2 * iso2022.c - support for ISO/IEC 2022 (alias ECMA-35).
4 * This isn't a complete implementation of ISO/IEC 2022, but it's
5 * close. It only handles decoding, because a fully general encoder
6 * isn't really useful. It can decode 8-bit and 7-bit versions, with
7 * support for single-byte and multi-byte character sets, all four
8 * containers (G0, G1, G2, and G3), using both single-shift and
9 * locking-shift sequences.
11 * The general principle is that any valid ISO/IEC 2022 sequence
12 * should either be correctly decoded or should emit an ERROR. The
13 * only exception to this is that the C0 and C1 sets are fixed as
14 * those of ISO/IEC 6429. Escape sequences for designating control
15 * sets are passed through, so a post-processor could fix them up if
18 * DOCS to UTF-8 works. Other DOCS sequences are ignored, which will
19 * produce surprising results.
36 enum {S4
, S6
, M4
, M6
};
38 static long int null_dbcs_to_unicode(int, int);
40 const struct iso2022_subcharset
{
43 const sbcs_data
*sbcs_base
;
44 long int (*dbcs_fn
)(int, int);
45 } iso2022_subcharsets
[] = {
46 { S4
, 0, 'B', 0x00, &sbcsdata_CS_ASCII
},
48 { S4
, 0, '<', 0x80, &sbcsdata_CS_DEC_MCS
},
49 { S4
, 0, 'I', 0x80, &sbcsdata_CS_JISX0201
},
50 { S4
, 0, 'J', 0x00, &sbcsdata_CS_JISX0201
},
52 { S6
, 0, 'A', 0x80, &sbcsdata_CS_ISO8859_1
},
53 { S6
, 0, 'B', 0x80, &sbcsdata_CS_ISO8859_2
},
54 { S6
, 0, 'C', 0x80, &sbcsdata_CS_ISO8859_3
},
55 { S6
, 0, 'D', 0x80, &sbcsdata_CS_ISO8859_4
},
56 { S6
, 0, 'F', 0x80, &sbcsdata_CS_ISO8859_7
},
57 { S6
, 0, 'G', 0x80, &sbcsdata_CS_ISO8859_6
},
58 { S6
, 0, 'H', 0x80, &sbcsdata_CS_ISO8859_8
},
59 { S6
, 0, 'L', 0x80, &sbcsdata_CS_ISO8859_5
},
60 { S6
, 0, 'M', 0x80, &sbcsdata_CS_ISO8859_9
},
61 { S6
, 0, 'T', 0x80, &sbcsdata_CS_ISO8859_11
},
62 { S6
, 0, 'V', 0x80, &sbcsdata_CS_ISO8859_10
},
63 { S6
, 0, 'Y', 0x80, &sbcsdata_CS_ISO8859_13
},
64 { S6
, 0, '_', 0x80, &sbcsdata_CS_ISO8859_14
},
65 { S6
, 0, 'b', 0x80, &sbcsdata_CS_ISO8859_15
},
66 { S6
, 0, 'f', 0x80, &sbcsdata_CS_ISO8859_16
},
67 { S6
, 0, '~' }, /* empty 96-set */
69 { M4
, 0, '@' }, /* JIS C 6226-1978 */
71 { M4
, 0, 'A', -0x21, 0, &gb2312_to_unicode
},
72 { M4
, 0, 'B', -0x21, 0, &jisx0208_to_unicode
},
73 { M4
, 0, 'C', -0x21, 0, &ksx1001_to_unicode
},
74 { M4
, 0, 'D', -0x21, 0, &jisx0212_to_unicode
},
75 { M4
, 0, '~', 0, 0, &null_dbcs_to_unicode
}, /* empty 94^n-set */
76 { M6
, 0, '~', 0, 0, &null_dbcs_to_unicode
}, /* empty 96^n-set */
79 static long int null_dbcs_to_unicode(int r
, int c
)
84 /* States, or "what we're currently accumulating". */
86 IDLE
, /* None of the below */
87 SS2CHAR
, /* Accumulating a character after SS2 */
88 SS3CHAR
, /* Accumulating a character after SS3 */
89 ESCSEQ
, /* Accumulating an escape sequence */
90 ESCDROP
, /* Discarding an escape sequence */
91 ESCPASS
, /* Passing through an escape sequence */
92 DOCSUTF8
/* DOCSed into UTF-8 */
97 static void dump_state(charset_state
*s
)
99 unsigned s0
= s
->s0
, s1
= s
->s1
;
100 char const * const modes
[] = { "IDLE", "SS2CHAR", "SS3CHAR",
101 "ESCSEQ", "ESCDROP", "ESCPASS",
104 fprintf(stderr
, "s0: %s", modes
[s0
>> 29]);
105 fprintf(stderr
, " %02x %02x %02x ", (s0
>> 16) & 0xff, (s0
>> 8) & 0xff,
107 fprintf(stderr
, "s1: LS%d LS%dR", (s1
>> 30) & 3, (s1
>> 28) & 3);
108 fprintf(stderr
, " %d %d %d %d\n", s1
& 0x7f, (s1
>> 7) & 0x7f,
109 (s1
>> 14) & 0x7f, (s1
>> 21) & 0x7f);
113 static void designate(charset_state
*state
, int container
,
114 int type
, int ibyte
, int fbyte
)
118 assert(container
>= 0 && container
<= 3);
119 assert(type
== S4
|| type
== S6
|| type
== M4
|| type
== M6
);
121 for (i
= 0; i
<= lenof(iso2022_subcharsets
); i
++) {
122 if (iso2022_subcharsets
[i
].type
== type
&&
123 iso2022_subcharsets
[i
].i
== ibyte
&&
124 iso2022_subcharsets
[i
].f
== fbyte
) {
125 state
->s1
&= ~(0x7fL
<< (container
* 7));
126 state
->s1
|= (i
<< (container
* 7));
131 * If we don't find the charset, invoke the empty one, so we
132 * output ERROR rather than garbage.
134 designate(state
, container
, type
, 0, '~');
137 static void do_utf8(long int input_chr
,
138 charset_state
*state
,
139 void (*emit
)(void *ctx
, long int output
),
142 charset_state ustate
;
143 charset_spec
const *utf8
;
146 ustate
.s0
= state
->s0
& 0x03ffffffL
;
147 read_utf8(NULL
, input_chr
, &ustate
, emit
, emitctx
);
148 state
->s0
= (state
->s0
& ~0x03ffffffL
) | (ustate
.s0
& 0x03ffffffL
);
151 static void docs_utf8(long int input_chr
,
152 charset_state
*state
,
153 void (*emit
)(void *ctx
, long int output
),
159 * Bits [25:0] of s0 are reserved for read_utf8().
160 * Bits [27:26] are a tiny state machine to recognise ESC % @.
162 retstate
= (state
->s0
& 0x0c000000L
) >> 26;
163 if (retstate
== 1 && input_chr
== '%')
165 else if (retstate
== 2 && input_chr
== '@') {
166 /* If we've got a partial UTF-8 sequence, complain. */
167 if (state
->s0
& 0x03ffffffL
)
168 emit(emitctx
, ERROR
);
172 if (retstate
>= 1) do_utf8(ESC
, state
, emit
, emitctx
);
173 if (retstate
>= 2) do_utf8('%', state
, emit
, emitctx
);
175 if (input_chr
== ESC
)
178 do_utf8(input_chr
, state
, emit
, emitctx
);
181 state
->s0
= (state
->s0
& ~0x0c000000L
) | (retstate
<< 26);
185 static void read_iso2022(charset_spec
const *charset
, long int input_chr
,
186 charset_state
*state
,
187 void (*emit
)(void *ctx
, long int output
),
191 /* dump_state(state); */
193 * We have to make fairly efficient use of the 64 bits of state
194 * available to us. Long-term state goes in s1, and consists of
195 * the identities of the character sets designated as G0/G1/G2/G3
196 * and the locking-shift states for GL and GR. Short-term state
197 * goes in s0: The bottom half of s0 accumulates characters for an
198 * escape sequence or a multi-byte character, while the top three
199 * bits indicate what they're being accumulated for. After DOCS,
200 * the bottom 29 bits of state are available for the DOCS function
201 * to use -- the UTF-8 one uses the bottom 26 for UTF-8 decoding
202 * and the top two to recognised ESC % @.
204 * s0[31:29] = state enum
205 * s0[24:0] = accumulated bytes
206 * s1[31:30] = GL locking-shift state
207 * s1[29:28] = GR locking-shift state
208 * s1[27:21] = G3 charset
209 * s1[20:14] = G2 charset
210 * s1[13:7] = G1 charset
211 * s1[6:0] = G0 charset
216 #define LOCKING_SHIFT(n,side) \
217 (state->s1 = (state->s1 & ~(3L<<(side))) | ((n ## L)<<(side)))
218 #define MODE ((state->s0 & 0xe0000000L) >> 29)
219 #define ENTER_MODE(m) (state->s0 = (state->s0 & ~0xe0000000L) | ((m)<<29))
220 #define SINGLE_SHIFT(n) ENTER_MODE(SS2CHAR - 2 + (n))
221 #define ASSERT_IDLE do { \
222 if (state->s0 != 0) emit(emitctx, ERROR); \
226 if (state
->s1
== 0) {
228 * Since there's no LS0R, this means we must just have started.
229 * Set up a sane initial state (LS0, LS1R, ASCII in G0/G1/G2/G3).
231 LOCKING_SHIFT(0, LEFT
);
232 LOCKING_SHIFT(1, RIGHT
);
233 designate(state
, 0, S4
, 0, 'B');
234 designate(state
, 1, S4
, 0, 'B');
235 designate(state
, 2, S4
, 0, 'B');
236 designate(state
, 3, S4
, 0, 'B');
239 if (MODE
== DOCSUTF8
) {
240 docs_utf8(input_chr
, state
, emit
, emitctx
);
244 if ((input_chr
& 0x60) == 0x00) {
245 /* C0 or C1 control */
252 LOCKING_SHIFT(0, LEFT
);
255 LOCKING_SHIFT(1, LEFT
);
264 emit(emitctx
, input_chr
);
267 } else if ((input_chr
& 0x80) || MODE
< ESCSEQ
) {
269 struct iso2022_subcharset
const *subcs
;
274 * Force idle state if we're in mid escape sequence, or in a
275 * multi-byte character with a different top bit.
277 if (MODE
>= ESCSEQ
||
278 ((state
->s0
& 0x00ff0000L
) != 0 &&
279 (((state
->s0
>> 16) ^ input_chr
) & 0x80)))
281 if (MODE
== SS2CHAR
|| MODE
== SS3CHAR
) /* Single-shift */
282 container
= MODE
- SS2CHAR
+ 2;
283 else if (input_chr
>= 0x80) /* GR */
284 container
= (state
->s1
>> 28) & 3;
286 container
= state
->s1
>> 30;
289 input_7bit
= input_chr
& ~0x80;
290 subcs
= &iso2022_subcharsets
[(state
->s1
>> (container
* 7)) & 0x7f];
291 if ((subcs
->type
== S4
|| subcs
->type
== M4
) &&
292 (input_7bit
== 0x20 || input_7bit
== 0x7f)) {
293 /* characters not in 94-char set */
294 if (is_gl
) emit(emitctx
, input_7bit
);
295 else emit(emitctx
, ERROR
);
296 } else if (subcs
->type
== M4
|| subcs
->type
== M6
) {
297 if ((state
->s0
& 0x00ff0000L
) == 0) {
298 state
->s0
|= input_chr
<< 16;
302 subcs
->dbcs_fn(((state
->s0
>> 16) & 0x7f) + subcs
->offset
,
303 input_7bit
+ subcs
->offset
));
306 if ((state
->s0
& 0x00ff0000L
) != 0)
307 emit(emitctx
, ERROR
);
308 emit(emitctx
, subcs
->sbcs_base ?
309 sbcs_to_unicode(subcs
->sbcs_base
, input_7bit
+ subcs
->offset
):
315 if (MODE
== ESCPASS
) {
316 emit(emitctx
, input_chr
);
317 if ((input_chr
& 0xf0) != 0x20)
323 * Intermediate bytes shall be any of the 16 positions of
324 * column 02 of the code table; they are denoted by the symbol
327 if ((input_chr
& 0xf0) == 0x20) {
328 if (((state
->s0
>> 16) & 0xff) == 0)
329 state
->s0
|= input_chr
<< 16;
330 else if (((state
->s0
>> 8) & 0xff) == 0)
331 state
->s0
|= input_chr
<< 8;
333 /* Long escape sequence. Switch to ESCPASS or ESCDROP. */
334 i1
= (state
->s0
>> 16) & 0xff;
335 i2
= (state
->s0
>> 8) & 0xff;
337 case '(': case ')': case '*': case '+':
338 case '-': case '.': case '/':
346 emit(emitctx
, input_chr
);
356 * Final bytes shall be any of the 79 positions of columns 03
357 * to 07 of the code table excluding position 07/15; they are
358 * denoted by the symbol F.
360 i1
= (state
->s0
>> 16) & 0xff;
361 i2
= (state
->s0
>> 8) & 0xff;
363 input_chr
= 0; /* Make sure it won't match. */
366 case 0: /* No intermediate bytes */
375 LOCKING_SHIFT(2, LEFT
);
378 LOCKING_SHIFT(3, LEFT
);
381 LOCKING_SHIFT(3, RIGHT
);
384 LOCKING_SHIFT(2, RIGHT
);
387 LOCKING_SHIFT(1, RIGHT
);
390 /* Unsupported escape sequence. Spit it back out. */
392 emit(emitctx
, input_chr
);
397 * Various coding structure facilities specify that designating
398 * a code element also invokes it. As far as I can see, invoking
399 * it now will have the same practical effect, since those
400 * facilities also ban the use of locking shifts.
403 case 'A': /* G0 element used and invoked into GL */
404 LOCKING_SHIFT(0, LEFT
);
406 case 'C': /* G0 in GL, G1 in GR */
407 case 'D': /* Ditto, at least for 8-bit codes */
408 case 'L': /* ISO 4873 (ECMA-43) level 1 */
409 case 'M': /* ISO 4873 (ECMA-43) level 2 */
410 LOCKING_SHIFT(0, LEFT
);
411 LOCKING_SHIFT(1, RIGHT
);
417 * IRR (Identify Revised Registration) is ignored here,
418 * since any revised registration must be
419 * upward-compatible with the old one, so either we'll
420 * support the new one or we'll emit ERROR when we run
421 * into a new character. In either case, there's nothing
425 case '(': /* GZD4 */ case ')': /* G1D4 */
426 case '*': /* G2D4 */ case '+': /* G3D4 */
427 designate(state
, i1
- '(', S4
, i2
, input_chr
);
429 case '-': /* G1D6 */ case '.': /* G2D6 */ case '/': /* G3D6 */
430 designate(state
, i1
- ',', S6
, i2
, input_chr
);
432 case '$': /* G?DM? */
434 case 0: /* Obsolete version of GZDM4 */
436 case '(': /* GZDM4 */ case ')': /* G1DM4 */
437 case '*': /* G2DM4 */ case '+': /* G3DM4 */
438 designate(state
, i2
- '(', M4
, 0, input_chr
);
440 case '-': /* G1DM6 */
441 case '.': /* G2DM6 */ case '/': /* G3DM6 */
442 designate(state
, i2
- ',', M6
, 0, input_chr
);
445 emit(emitctx
, ERROR
);
449 /* XXX What's a reasonable way to handle an unrecognised DOCS? */
454 ENTER_MODE(DOCSUTF8
);
461 /* Unsupported nF escape sequence. Re-emit it. */
464 if (i2
) emit(emitctx
, i2
);
465 emit(emitctx
, input_chr
);
471 static int write_iso2022(charset_spec
const *charset
, long int input_chr
,
472 charset_state
*state
,
473 void (*emit
)(void *ctx
, long int output
),
479 const charset_spec charset_CS_ISO2022
= {
480 CS_ISO2022
, read_iso2022
, write_iso2022
, NULL
491 void iso2022_emit(void *ctx
, long output
)
493 wchar_t **p
= (wchar_t **)ctx
;
497 void iso2022_read_test(int line
, char *input
, int inlen
, ...)
500 wchar_t *p
, str
[512];
505 state
.s0
= state
.s1
= 0;
508 for (i
= 0; i
< inlen
; i
++)
509 read_iso2022(NULL
, input
[i
] & 0xFF, &state
, iso2022_emit
, &p
);
513 for (i
= 0; i
< p
- str
; i
++) {
514 l
= va_arg(ap
, long int);
516 printf("%d: correct string shorter than output\n", line
);
521 printf("%d: char %d came out as %08x, should be %08lx\n",
527 l
= va_arg(ap
, long int);
529 printf("%d: correct string longer than output\n", line
);
536 /* Macro to concoct the first three parameters of iso2022_read_test. */
537 #define TESTSTR(x) __LINE__, x, lenof(x)
541 printf("read tests beginning\n");
542 /* Simple test (Emacs sample text for Japanese, in ISO-2022-JP) */
543 iso2022_read_test(TESTSTR("Japanese (\x1b$BF|K\\8l\x1b(B)\t"
544 "\x1b$B$3$s$K$A$O\x1b(B, "
545 "\x1b$B%3%s%K%A%O\x1b(B\n"),
546 'J','a','p','a','n','e','s','e',' ','(',
547 0x65E5, 0x672C, 0x8A9E, ')', '\t',
548 0x3053, 0x3093, 0x306b, 0x3061, 0x306f, ',', ' ',
549 0x30b3, 0x30f3, 0x30cb, 0x30c1, 0x30cf, '\n', 0, -1);
550 /* Same thing in EUC-JP (with designations, and half-width katakana) */
551 iso2022_read_test(TESTSTR("\x1b$)B\x1b*I\x1b$+D"
552 "Japanese (\xc6\xfc\xcb\xdc\xb8\xec)\t"
553 "\xa4\xb3\xa4\xf3\xa4\xcb\xa4\xc1\xa4\xcf, "
554 "\x8e\xba\x8e\xdd\x8e\xc6\x8e\xc1\x8e\xca\n"),
555 'J','a','p','a','n','e','s','e',' ','(',
556 0x65E5, 0x672C, 0x8A9E, ')', '\t',
557 0x3053, 0x3093, 0x306b, 0x3061, 0x306f, ',', ' ',
558 0xff7a, 0xff9d, 0xff86, 0xff81, 0xff8a, '\n', 0, -1);
559 /* Multibyte single-shift */
560 iso2022_read_test(TESTSTR("\x1b$)B\x1b*I\x1b$+D\x8f\"/!"),
562 /* Non-existent SBCS */
563 iso2022_read_test(TESTSTR("\x1b(!Zfnord\n"),
564 ERROR
, ERROR
, ERROR
, ERROR
, ERROR
, '\n', 0, -1);
565 /* Pass-through of ordinary escape sequences, including a long one */
566 iso2022_read_test(TESTSTR("\x1b""b\x1b#5\x1b#!!!5"),
567 0x1B, 'b', 0x1B, '#', '5',
568 0x1B, '#', '!', '!', '!', '5', 0, -1);
569 /* Non-existent DBCS (also 5-byte escape sequence) */
570 iso2022_read_test(TESTSTR("\x1b$(!Bfnord!"),
571 ERROR
, ERROR
, ERROR
, 0, -1);
572 /* Incomplete DB characters */
573 iso2022_read_test(TESTSTR("\x1b$B(,(\x1b(BHi\x1b$B(,(\n"),
574 0x2501, ERROR
, 'H', 'i', 0x2501, ERROR
, '\n', 0, -1);
575 iso2022_read_test(TESTSTR("\x1b$)B\x1b*I\x1b$+D\xa4""B"),
577 iso2022_read_test(TESTSTR("\x1b$)B\x1b*I\x1b$+D\x0e\x1b|$\xa2\xaf"),
578 ERROR
, 0x02D8, 0, -1);
579 /* Incomplete escape sequence */
580 iso2022_read_test(TESTSTR("\x1b\n"), ERROR
, '\n', 0, -1);
581 iso2022_read_test(TESTSTR("\x1b-A\x1b~\x1b\xa1"), ERROR
, 0xa1, 0, -1);
582 /* Incomplete single-shift */
583 iso2022_read_test(TESTSTR("\x8e\n"), ERROR
, '\n', 0, -1);
584 iso2022_read_test(TESTSTR("\x1b$*B\x8e(\n"), ERROR
, '\n', 0, -1);
585 /* Corner cases (02/00 and 07/15) */
586 iso2022_read_test(TESTSTR("\x1b(B\x20\x7f"), 0x20, 0x7f, 0, -1);
587 iso2022_read_test(TESTSTR("\x1b(I\x20\x7f"), 0x20, 0x7f, 0, -1);
588 iso2022_read_test(TESTSTR("\x1b$B\x20\x7f"), 0x20, 0x7f, 0, -1);
589 iso2022_read_test(TESTSTR("\x1b-A\x0e\x20\x7f"), 0xa0, 0xff, 0, -1);
590 iso2022_read_test(TESTSTR("\x1b$-~\x0e\x20\x7f"), ERROR
, 0, -1);
591 iso2022_read_test(TESTSTR("\x1b)B\xa0\xff"), ERROR
, ERROR
, 0, -1);
592 iso2022_read_test(TESTSTR("\x1b)I\xa0\xff"), ERROR
, ERROR
, 0, -1);
593 iso2022_read_test(TESTSTR("\x1b$)B\xa0\xff"), ERROR
, ERROR
, 0, -1);
594 iso2022_read_test(TESTSTR("\x1b-A\x1b~\xa0\xff"), 0xa0, 0xff, 0, -1);
595 iso2022_read_test(TESTSTR("\x1b$-~\x1b~\xa0\xff"), ERROR
, 0, -1);
596 /* Designate control sets */
597 iso2022_read_test(TESTSTR("\x1b!@"), 0x1b, '!', '@', 0, -1);
598 /* Designate other coding system */
599 iso2022_read_test(TESTSTR("\x1b%G"
600 "\xCE\xBA\xE1\xBD\xB9\xCF\x83\xCE\xBC\xCE\xB5"),
601 0x03BA, 0x1F79, 0x03C3, 0x03BC, 0x03B5, 0, -1);
602 iso2022_read_test(TESTSTR("\x1b-A\x1b%G\xCE\xBA\x1b%@\xa0"),
603 0x03BA, 0xA0, 0, -1);
604 iso2022_read_test(TESTSTR("\x1b%G\xCE\x1b%@"), ERROR
, 0, -1);
605 iso2022_read_test(TESTSTR("\x1b%G\xCE\xBA\x1b%\x1b%@"),
606 0x03BA, 0x1B, '%', 0, -1);
607 printf("read tests completed\n");
608 printf("total: %d errors\n", total_errs
);
609 return (total_errs
!= 0);
612 #endif /* TESTMODE */
614 #else /* ENUM_CHARSETS */
616 ENUM_CHARSET(CS_ISO2022
)