+ if (specials_p) { /* It might be "special" (sometimes, but not always, a
+ mapping to more than one character */
+
+ /* Construct an inverse mapping hash for the specials */
+ HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
+ HV * specials_inverse = newHV();
+ char *char_from; /* the lhs of the map */
+ I32 from_len; /* its byte length */
+ char *char_to; /* the rhs of the map */
+ I32 to_len; /* its byte length */
+ SV *sv_to; /* and in a sv */
+ AV* from_list; /* list of things that map to each 'to' */
+
+ hv_iterinit(specials_hv);
+
+ /* The keys are the characters (in utf8) that map to the corresponding
+ * utf8 string value. Iterate through the list creating the inverse
+ * list. */
+ while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
+ SV** listp;
+ if (! SvPOK(sv_to)) {
+ Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() unexpectedly is not a string");
+ }
+ /*DEBUG_U(PerlIO_printf(Perl_debug_log, "Found mapping from %"UVXf", First char of to is %"UVXf"\n", utf8_to_uvchr((U8*) char_from, 0), utf8_to_uvchr((U8*) SvPVX(sv_to), 0)));*/
+
+ /* Each key in the inverse list is a mapped-to value, and the key's
+ * hash value is a list of the strings (each in utf8) that map to
+ * it. Those strings are all one character long */
+ if ((listp = hv_fetch(specials_inverse,
+ SvPVX(sv_to),
+ SvCUR(sv_to), 0)))
+ {
+ from_list = (AV*) *listp;
+ }
+ else { /* No entry yet for it: create one */
+ from_list = newAV();
+ if (! hv_store(specials_inverse,
+ SvPVX(sv_to),
+ SvCUR(sv_to),
+ (SV*) from_list, 0))
+ {
+ Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
+ }
+ }
+
+ /* Here have the list associated with this 'to' (perhaps newly
+ * created and empty). Just add to it. Note that we ASSUME that
+ * the input is guaranteed to not have duplications, so we don't
+ * check for that. Duplications just slow down execution time. */
+ av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
+ }
+
+ /* Here, 'specials_inverse' contains the inverse mapping. Go through
+ * it looking for cases like the FB05/FB06 examples above. There would
+ * be an entry in the hash like
+ * 'st' => [ FB05, FB06 ]
+ * In this example we will create two lists that get stored in the
+ * returned hash, 'ret':
+ * FB05 => [ FB05, FB06 ]
+ * FB06 => [ FB05, FB06 ]
+ *
+ * Note that there is nothing to do if the array only has one element.
+ * (In the normal 1-1 case handled below, we don't have to worry about
+ * two lists, as everything gets tied to the single list that is
+ * generated for the single character 'to'. But here, we are omitting
+ * that list, ('st' in the example), so must have multiple lists.) */
+ while ((from_list = (AV *) hv_iternextsv(specials_inverse,
+ &char_to, &to_len)))
+ {
+ if (av_len(from_list) > 0) {
+ int i;
+
+ /* We iterate over all combinations of i,j to place each code
+ * point on each list */
+ for (i = 0; i <= av_len(from_list); i++) {
+ int j;
+ AV* i_list = newAV();
+ SV** entryp = av_fetch(from_list, i, FALSE);
+ if (entryp == NULL) {
+ Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
+ }
+ if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
+ Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
+ }
+ if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
+ (SV*) i_list, FALSE))
+ {
+ Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
+ }
+
+ /* For debugging: UV u = utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
+ for (j = 0; j <= av_len(from_list); j++) {
+ entryp = av_fetch(from_list, j, FALSE);
+ if (entryp == NULL) {
+ Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
+ }
+
+ /* When i==j this adds itself to the list */
+ av_push(i_list, newSVuv(utf8_to_uvchr(
+ (U8*) SvPVX(*entryp), 0)));
+ /*DEBUG_U(PerlIO_printf(Perl_debug_log, "Adding %"UVXf" to list for %"UVXf"\n", utf8_to_uvchr((U8*) SvPVX(*entryp), 0), u));*/
+ }
+ }
+ }
+ }
+ SvREFCNT_dec(specials_inverse); /* done with it */
+ } /* End of specials */
+