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.
37 enum {S4
, S6
, M4
, M6
};
39 static long int emacs_big5_1_to_unicode(int, int);
40 static long int emacs_big5_2_to_unicode(int, int);
41 static int unicode_to_emacs_big5(long int, int *, int *, int *);
42 static long int cns11643_1_to_unicode(int, int);
43 static long int cns11643_2_to_unicode(int, int);
44 static long int cns11643_3_to_unicode(int, int);
45 static long int cns11643_4_to_unicode(int, int);
46 static long int cns11643_5_to_unicode(int, int);
47 static long int cns11643_6_to_unicode(int, int);
48 static long int cns11643_7_to_unicode(int, int);
49 static long int null_dbcs_to_unicode(int, int);
50 static int unicode_to_null_dbcs(long int, int *, int *);
52 typedef int (*to_dbcs_t
)(long int, int *, int *);
53 typedef int (*to_dbcs_planar_t
)(long int, int *, int *, int *);
56 * Cast between to_dbcs_planar_t and to_dbcs_t.
58 * I (SGT) originally defined these two macros as follows:
60 #define DEPLANARISE(x) ( (x) == (to_dbcs_planar_t)NULL, (to_dbcs_t)(x) )
61 #define REPLANARISE(x) ( (x) == (to_dbcs_t)NULL, (to_dbcs_planar_t)(x) )
63 * When compiled with gcc, this had the effect of type-checking the
64 * input, so that DEPLANARISE would cast a to_dbcs_t to a
65 * to_dbcs_planar_t but cause a compile error if passed any other
66 * input type, and vice versa. However, MSVC felt that this was a
67 * non-constant expression and hence not legal to use in a static
68 * initialiser, and probably rightly so: I haven't had a chance to
69 * check with the C standard, but I'd be surprised if it _required_
70 * compilers to keep an open mind long enough to discover that the
71 * non-constant part of the expression has its result thrown away.
73 * I can't think of any other means of performing this type check
74 * which doesn't have the same problem, so I'm taking the type
75 * checks out, with regret.
77 #define DEPLANARISE(x) ( (to_dbcs_t)(x) )
78 #define REPLANARISE(x) ( (to_dbcs_planar_t)(x) )
81 * Values used in the `enable' field. Each of these identifies a
82 * class of character sets; we then have a bitmask indicating which
83 * classes are allowable in a given mode.
85 * These values are currently only checked on output: for input,
86 * any ISO 2022 we can comprehend at all is considered acceptable.
88 #define CCS 1 /* CTEXT standard */
89 #define COS 2 /* other standard */
90 #define CPU 3 /* private use */
91 #define CDC 4 /* DOCS for CTEXT */
92 #define CDU 5 /* DOCS for UTF-8 */
93 #define CNU 31 /* never used */
97 char ltype
, li
, lf
, rtype
, ri
, rf
;
100 const struct iso2022_subcharset
{
101 char type
, i
, f
, enable
;
103 const sbcs_data
*sbcs_base
;
104 long int (*from_dbcs
)(int, int);
107 * If to_dbcs_plane < 0, then to_dbcs is used as expected.
108 * However, if to_dbcs_plane >= 0, then to_dbcs is expected to
109 * be cast to a to_dbcs_planar_t before use, and the returned
110 * plane value (the first int *) must equal to_dbcs_plane.
112 * I'd have preferred to do this by means of a union, but you
113 * can't initialise a selected field of a union at compile
114 * time. Function pointer casts are guaranteed to work sensibly
115 * in ISO C (that is, it's undefined what happens if you call a
116 * function via the wrong type of pointer, but if you cast it
117 * back to the right type before calling it then it must work),
118 * so this is safe if ugly.
121 int to_dbcs_plane
; /* use to_dbcs_planar iff >= 0 */
122 } iso2022_subcharsets
[] = {
124 * We list these subcharsets in preference order for output.
125 * Since the best-defined use of ISO 2022 output is compound
126 * text, we'll use a preference order which matches that. So we
127 * begin with the charsets defined in the compound text spec.
129 { S4
, 0, 'B', CCS
, 0x00, &sbcsdata_CS_ASCII
},
130 { S6
, 0, 'A', CCS
, 0x80, &sbcsdata_CS_ISO8859_1
},
131 { S6
, 0, 'B', CCS
, 0x80, &sbcsdata_CS_ISO8859_2
},
132 { S6
, 0, 'C', CCS
, 0x80, &sbcsdata_CS_ISO8859_3
},
133 { S6
, 0, 'D', CCS
, 0x80, &sbcsdata_CS_ISO8859_4
},
134 { S6
, 0, 'F', CCS
, 0x80, &sbcsdata_CS_ISO8859_7
},
135 { S6
, 0, 'G', CCS
, 0x80, &sbcsdata_CS_ISO8859_6
},
136 { S6
, 0, 'H', CCS
, 0x80, &sbcsdata_CS_ISO8859_8
},
137 { S6
, 0, 'L', CCS
, 0x80, &sbcsdata_CS_ISO8859_5
},
138 { S6
, 0, 'M', CCS
, 0x80, &sbcsdata_CS_ISO8859_9
},
139 { S4
, 0, 'I', CCS
, 0x80, &sbcsdata_CS_JISX0201
},
140 { S4
, 0, 'J', CCS
, 0x00, &sbcsdata_CS_JISX0201
},
141 { M4
, 0, 'A', CCS
, -0x21, 0, &gb2312_to_unicode
, &unicode_to_gb2312
, -1 },
142 { M4
, 0, 'B', CCS
, -0x21, 0, &jisx0208_to_unicode
, &unicode_to_jisx0208
, -1 },
143 { M4
, 0, 'C', CCS
, -0x21, 0, &ksx1001_to_unicode
, &unicode_to_ksx1001
, -1 },
144 { M4
, 0, 'D', CCS
, -0x21, 0, &jisx0212_to_unicode
, &unicode_to_jisx0212
, -1 },
147 * Next, other reasonably standard things: the rest of the ISO
148 * 8859 sets, UK-ASCII, and CNS 11643.
150 { S6
, 0, 'T', COS
, 0x80, &sbcsdata_CS_ISO8859_11
},
151 { S6
, 0, 'V', COS
, 0x80, &sbcsdata_CS_ISO8859_10
},
152 { S6
, 0, 'Y', COS
, 0x80, &sbcsdata_CS_ISO8859_13
},
153 { S6
, 0, '_', COS
, 0x80, &sbcsdata_CS_ISO8859_14
},
154 { S6
, 0, 'b', COS
, 0x80, &sbcsdata_CS_ISO8859_15
},
155 { S6
, 0, 'f', COS
, 0x80, &sbcsdata_CS_ISO8859_16
},
156 { S4
, 0, 'A', COS
, 0x00, &sbcsdata_CS_BS4730
},
157 { M4
, 0, 'G', COS
, -0x21, 0, &cns11643_1_to_unicode
, DEPLANARISE(&unicode_to_cns11643
), 0 },
158 { M4
, 0, 'H', COS
, -0x21, 0, &cns11643_2_to_unicode
, DEPLANARISE(&unicode_to_cns11643
), 1 },
159 { M4
, 0, 'I', COS
, -0x21, 0, &cns11643_3_to_unicode
, DEPLANARISE(&unicode_to_cns11643
), 2 },
160 { M4
, 0, 'J', COS
, -0x21, 0, &cns11643_4_to_unicode
, DEPLANARISE(&unicode_to_cns11643
), 3 },
161 { M4
, 0, 'K', COS
, -0x21, 0, &cns11643_5_to_unicode
, DEPLANARISE(&unicode_to_cns11643
), 4 },
162 { M4
, 0, 'L', COS
, -0x21, 0, &cns11643_6_to_unicode
, DEPLANARISE(&unicode_to_cns11643
), 5 },
163 { M4
, 0, 'M', COS
, -0x21, 0, &cns11643_7_to_unicode
, DEPLANARISE(&unicode_to_cns11643
), 6 },
166 * Private-use designations: DEC private sets and Emacs's Big5
169 { S4
, 0, '0', CPU
, 0x00, &sbcsdata_CS_DEC_GRAPHICS
},
170 { S4
, 0, '<', CPU
, 0x80, &sbcsdata_CS_DEC_MCS
},
171 { M4
, 0, '0', CPU
, -0x21, 0, &emacs_big5_1_to_unicode
, DEPLANARISE(&unicode_to_emacs_big5
), 1 },
172 { M4
, 0, '1', CPU
, -0x21, 0, &emacs_big5_2_to_unicode
, DEPLANARISE(&unicode_to_emacs_big5
), 2 },
175 * Ben left this conditioned out without explanation,
176 * presumably on the grounds that we don't have a translation
180 { M4
, 0, '@', CNU
}, /* JIS C 6226-1978 */
184 * Finally, fallback entries for null character sets.
187 { S6
, 0, '~', CNU
}, /* empty 96-set */
188 { M4
, 0, '~', CNU
, 0, 0, &null_dbcs_to_unicode
, &unicode_to_null_dbcs
, -1 }, /* empty 94^n-set */
189 { M6
, 0, '~', CNU
, 0, 0, &null_dbcs_to_unicode
, &unicode_to_null_dbcs
, -1 }, /* empty 96^n-set */
192 static long int null_dbcs_to_unicode(int r
, int c
)
198 static int unicode_to_null_dbcs(long int unicode
, int *r
, int *c
)
203 return 0; /* failed to convert anything */
207 * Emacs encodes Big5 in COMPOUND_TEXT as two 94x94 character sets.
208 * We treat Big5 as a 94x191 character set with a bunch of undefined
209 * columns in the middle, so we have to mess around a bit to make
213 static long int emacs_big5_1_to_unicode(int r
, int c
)
219 if (c
>= 64) c
+= 34; /* Skip over the gap */
220 return big5_to_unicode(r
, c
);
223 static long int emacs_big5_2_to_unicode(int r
, int c
)
229 if (c
>= 64) c
+= 34; /* Skip over the gap */
230 return big5_to_unicode(r
, c
);
233 static int unicode_to_emacs_big5(long int unicode
, int *p
, int *r
, int *c
)
236 if (!unicode_to_big5(unicode
, &rr
, &cc
))
254 /* Wrappers for cns11643_to_unicode() */
255 static long int cns11643_1_to_unicode(int r
, int c
)
257 return cns11643_to_unicode(0, r
, c
);
259 static long int cns11643_2_to_unicode(int r
, int c
)
261 return cns11643_to_unicode(1, r
, c
);
263 static long int cns11643_3_to_unicode(int r
, int c
)
265 return cns11643_to_unicode(2, r
, c
);
267 static long int cns11643_4_to_unicode(int r
, int c
)
269 return cns11643_to_unicode(3, r
, c
);
271 static long int cns11643_5_to_unicode(int r
, int c
)
273 return cns11643_to_unicode(4, r
, c
);
275 static long int cns11643_6_to_unicode(int r
, int c
)
277 return cns11643_to_unicode(5, r
, c
);
279 static long int cns11643_7_to_unicode(int r
, int c
)
281 return cns11643_to_unicode(6, r
, c
);
284 /* States, or "what we're currently accumulating". */
286 IDLE
, /* None of the below */
287 SS2CHAR
, /* Accumulating a character after SS2 */
288 SS3CHAR
, /* Accumulating a character after SS3 */
289 ESCSEQ
, /* Accumulating an escape sequence */
290 ESCDROP
, /* Discarding an escape sequence */
291 ESCPASS
, /* Passing through an escape sequence */
292 DOCSUTF8
, /* DOCSed into UTF-8 */
293 DOCSCTEXT
/* DOCSed into a COMPOUND_TEXT extended segment */
298 static void dump_state(charset_state
*s
)
300 unsigned s0
= s
->s0
, s1
= s
->s1
;
301 char const * const modes
[] = { "IDLE", "SS2CHAR", "SS3CHAR",
302 "ESCSEQ", "ESCDROP", "ESCPASS",
305 fprintf(stderr
, "s0: %s", modes
[s0
>> 29]);
306 fprintf(stderr
, " %02x %02x %02x ", (s0
>> 16) & 0xff, (s0
>> 8) & 0xff,
308 fprintf(stderr
, "s1: LS%d LS%dR", (s1
>> 30) & 3, (s1
>> 28) & 3);
309 fprintf(stderr
, " %d %d %d %d\n", s1
& 0x7f, (s1
>> 7) & 0x7f,
310 (s1
>> 14) & 0x7f, (s1
>> 21) & 0x7f);
314 static void designate(charset_state
*state
, int container
,
315 int type
, int ibyte
, int fbyte
)
319 assert(container
>= 0 && container
<= 3);
320 assert(type
== S4
|| type
== S6
|| type
== M4
|| type
== M6
);
322 for (i
= 0; i
< lenof(iso2022_subcharsets
); i
++) {
323 if (iso2022_subcharsets
[i
].type
== type
&&
324 iso2022_subcharsets
[i
].i
== ibyte
&&
325 iso2022_subcharsets
[i
].f
== fbyte
) {
326 state
->s1
&= ~(0x7fL
<< (container
* 7));
327 state
->s1
|= (i
<< (container
* 7));
332 * If we don't find the charset, invoke the empty one, so we
333 * output ERROR rather than garbage.
335 designate(state
, container
, type
, 0, '~');
338 static void do_utf8(long int input_chr
,
339 charset_state
*state
,
340 void (*emit
)(void *ctx
, long int output
),
343 charset_state ustate
;
346 ustate
.s0
= state
->s0
& 0x03ffffffL
;
347 read_utf8(NULL
, input_chr
, &ustate
, emit
, emitctx
);
348 state
->s0
= (state
->s0
& ~0x03ffffffL
) | (ustate
.s0
& 0x03ffffffL
);
351 static void docs_utf8(long int input_chr
,
352 charset_state
*state
,
353 void (*emit
)(void *ctx
, long int output
),
359 * Bits [25:0] of s0 are reserved for read_utf8().
360 * Bits [27:26] are a tiny state machine to recognise ESC % @.
362 retstate
= (state
->s0
& 0x0c000000L
) >> 26;
363 if (retstate
== 1 && input_chr
== '%')
365 else if (retstate
== 2 && input_chr
== '@') {
366 /* If we've got a partial UTF-8 sequence, complain. */
367 if (state
->s0
& 0x03ffffffL
)
368 emit(emitctx
, ERROR
);
372 if (retstate
>= 1) do_utf8(ESC
, state
, emit
, emitctx
);
373 if (retstate
>= 2) do_utf8('%', state
, emit
, emitctx
);
375 if (input_chr
== ESC
)
378 do_utf8(input_chr
, state
, emit
, emitctx
);
381 state
->s0
= (state
->s0
& ~0x0c000000L
) | (retstate
<< 26);
384 struct ctext_encoding
{
386 char octets_per_char
, enable
;
387 charset_spec
const *subcs
;
391 * In theory, this list is in <ftp://ftp.x.org/pub/DOCS/registry>,
392 * but XLib appears to have its own ideas, and encodes these three
396 extern charset_spec
const charset_CS_ISO8859_14
;
397 extern charset_spec
const charset_CS_ISO8859_15
;
398 extern charset_spec
const charset_CS_BIG5
;
400 static struct ctext_encoding
const ctext_encodings
[] = {
401 { "big5-0\2", 0 /* variable */, CDC
, &charset_CS_BIG5
},
402 { "iso8859-14\2", 1, CDC
, &charset_CS_ISO8859_14
},
403 { "iso8859-15\2", 1, CDC
, &charset_CS_ISO8859_15
}
406 static void docs_ctext(long int input_chr
,
407 charset_state
*state
,
408 void (*emit
)(void *ctx
, long int output
),
412 * s0[27:26] = first entry in ctext_encodings that matches
413 * s0[25:22] = number of characters successfully matched, 0xf if all
414 * s0[21:8] count the number of octets left in the segment
415 * s0[7:0] are for sub-charset use
417 int n
= (state
->s0
>> 22) & 0xf, i
= (state
->s0
>> 26) & 3, oi
= i
, j
;
418 int length
= (state
->s0
>> 8) & 0x3fff;
421 * Note that we do not bother checking the octets-per-character
422 * byte against the selected charset when reading. It's
423 * extremely unlikely that this code will ever have to deal
424 * with two charset identifiers with the same name and
425 * different octets-per-character values! If it ever happens,
426 * we'll have to edit this file anyway so we can modify the
431 /* Haven't read length yet */
432 if ((state
->s0
& 0xff) == 0)
433 /* ... or even the first byte */
434 state
->s0
|= input_chr
;
436 length
= (state
->s0
& 0x7f) * 0x80 + (input_chr
& 0x7f);
440 state
->s0
= (state
->s0
& 0xf0000000) | (length
<< 8);
447 /* Skipping unknown encoding. Look out for STX. */
449 state
->s0
= (state
->s0
& 0xf0000000) | (i
<< 26) | (0xf << 22);
450 } else if (n
!= 0xf) {
451 while ((unsigned)j
< lenof(ctext_encodings
) &&
452 !memcmp(ctext_encodings
[j
].name
,
453 ctext_encodings
[oi
].name
, n
)) {
454 if (ctext_encodings
[j
].name
[n
] < input_chr
)
459 if ((unsigned)i
>= lenof(ctext_encodings
) ||
460 memcmp(ctext_encodings
[i
].name
,
461 ctext_encodings
[oi
].name
, n
) ||
462 ctext_encodings
[i
].name
[n
] != input_chr
) {
463 /* Doom! We haven't heard of this encoding */
464 i
= lenof(ctext_encodings
);
468 * Otherwise, we have found an additional character in our
469 * encoding name. See if we have reached the _end_ of our
473 if (!ctext_encodings
[i
].name
[n
])
477 * Failing _that_, we simply update our encoding-name-
480 assert(i
< 4 && n
< 16);
481 state
->s0
= (state
->s0
& 0xf0000000) | (i
<< 26) | (n
<< 22);
483 if ((unsigned)i
>= lenof(ctext_encodings
))
484 emit(emitctx
, ERROR
);
486 charset_state substate
;
487 charset_spec
const *subcs
= ctext_encodings
[i
].subcs
;
489 substate
.s0
= state
->s0
& 0xff;
490 subcs
->read(subcs
, input_chr
, &substate
, emit
, emitctx
);
491 state
->s0
= (state
->s0
& ~0xff) | (substate
.s0
& 0xff);
497 state
->s0
= (state
->s0
&~0x003fff00) | (length
<< 8);
500 static void read_iso2022(charset_spec
const *charset
, long int input_chr
,
501 charset_state
*state
,
502 void (*emit
)(void *ctx
, long int output
),
505 struct iso2022_mode
const *mode
= (struct iso2022_mode
*)charset
->data
;
507 /* dump_state(state); */
509 * We have to make fairly efficient use of the 64 bits of state
510 * available to us. Long-term state goes in s1, and consists of
511 * the identities of the character sets designated as G0/G1/G2/G3
512 * and the locking-shift states for GL and GR. Short-term state
513 * goes in s0: The bottom half of s0 accumulates characters for an
514 * escape sequence or a multi-byte character, while the top three
515 * bits indicate what they're being accumulated for. After DOCS,
516 * the bottom 29 bits of state are available for the DOCS function
517 * to use -- the UTF-8 one uses the bottom 26 for UTF-8 decoding
518 * and the top two to recognised ESC % @.
520 * s0[31:29] = state enum
521 * s0[24:0] = accumulated bytes
522 * s1[31:30] = GL locking-shift state
523 * s1[29:28] = GR locking-shift state
524 * s1[27:21] = G3 charset
525 * s1[20:14] = G2 charset
526 * s1[13:7] = G1 charset
527 * s1[6:0] = G0 charset
532 #define LOCKING_SHIFT(n,side) \
533 (state->s1 = (state->s1 & ~(3L<<(side))) | ((n ## L)<<(side)))
534 #define MODE ((state->s0 & 0xe0000000L) >> 29)
535 #define ENTER_MODE(m) (state->s0 = (state->s0 & ~0xe0000000L) | ((m)<<29))
536 #define SINGLE_SHIFT(n) ENTER_MODE(SS2CHAR - 2 + (n))
537 #define ASSERT_IDLE do { \
538 if (state->s0 != 0) emit(emitctx, ERROR); \
542 if (state
->s1
== 0) {
544 * Since there's no LS0R, this means we must just have started.
545 * Set up a sane initial state (LS0, LS1R, ASCII in G0/G1/G2/G3).
547 LOCKING_SHIFT(0, LEFT
);
548 LOCKING_SHIFT(1, RIGHT
);
549 designate(state
, 0, mode
->ltype
, mode
->li
, mode
->lf
);
550 designate(state
, 1, mode
->rtype
, mode
->ri
, mode
->rf
);
551 designate(state
, 2, S4
, 0, 'B');
552 designate(state
, 3, S4
, 0, 'B');
555 if (MODE
== DOCSUTF8
) {
556 docs_utf8(input_chr
, state
, emit
, emitctx
);
559 if (MODE
== DOCSCTEXT
) {
560 docs_ctext(input_chr
, state
, emit
, emitctx
);
564 if ((input_chr
& 0x60) == 0x00) {
565 /* C0 or C1 control */
572 LOCKING_SHIFT(0, LEFT
);
575 LOCKING_SHIFT(1, LEFT
);
584 emit(emitctx
, input_chr
);
587 } else if ((input_chr
& 0x80) || MODE
< ESCSEQ
) {
589 struct iso2022_subcharset
const *subcs
;
594 * Force idle state if we're in mid escape sequence, or in a
595 * multi-byte character with a different top bit.
597 if (MODE
>= ESCSEQ
||
598 ((state
->s0
& 0x00ff0000L
) != 0 &&
599 (((state
->s0
>> 16) ^ input_chr
) & 0x80)))
601 if (MODE
== SS2CHAR
|| MODE
== SS3CHAR
) /* Single-shift */
602 container
= MODE
- SS2CHAR
+ 2;
603 else if (input_chr
>= 0x80) /* GR */
604 container
= (state
->s1
>> 28) & 3;
606 container
= state
->s1
>> 30;
609 input_7bit
= input_chr
& ~0x80;
610 subcs
= &iso2022_subcharsets
[(state
->s1
>> (container
* 7)) & 0x7f];
611 if ((subcs
->type
== S4
|| subcs
->type
== M4
) &&
612 (input_7bit
== 0x20 || input_7bit
== 0x7f)) {
613 /* characters not in 94-char set */
614 if (is_gl
) emit(emitctx
, input_7bit
);
615 else emit(emitctx
, ERROR
);
616 } else if (subcs
->type
== M4
|| subcs
->type
== M6
) {
617 if ((state
->s0
& 0x00ff0000L
) == 0) {
618 state
->s0
|= input_chr
<< 16;
622 subcs
->from_dbcs(((state
->s0
>> 16) & 0x7f) +
624 input_7bit
+ subcs
->offset
));
627 if ((state
->s0
& 0x00ff0000L
) != 0)
628 emit(emitctx
, ERROR
);
629 emit(emitctx
, subcs
->sbcs_base ?
630 sbcs_to_unicode(subcs
->sbcs_base
, input_7bit
+ subcs
->offset
):
636 if (MODE
== ESCPASS
) {
637 emit(emitctx
, input_chr
);
638 if ((input_chr
& 0xf0) != 0x20)
644 * Intermediate bytes shall be any of the 16 positions of
645 * column 02 of the code table; they are denoted by the symbol
648 if ((input_chr
& 0xf0) == 0x20) {
649 if (((state
->s0
>> 16) & 0xff) == 0)
650 state
->s0
|= input_chr
<< 16;
651 else if (((state
->s0
>> 8) & 0xff) == 0)
652 state
->s0
|= input_chr
<< 8;
654 /* Long escape sequence. Switch to ESCPASS or ESCDROP. */
655 i1
= (state
->s0
>> 16) & 0xff;
656 i2
= (state
->s0
>> 8) & 0xff;
658 case '(': case ')': case '*': case '+':
659 case '-': case '.': case '/':
667 emit(emitctx
, input_chr
);
677 * Final bytes shall be any of the 79 positions of columns 03
678 * to 07 of the code table excluding position 07/15; they are
679 * denoted by the symbol F.
681 i1
= (state
->s0
>> 16) & 0xff;
682 i2
= (state
->s0
>> 8) & 0xff;
684 input_chr
= 0; /* Make sure it won't match. */
687 case 0: /* No intermediate bytes */
696 LOCKING_SHIFT(2, LEFT
);
699 LOCKING_SHIFT(3, LEFT
);
702 LOCKING_SHIFT(3, RIGHT
);
705 LOCKING_SHIFT(2, RIGHT
);
708 LOCKING_SHIFT(1, RIGHT
);
711 /* Unsupported escape sequence. Spit it back out. */
713 emit(emitctx
, input_chr
);
718 * Various coding structure facilities specify that designating
719 * a code element also invokes it. As far as I can see, invoking
720 * it now will have the same practical effect, since those
721 * facilities also ban the use of locking shifts.
724 case 'A': /* G0 element used and invoked into GL */
725 LOCKING_SHIFT(0, LEFT
);
727 case 'C': /* G0 in GL, G1 in GR */
728 case 'D': /* Ditto, at least for 8-bit codes */
729 case 'L': /* ISO 4873 (ECMA-43) level 1 */
730 case 'M': /* ISO 4873 (ECMA-43) level 2 */
731 LOCKING_SHIFT(0, LEFT
);
732 LOCKING_SHIFT(1, RIGHT
);
738 * IRR (Identify Revised Registration) is ignored here,
739 * since any revised registration must be
740 * upward-compatible with the old one, so either we'll
741 * support the new one or we'll emit ERROR when we run
742 * into a new character. In either case, there's nothing
746 case '(': /* GZD4 */ case ')': /* G1D4 */
747 case '*': /* G2D4 */ case '+': /* G3D4 */
748 designate(state
, i1
- '(', S4
, i2
, input_chr
);
750 case '-': /* G1D6 */ case '.': /* G2D6 */ case '/': /* G3D6 */
751 designate(state
, i1
- ',', S6
, i2
, input_chr
);
753 case '$': /* G?DM? */
755 case 0: /* Obsolete version of GZDM4 */
757 case '(': /* GZDM4 */ case ')': /* G1DM4 */
758 case '*': /* G2DM4 */ case '+': /* G3DM4 */
759 designate(state
, i2
- '(', M4
, 0, input_chr
);
761 case '-': /* G1DM6 */
762 case '.': /* G2DM6 */ case '/': /* G3DM6 */
763 designate(state
, i2
- ',', M6
, 0, input_chr
);
766 emit(emitctx
, ERROR
);
770 /* XXX What's a reasonable way to handle an unrecognised DOCS? */
775 ENTER_MODE(DOCSUTF8
);
782 ENTER_MODE(DOCSCTEXT
);
789 /* Unsupported nF escape sequence. Re-emit it. */
792 if (i2
) emit(emitctx
, i2
);
793 emit(emitctx
, input_chr
);
799 static void oselect(charset_state
*state
, int i
, int right
,
800 void (*emit
)(void *ctx
, long int output
),
803 int shift
= (right ?
31-7 : 31-7-7);
804 struct iso2022_subcharset
const *subcs
= &iso2022_subcharsets
[i
];
806 if (((state
->s1
>> shift
) & 0x7F) != (unsigned)i
) {
807 state
->s1
&= ~(0x7FL
<< shift
);
808 state
->s1
|= (i
<< shift
);
812 if (subcs
->type
== M4
|| subcs
->type
== M6
)
814 if (subcs
->type
== S6
|| subcs
->type
== M6
) {
823 emit(emitctx
, subcs
->i
);
824 emit(emitctx
, subcs
->f
);
829 static void docs_char(charset_state
*state
,
830 void (*emit
)(void *ctx
, long int output
),
831 void *emitctx
, int cset
, char *data
, int datalen
)
833 int curr_cset
, currlen
, i
;
836 * cset is the index into ctext_encodings[]. It can also be -1
837 * to mean DOCS UTF-8, or -2 to mean no DOCS (ordinary 2022).
838 * In the latter case, `chr' is ignored.
842 * First, terminate a DOCS segment if necessary. We always have
843 * to terminate a DOCS segment if one is active and we're about
844 * to switch to a different one; we might also have to
845 * terminate a length-encoded DOCS segment if we've run out of
846 * storage space to accumulate characters in it.
848 curr_cset
= ((state
->s1
>> 14) & 7) - 2;
849 currlen
= ((state
->s1
>> 11) & 7);
850 if ((curr_cset
!= -2 && curr_cset
!= cset
) ||
851 (curr_cset
>= 0 && currlen
+ datalen
> 5)) {
852 if (curr_cset
== -1) {
854 * Terminating DOCS UTF-8 is easy.
863 * To terminate a length-encoded DOCS segment we must
864 * actually output the whole thing.
869 emit(emitctx
, '0' + ctext_encodings
[curr_cset
].octets_per_char
);
870 len
= currlen
+ datalen
+
871 strlen(ctext_encodings
[curr_cset
].name
);
872 assert(len
< (1 << 14));
873 emit(emitctx
, 0x80 | ((len
>> 7) & 0x7F));
874 emit(emitctx
, 0x80 | ((len
) & 0x7F));
875 /* The name stored in ctext_encodings[] includes the trailing \2 */
876 for (i
= 0; ctext_encodings
[curr_cset
].name
[i
]; i
++)
877 emit(emitctx
, ctext_encodings
[curr_cset
].name
[i
]);
878 for (i
= 0; i
< currlen
; i
++)
880 (i
== 0 ? state
->s1
: state
->s0
>> (8*(4-i
))) & 0xFF);
881 for (i
= 0; i
< datalen
; i
++)
882 emit(emitctx
, data
[i
]);
885 * We've now dealt with the input data, so clear it so
886 * we don't try to do so again below.
894 * Now, start a DOCS segment if necessary.
896 if (curr_cset
!= cset
) {
907 * Starting a length-encoded DOCS segment is simply a
908 * matter of setting our stored length counter to zero.
911 state
->s1
&= ~(7 << 11);
916 state
->s1
&= ~(7 << 14);
917 assert((cset
+2) >= 0 && (cset
+2) < 8);
918 state
->s1
|= ((cset
+2) << 14);
921 * Now we're in the right DOCS state. Actually deal with the
922 * input data, if we haven't already done so above.
928 * In DOCS UTF-8, we output data as soon as we get it.
930 for (i
= 0; i
< datalen
; i
++)
931 emit(emitctx
, data
[i
]);
934 * In length-encoded DOCS, we just store our data and
935 * bide our time. It'll all be output when we fill up
936 * or switch to another character set.
938 assert(currlen
+ datalen
<= 5); /* overflow handled already */
939 for (i
= 0; i
< datalen
; i
++) {
940 if (currlen
+ i
== 0)
941 state
->s1
|= data
[i
] & 0xFF;
943 state
->s0
|= (data
[i
] & 0xFF) << (8*(4-(currlen
+i
)));
946 assert(currlen
>= 0 && currlen
< 8);
947 state
->s1
&= ~(7 << 11);
948 state
->s1
|= (currlen
<< 11);
953 static void write_to_pointer(void *ctx
, long int output
)
955 char **ptr
= (char **)ctx
;
960 * Writing full ISO-2022 is not useful in very many circumstances.
961 * One of the few situations in which it _is_ useful is generating
962 * X11 COMPOUND_TEXT; therefore, this writing function will obey
963 * the compound text restrictions and hence output the subset of
964 * ISO-2022 that's usable in that context.
966 * The subset in question is roughly that we use GL/GR for G0/G1
967 * always, and that the _only_ escape sequences we output (other
968 * than the occasional DOCS) are those which designate different
969 * subcharsets into G0 and G1. There are additional constraints
970 * about which things go in which container; see below.
972 * FIXME: this wants some decent tests to be written, and also the
973 * exact output policy for compound text wants thinking about more
976 static int write_iso2022(charset_spec
const *charset
, long int input_chr
,
977 charset_state
*state
,
978 void (*emit
)(void *ctx
, long int output
),
982 struct iso2022_subcharset
const *subcs
;
983 struct iso2022_mode
const *mode
= (struct iso2022_mode
*)charset
->data
;
984 to_dbcs_planar_t last_planar_dbcs
= NULL
;
985 int last_p
, last_r
, last_c
;
989 * For output, I allocate the state variables as follows:
991 * s1[31] == 1 if output state has been initialised
992 * s1[30:24] == G1 charset (always in GR)
993 * s1[23:17] == G0 charset (always in GL)
994 * s1[16:14] == DOCS index plus 2 (because -1 and -2 are special)
995 * s1[13:11] == number of DOCS accumulated characters (up to five)
996 * s1[7:0] + s0[31:0] == DOCS collected characters
1000 state
->s0
= 0x00000000UL
;
1001 state
->s1
= 0x80000000UL
;
1003 * Start with US-ASCII in GL and also in GR.
1005 for (i
= 0; (unsigned)i
< lenof(iso2022_subcharsets
); i
++) {
1006 subcs
= &iso2022_subcharsets
[i
];
1007 if (subcs
->type
== mode
->ltype
&&
1008 subcs
->i
== mode
->li
&&
1009 subcs
->f
== mode
->lf
)
1010 oselect(state
, i
, FALSE
, NULL
, NULL
);
1011 if (subcs
->type
== mode
->rtype
&&
1012 subcs
->i
== mode
->ri
&&
1013 subcs
->f
== mode
->rf
)
1014 oselect(state
, i
, TRUE
, NULL
, NULL
);
1018 if (input_chr
== -1) {
1020 * Special case: reset encoding state.
1022 docs_char(state
, emit
, emitctx
, -2, NULL
, 0); /* leave DOCS */
1024 for (i
= 0; (unsigned)i
< lenof(iso2022_subcharsets
); i
++) {
1025 subcs
= &iso2022_subcharsets
[i
];
1026 if (subcs
->type
== mode
->ltype
&&
1027 subcs
->i
== mode
->li
&&
1028 subcs
->f
== mode
->lf
)
1029 oselect(state
, i
, FALSE
, emit
, emitctx
);
1030 if (subcs
->type
== mode
->rtype
&&
1031 subcs
->i
== mode
->ri
&&
1032 subcs
->f
== mode
->rf
)
1033 oselect(state
, i
, TRUE
, emit
, emitctx
);
1039 * Special-case characters: Space, Delete, and anything in C0
1040 * or C1 are output unchanged.
1042 if (input_chr
<= 0x20 || (input_chr
>= 0x7F && input_chr
< 0xA0)) {
1043 emit(emitctx
, input_chr
);
1048 * Analyse the input character and work out which subcharset it
1051 for (i
= 0; (unsigned)i
< lenof(iso2022_subcharsets
); i
++) {
1052 subcs
= &iso2022_subcharsets
[i
];
1053 if (!(mode
->enable_mask
& (1 << subcs
->enable
)))
1054 continue; /* this charset is disabled */
1055 if (subcs
->sbcs_base
) {
1056 c1
= sbcs_from_unicode(subcs
->sbcs_base
, input_chr
);
1057 c1
-= subcs
->offset
;
1058 if (c1
>= 0x20 && c1
<= 0x7f) {
1062 } else if (subcs
->to_dbcs
) {
1063 if (subcs
->to_dbcs_plane
>= 0) {
1065 * Since multiplanar DBCSes almost by definition
1066 * involve several entries in iso2022_subcharsets
1067 * with the same to_dbcs function and different
1068 * plane values, we remember the last such function
1069 * we called and what its result was, so that we
1070 * don't (for example) have to call
1071 * unicode_to_cns11643 seven times.
1073 if (last_planar_dbcs
!= REPLANARISE(subcs
->to_dbcs
)) {
1074 last_planar_dbcs
= REPLANARISE(subcs
->to_dbcs
);
1075 if (!last_planar_dbcs(input_chr
,
1076 &last_p
, &last_r
, &last_c
))
1080 last_p
= subcs
->to_dbcs_plane
;
1081 if (!subcs
->to_dbcs(input_chr
, &last_r
, &last_c
))
1082 last_p
= 0; /* cannot match since to_dbcs_plane<0 */
1085 if (last_p
== subcs
->to_dbcs_plane
) {
1086 c1
= last_r
- subcs
->offset
;
1087 c2
= last_c
- subcs
->offset
;
1088 assert(c1
>= 0x20 && c1
<= 0x7f);
1089 assert(c2
>= 0x20 && c2
<= 0x7f);
1095 if ((unsigned)i
< lenof(iso2022_subcharsets
)) {
1099 * Our character is represented by c1 (and possibly also
1100 * c2) in subcharset `subcs'. So now we must decide whether
1101 * to designate that character set into G0/GL or G1/GR.
1103 * Any S6 or M6 subcharset has to go in GR because it won't
1104 * fit in GL. In addition, the compound text rules state
1105 * that any single-byte subcharset defined as the
1106 * right-hand half of some SBCS must go in GR.
1108 * M4 subcharsets can go in either half according to the
1109 * rules. I choose to put them in GR always because it's a
1110 * simple policy with reasonable behaviour (facilitates
1111 * switching between them and ASCII).
1113 right
= (subcs
->type
== S6
|| subcs
->type
== M6
|| subcs
->type
== M4
||
1114 (subcs
->sbcs_base
&& subcs
->offset
== 0x80));
1117 * If we're in a DOCS mode, leave it.
1119 docs_char(state
, emit
, emitctx
, -2, NULL
, 0);
1122 * If this subcharset is not already selected in that
1123 * container, select it.
1125 oselect(state
, i
, right
, emit
, emitctx
);
1128 * Now emit the actual characters.
1131 assert(c1
>= 0x20 && c1
<= 0x7f);
1132 emit(emitctx
, c1
| 0x80);
1134 assert(c2
>= 0x20 && c2
<= 0x7f);
1135 emit(emitctx
, c2
| 0x80);
1138 assert(c1
> 0x20 && c1
< 0x7f);
1141 assert(c2
> 0x20 && c2
< 0x7f);
1150 * Fall back to DOCS.
1157 cs
= -2; /* means failure */
1159 for (i
= 0; (unsigned)i
<= lenof(ctext_encodings
); i
++) {
1160 charset_state substate
;
1161 charset_spec
const *subcs
= ctext_encodings
[i
].subcs
;
1164 * We assume that all character sets dealt with by DOCS
1165 * are stateless for output purposes.
1167 substate
.s1
= substate
.s0
= 0;
1170 if ((unsigned)i
< lenof(ctext_encodings
)) {
1171 if ((mode
->enable_mask
& (1 << ctext_encodings
[i
].enable
)) &&
1172 subcs
->write(subcs
, input_chr
, &substate
,
1173 write_to_pointer
, &p
)) {
1178 if ((mode
->enable_mask
& (1 << CDU
)) &&
1179 write_utf8(NULL
, input_chr
, NULL
, write_to_pointer
, &p
)) {
1187 docs_char(state
, emit
, emitctx
, cs
, data
, p
- data
);
1196 * Full ISO 2022 output with all options on. Not entirely sure what
1197 * if anything this is useful for, but here it is anyway. All
1198 * output character sets and DOCS variants are permitted; all
1199 * containers start out with ASCII in them.
1201 static const struct iso2022_mode iso2022_all
= {
1202 (1<<CCS
) | (1<<COS
) | (1<<CPU
) | (1<<CDC
) | (1<<CDU
),
1203 S4
, 0, 'B', S4
, 0, 'B',
1206 const charset_spec charset_CS_ISO2022
= {
1207 CS_ISO2022
, read_iso2022
, write_iso2022
, &iso2022_all
1211 * X11 compound text. A subset of output charsets is permitted, and
1212 * G1/GR starts off in ISO8859-1.
1214 static const struct iso2022_mode iso2022_ctext
= {
1215 (1<<CCS
) | (1<<CDC
),
1216 S4
, 0, 'B', S6
, 0, 'A',
1219 const charset_spec charset_CS_CTEXT
= {
1220 CS_CTEXT
, read_iso2022
, write_iso2022
, &iso2022_ctext
1231 void iso2022_emit(void *ctx
, long output
)
1233 wchar_t **p
= (wchar_t **)ctx
;
1237 void iso2022_read_test(int line
, char *input
, int inlen
, ...)
1240 wchar_t *p
, str
[512];
1242 charset_state state
;
1245 state
.s0
= state
.s1
= 0;
1248 for (i
= 0; i
< inlen
; i
++)
1249 read_iso2022(NULL
, input
[i
] & 0xFF, &state
, iso2022_emit
, &p
);
1251 va_start(ap
, inlen
);
1253 for (i
= 0; i
< p
- str
; i
++) {
1254 l
= va_arg(ap
, long int);
1256 printf("%d: correct string shorter than output\n", line
);
1261 printf("%d: char %d came out as %08x, should be %08lx\n",
1262 line
, i
, str
[i
], l
);
1267 l
= va_arg(ap
, long int);
1269 printf("%d: correct string longer than output\n", line
);
1276 /* Macro to concoct the first three parameters of iso2022_read_test. */
1277 #define TESTSTR(x) __LINE__, x, lenof(x)
1281 printf("read tests beginning\n");
1282 /* Simple test (Emacs sample text for Japanese, in ISO-2022-JP) */
1283 iso2022_read_test(TESTSTR("Japanese (\x1b$BF|K\\8l\x1b(B)\t"
1284 "\x1b$B$3$s$K$A$O\x1b(B, "
1285 "\x1b$B%3%s%K%A%O\x1b(B\n"),
1286 'J','a','p','a','n','e','s','e',' ','(',
1287 0x65E5, 0x672C, 0x8A9E, ')', '\t',
1288 0x3053, 0x3093, 0x306b, 0x3061, 0x306f, ',', ' ',
1289 0x30b3, 0x30f3, 0x30cb, 0x30c1, 0x30cf, '\n', 0, -1);
1290 /* Same thing in EUC-JP (with designations, and half-width katakana) */
1291 iso2022_read_test(TESTSTR("\x1b$)B\x1b*I\x1b$+D"
1292 "Japanese (\xc6\xfc\xcb\xdc\xb8\xec)\t"
1293 "\xa4\xb3\xa4\xf3\xa4\xcb\xa4\xc1\xa4\xcf, "
1294 "\x8e\xba\x8e\xdd\x8e\xc6\x8e\xc1\x8e\xca\n"),
1295 'J','a','p','a','n','e','s','e',' ','(',
1296 0x65E5, 0x672C, 0x8A9E, ')', '\t',
1297 0x3053, 0x3093, 0x306b, 0x3061, 0x306f, ',', ' ',
1298 0xff7a, 0xff9d, 0xff86, 0xff81, 0xff8a, '\n', 0, -1);
1299 /* Multibyte single-shift */
1300 iso2022_read_test(TESTSTR("\x1b$)B\x1b*I\x1b$+D\x8f\"/!"),
1301 0x02D8, '!', 0, -1);
1302 /* Non-existent SBCS */
1303 iso2022_read_test(TESTSTR("\x1b(!Zfnord\n"),
1304 ERROR
, ERROR
, ERROR
, ERROR
, ERROR
, '\n', 0, -1);
1305 /* Pass-through of ordinary escape sequences, including a long one */
1306 iso2022_read_test(TESTSTR("\x1b""b\x1b#5\x1b#!!!5"),
1307 0x1B, 'b', 0x1B, '#', '5',
1308 0x1B, '#', '!', '!', '!', '5', 0, -1);
1309 /* Non-existent DBCS (also 5-byte escape sequence) */
1310 iso2022_read_test(TESTSTR("\x1b$(!Bfnord!"),
1311 ERROR
, ERROR
, ERROR
, 0, -1);
1312 /* Incomplete DB characters */
1313 iso2022_read_test(TESTSTR("\x1b$B(,(\x1b(BHi\x1b$B(,(\n"),
1314 0x2501, ERROR
, 'H', 'i', 0x2501, ERROR
, '\n', 0, -1);
1315 iso2022_read_test(TESTSTR("\x1b$)B\x1b*I\x1b$+D\xa4""B"),
1317 iso2022_read_test(TESTSTR("\x1b$)B\x1b*I\x1b$+D\x0e\x1b|$\xa2\xaf"),
1318 ERROR
, 0x02D8, 0, -1);
1319 /* Incomplete escape sequence */
1320 iso2022_read_test(TESTSTR("\x1b\n"), ERROR
, '\n', 0, -1);
1321 iso2022_read_test(TESTSTR("\x1b-A\x1b~\x1b\xa1"), ERROR
, 0xa1, 0, -1);
1322 /* Incomplete single-shift */
1323 iso2022_read_test(TESTSTR("\x8e\n"), ERROR
, '\n', 0, -1);
1324 iso2022_read_test(TESTSTR("\x1b$*B\x8e(\n"), ERROR
, '\n', 0, -1);
1325 /* Corner cases (02/00 and 07/15) */
1326 iso2022_read_test(TESTSTR("\x1b(B\x20\x7f"), 0x20, 0x7f, 0, -1);
1327 iso2022_read_test(TESTSTR("\x1b(I\x20\x7f"), 0x20, 0x7f, 0, -1);
1328 iso2022_read_test(TESTSTR("\x1b$B\x20\x7f"), 0x20, 0x7f, 0, -1);
1329 iso2022_read_test(TESTSTR("\x1b-A\x0e\x20\x7f"), 0xa0, 0xff, 0, -1);
1330 iso2022_read_test(TESTSTR("\x1b$-~\x0e\x20\x7f"), ERROR
, 0, -1);
1331 iso2022_read_test(TESTSTR("\x1b)B\xa0\xff"), ERROR
, ERROR
, 0, -1);
1332 iso2022_read_test(TESTSTR("\x1b)I\xa0\xff"), ERROR
, ERROR
, 0, -1);
1333 iso2022_read_test(TESTSTR("\x1b$)B\xa0\xff"), ERROR
, ERROR
, 0, -1);
1334 iso2022_read_test(TESTSTR("\x1b-A\x1b~\xa0\xff"), 0xa0, 0xff, 0, -1);
1335 iso2022_read_test(TESTSTR("\x1b$-~\x1b~\xa0\xff"), ERROR
, 0, -1);
1336 /* Designate control sets */
1337 iso2022_read_test(TESTSTR("\x1b!@"), 0x1b, '!', '@', 0, -1);
1338 /* Designate other coding system (UTF-8) */
1339 iso2022_read_test(TESTSTR("\x1b%G"
1340 "\xCE\xBA\xE1\xBD\xB9\xCF\x83\xCE\xBC\xCE\xB5"),
1341 0x03BA, 0x1F79, 0x03C3, 0x03BC, 0x03B5, 0, -1);
1342 iso2022_read_test(TESTSTR("\x1b-A\x1b%G\xCE\xBA\x1b%@\xa0"),
1343 0x03BA, 0xA0, 0, -1);
1344 iso2022_read_test(TESTSTR("\x1b%G\xCE\x1b%@"), ERROR
, 0, -1);
1345 iso2022_read_test(TESTSTR("\x1b%G\xCE\xBA\x1b%\x1b%@"),
1346 0x03BA, 0x1B, '%', 0, -1);
1347 /* DOCS (COMPOUND_TEXT extended segment) */
1348 iso2022_read_test(TESTSTR("\x1b%/1\x80\x80"), 0, -1);
1349 iso2022_read_test(TESTSTR("\x1b%/1\x80\x8fiso-8859-15\2xyz\x1b(B"),
1350 ERROR
, ERROR
, ERROR
, 0, -1);
1351 iso2022_read_test(TESTSTR("\x1b%/1\x80\x8eiso8859-15\2xyz\x1b(B"),
1352 'x', 'y', 'z', 0, -1);
1353 iso2022_read_test(TESTSTR("\x1b-A\x1b%/2\x80\x89"
1354 "big5-0\2\xa1\x40\xa1\x40"),
1355 0x3000, 0xa1, 0x40, 0, -1);
1356 /* Emacs Big5-in-ISO-2022 mapping */
1357 iso2022_read_test(TESTSTR("\x1b$(0&x86\x1b(B \x1b$(0DeBv"),
1358 0x5143, 0x6c23, ' ', ' ', 0x958b, 0x767c, 0, -1);
1359 /* Test from RFC 1922 (ISO-2022-CN) */
1360 iso2022_read_test(TESTSTR("\x1b$)A\x0e=;;;\x1b$)GG(_P\x0f"),
1361 0x4EA4, 0x6362, 0x4EA4, 0x63db, 0, -1);
1363 printf("read tests completed\n");
1364 printf("total: %d errors\n", total_errs
);
1365 return (total_errs
!= 0);
1368 #endif /* TESTMODE */
1370 #else /* ENUM_CHARSETS */
1372 ENUM_CHARSET(CS_ISO2022
)