+static void oselect(charset_state *state, int i, int right,
+ void (*emit)(void *ctx, long int output),
+ void *emitctx)
+{
+ int shift = (right ? 31-7 : 31-7-7);
+ struct iso2022_subcharset const *subcs = &iso2022_subcharsets[i];
+
+ if (((state->s1 >> shift) & 0x7F) != (unsigned)i) {
+ state->s1 &= ~(0x7FL << shift);
+ state->s1 |= (i << shift);
+
+ if (emit) {
+ emit(emitctx, ESC);
+ if (subcs->type == M4 || subcs->type == M6)
+ emit(emitctx, '$');
+ if (subcs->type == S6 || subcs->type == M6) {
+ assert(right);
+ emit(emitctx, '-');
+ } else if (right) {
+ emit(emitctx, ')');
+ } else {
+ emit(emitctx, '(');
+ }
+ if (subcs->i)
+ emit(emitctx, subcs->i);
+ emit(emitctx, subcs->f);
+ }
+ }
+}
+
+static void docs_char(charset_state *state,
+ void (*emit)(void *ctx, long int output),
+ void *emitctx, int cset, char *data, int datalen)
+{
+ int curr_cset, currlen, i;
+
+ /*
+ * cset is the index into ctext_encodings[]. It can also be -1
+ * to mean DOCS UTF-8, or -2 to mean no DOCS (ordinary 2022).
+ * In the latter case, `chr' is ignored.
+ */
+
+ /*
+ * First, terminate a DOCS segment if necessary. We always have
+ * to terminate a DOCS segment if one is active and we're about
+ * to switch to a different one; we might also have to
+ * terminate a length-encoded DOCS segment if we've run out of
+ * storage space to accumulate characters in it.
+ */
+ curr_cset = ((state->s1 >> 14) & 7) - 2;
+ currlen = ((state->s1 >> 11) & 7);
+ if ((curr_cset != -2 && curr_cset != cset) ||
+ (curr_cset >= 0 && currlen + datalen > 5)) {
+ if (curr_cset == -1) {
+ /*
+ * Terminating DOCS UTF-8 is easy.
+ */
+ emit(emitctx, ESC);
+ emit(emitctx, '%');
+ emit(emitctx, '@');
+ } else {
+ int len;
+
+ /*
+ * To terminate a length-encoded DOCS segment we must
+ * actually output the whole thing.
+ */
+ emit(emitctx, ESC);
+ emit(emitctx, '%');
+ emit(emitctx, '/');
+ emit(emitctx, '0' + ctext_encodings[curr_cset].octets_per_char);
+ len = currlen + datalen +
+ strlen(ctext_encodings[curr_cset].name);
+ assert(len < (1 << 14));
+ emit(emitctx, 0x80 | ((len >> 7) & 0x7F));
+ emit(emitctx, 0x80 | ((len ) & 0x7F));
+ /* The name stored in ctext_encodings[] includes the trailing \2 */
+ for (i = 0; ctext_encodings[curr_cset].name[i]; i++)
+ emit(emitctx, ctext_encodings[curr_cset].name[i]);
+ for (i = 0; i < currlen; i++)
+ emit(emitctx,
+ (i == 0 ? state->s1 : state->s0 >> (8*(4-i))) & 0xFF);
+ for (i = 0; i < datalen; i++)
+ emit(emitctx, data[i]);
+
+ /*
+ * We've now dealt with the input data, so clear it so
+ * we don't try to do so again below.
+ */
+ datalen = 0;
+ }
+ curr_cset = -2;
+ }
+
+ /*
+ * Now, start a DOCS segment if necessary.
+ */
+ if (curr_cset != cset) {
+ assert(cset != -2);
+ if (cset == -1) {
+ /*
+ * Start DOCS UTF-8.
+ */
+ emit(emitctx, ESC);
+ emit(emitctx, '%');
+ emit(emitctx, 'G');
+ } else {
+ /*
+ * Starting a length-encoded DOCS segment is simply a
+ * matter of setting our stored length counter to zero.
+ */
+ currlen = 0;
+ state->s1 &= ~(7 << 11);
+ state->s1 &= ~0xFF;
+ state->s0 = 0;
+ }
+ }
+ state->s1 &= ~(7 << 14);
+ assert((cset+2) >= 0 && (cset+2) < 8);
+ state->s1 |= ((cset+2) << 14);
+
+ /*
+ * Now we're in the right DOCS state. Actually deal with the
+ * input data, if we haven't already done so above.
+ */
+ if (datalen > 0) {
+ assert(cset != 2);
+ if (cset == -1) {
+ /*
+ * In DOCS UTF-8, we output data as soon as we get it.
+ */
+ for (i = 0; i < datalen; i++)
+ emit(emitctx, data[i]);
+ } else {
+ /*
+ * In length-encoded DOCS, we just store our data and
+ * bide our time. It'll all be output when we fill up
+ * or switch to another character set.
+ */
+ assert(currlen + datalen <= 5); /* overflow handled already */
+ for (i = 0; i < datalen; i++) {
+ if (currlen + i == 0)
+ state->s1 |= data[i] & 0xFF;
+ else
+ state->s0 |= (data[i] & 0xFF) << (8*(4-(currlen+i)));
+ }
+ currlen += datalen;
+ assert(currlen >= 0 && currlen < 8);
+ state->s1 &= ~(7 << 11);
+ state->s1 |= (currlen << 11);
+ }
+ }
+}
+
+static void write_to_pointer(void *ctx, long int output)
+{
+ char **ptr = (char **)ctx;
+ *(*ptr)++ = output;
+}
+
+/*
+ * Writing full ISO-2022 is not useful in very many circumstances.
+ * One of the few situations in which it _is_ useful is generating
+ * X11 COMPOUND_TEXT; therefore, this writing function will obey
+ * the compound text restrictions and hence output the subset of
+ * ISO-2022 that's usable in that context.
+ *
+ * The subset in question is roughly that we use GL/GR for G0/G1
+ * always, and that the _only_ escape sequences we output (other
+ * than the occasional DOCS) are those which designate different
+ * subcharsets into G0 and G1. There are additional constraints
+ * about which things go in which container; see below.
+ *
+ * FIXME: this wants some decent tests to be written, and also the
+ * exact output policy for compound text wants thinking about more
+ * carefully.
+ */
+static int write_iso2022(charset_spec const *charset, long int input_chr,
+ charset_state *state,
+ void (*emit)(void *ctx, long int output),
+ void *emitctx)
+{
+ int i;
+ struct iso2022_subcharset const *subcs;
+ struct iso2022_mode const *mode = (struct iso2022_mode *)charset->data;
+ to_dbcs_planar_t last_planar_dbcs = NULL;
+ int last_p, last_r, last_c;
+ long int c1, c2;
+
+ /*
+ * For output, I allocate the state variables as follows:
+ *
+ * s1[31] == 1 if output state has been initialised
+ * s1[30:24] == G1 charset (always in GR)
+ * s1[23:17] == G0 charset (always in GL)
+ * s1[16:14] == DOCS index plus 2 (because -1 and -2 are special)
+ * s1[13:11] == number of DOCS accumulated characters (up to five)
+ * s1[7:0] + s0[31:0] == DOCS collected characters
+ */
+
+ if (!state->s1) {
+ state->s0 = 0x00000000UL;
+ state->s1 = 0x80000000UL;
+ /*
+ * Start with US-ASCII in GL and also in GR.
+ */
+ for (i = 0; (unsigned)i < lenof(iso2022_subcharsets); i++) {
+ subcs = &iso2022_subcharsets[i];
+ if (subcs->type == mode->ltype &&
+ subcs->i == mode->li &&
+ subcs->f == mode->lf)
+ oselect(state, i, FALSE, NULL, NULL);
+ if (subcs->type == mode->rtype &&
+ subcs->i == mode->ri &&
+ subcs->f == mode->rf)
+ oselect(state, i, TRUE, NULL, NULL);
+ }
+ }
+
+ if (input_chr == -1) {
+ /*
+ * Special case: reset encoding state.
+ */
+ docs_char(state, emit, emitctx, -2, NULL, 0); /* leave DOCS */
+
+ for (i = 0; (unsigned)i < lenof(iso2022_subcharsets); i++) {
+ subcs = &iso2022_subcharsets[i];
+ if (subcs->type == mode->ltype &&
+ subcs->i == mode->li &&
+ subcs->f == mode->lf)
+ oselect(state, i, FALSE, emit, emitctx);
+ if (subcs->type == mode->rtype &&
+ subcs->i == mode->ri &&
+ subcs->f == mode->rf)
+ oselect(state, i, TRUE, emit, emitctx);
+ }
+ return TRUE;
+ }
+
+ /*
+ * Special-case characters: Space, Delete, and anything in C0
+ * or C1 are output unchanged.
+ */
+ if (input_chr <= 0x20 || (input_chr >= 0x7F && input_chr < 0xA0)) {
+ emit(emitctx, input_chr);
+ return TRUE;
+ }
+
+ /*
+ * Analyse the input character and work out which subcharset it
+ * belongs to.
+ */
+ for (i = 0; (unsigned)i < lenof(iso2022_subcharsets); i++) {
+ subcs = &iso2022_subcharsets[i];
+ if (!(mode->enable_mask & (1 << subcs->enable)))
+ continue; /* this charset is disabled */
+ if (subcs->sbcs_base) {
+ c1 = sbcs_from_unicode(subcs->sbcs_base, input_chr);
+ c1 -= subcs->offset;
+ if (c1 >= 0x20 && c1 <= 0x7f) {
+ c2 = 0;
+ break;
+ }
+ } else if (subcs->to_dbcs) {
+ if (subcs->to_dbcs_plane >= 0) {
+ /*
+ * Since multiplanar DBCSes almost by definition
+ * involve several entries in iso2022_subcharsets
+ * with the same to_dbcs function and different
+ * plane values, we remember the last such function
+ * we called and what its result was, so that we
+ * don't (for example) have to call
+ * unicode_to_cns11643 seven times.
+ */
+ if (last_planar_dbcs != REPLANARISE(subcs->to_dbcs)) {
+ last_planar_dbcs = REPLANARISE(subcs->to_dbcs);
+ if (!last_planar_dbcs(input_chr,
+ &last_p, &last_r, &last_c))
+ last_p = -1;
+ }
+ } else {
+ last_p = subcs->to_dbcs_plane;
+ if (!subcs->to_dbcs(input_chr, &last_r, &last_c))
+ last_p = 0; /* cannot match since to_dbcs_plane<0 */
+ }
+
+ if (last_p == subcs->to_dbcs_plane) {
+ c1 = last_r - subcs->offset;
+ c2 = last_c - subcs->offset;
+ assert(c1 >= 0x20 && c1 <= 0x7f);
+ assert(c2 >= 0x20 && c2 <= 0x7f);
+ break;
+ }
+ }
+ }
+
+ if ((unsigned)i < lenof(iso2022_subcharsets)) {
+ int right;
+
+ /*
+ * Our character is represented by c1 (and possibly also
+ * c2) in subcharset `subcs'. So now we must decide whether
+ * to designate that character set into G0/GL or G1/GR.
+ *
+ * Any S6 or M6 subcharset has to go in GR because it won't
+ * fit in GL. In addition, the compound text rules state
+ * that any single-byte subcharset defined as the
+ * right-hand half of some SBCS must go in GR.
+ *
+ * M4 subcharsets can go in either half according to the
+ * rules. I choose to put them in GR always because it's a
+ * simple policy with reasonable behaviour (facilitates
+ * switching between them and ASCII).
+ */
+ right = (subcs->type == S6 || subcs->type == M6 || subcs->type == M4 ||
+ (subcs->sbcs_base && subcs->offset == 0x80));
+
+ /*
+ * If we're in a DOCS mode, leave it.
+ */
+ docs_char(state, emit, emitctx, -2, NULL, 0);
+
+ /*
+ * If this subcharset is not already selected in that
+ * container, select it.
+ */
+ oselect(state, i, right, emit, emitctx);
+
+ /*
+ * Now emit the actual characters.
+ */
+ if (right) {
+ assert(c1 >= 0x20 && c1 <= 0x7f);
+ emit(emitctx, c1 | 0x80);
+ if (c2) {
+ assert(c2 >= 0x20 && c2 <= 0x7f);
+ emit(emitctx, c2 | 0x80);
+ }
+ } else {
+ assert(c1 > 0x20 && c1 < 0x7f);
+ emit(emitctx, c1);
+ if (c2) {
+ assert(c2 > 0x20 && c2 < 0x7f);
+ emit(emitctx, c2);
+ }
+ }
+
+ return TRUE;
+ }
+
+ /*
+ * Fall back to DOCS.
+ */
+ {
+ char data[10];
+ char *p = data;
+ int i, cs;
+
+ cs = -2; /* means failure */
+
+ for (i = 0; (unsigned)i <= lenof(ctext_encodings); i++) {
+ charset_state substate;
+ charset_spec const *subcs = ctext_encodings[i].subcs;
+
+ /*
+ * We assume that all character sets dealt with by DOCS
+ * are stateless for output purposes.
+ */
+ substate.s1 = substate.s0 = 0;
+ p = data;
+
+ if ((unsigned)i < lenof(ctext_encodings)) {
+ if ((mode->enable_mask & (1 << ctext_encodings[i].enable)) &&
+ subcs->write(subcs, input_chr, &substate,
+ write_to_pointer, &p)) {
+ cs = i;
+ break;
+ }
+ } else {
+ if ((mode->enable_mask & (1 << CDU)) &&
+ write_utf8(NULL, input_chr, NULL, write_to_pointer, &p)) {
+ cs = -1;
+ break;
+ }
+ }
+ }
+
+ if (cs != -2) {
+ docs_char(state, emit, emitctx, cs, data, p - data);
+ return TRUE;
+ }
+ }
+
+ return FALSE;
+}
+
+/*
+ * Full ISO 2022 output with all options on. Not entirely sure what
+ * if anything this is useful for, but here it is anyway. All
+ * output character sets and DOCS variants are permitted; all
+ * containers start out with ASCII in them.
+ */
+static const struct iso2022_mode iso2022_all = {
+ (1<<CCS) | (1<<COS) | (1<<CPU) | (1<<CDC) | (1<<CDU),
+ S4, 0, 'B', S4, 0, 'B',
+};
+