content))
reg[RRR] = -1;
else if (NUMBERP (content))
- reg[rrr] = XUINT (content);
- else if (EQ (content, Qt))
- reg[RRR] = i;
+ reg[rrr] = XINT (content);
+ else if (EQ (content, Qt));
else if (CONSP (content))
{
- attrib = XCONS (content)->car;
- value = XCONS (content)->cdr;
+ attrib = XCAR (content);
+ value = XCDR (content);
if (!NUMBERP (attrib) || !NUMBERP (value))
continue;
reg[rrr] = XUINT(value);
break;
}
+ else if (SYMBOLP (content))
+ CCL_CALL_FOR_MAP_INSTRUCTION (content, ic);
else
reg[RRR] = -1;
}
}
break;
-
+
default:
CCL_INVALID_CMD;
}
}
ccl_error_handler:
- if (destination)
+ /* The suppress_error member is set when e.g. a CCL-based coding
+ system is used for terminal output. */
+ if (!ccl->suppress_error && destination)
{
/* We can insert an error message only if DESTINATION is
specified and we still have a room to store the message
char msg[256];
int msglen;
+ if (!dst)
+ dst = destination;
+
switch (ccl->status)
{
case CCL_STAT_INVALID_CMD:
sprintf(msg, "\nCCL: Invalid command %x (ccl_code = %x) at %d.",
- code & 0x1F, code, ic);
+ code & 0x1F, code, this_ic);
#ifdef CCL_DEBUG
{
int i = ccl_backtrace_idx - 1;
bcopy (msg, dst, msglen);
dst += msglen;
}
+ goto ccl_finish;
}
#endif
- goto ccl_finish;
+ break;
case CCL_STAT_QUIT:
sprintf(msg, "\nCCL: Quited.");
break;
default:
- sprintf(msg, "\nCCL: Unknown error type (%d).", ccl->status);
+ sprintf(msg, "\nCCL: Unknown error type (%d)", ccl->status);
}
msglen = strlen (msg);
bcopy (msg, dst, msglen);
dst += msglen;
}
+
+ if (ccl->status == CCL_STAT_INVALID_CMD)
+ {
+#if 0 /* If the remaining bytes contain 0x80..0x9F, copying them
+ results in an invalid multibyte sequence. */
+
+ /* Copy the remaining source data. */
+ int i = src_end - src;
+ if (dst_bytes && (dst_end - dst) < i)
+ i = dst_end - dst;
+ bcopy (src, dst, i);
+ src += i;
+ dst += i;
+#else
+ /* Signal that we've consumed everything. */
+ src = src_end;
+#endif
+ }
}
ccl_finish:
ccl->ic = ic;
- if (consumed) *consumed = src - source;
- return dst - destination;
+ ccl->stack_idx = stack_idx;
+ ccl->prog = ccl_prog;
+ ccl->eight_bit_control = (extra_bytes > 1);
+ if (consumed)
+ *consumed = src - source;
+ return (dst ? dst - destination : 0);
+}
+
+/* Resolve symbols in the specified CCL code (Lisp vector). This
+ function converts symbols of code conversion maps and character
+ translation tables embeded in the CCL code into their ID numbers.
+
+ The return value is a vector (CCL itself or a new vector in which
+ all symbols are resolved), Qt if resolving of some symbol failed,
+ or nil if CCL contains invalid data. */
+
+static Lisp_Object
+resolve_symbol_ccl_program (ccl)
+ Lisp_Object ccl;
+{
+ int i, veclen, unresolved = 0;
+ Lisp_Object result, contents, val;
+
+ result = ccl;
+ veclen = ASIZE (result);
+
+ for (i = 0; i < veclen; i++)
+ {
+ contents = AREF (result, i);
+ if (INTEGERP (contents))
+ continue;
+ else if (CONSP (contents)
+ && SYMBOLP (XCAR (contents))
+ && SYMBOLP (XCDR (contents)))
+ {
+ /* This is the new style for embedding symbols. The form is
+ (SYMBOL . PROPERTY). (get SYMBOL PROPERTY) should give
+ an index number. */
+
+ if (EQ (result, ccl))
+ result = Fcopy_sequence (ccl);
+
+ val = Fget (XCAR (contents), XCDR (contents));
+ if (NATNUMP (val))
+ AREF (result, i) = val;
+ else
+ unresolved = 1;
+ continue;
+ }
+ else if (SYMBOLP (contents))
+ {
+ /* This is the old style for embedding symbols. This style
+ may lead to a bug if, for instance, a translation table
+ and a code conversion map have the same name. */
+ if (EQ (result, ccl))
+ result = Fcopy_sequence (ccl);
+
+ val = Fget (contents, Qtranslation_table_id);
+ if (NATNUMP (val))
+ AREF (result, i) = val;
+ else
+ {
+ val = Fget (contents, Qcode_conversion_map_id);
+ if (NATNUMP (val))
+ AREF (result, i) = val;
+ else
+ {
+ val = Fget (contents, Qccl_program_idx);
+ if (NATNUMP (val))
+ AREF (result, i) = val;
+ else
+ unresolved = 1;
+ }
+ }
+ continue;
+ }
+ return Qnil;
+ }
+
+ return (unresolved ? Qt : result);
+}
+
+/* Return the compiled code (vector) of CCL program CCL_PROG.
+ CCL_PROG is a name (symbol) of the program or already compiled
+ code. If necessary, resolve symbols in the compiled code to index
+ numbers. If we failed to get the compiled code or to resolve
+ symbols, return Qnil. */
+
+static Lisp_Object
+ccl_get_compiled_code (ccl_prog)
+ Lisp_Object ccl_prog;
+{
+ Lisp_Object val, slot;
+
+ if (VECTORP (ccl_prog))
+ {
+ val = resolve_symbol_ccl_program (ccl_prog);
+ return (VECTORP (val) ? val : Qnil);
+ }
+ if (!SYMBOLP (ccl_prog))
+ return Qnil;
+
+ val = Fget (ccl_prog, Qccl_program_idx);
+ if (! NATNUMP (val)
+ || XINT (val) >= ASIZE (Vccl_program_table))
+ return Qnil;
+ slot = AREF (Vccl_program_table, XINT (val));
+ if (! VECTORP (slot)
+ || ASIZE (slot) != 3
+ || ! VECTORP (AREF (slot, 1)))
+ return Qnil;
+ if (NILP (AREF (slot, 2)))
+ {
+ val = resolve_symbol_ccl_program (AREF (slot, 1));
+ if (! VECTORP (val))
+ return Qnil;
+ AREF (slot, 1) = val;
+ AREF (slot, 2) = Qt;
+ }
+ return AREF (slot, 1);
}
/* Setup fields of the structure pointed by CCL appropriately for the
- execution of compiled CCL code in VEC (vector of integer). */
-void
-setup_ccl_program (ccl, vec)
+ execution of CCL program CCL_PROG. CCL_PROG is the name (symbol)
+ of the CCL program or the already compiled code (vector).
+ Return 0 if we succeed this setup, else return -1.
+
+ If CCL_PROG is nil, we just reset the structure pointed by CCL. */
+int
+setup_ccl_program (ccl, ccl_prog)
struct ccl_program *ccl;
- Lisp_Object vec;
+ Lisp_Object ccl_prog;
{
int i;
- ccl->size = XVECTOR (vec)->size;
- ccl->prog = XVECTOR (vec)->contents;
+ if (! NILP (ccl_prog))
+ {
+ struct Lisp_Vector *vp;
+
+ ccl_prog = ccl_get_compiled_code (ccl_prog);
+ if (! VECTORP (ccl_prog))
+ return -1;
+ vp = XVECTOR (ccl_prog);
+ ccl->size = vp->size;
+ ccl->prog = vp->contents;
+ ccl->eof_ic = XINT (vp->contents[CCL_HEADER_EOF]);
+ ccl->buf_magnification = XINT (vp->contents[CCL_HEADER_BUF_MAG]);
+ }
ccl->ic = CCL_HEADER_MAIN;
- ccl->eof_ic = XINT (XVECTOR (vec)->contents[CCL_HEADER_EOF]);
- ccl->buf_magnification = XINT (XVECTOR (vec)->contents[CCL_HEADER_BUF_MAG]);
for (i = 0; i < 8; i++)
ccl->reg[i] = 0;
ccl->last_block = 0;
ccl->private_state = 0;
ccl->status = 0;
+ ccl->stack_idx = 0;
+ ccl->eol_type = CODING_EOL_LF;
+ ccl->suppress_error = 0;
+ ccl->eight_bit_control = 0;
+ return 0;
}
-#ifdef emacs
+DEFUN ("ccl-program-p", Fccl_program_p, Sccl_program_p, 1, 1, 0,
+ doc: /* Return t if OBJECT is a CCL program name or a compiled CCL program code.
+See the documentation of `define-ccl-program' for the detail of CCL program. */)
+ (object)
+ Lisp_Object object;
+{
+ Lisp_Object val;
+
+ if (VECTORP (object))
+ {
+ val = resolve_symbol_ccl_program (object);
+ return (VECTORP (val) ? Qt : Qnil);
+ }
+ if (!SYMBOLP (object))
+ return Qnil;
+
+ val = Fget (object, Qccl_program_idx);
+ return ((! NATNUMP (val)
+ || XINT (val) >= ASIZE (Vccl_program_table))
+ ? Qnil : Qt);
+}
DEFUN ("ccl-execute", Fccl_execute, Sccl_execute, 2, 2, 0,
- "Execute CCL-PROGRAM with registers initialized by REGISTERS.\n\
-CCL-PROGRAM is a compiled code generated by `ccl-compile',\n\
- no I/O commands should appear in the CCL program.\n\
-REGISTERS is a vector of [R0 R1 ... R7] where RN is an initial value\n\
- of Nth register.\n\
-As side effect, each element of REGISTER holds the value of\n\
- corresponding register after the execution.")
- (ccl_prog, reg)
+ doc: /* Execute CCL-PROGRAM with registers initialized by REGISTERS.
+
+CCL-PROGRAM is a CCL program name (symbol)
+or compiled code generated by `ccl-compile' (for backward compatibility.
+In the latter case, the execution overhead is bigger than in the former).
+No I/O commands should appear in CCL-PROGRAM.
+
+REGISTERS is a vector of [R0 R1 ... R7] where RN is an initial value
+for the Nth register.
+
+As side effect, each element of REGISTERS holds the value of
+the corresponding register after the execution.
+
+See the documentation of `define-ccl-program' for a definition of CCL
+programs. */)
+ (ccl_prog, reg)
Lisp_Object ccl_prog, reg;
{
struct ccl_program ccl;
int i;
- CHECK_VECTOR (ccl_prog, 0);
- CHECK_VECTOR (reg, 1);
- if (XVECTOR (reg)->size != 8)
- error ("Invalid length of vector REGISTERS");
+ if (setup_ccl_program (&ccl, ccl_prog) < 0)
+ error ("Invalid CCL program");
+
+ CHECK_VECTOR (reg);
+ if (ASIZE (reg) != 8)
+ error ("Length of vector REGISTERS is not 8");
- setup_ccl_program (&ccl, ccl_prog);
for (i = 0; i < 8; i++)
- ccl.reg[i] = (INTEGERP (XVECTOR (reg)->contents[i])
- ? XINT (XVECTOR (reg)->contents[i])
+ ccl.reg[i] = (INTEGERP (AREF (reg, i))
+ ? XINT (AREF (reg, i))
: 0);
- ccl_driver (&ccl, (char *)0, (char *)0, 0, 0, (int *)0);
+ ccl_driver (&ccl, (unsigned char *)0, (unsigned char *)0, 0, 0, (int *)0);
QUIT;
if (ccl.status != CCL_STAT_SUCCESS)
error ("Error in CCL program at %dth code", ccl.ic);
for (i = 0; i < 8; i++)
- XSETINT (XVECTOR (reg)->contents[i], ccl.reg[i]);
+ XSETINT (AREF (reg, i), ccl.reg[i]);
return Qnil;
}
DEFUN ("ccl-execute-on-string", Fccl_execute_on_string, Sccl_execute_on_string,
- 3, 4, 0,
- "Execute CCL-PROGRAM with initial STATUS on STRING.\n\
-CCL-PROGRAM is a compiled code generated by `ccl-compile'.\n\
-Read buffer is set to STRING, and write buffer is allocated automatically.\n\
-STATUS is a vector of [R0 R1 ... R7 IC], where\n\
- R0..R7 are initial values of corresponding registers,\n\
- IC is the instruction counter specifying from where to start the program.\n\
-If R0..R7 are nil, they are initialized to 0.\n\
-If IC is nil, it is initialized to head of the CCL program.\n\
-Returns the contents of write buffer as a string,\n\
- and as side effect, STATUS is updated.\n\
-If optional 4th arg CONTINUE is non-nil, keep IC on read operation\n\
-when read buffer is exausted, else, IC is always set to the end of\n\
-CCL-PROGRAM on exit.")
- (ccl_prog, status, str, contin)
- Lisp_Object ccl_prog, status, str, contin;
+ 3, 5, 0,
+ doc: /* Execute CCL-PROGRAM with initial STATUS on STRING.
+
+CCL-PROGRAM is a symbol registered by register-ccl-program,
+or a compiled code generated by `ccl-compile' (for backward compatibility,
+in this case, the execution is slower).
+
+Read buffer is set to STRING, and write buffer is allocated automatically.
+
+STATUS is a vector of [R0 R1 ... R7 IC], where
+ R0..R7 are initial values of corresponding registers,
+ IC is the instruction counter specifying from where to start the program.
+If R0..R7 are nil, they are initialized to 0.
+If IC is nil, it is initialized to head of the CCL program.
+
+If optional 4th arg CONTINUE is non-nil, keep IC on read operation
+when read buffer is exausted, else, IC is always set to the end of
+CCL-PROGRAM on exit.
+
+It returns the contents of write buffer as a string,
+ and as side effect, STATUS is updated.
+If the optional 5th arg UNIBYTE-P is non-nil, the returned string
+is a unibyte string. By default it is a multibyte string.
+
+See the documentation of `define-ccl-program' for the detail of CCL program. */)
+ (ccl_prog, status, str, contin, unibyte_p)
+ Lisp_Object ccl_prog, status, str, contin, unibyte_p;
{
Lisp_Object val;
struct ccl_program ccl;
int i, produced;
int outbufsize;
char *outbuf;
- struct gcpro gcpro1, gcpro2, gcpro3;
+ struct gcpro gcpro1, gcpro2;
+
+ if (setup_ccl_program (&ccl, ccl_prog) < 0)
+ error ("Invalid CCL program");
- CHECK_VECTOR (ccl_prog, 0);
- CHECK_VECTOR (status, 1);
- if (XVECTOR (status)->size != 9)
- error ("Invalid length of vector STATUS");
- CHECK_STRING (str, 2);
- GCPRO3 (ccl_prog, status, str);
+ CHECK_VECTOR (status);
+ if (ASIZE (status) != 9)
+ error ("Length of vector STATUS is not 9");
+ CHECK_STRING (str);
+
+ GCPRO2 (status, str);
- setup_ccl_program (&ccl, ccl_prog);
for (i = 0; i < 8; i++)
{
- if (NILP (XVECTOR (status)->contents[i]))
- XSETINT (XVECTOR (status)->contents[i], 0);
- if (INTEGERP (XVECTOR (status)->contents[i]))
- ccl.reg[i] = XINT (XVECTOR (status)->contents[i]);
+ if (NILP (AREF (status, i)))
+ XSETINT (AREF (status, i), 0);
+ if (INTEGERP (AREF (status, i)))
+ ccl.reg[i] = XINT (AREF (status, i));
}
- if (INTEGERP (XVECTOR (status)->contents[i]))
+ if (INTEGERP (AREF (status, i)))
{
- i = XFASTINT (XVECTOR (status)->contents[8]);
+ i = XFASTINT (AREF (status, 8));
if (ccl.ic < i && i < ccl.size)
ccl.ic = i;
}
- outbufsize = XSTRING (str)->size_byte * ccl.buf_magnification + 256;
+ outbufsize = SBYTES (str) * ccl.buf_magnification + 256;
outbuf = (char *) xmalloc (outbufsize);
- if (!outbuf)
- error ("Not enough memory");
ccl.last_block = NILP (contin);
- produced = ccl_driver (&ccl, XSTRING (str)->data, outbuf,
- XSTRING (str)->size_byte, outbufsize, (int *)0);
+ ccl.multibyte = STRING_MULTIBYTE (str);
+ produced = ccl_driver (&ccl, SDATA (str), outbuf,
+ SBYTES (str), outbufsize, (int *) 0);
for (i = 0; i < 8; i++)
- XSET (XVECTOR (status)->contents[i], Lisp_Int, ccl.reg[i]);
- XSETINT (XVECTOR (status)->contents[8], ccl.ic);
+ ASET (status, i, make_number (ccl.reg[i]));
+ ASET (status, 8, make_number (ccl.ic));
UNGCPRO;
- val = make_string (outbuf, produced);
- free (outbuf);
+ if (NILP (unibyte_p))
+ {
+ int nchars;
+
+ produced = str_as_multibyte (outbuf, outbufsize, produced, &nchars);
+ val = make_multibyte_string (outbuf, nchars, produced);
+ }
+ else
+ val = make_unibyte_string (outbuf, produced);
+ xfree (outbuf);
QUIT;
+ if (ccl.status == CCL_STAT_SUSPEND_BY_DST)
+ error ("Output buffer for the CCL programs overflow");
if (ccl.status != CCL_STAT_SUCCESS
- && ccl.status != CCL_STAT_SUSPEND_BY_SRC
- && ccl.status != CCL_STAT_SUSPEND_BY_DST)
+ && ccl.status != CCL_STAT_SUSPEND_BY_SRC)
error ("Error in CCL program at %dth code", ccl.ic);
return val;
DEFUN ("register-ccl-program", Fregister_ccl_program, Sregister_ccl_program,
2, 2, 0,
- "Register CCL program PROGRAM of NAME in `ccl-program-table'.\n\
-PROGRAM should be a compiled code of CCL program, or nil.\n\
-Return index number of the registered CCL program.")
- (name, ccl_prog)
+ doc: /* Register CCL program CCL_PROG as NAME in `ccl-program-table'.
+CCL_PROG should be a compiled CCL program (vector), or nil.
+If it is nil, just reserve NAME as a CCL program name.
+Return index number of the registered CCL program. */)
+ (name, ccl_prog)
Lisp_Object name, ccl_prog;
{
- int len = XVECTOR (Vccl_program_table)->size;
- int i;
+ int len = ASIZE (Vccl_program_table);
+ int idx;
+ Lisp_Object resolved;
- CHECK_SYMBOL (name, 0);
+ CHECK_SYMBOL (name);
+ resolved = Qnil;
if (!NILP (ccl_prog))
- CHECK_VECTOR (ccl_prog, 1);
-
- for (i = 0; i < len; i++)
{
- Lisp_Object slot = XVECTOR (Vccl_program_table)->contents[i];
+ CHECK_VECTOR (ccl_prog);
+ resolved = resolve_symbol_ccl_program (ccl_prog);
+ if (NILP (resolved))
+ error ("Error in CCL program");
+ if (VECTORP (resolved))
+ {
+ ccl_prog = resolved;
+ resolved = Qt;
+ }
+ else
+ resolved = Qnil;
+ }
- if (!CONSP (slot))
+ for (idx = 0; idx < len; idx++)
+ {
+ Lisp_Object slot;
+
+ slot = AREF (Vccl_program_table, idx);
+ if (!VECTORP (slot))
+ /* This is the first unsed slot. Register NAME here. */
break;
- if (EQ (name, XCONS (slot)->car))
+ if (EQ (name, AREF (slot, 0)))
{
- XCONS (slot)->cdr = ccl_prog;
- return make_number (i);
+ /* Update this slot. */
+ AREF (slot, 1) = ccl_prog;
+ AREF (slot, 2) = resolved;
+ return make_number (idx);
}
}
- if (i == len)
+ if (idx == len)
{
- Lisp_Object new_table = Fmake_vector (make_number (len * 2), Qnil);
+ /* Extend the table. */
+ Lisp_Object new_table;
int j;
+ new_table = Fmake_vector (make_number (len * 2), Qnil);
for (j = 0; j < len; j++)
- XVECTOR (new_table)->contents[j]
- = XVECTOR (Vccl_program_table)->contents[j];
+ AREF (new_table, j)
+ = AREF (Vccl_program_table, j);
Vccl_program_table = new_table;
}
- XVECTOR (Vccl_program_table)->contents[i] = Fcons (name, ccl_prog);
- return make_number (i);
+ {
+ Lisp_Object elt;
+
+ elt = Fmake_vector (make_number (3), Qnil);
+ AREF (elt, 0) = name;
+ AREF (elt, 1) = ccl_prog;
+ AREF (elt, 2) = resolved;
+ AREF (Vccl_program_table, idx) = elt;
+ }
+
+ Fput (name, Qccl_program_idx, make_number (idx));
+ return make_number (idx);
}
-/* register CCL translation table.
- CCL translation table consists of numbers and Qt and Qnil and Qlambda.
- The first element is start code point.
- The rest elements are translated numbers.
- Qt shows that an original number before translation.
- Qnil shows that an empty element.
- Qlambda makes translation stopped.
+/* Register code conversion map.
+ A code conversion map consists of numbers, Qt, Qnil, and Qlambda.
+ The first element is the start code point.
+ The other elements are mapped numbers.
+ Symbol t means to map to an original number before mapping.
+ Symbol nil means that the corresponding element is empty.
+ Symbol lambda means to terminate mapping here.
*/
-DEFUN ("register-ccl-translation-table", Fregister_ccl_translation_table,
- Sregister_ccl_translation_table,
+DEFUN ("register-code-conversion-map", Fregister_code_conversion_map,
+ Sregister_code_conversion_map,
2, 2, 0,
- "Register CCL translation table.\n\
-TABLE should be a vector. SYMBOL is used for pointing the translation table out.\n\
-Return index number of the registered translation table.")
- (symbol, table)
- Lisp_Object symbol, table;
+ doc: /* Register SYMBOL as code conversion map MAP.
+Return index number of the registered map. */)
+ (symbol, map)
+ Lisp_Object symbol, map;
{
- int len = XVECTOR (Vccl_translation_table_vector)->size;
+ int len = ASIZE (Vcode_conversion_map_vector);
int i;
Lisp_Object index;
- CHECK_SYMBOL (symbol, 0);
- CHECK_VECTOR (table, 1);
-
+ CHECK_SYMBOL (symbol);
+ CHECK_VECTOR (map);
+
for (i = 0; i < len; i++)
{
- Lisp_Object slot = XVECTOR (Vccl_translation_table_vector)->contents[i];
+ Lisp_Object slot = AREF (Vcode_conversion_map_vector, i);
if (!CONSP (slot))
break;
- if (EQ (symbol, XCONS (slot)->car))
+ if (EQ (symbol, XCAR (slot)))
{
index = make_number (i);
- XCONS (slot)->cdr = table;
- Fput (symbol, Qccl_translation_table, table);
- Fput (symbol, Qccl_translation_table_id, index);
+ XSETCDR (slot, map);
+ Fput (symbol, Qcode_conversion_map, map);
+ Fput (symbol, Qcode_conversion_map_id, index);
return index;
}
}
int j;
for (j = 0; j < len; j++)
- XVECTOR (new_vector)->contents[j]
- = XVECTOR (Vccl_translation_table_vector)->contents[j];
- Vccl_translation_table_vector = new_vector;
+ AREF (new_vector, j)
+ = AREF (Vcode_conversion_map_vector, j);
+ Vcode_conversion_map_vector = new_vector;
}
index = make_number (i);
- Fput (symbol, Qccl_translation_table, table);
- Fput (symbol, Qccl_translation_table_id, index);
- XVECTOR (Vccl_translation_table_vector)->contents[i] = Fcons (symbol, table);
+ Fput (symbol, Qcode_conversion_map, map);
+ Fput (symbol, Qcode_conversion_map_id, index);
+ AREF (Vcode_conversion_map_vector, i) = Fcons (symbol, map);
return index;
}
+void
syms_of_ccl ()
{
staticpro (&Vccl_program_table);
Vccl_program_table = Fmake_vector (make_number (32), Qnil);
- Qccl_program = intern("ccl-program");
- staticpro(&Qccl_program);
+ Qccl_program = intern ("ccl-program");
+ staticpro (&Qccl_program);
- Qccl_translation_table = intern ("ccl-translation-table");
- staticpro (&Qccl_translation_table);
+ Qccl_program_idx = intern ("ccl-program-idx");
+ staticpro (&Qccl_program_idx);
- Qccl_translation_table_id = intern ("ccl-translation-table-id");
- staticpro (&Qccl_translation_table_id);
+ Qcode_conversion_map = intern ("code-conversion-map");
+ staticpro (&Qcode_conversion_map);
- DEFVAR_LISP ("ccl-translation-table-vector", &Vccl_translation_table_vector,
- "Where is stored translation tables for CCL program.\n\
-Because CCL program can't access these tables except by the index of the vector.");
- Vccl_translation_table_vector = Fmake_vector (XFASTINT (16), Qnil);
+ Qcode_conversion_map_id = intern ("code-conversion-map-id");
+ staticpro (&Qcode_conversion_map_id);
+
+ DEFVAR_LISP ("code-conversion-map-vector", &Vcode_conversion_map_vector,
+ doc: /* Vector of code conversion maps. */);
+ Vcode_conversion_map_vector = Fmake_vector (make_number (16), Qnil);
DEFVAR_LISP ("font-ccl-encoder-alist", &Vfont_ccl_encoder_alist,
- "Alist of fontname patterns vs corresponding CCL program.\n\
-Each element looks like (REGEXP . CCL-CODE),\n\
- where CCL-CODE is a compiled CCL program.\n\
-When a font whose name matches REGEXP is used for displaying a character,\n\
- CCL-CODE is executed to calculate the code point in the font\n\
- from the charset number and position code(s) of the character which are set\n\
- in CCL registers R0, R1, and R2 before the execution.\n\
-The code point in the font is set in CCL registers R1 and R2\n\
- when the execution terminated.\n\
-If the font is single-byte font, the register R2 is not used.");
+ doc: /* Alist of fontname patterns vs corresponding CCL program.
+Each element looks like (REGEXP . CCL-CODE),
+ where CCL-CODE is a compiled CCL program.
+When a font whose name matches REGEXP is used for displaying a character,
+ CCL-CODE is executed to calculate the code point in the font
+ from the charset number and position code(s) of the character which are set
+ in CCL registers R0, R1, and R2 before the execution.
+The code point in the font is set in CCL registers R1 and R2
+ when the execution terminated.
+ If the font is single-byte font, the register R2 is not used. */);
Vfont_ccl_encoder_alist = Qnil;
+ DEFVAR_LISP ("translation-hash-table-vector", &Vtranslation_hash_table_vector,
+ doc: /* Vector containing all translation hash tables ever defined.
+Comprises pairs (SYMBOL . TABLE) where SYMBOL and TABLE were set up by calls
+to `define-translation-hash-table'. The vector is indexed by the table id
+used by CCL. */);
+ Vtranslation_hash_table_vector = Qnil;
+
+ defsubr (&Sccl_program_p);
defsubr (&Sccl_execute);
defsubr (&Sccl_execute_on_string);
defsubr (&Sregister_ccl_program);
- defsubr (&Sregister_ccl_translation_table);
+ defsubr (&Sregister_code_conversion_map);
}
-#endif /* emacs */
+/* arch-tag: bb9a37be-68ce-4576-8d3d-15d750e4a860
+ (do not change this comment) */