This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Removed set_emergency_buffer from malloc.c
[perl5.git] / regcomp.c
index df56350..e7c6662 100644 (file)
--- a/regcomp.c
+++ b/regcomp.c
@@ -93,6 +93,8 @@ EXTERN_C const struct regexp_engine my_reg_engine;
 
 #define HAS_NONLATIN1_FOLD_CLOSURE(i) \
  _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
+#define HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(i) \
+ _HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
 #define IS_NON_FINAL_FOLD(c) _IS_NON_FINAL_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
 #define IS_IN_SOME_FOLD_L1(c) _IS_IN_SOME_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
 
@@ -1718,7 +1720,7 @@ S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
               May be the same as tail.
   tail       : item following the branch sequence
   count      : words in the sequence
-  flags      : currently the OP() type we will be building one of /EXACT(|F|Fl)/
+  flags      : currently the OP() type we will be building one of /EXACT(|F|FA|FU|FU_SS)/
   depth      : indent depth
 
 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
@@ -2982,8 +2984,8 @@ S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch,
              : MADE_TRIE;
 }
 
-STATIC void
-S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source,  regnode *stclass, U32 depth)
+STATIC regnode *
+S_construct_ahocorasick_from_trie(pTHX_ RExC_state_t *pRExC_state, regnode *source, U32 depth)
 {
 /* The Trie is constructed and compressed now so we can build a fail array if
  * it's needed
@@ -3021,13 +3023,26 @@ S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source,  regnode
     U32 *fail;
     reg_ac_data *aho;
     const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("T"));
+    regnode *stclass;
     GET_RE_DEBUG_FLAGS_DECL;
 
-    PERL_ARGS_ASSERT_MAKE_TRIE_FAILTABLE;
+    PERL_ARGS_ASSERT_CONSTRUCT_AHOCORASICK_FROM_TRIE;
 #ifndef DEBUGGING
     PERL_UNUSED_ARG(depth);
 #endif
 
+    if ( OP(source) == TRIE ) {
+        struct regnode_1 *op = (struct regnode_1 *)
+            PerlMemShared_calloc(1, sizeof(struct regnode_1));
+        StructCopy(source,op,struct regnode_1);
+        stclass = (regnode *)op;
+    } else {
+        struct regnode_charclass *op = (struct regnode_charclass *)
+            PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
+        StructCopy(source,op,struct regnode_charclass);
+        stclass = (regnode *)op;
+    }
+    OP(stclass)+=2; /* covert the TRIE type to its AHO-CORASICK equivalent */
 
     ARG_SET( stclass, data_slot );
     aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
@@ -3094,6 +3109,7 @@ S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source,  regnode
     });
     Safefree(q);
     /*RExC_seen |= REG_TRIEDFA_SEEN;*/
+    return stclass;
 }
 
 
@@ -3417,9 +3433,9 @@ S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan,
                 else { /* Here is a generic multi-char fold. */
                     U8* multi_end  = s + len;
 
-                    /* Count how many characters in it.  In the case of /aa, no
-                     * folds which contain ASCII code points are allowed, so
-                     * check for those, and skip if found. */
+                    /* Count how many characters are in it.  In the case of
+                     * /aa, no folds which contain ASCII code points are
+                     * allowed, so check for those, and skip if found. */
                     if (OP(scan) != EXACTFA && OP(scan) != EXACTFA_NO_TRIE) {
                         count = utf8_length(s, multi_end);
                         s = multi_end;
@@ -4249,11 +4265,13 @@ S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
            }
            flags &= ~SCF_DO_STCLASS;
        }
-       else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
+        else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT!, so is
+                                                     EXACTFish */
            SSize_t l = STR_LEN(scan);
            UV uc = *((U8*)STRING(scan));
             SV* EXACTF_invlist = _new_invlist(4); /* Start out big enough for 2
                                                      separate code points */
+            const U8 * s = (U8*)STRING(scan);
 
            /* Search for fixed substrings supports EXACT only. */
            if (flags & SCF_DO_SUBSTR) {
@@ -4261,7 +4279,6 @@ S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
                 scan_commit(pRExC_state, data, minlenp, is_inf);
            }
            if (UTF) {
-               const U8 * const s = (U8 *)STRING(scan);
                uc = utf8_to_uvchr_buf(s, s + l, NULL);
                l = utf8_length(s, s + l);
            }
@@ -4281,71 +4298,140 @@ S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
                    data->longest = &(data->longest_float);
                }
            }
-            if (OP(scan) == EXACTFL) {
 
-                /* We don't know what the folds are; it could be anything. XXX
-                 * Actually, we only support UTF-8 encoding for code points
-                 * above Latin1, so we could know what those folds are. */
-                EXACTF_invlist = _add_range_to_invlist(EXACTF_invlist,
-                                                       0,
-                                                       UV_MAX);
+            if (OP(scan) != EXACTFL && flags & SCF_DO_STCLASS_AND) {
+                ssc_clear_locale(data->start_class);
             }
-            else {  /* Non-locale EXACTFish */
-                EXACTF_invlist = add_cp_to_invlist(EXACTF_invlist, uc);
-                if (flags & SCF_DO_STCLASS_AND) {
-                    ssc_clear_locale(data->start_class);
+
+            if (! UTF) {
+
+                /* We punt and assume can match anything if the node begins
+                 * with a multi-character fold.  Things are complicated.  For
+                 * example, /ffi/i could match any of:
+                 *  "\N{LATIN SMALL LIGATURE FFI}"
+                 *  "\N{LATIN SMALL LIGATURE FF}I"
+                 *  "F\N{LATIN SMALL LIGATURE FI}"
+                 *  plus several other things; and making sure we have all the
+                 *  possibilities is hard. */
+                if (is_MULTI_CHAR_FOLD_latin1_safe(s, s + STR_LEN(scan))) {
+                    EXACTF_invlist =
+                             _add_range_to_invlist(EXACTF_invlist, 0, UV_MAX);
                 }
-                if (uc < 256) { /* We know what the Latin1 folds are ... */
-                    if (IS_IN_SOME_FOLD_L1(uc)) {   /* For instance, we
-                                                       know if anything folds
-                                                       with this */
-                        EXACTF_invlist = add_cp_to_invlist(EXACTF_invlist,
+                else {
+
+                    /* Any Latin1 range character can potentially match any
+                     * other depending on the locale */
+                    if (OP(scan) == EXACTFL) {
+                        _invlist_union(EXACTF_invlist, PL_Latin1,
+                                                              &EXACTF_invlist);
+                    }
+                    else {
+                        /* But otherwise, it matches at least itself.  We can
+                         * quickly tell if it has a distinct fold, and if so,
+                         * it matches that as well */
+                        EXACTF_invlist = add_cp_to_invlist(EXACTF_invlist, uc);
+                        if (IS_IN_SOME_FOLD_L1(uc)) {
+                            EXACTF_invlist = add_cp_to_invlist(EXACTF_invlist,
                                                            PL_fold_latin1[uc]);
-                        if (OP(scan) != EXACTFA) { /* The folds below aren't
-                                                      legal under /iaa */
-                            if (isARG2_lower_or_UPPER_ARG1('s', uc)) {
-                                EXACTF_invlist
-                                    = add_cp_to_invlist(EXACTF_invlist,
-                                                LATIN_SMALL_LETTER_SHARP_S);
-                            }
-                            else if (uc == LATIN_SMALL_LETTER_SHARP_S) {
-                                EXACTF_invlist
-                                    = add_cp_to_invlist(EXACTF_invlist, 's');
-                                EXACTF_invlist
-                                    = add_cp_to_invlist(EXACTF_invlist, 'S');
-                            }
                         }
+                    }
 
-                        /* We also know if there are above-Latin1 code points
-                         * that fold to this (none legal for ASCII and /iaa) */
-                        if ((! isASCII(uc) || OP(scan) != EXACTFA)
-                            && HAS_NONLATIN1_FOLD_CLOSURE(uc))
-                        {
-                            /* XXX We could know exactly what does fold to this
-                             * if the reverse folds are loaded, as currently in
-                             * S_regclass() */
-                            _invlist_union(EXACTF_invlist,
-                                           PL_AboveLatin1,
-                                           &EXACTF_invlist);
+                    /* Some characters match above-Latin1 ones under /i.  This
+                     * is true of EXACTFL ones when the locale is UTF-8 */
+                    if (HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(uc)
+                        && (! isASCII(uc) || (OP(scan) != EXACTFA
+                                            && OP(scan) != EXACTFA_NO_TRIE)))
+                    {
+                        add_above_Latin1_folds(pRExC_state,
+                                               (U8) uc,
+                                               &EXACTF_invlist);
+                    }
+                }
+            }
+            else {  /* Pattern is UTF-8 */
+                U8 folded[UTF8_MAX_FOLD_CHAR_EXPAND * UTF8_MAXBYTES_CASE + 1] = { '\0' };
+                STRLEN foldlen = UTF8SKIP(s);
+                const U8* e = s + STR_LEN(scan);
+                SV** listp;
+
+                /* The only code points that aren't folded in a UTF EXACTFish
+                 * node are are the problematic ones in EXACTFL nodes */
+                if (OP(scan) == EXACTFL
+                    && is_PROBLEMATIC_LOCALE_FOLDEDS_START_cp(uc))
+                {
+                    /* We need to check for the possibility that this EXACTFL
+                     * node begins with a multi-char fold.  Therefore we fold
+                     * the first few characters of it so that we can make that
+                     * check */
+                    U8 *d = folded;
+                    int i;
+
+                    for (i = 0; i < UTF8_MAX_FOLD_CHAR_EXPAND && s < e; i++) {
+                        if (isASCII(*s)) {
+                            *(d++) = (U8) toFOLD(*s);
+                            s++;
+                        }
+                        else {
+                            STRLEN len;
+                            to_utf8_fold(s, d, &len);
+                            d += len;
+                            s += UTF8SKIP(s);
                         }
                     }
+
+                    /* And set up so the code below that looks in this folded
+                     * buffer instead of the node's string */
+                    e = d;
+                    foldlen = UTF8SKIP(folded);
+                    s = folded;
                 }
-                else {  /* Non-locale, above Latin1.  XXX We don't currently
-                           know what participates in folds with this, so have
-                           to assume anything could */
-
-                    /* XXX We could know exactly what does fold to this if the
-                     * reverse folds are loaded, as currently in S_regclass().
-                     * But we do know that under /iaa nothing in the ASCII
-                     * range can participate */
-                    if (OP(scan) == EXACTFA) {
-                        _invlist_union_complement_2nd(EXACTF_invlist,
-                                                      PL_XPosix_ptrs[_CC_ASCII],
-                                                      &EXACTF_invlist);
+
+                /* When we reach here 's' points to the fold of the first
+                 * character(s) of the node; and 'e' points to far enough along
+                 * the folded string to be just past any possible multi-char
+                 * fold. 'foldlen' is the length in bytes of the first
+                 * character in 's'
+                 *
+                 * Unlike the non-UTF-8 case, the macro for determining if a
+                 * string is a multi-char fold requires all the characters to
+                 * already be folded.  This is because of all the complications
+                 * if not.  Note that they are folded anyway, except in EXACTFL
+                 * nodes.  Like the non-UTF case above, we punt if the node
+                 * begins with a multi-char fold  */
+
+                if (is_MULTI_CHAR_FOLD_utf8_safe(s, e)) {
+                    EXACTF_invlist =
+                             _add_range_to_invlist(EXACTF_invlist, 0, UV_MAX);
+                }
+                else {  /* Single char fold */
+
+                    /* It matches all the things that fold to it, which are
+                     * found in PL_utf8_foldclosures (including itself) */
+                    EXACTF_invlist = add_cp_to_invlist(EXACTF_invlist, uc);
+                    if (! PL_utf8_foldclosures) {
+                        _load_PL_utf8_foldclosures();
                     }
-                    else {
-                        EXACTF_invlist = _add_range_to_invlist(EXACTF_invlist,
-                                                               0, UV_MAX);
+                    if ((listp = hv_fetch(PL_utf8_foldclosures,
+                                        (char *) s, foldlen, FALSE)))
+                    {
+                        AV* list = (AV*) *listp;
+                        IV k;
+                        for (k = 0; k <= av_tindex(list); k++) {
+                            SV** c_p = av_fetch(list, k, FALSE);
+                            UV c;
+                            assert(c_p);
+
+                            c = SvUV(*c_p);
+
+                            /* /aa doesn't allow folds between ASCII and non- */
+                            if ((OP(scan) == EXACTFA || OP(scan) == EXACTFA_NO_TRIE)
+                                && isASCII(c) != isASCII(uc))
+                            {
+                                continue;
+                            }
+
+                            EXACTF_invlist = add_cp_to_invlist(EXACTF_invlist, c);
+                        }
                     }
                 }
             }
@@ -4693,13 +4779,13 @@ S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
                    /* It is counted once already... */
                    data->pos_min += minnext * (mincount - counted);
 #if 0
-PerlIO_printf(Perl_debug_log, "counted=%"UVdf" deltanext=%"UVdf
-                              " SSize_t_MAX=%"UVdf" minnext=%"UVdf
-                              " maxcount=%"UVdf" mincount=%"UVdf"\n",
+PerlIO_printf(Perl_debug_log, "counted=%"UVuf" deltanext=%"UVuf
+                              " SSize_t_MAX=%"UVuf" minnext=%"UVuf
+                              " maxcount=%"UVuf" mincount=%"UVuf"\n",
     (UV)counted, (UV)deltanext, (UV)SSize_t_MAX, (UV)minnext, (UV)maxcount,
     (UV)mincount);
 if (deltanext != SSize_t_MAX)
-PerlIO_printf(Perl_debug_log, "LHS=%"UVdf" RHS=%"UVdf"\n",
+PerlIO_printf(Perl_debug_log, "LHS=%"UVuf" RHS=%"UVuf"\n",
     (UV)(-counted * deltanext + (minnext + deltanext) * maxcount
           - minnext * mincount), (UV)(SSize_t_MAX - data->pos_delta));
 #endif
@@ -5430,7 +5516,8 @@ S_add_data(RExC_state_t* const pRExC_state, const char* const s, const U32 n)
     return count;
 }
 
-/*XXX: todo make this not included in a non debugging perl */
+/*XXX: todo make this not included in a non debugging perl, but appears to be
+ * used anyway there, in 'use re' */
 #ifndef PERL_IN_XSUB_RE
 void
 Perl_reginitcolors(pTHX)
@@ -5633,7 +5720,7 @@ S_concat_pat(pTHX_ RExC_state_t * const pRExC_state,
     /* if we know we have at least two args, create an empty string,
      * then concatenate args to that. For no args, return an empty string */
     if (!pat && pat_count != 1) {
-        pat = newSVpvn("", 0);
+        pat = newSVpvs("");
         SAVEFREESV(pat);
         alloced = TRUE;
     }
@@ -6751,22 +6838,8 @@ reStudy:
        else if (PL_regkind[OP(first)] == TRIE &&
                ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
        {
-           regnode *trie_op;
-           /* this can happen only on restudy */
-           if ( OP(first) == TRIE ) {
-                struct regnode_1 *trieop = (struct regnode_1 *)
-                   PerlMemShared_calloc(1, sizeof(struct regnode_1));
-                StructCopy(first,trieop,struct regnode_1);
-                trie_op=(regnode *)trieop;
-            } else {
-                struct regnode_charclass *trieop = (struct regnode_charclass *)
-                   PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
-                StructCopy(first,trieop,struct regnode_charclass);
-                trie_op=(regnode *)trieop;
-            }
-            OP(trie_op)+=2;
-            make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
-           ri->regstclass = trie_op;
+            /* this can happen only on restudy */
+            ri->regstclass = construct_ahocorasick_from_trie(pRExC_state, (regnode *)first, 0);
        }
 #endif
        else if (REGNODE_SIMPLE(OP(first)))
@@ -9706,18 +9779,6 @@ S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
            case '@':           /* (?@...) */
                vFAIL2("Sequence (?%c...) not implemented", (int)paren);
                break;
-            case '#':           /* (?#...) */
-                /* XXX As soon as we disallow separating the '?' and '*' (by
-                 * spaces or (?#...) comment), it is believed that this case
-                 * will be unreachable and can be removed.  See
-                 * [perl #117327] */
-                while (*RExC_parse && *RExC_parse != ')')
-                   RExC_parse++;
-               if (*RExC_parse != ')')
-                   FAIL("Sequence (?#... not terminated");
-               nextchar(pRExC_state);
-               *flagp = TRYAGAIN;
-               return NULL;
            case '0' :           /* (?0) */
            case 'R' :           /* (?R) */
                if (*RExC_parse != ')')
@@ -13131,6 +13192,74 @@ S_handle_regex_sets(pTHX_ RExC_state_t *pRExC_state, SV** return_invlist,
 }
 #undef IS_OPERAND
 
+STATIC void
+S_add_above_Latin1_folds(pTHX_ RExC_state_t *pRExC_state, const U8 cp, SV** invlist)
+{
+    /* This hard-codes the Latin1/above-Latin1 folding rules, so that an
+     * innocent-looking character class, like /[ks]/i won't have to go out to
+     * disk to find the possible matches.
+     *
+     * This should be called only for a Latin1-range code points, cp, which is
+     * known to be involved in a fold with other code points above Latin1.  It
+     * would give false results if /aa has been specified.  Multi-char folds
+     * are outside the scope of this, and must be handled specially.
+     *
+     * XXX It would be better to generate these via regen, in case a new
+     * version of the Unicode standard adds new mappings, though that is not
+     * really likely, and may be caught by the default: case of the switch
+     * below. */
+
+    PERL_ARGS_ASSERT_ADD_ABOVE_LATIN1_FOLDS;
+
+    switch (cp) {
+        case 'k':
+        case 'K':
+          *invlist =
+             add_cp_to_invlist(*invlist, KELVIN_SIGN);
+            break;
+        case 's':
+        case 'S':
+          *invlist = add_cp_to_invlist(*invlist, LATIN_SMALL_LETTER_LONG_S);
+            break;
+        case MICRO_SIGN:
+          *invlist = add_cp_to_invlist(*invlist, GREEK_CAPITAL_LETTER_MU);
+          *invlist = add_cp_to_invlist(*invlist, GREEK_SMALL_LETTER_MU);
+            break;
+        case LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE:
+        case LATIN_SMALL_LETTER_A_WITH_RING_ABOVE:
+          *invlist = add_cp_to_invlist(*invlist, ANGSTROM_SIGN);
+            break;
+        case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
+          *invlist = add_cp_to_invlist(*invlist,
+                                        LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS);
+            break;
+        case LATIN_SMALL_LETTER_SHARP_S:
+          *invlist = add_cp_to_invlist(*invlist, LATIN_CAPITAL_LETTER_SHARP_S);
+            break;
+        case 'F': case 'f':
+        case 'I': case 'i':
+        case 'L': case 'l':
+        case 'T': case 't':
+        case 'A': case 'a':
+        case 'H': case 'h':
+        case 'J': case 'j':
+        case 'N': case 'n':
+        case 'W': case 'w':
+        case 'Y': case 'y':
+            /* These all are targets of multi-character folds from code points
+             * that require UTF8 to express, so they can't match unless the
+             * target string is in UTF-8, so no action here is necessary, as
+             * regexec.c properly handles the general case for UTF-8 matching
+             * and multi-char folds */
+            break;
+        default:
+            /* Use deprecated warning to increase the chances of this being
+             * output */
+            ckWARN2reg_d(RExC_parse, "Perl folding rules are not up-to-date for 0x%02X; please use the perlbug utility to report;", cp);
+            break;
+    }
+}
+
 /* The names of properties whose definitions are not known at compile time are
  * stored in this SV, after a constant heading.  So if the length has been
  * changed since initialization, then there is a run-time definition. */
@@ -13437,7 +13566,6 @@ parseit:
                }
                if (!SIZE_ONLY) {
                     SV* invlist;
-                    char* formatted;
                     char* name;
 
                    if (UCHARAT(RExC_parse) == '^') {
@@ -13458,14 +13586,13 @@ parseit:
                      * will have its name be <__NAME_i>.  The design is
                      * discussed in commit
                      * 2f833f5208e26b208886e51e09e2c072b5eabb46 */
-                    formatted = Perl_form(aTHX_
+                    name = savepv(Perl_form(aTHX_
                                           "%s%.*s%s\n",
                                           (FOLD) ? "__" : "",
                                           (int)n,
                                           RExC_parse,
                                           (FOLD) ? "_i" : ""
-                                );
-                    name = savepvn(formatted, strlen(formatted));
+                                ));
 
                     /* Look up the property name, and get its swash and
                      * inversion list, if the property is found  */
@@ -13494,6 +13621,19 @@ parseit:
                                 "Property '%"UTF8f"' is unknown",
                                 UTF8fARG(UTF, n, name));
                         }
+
+                        /* If the property name doesn't already have a package
+                         * name, add the current one to it so that it can be
+                         * referred to outside it. [perl #121777] */
+                        if (! instr(name, "::") && PL_curstash) {
+                            char* full_name = Perl_form(aTHX_
+                                                        "%s::%s",
+                                                        HvNAME(PL_curstash),
+                                                        name);
+                            n = strlen(full_name);
+                            Safefree(name);
+                            name = savepvn(full_name, n);
+                        }
                         Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%"UTF8f"\n",
                                         (value == 'p' ? '+' : '!'),
                                         UTF8fARG(UTF, n, name));
@@ -13733,6 +13873,11 @@ parseit:
                     ANYOF_POSIXL_ZERO(ret);
                 }
 
+                /* Coverity thinks it is possible for this to be negative; both
+                 * jhi and khw think it's not, but be safer */
+                assert(! (ANYOF_FLAGS(ret) & ANYOF_POSIXL)
+                       || (namedclass + ((namedclass % 2) ? -1 : 1)) >= 0);
+
                 /* See if it already matches the complement of this POSIX
                  * class */
                 if ((ANYOF_FLAGS(ret) & ANYOF_POSIXL)
@@ -13943,7 +14088,7 @@ parseit:
                         AV* this_array;
                         STRLEN cp_count = utf8_length(foldbuf,
                                                       foldbuf + foldlen);
-                        SV* multi_fold = sv_2mortal(newSVpvn("", 0));
+                        SV* multi_fold = sv_2mortal(newSVpvs(""));
 
                         Perl_sv_catpvf(aTHX_ multi_fold, "\\x{%"UVXf"}", value);
 
@@ -14010,7 +14155,7 @@ parseit:
                 && ((prevvalue >= 'a' && value <= 'z')
                     || (prevvalue >= 'A' && value <= 'Z')))
             {
-                _invlist_intersection(this_range, PL_ASCII,
+                _invlist_intersection(this_range, PL_XPosix_ptrs[_CC_ASCII],
                                       &this_range);
 
                 /* Since this above only contains ascii, the intersection of it
@@ -14350,15 +14495,6 @@ parseit:
 
                     if (j < 256) {
 
-                        /* We have the latin1 folding rules hard-coded here so
-                         * that an innocent-looking character class, like
-                         * /[ks]/i won't have to go out to disk to find the
-                         * possible matches.  XXX It would be better to
-                         * generate these via regen, in case a new version of
-                         * the Unicode standard adds new mappings, though that
-                         * is not really likely, and may be caught by the
-                         * default: case of the switch below. */
-
                         if (IS_IN_SOME_FOLD_L1(j)) {
 
                             /* ASCII is always matched; non-ASCII is matched
@@ -14375,72 +14511,12 @@ parseit:
                             }
                         }
 
-                        if (HAS_NONLATIN1_FOLD_CLOSURE(j)
+                        if (HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(j)
                             && (! isASCII(j) || ! ASCII_FOLD_RESTRICTED))
                         {
-                            /* Certain Latin1 characters have matches outside
-                            * Latin1.  To get here, <j> is one of those
-                            * characters.   None of these matches is valid for
-                            * ASCII characters under /aa, which is why the 'if'
-                            * just above excludes those.  These matches only
-                            * happen when the target string is utf8.  The code
-                            * below adds the single fold closures for <j> to the
-                            * inversion list. */
-
-                            switch (j) {
-                                case 'k':
-                                case 'K':
-                                  *use_list =
-                                     add_cp_to_invlist(*use_list, KELVIN_SIGN);
-                                    break;
-                                case 's':
-                                case 'S':
-                                  *use_list = add_cp_to_invlist(*use_list,
-                                                    LATIN_SMALL_LETTER_LONG_S);
-                                    break;
-                                case MICRO_SIGN:
-                                  *use_list = add_cp_to_invlist(*use_list,
-                                                      GREEK_CAPITAL_LETTER_MU);
-                                  *use_list = add_cp_to_invlist(*use_list,
-                                                        GREEK_SMALL_LETTER_MU);
-                                    break;
-                                case LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE:
-                                case LATIN_SMALL_LETTER_A_WITH_RING_ABOVE:
-                                  *use_list =
-                                   add_cp_to_invlist(*use_list, ANGSTROM_SIGN);
-                                    break;
-                                case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
-                                  *use_list = add_cp_to_invlist(*use_list,
-                                        LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS);
-                                    break;
-                                case LATIN_SMALL_LETTER_SHARP_S:
-                                  *use_list = add_cp_to_invlist(*use_list,
-                                                 LATIN_CAPITAL_LETTER_SHARP_S);
-                                    break;
-                                case 'F': case 'f':
-                                case 'I': case 'i':
-                                case 'L': case 'l':
-                                case 'T': case 't':
-                                case 'A': case 'a':
-                                case 'H': case 'h':
-                                case 'J': case 'j':
-                                case 'N': case 'n':
-                                case 'W': case 'w':
-                                case 'Y': case 'y':
-                                    /* These all are targets of multi-character
-                                     * folds from code points that require UTF8
-                                     * to express, so they can't match unless
-                                     * the target string is in UTF-8, so no
-                                     * action here is necessary, as regexec.c
-                                     * properly handles the general case for
-                                     * UTF-8 matching and multi-char folds */
-                                    break;
-                                default:
-                                    /* Use deprecated warning to increase the
-                                    * chances of this being output */
-                                    ckWARN2reg_d(RExC_parse, "Perl folding rules are not up-to-date for 0x%"UVXf"; please use the perlbug utility to report;", j);
-                                    break;
-                            }
+                            add_above_Latin1_folds(pRExC_state,
+                                                   (U8) j,
+                                                   use_list);
                         }
                         continue;
                     }
@@ -14469,9 +14545,8 @@ parseit:
                         for (k = 0; k <= av_tindex(list); k++) {
                             SV** c_p = av_fetch(list, k, FALSE);
                             UV c;
-                            if (c_p == NULL) {
-                                Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
-                            }
+                            assert(c_p);
+
                             c = SvUV(*c_p);
 
                             /* /aa doesn't allow folds between ASCII and non- */
@@ -16117,7 +16192,16 @@ Perl_regfree_internal(pTHX_ REGEXP * const rx)
                         PerlMemShared_free(aho->fail);
                         /* do this last!!!! */
                         PerlMemShared_free(ri->data->data[n]);
-                        PerlMemShared_free(ri->regstclass);
+                        /* we should only ever get called once, so
+                         * assert as much, and also guard the free
+                         * which /might/ happen twice. At the least
+                         * it will make code anlyzers happy and it
+                         * doesn't cost much. - Yves */
+                        assert(ri->regstclass);
+                        if (ri->regstclass) {
+                            PerlMemShared_free(ri->regstclass);
+                            ri->regstclass = 0;
+                        }
                     }
                 }
                 break;