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 emacs_big5_1_to_unicode(int, int);
39 static long int emacs_big5_2_to_unicode(int, int);
40 static int unicode_to_emacs_big5(long int, int *, int *, int *);
41 static long int cns11643_1_to_unicode(int, int);
42 static long int cns11643_2_to_unicode(int, int);
43 static long int cns11643_3_to_unicode(int, int);
44 static long int cns11643_4_to_unicode(int, int);
45 static long int cns11643_5_to_unicode(int, int);
46 static long int cns11643_6_to_unicode(int, int);
47 static long int cns11643_7_to_unicode(int, int);
48 static long int null_dbcs_to_unicode(int, int);
49 static int unicode_to_null_dbcs(long int, int *, int *);
51 typedef int (*to_dbcs_t
)(long int, int *, int *);
52 typedef int (*to_dbcs_planar_t
)(long int, int *, int *, int *);
54 /* Cast between to_dbcs_planar_t and to_dbcs_t, type-checking first */
55 #define DEPLANARISE(x) ( (x) == (to_dbcs_planar_t)NULL, (to_dbcs_t)(x) )
56 #define REPLANARISE(x) ( (x) == (to_dbcs_t)NULL, (to_dbcs_planar_t)(x) )
59 * Values used in the `enable' field. Each of these identifies a
60 * class of character sets; we then have a bitmask indicating which
61 * classes are allowable in a given mode.
63 * These values are currently only checked on output: for input,
64 * any ISO 2022 we can comprehend at all is considered acceptable.
66 #define CCS 1 /* CTEXT standard */
67 #define COS 2 /* other standard */
68 #define CPU 3 /* private use */
69 #define CDC 4 /* DOCS for CTEXT */
70 #define CDU 5 /* DOCS for UTF-8 */
71 #define CNU 31 /* never used */
75 char ltype
, li
, lf
, rtype
, ri
, rf
;
78 const struct iso2022_subcharset
{
79 char type
, i
, f
, enable
;
81 const sbcs_data
*sbcs_base
;
82 long int (*from_dbcs
)(int, int);
85 * If to_dbcs_plane < 0, then to_dbcs is used as expected.
86 * However, if to_dbcs_plane >= 0, then to_dbcs is expected to
87 * be cast to a to_dbcs_planar_t before use, and the returned
88 * plane value (the first int *) must equal to_dbcs_plane.
90 * I'd have preferred to do this by means of a union, but you
91 * can't initialise a selected field of a union at compile
92 * time. Function pointer casts are guaranteed to work sensibly
93 * in ISO C (that is, it's undefined what happens if you call a
94 * function via the wrong type of pointer, but if you cast it
95 * back to the right type before calling it then it must work),
96 * so this is safe if ugly.
99 int to_dbcs_plane
; /* use to_dbcs_planar iff >= 0 */
100 } iso2022_subcharsets
[] = {
102 * We list these subcharsets in preference order for output.
103 * Since the best-defined use of ISO 2022 output is compound
104 * text, we'll use a preference order which matches that. So we
105 * begin with the charsets defined in the compound text spec.
107 { S4
, 0, 'B', CCS
, 0x00, &sbcsdata_CS_ASCII
},
108 { S6
, 0, 'A', CCS
, 0x80, &sbcsdata_CS_ISO8859_1
},
109 { S6
, 0, 'B', CCS
, 0x80, &sbcsdata_CS_ISO8859_2
},
110 { S6
, 0, 'C', CCS
, 0x80, &sbcsdata_CS_ISO8859_3
},
111 { S6
, 0, 'D', CCS
, 0x80, &sbcsdata_CS_ISO8859_4
},
112 { S6
, 0, 'F', CCS
, 0x80, &sbcsdata_CS_ISO8859_7
},
113 { S6
, 0, 'G', CCS
, 0x80, &sbcsdata_CS_ISO8859_6
},
114 { S6
, 0, 'H', CCS
, 0x80, &sbcsdata_CS_ISO8859_8
},
115 { S6
, 0, 'L', CCS
, 0x80, &sbcsdata_CS_ISO8859_5
},
116 { S6
, 0, 'M', CCS
, 0x80, &sbcsdata_CS_ISO8859_9
},
117 { S4
, 0, 'I', CCS
, 0x80, &sbcsdata_CS_JISX0201
},
118 { S4
, 0, 'J', CCS
, 0x00, &sbcsdata_CS_JISX0201
},
119 { M4
, 0, 'A', CCS
, -0x21, 0, &gb2312_to_unicode
, &unicode_to_gb2312
, -1 },
120 { M4
, 0, 'B', CCS
, -0x21, 0, &jisx0208_to_unicode
, &unicode_to_jisx0208
, -1 },
121 { M4
, 0, 'C', CCS
, -0x21, 0, &ksx1001_to_unicode
, &unicode_to_ksx1001
, -1 },
122 { M4
, 0, 'D', CCS
, -0x21, 0, &jisx0212_to_unicode
, &unicode_to_jisx0212
, -1 },
125 * Next, other reasonably standard things: the rest of the ISO
126 * 8859 sets, UK-ASCII, and CNS 11643.
128 { S6
, 0, 'T', COS
, 0x80, &sbcsdata_CS_ISO8859_11
},
129 { S6
, 0, 'V', COS
, 0x80, &sbcsdata_CS_ISO8859_10
},
130 { S6
, 0, 'Y', COS
, 0x80, &sbcsdata_CS_ISO8859_13
},
131 { S6
, 0, '_', COS
, 0x80, &sbcsdata_CS_ISO8859_14
},
132 { S6
, 0, 'b', COS
, 0x80, &sbcsdata_CS_ISO8859_15
},
133 { S6
, 0, 'f', COS
, 0x80, &sbcsdata_CS_ISO8859_16
},
134 { S4
, 0, 'A', COS
, 0x00, &sbcsdata_CS_BS4730
},
135 { M4
, 0, 'G', COS
, -0x21, 0, &cns11643_1_to_unicode
, DEPLANARISE(&unicode_to_cns11643
), 0 },
136 { M4
, 0, 'H', COS
, -0x21, 0, &cns11643_2_to_unicode
, DEPLANARISE(&unicode_to_cns11643
), 1 },
137 { M4
, 0, 'I', COS
, -0x21, 0, &cns11643_3_to_unicode
, DEPLANARISE(&unicode_to_cns11643
), 2 },
138 { M4
, 0, 'J', COS
, -0x21, 0, &cns11643_4_to_unicode
, DEPLANARISE(&unicode_to_cns11643
), 3 },
139 { M4
, 0, 'K', COS
, -0x21, 0, &cns11643_5_to_unicode
, DEPLANARISE(&unicode_to_cns11643
), 4 },
140 { M4
, 0, 'L', COS
, -0x21, 0, &cns11643_6_to_unicode
, DEPLANARISE(&unicode_to_cns11643
), 5 },
141 { M4
, 0, 'M', COS
, -0x21, 0, &cns11643_7_to_unicode
, DEPLANARISE(&unicode_to_cns11643
), 6 },
144 * Private-use designations: DEC private sets and Emacs's Big5
147 { S4
, 0, '0', CPU
, 0x00, &sbcsdata_CS_DEC_GRAPHICS
},
148 { S4
, 0, '<', CPU
, 0x80, &sbcsdata_CS_DEC_MCS
},
149 { M4
, 0, '0', CPU
, -0x21, 0, &emacs_big5_1_to_unicode
, DEPLANARISE(&unicode_to_emacs_big5
), 1 },
150 { M4
, 0, '1', CPU
, -0x21, 0, &emacs_big5_2_to_unicode
, DEPLANARISE(&unicode_to_emacs_big5
), 2 },
153 * Ben left this conditioned out without explanation,
154 * presumably on the grounds that we don't have a translation
158 { M4
, 0, '@', CNU
}, /* JIS C 6226-1978 */
162 * Finally, fallback entries for null character sets.
165 { S6
, 0, '~', CNU
}, /* empty 96-set */
166 { M4
, 0, '~', CNU
, 0, 0, &null_dbcs_to_unicode
, &unicode_to_null_dbcs
, -1 }, /* empty 94^n-set */
167 { M6
, 0, '~', CNU
, 0, 0, &null_dbcs_to_unicode
, &unicode_to_null_dbcs
, -1 }, /* empty 96^n-set */
170 static long int null_dbcs_to_unicode(int r
, int c
)
174 static int unicode_to_null_dbcs(long int unicode
, int *r
, int *c
)
176 return 0; /* failed to convert anything */
180 * Emacs encodes Big5 in COMPOUND_TEXT as two 94x94 character sets.
181 * We treat Big5 as a 94x191 character set with a bunch of undefined
182 * columns in the middle, so we have to mess around a bit to make
186 static long int emacs_big5_1_to_unicode(int r
, int c
)
192 if (c
>= 64) c
+= 34; /* Skip over the gap */
193 return big5_to_unicode(r
, c
);
196 static long int emacs_big5_2_to_unicode(int r
, int c
)
202 if (c
>= 64) c
+= 34; /* Skip over the gap */
203 return big5_to_unicode(r
, c
);
206 static int unicode_to_emacs_big5(long int unicode
, int *p
, int *r
, int *c
)
209 if (!unicode_to_big5(unicode
, &rr
, &cc
))
227 /* Wrappers for cns11643_to_unicode() */
228 static long int cns11643_1_to_unicode(int r
, int c
)
230 return cns11643_to_unicode(0, r
, c
);
232 static long int cns11643_2_to_unicode(int r
, int c
)
234 return cns11643_to_unicode(1, r
, c
);
236 static long int cns11643_3_to_unicode(int r
, int c
)
238 return cns11643_to_unicode(2, r
, c
);
240 static long int cns11643_4_to_unicode(int r
, int c
)
242 return cns11643_to_unicode(3, r
, c
);
244 static long int cns11643_5_to_unicode(int r
, int c
)
246 return cns11643_to_unicode(4, r
, c
);
248 static long int cns11643_6_to_unicode(int r
, int c
)
250 return cns11643_to_unicode(5, r
, c
);
252 static long int cns11643_7_to_unicode(int r
, int c
)
254 return cns11643_to_unicode(6, r
, c
);
257 /* States, or "what we're currently accumulating". */
259 IDLE
, /* None of the below */
260 SS2CHAR
, /* Accumulating a character after SS2 */
261 SS3CHAR
, /* Accumulating a character after SS3 */
262 ESCSEQ
, /* Accumulating an escape sequence */
263 ESCDROP
, /* Discarding an escape sequence */
264 ESCPASS
, /* Passing through an escape sequence */
265 DOCSUTF8
, /* DOCSed into UTF-8 */
266 DOCSCTEXT
/* DOCSed into a COMPOUND_TEXT extended segment */
271 static void dump_state(charset_state
*s
)
273 unsigned s0
= s
->s0
, s1
= s
->s1
;
274 char const * const modes
[] = { "IDLE", "SS2CHAR", "SS3CHAR",
275 "ESCSEQ", "ESCDROP", "ESCPASS",
278 fprintf(stderr
, "s0: %s", modes
[s0
>> 29]);
279 fprintf(stderr
, " %02x %02x %02x ", (s0
>> 16) & 0xff, (s0
>> 8) & 0xff,
281 fprintf(stderr
, "s1: LS%d LS%dR", (s1
>> 30) & 3, (s1
>> 28) & 3);
282 fprintf(stderr
, " %d %d %d %d\n", s1
& 0x7f, (s1
>> 7) & 0x7f,
283 (s1
>> 14) & 0x7f, (s1
>> 21) & 0x7f);
287 static void designate(charset_state
*state
, int container
,
288 int type
, int ibyte
, int fbyte
)
292 assert(container
>= 0 && container
<= 3);
293 assert(type
== S4
|| type
== S6
|| type
== M4
|| type
== M6
);
295 for (i
= 0; i
< lenof(iso2022_subcharsets
); i
++) {
296 if (iso2022_subcharsets
[i
].type
== type
&&
297 iso2022_subcharsets
[i
].i
== ibyte
&&
298 iso2022_subcharsets
[i
].f
== fbyte
) {
299 state
->s1
&= ~(0x7fL
<< (container
* 7));
300 state
->s1
|= (i
<< (container
* 7));
305 * If we don't find the charset, invoke the empty one, so we
306 * output ERROR rather than garbage.
308 designate(state
, container
, type
, 0, '~');
311 static void do_utf8(long int input_chr
,
312 charset_state
*state
,
313 void (*emit
)(void *ctx
, long int output
),
316 charset_state ustate
;
319 ustate
.s0
= state
->s0
& 0x03ffffffL
;
320 read_utf8(NULL
, input_chr
, &ustate
, emit
, emitctx
);
321 state
->s0
= (state
->s0
& ~0x03ffffffL
) | (ustate
.s0
& 0x03ffffffL
);
324 static void docs_utf8(long int input_chr
,
325 charset_state
*state
,
326 void (*emit
)(void *ctx
, long int output
),
332 * Bits [25:0] of s0 are reserved for read_utf8().
333 * Bits [27:26] are a tiny state machine to recognise ESC % @.
335 retstate
= (state
->s0
& 0x0c000000L
) >> 26;
336 if (retstate
== 1 && input_chr
== '%')
338 else if (retstate
== 2 && input_chr
== '@') {
339 /* If we've got a partial UTF-8 sequence, complain. */
340 if (state
->s0
& 0x03ffffffL
)
341 emit(emitctx
, ERROR
);
345 if (retstate
>= 1) do_utf8(ESC
, state
, emit
, emitctx
);
346 if (retstate
>= 2) do_utf8('%', state
, emit
, emitctx
);
348 if (input_chr
== ESC
)
351 do_utf8(input_chr
, state
, emit
, emitctx
);
354 state
->s0
= (state
->s0
& ~0x0c000000L
) | (retstate
<< 26);
357 struct ctext_encoding
{
359 char octets_per_char
, enable
;
360 charset_spec
const *subcs
;
364 * In theory, this list is in <ftp://ftp.x.org/pub/DOCS/registry>,
365 * but XLib appears to have its own ideas, and encodes these three
369 extern charset_spec
const charset_CS_ISO8859_14
;
370 extern charset_spec
const charset_CS_ISO8859_15
;
371 extern charset_spec
const charset_CS_BIG5
;
373 static struct ctext_encoding
const ctext_encodings
[] = {
374 { "big5-0\2", 0 /* variable */, CDC
, &charset_CS_BIG5
},
375 { "iso8859-14\2", 1, CDC
, &charset_CS_ISO8859_14
},
376 { "iso8859-15\2", 1, CDC
, &charset_CS_ISO8859_15
}
379 static void docs_ctext(long int input_chr
,
380 charset_state
*state
,
381 void (*emit
)(void *ctx
, long int output
),
385 * s0[27:26] = first entry in ctext_encodings that matches
386 * s0[25:22] = number of characters successfully matched, 0xf if all
387 * s0[21:8] count the number of octets left in the segment
388 * s0[7:0] are for sub-charset use
390 int n
= (state
->s0
>> 22) & 0xf, i
= (state
->s0
>> 26) & 3, oi
= i
, j
;
391 int length
= (state
->s0
>> 8) & 0x3fff;
394 * Note that we do not bother checking the octets-per-character
395 * byte against the selected charset when reading. It's
396 * extremely unlikely that this code will ever have to deal
397 * with two charset identifiers with the same name and
398 * different octets-per-character values! If it ever happens,
399 * we'll have to edit this file anyway so we can modify the
404 /* Haven't read length yet */
405 if ((state
->s0
& 0xff) == 0)
406 /* ... or even the first byte */
407 state
->s0
|= input_chr
;
409 length
= (state
->s0
& 0x7f) * 0x80 + (input_chr
& 0x7f);
413 state
->s0
= (state
->s0
& 0xf0000000) | (length
<< 8);
420 /* Skipping unknown encoding. Look out for STX. */
422 state
->s0
= (state
->s0
& 0xf0000000) | (i
<< 26) | (0xf << 22);
423 } else if (n
!= 0xf) {
424 while (j
< lenof(ctext_encodings
) &&
425 !memcmp(ctext_encodings
[j
].name
,
426 ctext_encodings
[oi
].name
, n
)) {
427 if (ctext_encodings
[j
].name
[n
] < input_chr
)
432 if (i
>= lenof(ctext_encodings
) ||
433 memcmp(ctext_encodings
[i
].name
,
434 ctext_encodings
[oi
].name
, n
) ||
435 ctext_encodings
[i
].name
[n
] != input_chr
) {
436 /* Doom! We haven't heard of this encoding */
437 i
= lenof(ctext_encodings
);
441 * Otherwise, we have found an additional character in our
442 * encoding name. See if we have reached the _end_ of our
446 if (!ctext_encodings
[i
].name
[n
])
450 * Failing _that_, we simply update our encoding-name-
453 assert(i
< 4 && n
< 16);
454 state
->s0
= (state
->s0
& 0xf0000000) | (i
<< 26) | (n
<< 22);
456 if (i
>= lenof(ctext_encodings
))
457 emit(emitctx
, ERROR
);
459 charset_state substate
;
460 charset_spec
const *subcs
= ctext_encodings
[i
].subcs
;
462 substate
.s0
= state
->s0
& 0xff;
463 subcs
->read(subcs
, input_chr
, &substate
, emit
, emitctx
);
464 state
->s0
= (state
->s0
& ~0xff) | (substate
.s0
& 0xff);
470 state
->s0
= (state
->s0
&~0x003fff00) | (length
<< 8);
473 static void read_iso2022(charset_spec
const *charset
, long int input_chr
,
474 charset_state
*state
,
475 void (*emit
)(void *ctx
, long int output
),
478 struct iso2022_mode
const *mode
= (struct iso2022_mode
*)charset
->data
;
480 /* dump_state(state); */
482 * We have to make fairly efficient use of the 64 bits of state
483 * available to us. Long-term state goes in s1, and consists of
484 * the identities of the character sets designated as G0/G1/G2/G3
485 * and the locking-shift states for GL and GR. Short-term state
486 * goes in s0: The bottom half of s0 accumulates characters for an
487 * escape sequence or a multi-byte character, while the top three
488 * bits indicate what they're being accumulated for. After DOCS,
489 * the bottom 29 bits of state are available for the DOCS function
490 * to use -- the UTF-8 one uses the bottom 26 for UTF-8 decoding
491 * and the top two to recognised ESC % @.
493 * s0[31:29] = state enum
494 * s0[24:0] = accumulated bytes
495 * s1[31:30] = GL locking-shift state
496 * s1[29:28] = GR locking-shift state
497 * s1[27:21] = G3 charset
498 * s1[20:14] = G2 charset
499 * s1[13:7] = G1 charset
500 * s1[6:0] = G0 charset
505 #define LOCKING_SHIFT(n,side) \
506 (state->s1 = (state->s1 & ~(3L<<(side))) | ((n ## L)<<(side)))
507 #define MODE ((state->s0 & 0xe0000000L) >> 29)
508 #define ENTER_MODE(m) (state->s0 = (state->s0 & ~0xe0000000L) | ((m)<<29))
509 #define SINGLE_SHIFT(n) ENTER_MODE(SS2CHAR - 2 + (n))
510 #define ASSERT_IDLE do { \
511 if (state->s0 != 0) emit(emitctx, ERROR); \
515 if (state
->s1
== 0) {
517 * Since there's no LS0R, this means we must just have started.
518 * Set up a sane initial state (LS0, LS1R, ASCII in G0/G1/G2/G3).
520 LOCKING_SHIFT(0, LEFT
);
521 LOCKING_SHIFT(1, RIGHT
);
522 designate(state
, 0, mode
->ltype
, mode
->li
, mode
->lf
);
523 designate(state
, 1, mode
->rtype
, mode
->ri
, mode
->rf
);
524 designate(state
, 2, S4
, 0, 'B');
525 designate(state
, 3, S4
, 0, 'B');
528 if (MODE
== DOCSUTF8
) {
529 docs_utf8(input_chr
, state
, emit
, emitctx
);
532 if (MODE
== DOCSCTEXT
) {
533 docs_ctext(input_chr
, state
, emit
, emitctx
);
537 if ((input_chr
& 0x60) == 0x00) {
538 /* C0 or C1 control */
545 LOCKING_SHIFT(0, LEFT
);
548 LOCKING_SHIFT(1, LEFT
);
557 emit(emitctx
, input_chr
);
560 } else if ((input_chr
& 0x80) || MODE
< ESCSEQ
) {
562 struct iso2022_subcharset
const *subcs
;
567 * Force idle state if we're in mid escape sequence, or in a
568 * multi-byte character with a different top bit.
570 if (MODE
>= ESCSEQ
||
571 ((state
->s0
& 0x00ff0000L
) != 0 &&
572 (((state
->s0
>> 16) ^ input_chr
) & 0x80)))
574 if (MODE
== SS2CHAR
|| MODE
== SS3CHAR
) /* Single-shift */
575 container
= MODE
- SS2CHAR
+ 2;
576 else if (input_chr
>= 0x80) /* GR */
577 container
= (state
->s1
>> 28) & 3;
579 container
= state
->s1
>> 30;
582 input_7bit
= input_chr
& ~0x80;
583 subcs
= &iso2022_subcharsets
[(state
->s1
>> (container
* 7)) & 0x7f];
584 if ((subcs
->type
== S4
|| subcs
->type
== M4
) &&
585 (input_7bit
== 0x20 || input_7bit
== 0x7f)) {
586 /* characters not in 94-char set */
587 if (is_gl
) emit(emitctx
, input_7bit
);
588 else emit(emitctx
, ERROR
);
589 } else if (subcs
->type
== M4
|| subcs
->type
== M6
) {
590 if ((state
->s0
& 0x00ff0000L
) == 0) {
591 state
->s0
|= input_chr
<< 16;
595 subcs
->from_dbcs(((state
->s0
>> 16) & 0x7f) +
597 input_7bit
+ subcs
->offset
));
600 if ((state
->s0
& 0x00ff0000L
) != 0)
601 emit(emitctx
, ERROR
);
602 emit(emitctx
, subcs
->sbcs_base ?
603 sbcs_to_unicode(subcs
->sbcs_base
, input_7bit
+ subcs
->offset
):
609 if (MODE
== ESCPASS
) {
610 emit(emitctx
, input_chr
);
611 if ((input_chr
& 0xf0) != 0x20)
617 * Intermediate bytes shall be any of the 16 positions of
618 * column 02 of the code table; they are denoted by the symbol
621 if ((input_chr
& 0xf0) == 0x20) {
622 if (((state
->s0
>> 16) & 0xff) == 0)
623 state
->s0
|= input_chr
<< 16;
624 else if (((state
->s0
>> 8) & 0xff) == 0)
625 state
->s0
|= input_chr
<< 8;
627 /* Long escape sequence. Switch to ESCPASS or ESCDROP. */
628 i1
= (state
->s0
>> 16) & 0xff;
629 i2
= (state
->s0
>> 8) & 0xff;
631 case '(': case ')': case '*': case '+':
632 case '-': case '.': case '/':
640 emit(emitctx
, input_chr
);
650 * Final bytes shall be any of the 79 positions of columns 03
651 * to 07 of the code table excluding position 07/15; they are
652 * denoted by the symbol F.
654 i1
= (state
->s0
>> 16) & 0xff;
655 i2
= (state
->s0
>> 8) & 0xff;
657 input_chr
= 0; /* Make sure it won't match. */
660 case 0: /* No intermediate bytes */
669 LOCKING_SHIFT(2, LEFT
);
672 LOCKING_SHIFT(3, LEFT
);
675 LOCKING_SHIFT(3, RIGHT
);
678 LOCKING_SHIFT(2, RIGHT
);
681 LOCKING_SHIFT(1, RIGHT
);
684 /* Unsupported escape sequence. Spit it back out. */
686 emit(emitctx
, input_chr
);
691 * Various coding structure facilities specify that designating
692 * a code element also invokes it. As far as I can see, invoking
693 * it now will have the same practical effect, since those
694 * facilities also ban the use of locking shifts.
697 case 'A': /* G0 element used and invoked into GL */
698 LOCKING_SHIFT(0, LEFT
);
700 case 'C': /* G0 in GL, G1 in GR */
701 case 'D': /* Ditto, at least for 8-bit codes */
702 case 'L': /* ISO 4873 (ECMA-43) level 1 */
703 case 'M': /* ISO 4873 (ECMA-43) level 2 */
704 LOCKING_SHIFT(0, LEFT
);
705 LOCKING_SHIFT(1, RIGHT
);
711 * IRR (Identify Revised Registration) is ignored here,
712 * since any revised registration must be
713 * upward-compatible with the old one, so either we'll
714 * support the new one or we'll emit ERROR when we run
715 * into a new character. In either case, there's nothing
719 case '(': /* GZD4 */ case ')': /* G1D4 */
720 case '*': /* G2D4 */ case '+': /* G3D4 */
721 designate(state
, i1
- '(', S4
, i2
, input_chr
);
723 case '-': /* G1D6 */ case '.': /* G2D6 */ case '/': /* G3D6 */
724 designate(state
, i1
- ',', S6
, i2
, input_chr
);
726 case '$': /* G?DM? */
728 case 0: /* Obsolete version of GZDM4 */
730 case '(': /* GZDM4 */ case ')': /* G1DM4 */
731 case '*': /* G2DM4 */ case '+': /* G3DM4 */
732 designate(state
, i2
- '(', M4
, 0, input_chr
);
734 case '-': /* G1DM6 */
735 case '.': /* G2DM6 */ case '/': /* G3DM6 */
736 designate(state
, i2
- ',', M6
, 0, input_chr
);
739 emit(emitctx
, ERROR
);
743 /* XXX What's a reasonable way to handle an unrecognised DOCS? */
748 ENTER_MODE(DOCSUTF8
);
755 ENTER_MODE(DOCSCTEXT
);
762 /* Unsupported nF escape sequence. Re-emit it. */
765 if (i2
) emit(emitctx
, i2
);
766 emit(emitctx
, input_chr
);
772 static void oselect(charset_state
*state
, int i
, int right
,
773 void (*emit
)(void *ctx
, long int output
),
776 int shift
= (right ?
31-7 : 31-7-7);
777 struct iso2022_subcharset
const *subcs
= &iso2022_subcharsets
[i
];
779 if (((state
->s1
>> shift
) & 0x7F) != i
) {
780 state
->s1
&= ~(0x7FL
<< shift
);
781 state
->s1
|= (i
<< shift
);
785 if (subcs
->type
== M4
|| subcs
->type
== M6
)
787 if (subcs
->type
== S6
|| subcs
->type
== M6
) {
796 emit(emitctx
, subcs
->i
);
797 emit(emitctx
, subcs
->f
);
802 static void docs_char(charset_state
*state
,
803 void (*emit
)(void *ctx
, long int output
),
804 void *emitctx
, int cset
, char *data
, int datalen
)
806 int curr_cset
, currlen
, i
;
809 * cset is the index into ctext_encodings[]. It can also be -1
810 * to mean DOCS UTF-8, or -2 to mean no DOCS (ordinary 2022).
811 * In the latter case, `chr' is ignored.
815 * First, terminate a DOCS segment if necessary. We always have
816 * to terminate a DOCS segment if one is active and we're about
817 * to switch to a different one; we might also have to
818 * terminate a length-encoded DOCS segment if we've run out of
819 * storage space to accumulate characters in it.
821 curr_cset
= ((state
->s1
>> 14) & 7) - 2;
822 currlen
= ((state
->s1
>> 11) & 7);
823 if ((curr_cset
!= -2 && curr_cset
!= cset
) ||
824 (curr_cset
>= 0 && currlen
+ datalen
> 5)) {
825 if (curr_cset
== -1) {
827 * Terminating DOCS UTF-8 is easy.
836 * To terminate a length-encoded DOCS segment we must
837 * actually output the whole thing.
842 emit(emitctx
, '0' + ctext_encodings
[curr_cset
].octets_per_char
);
843 len
= currlen
+ datalen
+
844 strlen(ctext_encodings
[curr_cset
].name
);
845 assert(len
< (1 << 14));
846 emit(emitctx
, 0x80 | ((len
>> 7) & 0x7F));
847 emit(emitctx
, 0x80 | ((len
) & 0x7F));
848 /* The name stored in ctext_encodings[] includes the trailing \2 */
849 for (i
= 0; ctext_encodings
[curr_cset
].name
[i
]; i
++)
850 emit(emitctx
, ctext_encodings
[curr_cset
].name
[i
]);
851 for (i
= 0; i
< currlen
; i
++)
853 (i
== 0 ? state
->s1
: state
->s0
>> (8*(4-i
))) & 0xFF);
854 for (i
= 0; i
< datalen
; i
++)
855 emit(emitctx
, data
[i
]);
858 * We've now dealt with the input data, so clear it so
859 * we don't try to do so again below.
867 * Now, start a DOCS segment if necessary.
869 if (curr_cset
!= cset
) {
880 * Starting a length-encoded DOCS segment is simply a
881 * matter of setting our stored length counter to zero.
884 state
->s1
&= ~(7 << 11);
889 state
->s1
&= ~(7 << 14);
890 assert((cset
+2) >= 0 && (cset
+2) < 8);
891 state
->s1
|= ((cset
+2) << 14);
894 * Now we're in the right DOCS state. Actually deal with the
895 * input data, if we haven't already done so above.
901 * In DOCS UTF-8, we output data as soon as we get it.
903 for (i
= 0; i
< datalen
; i
++)
904 emit(emitctx
, data
[i
]);
907 * In length-encoded DOCS, we just store our data and
908 * bide our time. It'll all be output when we fill up
909 * or switch to another character set.
911 assert(currlen
+ datalen
<= 5); /* overflow handled already */
912 for (i
= 0; i
< datalen
; i
++) {
913 if (currlen
+ i
== 0)
914 state
->s1
|= data
[i
] & 0xFF;
916 state
->s0
|= (data
[i
] & 0xFF) << (8*(4-(currlen
+i
)));
919 assert(currlen
>= 0 && currlen
< 8);
920 state
->s1
&= ~(7 << 11);
921 state
->s1
|= (currlen
<< 11);
926 static void write_to_pointer(void *ctx
, long int output
)
928 char **ptr
= (char **)ctx
;
933 * Writing full ISO-2022 is not useful in very many circumstances.
934 * One of the few situations in which it _is_ useful is generating
935 * X11 COMPOUND_TEXT; therefore, this writing function will obey
936 * the compound text restrictions and hence output the subset of
937 * ISO-2022 that's usable in that context.
939 * The subset in question is roughly that we use GL/GR for G0/G1
940 * always, and that the _only_ escape sequences we output (other
941 * than the occasional DOCS) are those which designate different
942 * subcharsets into G0 and G1. There are additional constraints
943 * about which things go in which container; see below.
945 * FIXME: this wants some decent tests to be written, and also the
946 * exact output policy for compound text wants thinking about more
949 static int write_iso2022(charset_spec
const *charset
, long int input_chr
,
950 charset_state
*state
,
951 void (*emit
)(void *ctx
, long int output
),
955 struct iso2022_subcharset
const *subcs
;
956 struct iso2022_mode
const *mode
= (struct iso2022_mode
*)charset
->data
;
957 to_dbcs_planar_t last_planar_dbcs
= NULL
;
958 int last_p
, last_r
, last_c
;
962 * For output, I allocate the state variables as follows:
964 * s1[31] == 1 if output state has been initialised
965 * s1[30:24] == G1 charset (always in GR)
966 * s1[23:17] == G0 charset (always in GL)
967 * s1[16:14] == DOCS index plus 2 (because -1 and -2 are special)
968 * s1[13:11] == number of DOCS accumulated characters (up to five)
969 * s1[7:0] + s0[31:0] == DOCS collected characters
973 state
->s0
= 0x00000000UL
;
974 state
->s1
= 0x80000000UL
;
976 * Start with US-ASCII in GL and also in GR.
978 for (i
= 0; i
< lenof(iso2022_subcharsets
); i
++) {
979 subcs
= &iso2022_subcharsets
[i
];
980 if (subcs
->type
== mode
->ltype
&&
981 subcs
->i
== mode
->li
&&
982 subcs
->f
== mode
->lf
)
983 oselect(state
, i
, FALSE
, NULL
, NULL
);
984 if (subcs
->type
== mode
->rtype
&&
985 subcs
->i
== mode
->ri
&&
986 subcs
->f
== mode
->rf
)
987 oselect(state
, i
, TRUE
, NULL
, NULL
);
991 if (input_chr
== -1) {
993 * Special case: reset encoding state.
995 docs_char(state
, emit
, emitctx
, -2, NULL
, 0); /* leave DOCS */
997 for (i
= 0; i
< lenof(iso2022_subcharsets
); i
++) {
998 subcs
= &iso2022_subcharsets
[i
];
999 if (subcs
->type
== mode
->ltype
&&
1000 subcs
->i
== mode
->li
&&
1001 subcs
->f
== mode
->lf
)
1002 oselect(state
, i
, FALSE
, emit
, emitctx
);
1003 if (subcs
->type
== mode
->rtype
&&
1004 subcs
->i
== mode
->ri
&&
1005 subcs
->f
== mode
->rf
)
1006 oselect(state
, i
, TRUE
, emit
, emitctx
);
1012 * Special-case characters: Space, Delete, and anything in C0
1013 * or C1 are output unchanged.
1015 if (input_chr
<= 0x20 || (input_chr
>= 0x7F && input_chr
< 0xA0)) {
1016 emit(emitctx
, input_chr
);
1021 * Analyse the input character and work out which subcharset it
1024 for (i
= 0; i
< lenof(iso2022_subcharsets
); i
++) {
1025 subcs
= &iso2022_subcharsets
[i
];
1026 if (!(mode
->enable_mask
& (1 << subcs
->enable
)))
1027 continue; /* this charset is disabled */
1028 if (subcs
->sbcs_base
) {
1029 c1
= sbcs_from_unicode(subcs
->sbcs_base
, input_chr
);
1030 c1
-= subcs
->offset
;
1031 if (c1
>= 0x20 && c1
<= 0x7f) {
1035 } else if (subcs
->to_dbcs
) {
1036 if (subcs
->to_dbcs_plane
>= 0) {
1038 * Since multiplanar DBCSes almost by definition
1039 * involve several entries in iso2022_subcharsets
1040 * with the same to_dbcs function and different
1041 * plane values, we remember the last such function
1042 * we called and what its result was, so that we
1043 * don't (for example) have to call
1044 * unicode_to_cns11643 seven times.
1046 if (last_planar_dbcs
!= REPLANARISE(subcs
->to_dbcs
)) {
1047 last_planar_dbcs
= REPLANARISE(subcs
->to_dbcs
);
1048 if (!last_planar_dbcs(input_chr
,
1049 &last_p
, &last_r
, &last_c
))
1053 last_p
= subcs
->to_dbcs_plane
;
1054 if (!subcs
->to_dbcs(input_chr
, &last_r
, &last_c
))
1055 last_p
= 0; /* cannot match since to_dbcs_plane<0 */
1058 if (last_p
== subcs
->to_dbcs_plane
) {
1059 c1
= last_r
- subcs
->offset
;
1060 c2
= last_c
- subcs
->offset
;
1061 assert(c1
>= 0x20 && c1
<= 0x7f);
1062 assert(c2
>= 0x20 && c2
<= 0x7f);
1068 if (i
< lenof(iso2022_subcharsets
)) {
1072 * Our character is represented by c1 (and possibly also
1073 * c2) in subcharset `subcs'. So now we must decide whether
1074 * to designate that character set into G0/GL or G1/GR.
1076 * Any S6 or M6 subcharset has to go in GR because it won't
1077 * fit in GL. In addition, the compound text rules state
1078 * that any single-byte subcharset defined as the
1079 * right-hand half of some SBCS must go in GR.
1081 * M4 subcharsets can go in either half according to the
1082 * rules. I choose to put them in GR always because it's a
1083 * simple policy with reasonable behaviour (facilitates
1084 * switching between them and ASCII).
1086 right
= (subcs
->type
== S6
|| subcs
->type
== M6
|| subcs
->type
== M4
||
1087 (subcs
->sbcs_base
&& subcs
->offset
== 0x80));
1090 * If we're in a DOCS mode, leave it.
1092 docs_char(state
, emit
, emitctx
, -2, NULL
, 0);
1095 * If this subcharset is not already selected in that
1096 * container, select it.
1098 oselect(state
, i
, right
, emit
, emitctx
);
1101 * Now emit the actual characters.
1104 assert(c1
>= 0x20 && c1
<= 0x7f);
1105 emit(emitctx
, c1
| 0x80);
1107 assert(c2
>= 0x20 && c2
<= 0x7f);
1108 emit(emitctx
, c2
| 0x80);
1111 assert(c1
> 0x20 && c1
< 0x7f);
1114 assert(c2
> 0x20 && c2
< 0x7f);
1123 * Fall back to DOCS.
1130 cs
= -2; /* means failure */
1132 for (i
= 0; i
<= lenof(ctext_encodings
); i
++) {
1133 charset_state substate
;
1134 charset_spec
const *subcs
= ctext_encodings
[i
].subcs
;
1137 * We assume that all character sets dealt with by DOCS
1138 * are stateless for output purposes.
1140 substate
.s1
= substate
.s0
= 0;
1143 if (i
< lenof(ctext_encodings
)) {
1144 if ((mode
->enable_mask
& (1 << ctext_encodings
[i
].enable
)) &&
1145 subcs
->write(subcs
, input_chr
, &substate
,
1146 write_to_pointer
, &p
)) {
1151 if ((mode
->enable_mask
& (1 << CDU
)) &&
1152 write_utf8(NULL
, input_chr
, NULL
, write_to_pointer
, &p
)) {
1160 docs_char(state
, emit
, emitctx
, cs
, data
, p
- data
);
1169 * Full ISO 2022 output with all options on. Not entirely sure what
1170 * if anything this is useful for, but here it is anyway. All
1171 * output character sets and DOCS variants are permitted; all
1172 * containers start out with ASCII in them.
1174 static const struct iso2022_mode iso2022_all
= {
1175 (1<<CCS
) | (1<<COS
) | (1<<CPU
) | (1<<CDC
) | (1<<CDU
),
1176 S4
, 0, 'B', S4
, 0, 'B',
1179 const charset_spec charset_CS_ISO2022
= {
1180 CS_ISO2022
, read_iso2022
, write_iso2022
, &iso2022_all
1184 * X11 compound text. A subset of output charsets is permitted, and
1185 * G1/GR starts off in ISO8859-1.
1187 static const struct iso2022_mode iso2022_ctext
= {
1188 (1<<CCS
) | (1<<CDC
),
1189 S4
, 0, 'B', S6
, 0, 'A',
1192 const charset_spec charset_CS_CTEXT
= {
1193 CS_CTEXT
, read_iso2022
, write_iso2022
, &iso2022_ctext
1204 void iso2022_emit(void *ctx
, long output
)
1206 wchar_t **p
= (wchar_t **)ctx
;
1210 void iso2022_read_test(int line
, char *input
, int inlen
, ...)
1213 wchar_t *p
, str
[512];
1215 charset_state state
;
1218 state
.s0
= state
.s1
= 0;
1221 for (i
= 0; i
< inlen
; i
++)
1222 read_iso2022(NULL
, input
[i
] & 0xFF, &state
, iso2022_emit
, &p
);
1224 va_start(ap
, inlen
);
1226 for (i
= 0; i
< p
- str
; i
++) {
1227 l
= va_arg(ap
, long int);
1229 printf("%d: correct string shorter than output\n", line
);
1234 printf("%d: char %d came out as %08x, should be %08lx\n",
1235 line
, i
, str
[i
], l
);
1240 l
= va_arg(ap
, long int);
1242 printf("%d: correct string longer than output\n", line
);
1249 /* Macro to concoct the first three parameters of iso2022_read_test. */
1250 #define TESTSTR(x) __LINE__, x, lenof(x)
1254 printf("read tests beginning\n");
1255 /* Simple test (Emacs sample text for Japanese, in ISO-2022-JP) */
1256 iso2022_read_test(TESTSTR("Japanese (\x1b$BF|K\\8l\x1b(B)\t"
1257 "\x1b$B$3$s$K$A$O\x1b(B, "
1258 "\x1b$B%3%s%K%A%O\x1b(B\n"),
1259 'J','a','p','a','n','e','s','e',' ','(',
1260 0x65E5, 0x672C, 0x8A9E, ')', '\t',
1261 0x3053, 0x3093, 0x306b, 0x3061, 0x306f, ',', ' ',
1262 0x30b3, 0x30f3, 0x30cb, 0x30c1, 0x30cf, '\n', 0, -1);
1263 /* Same thing in EUC-JP (with designations, and half-width katakana) */
1264 iso2022_read_test(TESTSTR("\x1b$)B\x1b*I\x1b$+D"
1265 "Japanese (\xc6\xfc\xcb\xdc\xb8\xec)\t"
1266 "\xa4\xb3\xa4\xf3\xa4\xcb\xa4\xc1\xa4\xcf, "
1267 "\x8e\xba\x8e\xdd\x8e\xc6\x8e\xc1\x8e\xca\n"),
1268 'J','a','p','a','n','e','s','e',' ','(',
1269 0x65E5, 0x672C, 0x8A9E, ')', '\t',
1270 0x3053, 0x3093, 0x306b, 0x3061, 0x306f, ',', ' ',
1271 0xff7a, 0xff9d, 0xff86, 0xff81, 0xff8a, '\n', 0, -1);
1272 /* Multibyte single-shift */
1273 iso2022_read_test(TESTSTR("\x1b$)B\x1b*I\x1b$+D\x8f\"/!"),
1274 0x02D8, '!', 0, -1);
1275 /* Non-existent SBCS */
1276 iso2022_read_test(TESTSTR("\x1b(!Zfnord\n"),
1277 ERROR
, ERROR
, ERROR
, ERROR
, ERROR
, '\n', 0, -1);
1278 /* Pass-through of ordinary escape sequences, including a long one */
1279 iso2022_read_test(TESTSTR("\x1b""b\x1b#5\x1b#!!!5"),
1280 0x1B, 'b', 0x1B, '#', '5',
1281 0x1B, '#', '!', '!', '!', '5', 0, -1);
1282 /* Non-existent DBCS (also 5-byte escape sequence) */
1283 iso2022_read_test(TESTSTR("\x1b$(!Bfnord!"),
1284 ERROR
, ERROR
, ERROR
, 0, -1);
1285 /* Incomplete DB characters */
1286 iso2022_read_test(TESTSTR("\x1b$B(,(\x1b(BHi\x1b$B(,(\n"),
1287 0x2501, ERROR
, 'H', 'i', 0x2501, ERROR
, '\n', 0, -1);
1288 iso2022_read_test(TESTSTR("\x1b$)B\x1b*I\x1b$+D\xa4""B"),
1290 iso2022_read_test(TESTSTR("\x1b$)B\x1b*I\x1b$+D\x0e\x1b|$\xa2\xaf"),
1291 ERROR
, 0x02D8, 0, -1);
1292 /* Incomplete escape sequence */
1293 iso2022_read_test(TESTSTR("\x1b\n"), ERROR
, '\n', 0, -1);
1294 iso2022_read_test(TESTSTR("\x1b-A\x1b~\x1b\xa1"), ERROR
, 0xa1, 0, -1);
1295 /* Incomplete single-shift */
1296 iso2022_read_test(TESTSTR("\x8e\n"), ERROR
, '\n', 0, -1);
1297 iso2022_read_test(TESTSTR("\x1b$*B\x8e(\n"), ERROR
, '\n', 0, -1);
1298 /* Corner cases (02/00 and 07/15) */
1299 iso2022_read_test(TESTSTR("\x1b(B\x20\x7f"), 0x20, 0x7f, 0, -1);
1300 iso2022_read_test(TESTSTR("\x1b(I\x20\x7f"), 0x20, 0x7f, 0, -1);
1301 iso2022_read_test(TESTSTR("\x1b$B\x20\x7f"), 0x20, 0x7f, 0, -1);
1302 iso2022_read_test(TESTSTR("\x1b-A\x0e\x20\x7f"), 0xa0, 0xff, 0, -1);
1303 iso2022_read_test(TESTSTR("\x1b$-~\x0e\x20\x7f"), ERROR
, 0, -1);
1304 iso2022_read_test(TESTSTR("\x1b)B\xa0\xff"), ERROR
, ERROR
, 0, -1);
1305 iso2022_read_test(TESTSTR("\x1b)I\xa0\xff"), ERROR
, ERROR
, 0, -1);
1306 iso2022_read_test(TESTSTR("\x1b$)B\xa0\xff"), ERROR
, ERROR
, 0, -1);
1307 iso2022_read_test(TESTSTR("\x1b-A\x1b~\xa0\xff"), 0xa0, 0xff, 0, -1);
1308 iso2022_read_test(TESTSTR("\x1b$-~\x1b~\xa0\xff"), ERROR
, 0, -1);
1309 /* Designate control sets */
1310 iso2022_read_test(TESTSTR("\x1b!@"), 0x1b, '!', '@', 0, -1);
1311 /* Designate other coding system (UTF-8) */
1312 iso2022_read_test(TESTSTR("\x1b%G"
1313 "\xCE\xBA\xE1\xBD\xB9\xCF\x83\xCE\xBC\xCE\xB5"),
1314 0x03BA, 0x1F79, 0x03C3, 0x03BC, 0x03B5, 0, -1);
1315 iso2022_read_test(TESTSTR("\x1b-A\x1b%G\xCE\xBA\x1b%@\xa0"),
1316 0x03BA, 0xA0, 0, -1);
1317 iso2022_read_test(TESTSTR("\x1b%G\xCE\x1b%@"), ERROR
, 0, -1);
1318 iso2022_read_test(TESTSTR("\x1b%G\xCE\xBA\x1b%\x1b%@"),
1319 0x03BA, 0x1B, '%', 0, -1);
1320 /* DOCS (COMPOUND_TEXT extended segment) */
1321 iso2022_read_test(TESTSTR("\x1b%/1\x80\x80"), 0, -1);
1322 iso2022_read_test(TESTSTR("\x1b%/1\x80\x8fiso-8859-15\2xyz\x1b(B"),
1323 ERROR
, ERROR
, ERROR
, 0, -1);
1324 iso2022_read_test(TESTSTR("\x1b%/1\x80\x8eiso8859-15\2xyz\x1b(B"),
1325 'x', 'y', 'z', 0, -1);
1326 iso2022_read_test(TESTSTR("\x1b-A\x1b%/2\x80\x89"
1327 "big5-0\2\xa1\x40\xa1\x40"),
1328 0x3000, 0xa1, 0x40, 0, -1);
1329 /* Emacs Big5-in-ISO-2022 mapping */
1330 iso2022_read_test(TESTSTR("\x1b$(0&x86\x1b(B \x1b$(0DeBv"),
1331 0x5143, 0x6c23, ' ', ' ', 0x958b, 0x767c, 0, -1);
1332 /* Test from RFC 1922 (ISO-2022-CN) */
1333 iso2022_read_test(TESTSTR("\x1b$)A\x0e=;;;\x1b$)GG(_P\x0f"),
1334 0x4EA4, 0x6362, 0x4EA4, 0x63db, 0, -1);
1336 printf("read tests completed\n");
1337 printf("total: %d errors\n", total_errs
);
1338 return (total_errs
!= 0);
1341 #endif /* TESTMODE */
1343 #else /* ENUM_CHARSETS */
1345 ENUM_CHARSET(CS_ISO2022
)