PATCH: [perl #134034] Assert fail in pattern match

[perl5.git] / regexec.c

/*    regexec.c
 */

/*
 *      One Ring to rule them all, One Ring to find them
 *
 *     [p.v of _The Lord of the Rings_, opening poem]
 *     [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
 *     [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
 */

/* This file contains functions for executing a regular expression.  See
 * also regcomp.c which funnily enough, contains functions for compiling
 * a regular expression.
 *
 * This file is also copied at build time to ext/re/re_exec.c, where
 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
 * This causes the main functions to be compiled under new names and with
 * debugging support added, which makes "use re 'debug'" work.
 */

/* NOTE: this is derived from Henry Spencer's regexp code, and should not
 * confused with the original package (see point 3 below).  Thanks, Henry!
 */

/* Additional note: this code is very heavily munged from Henry's version
 * in places.  In some spots I've traded clarity for efficiency, so don't
 * blame Henry for some of the lack of readability.
 */

/* The names of the functions have been changed from regcomp and
 * regexec to  pregcomp and pregexec in order to avoid conflicts
 * with the POSIX routines of the same names.
*/

#ifdef PERL_EXT_RE_BUILD
#include "re_top.h"
#endif

/*
 * pregcomp and pregexec -- regsub and regerror are not used in perl
 *
 *      Copyright (c) 1986 by University of Toronto.
 *      Written by Henry Spencer.  Not derived from licensed software.
 *
 *      Permission is granted to anyone to use this software for any
 *      purpose on any computer system, and to redistribute it freely,
 *      subject to the following restrictions:
 *
 *      1. The author is not responsible for the consequences of use of
 *              this software, no matter how awful, even if they arise
 *              from defects in it.
 *
 *      2. The origin of this software must not be misrepresented, either
 *              by explicit claim or by omission.
 *
 *      3. Altered versions must be plainly marked as such, and must not
 *              be misrepresented as being the original software.
 *
 ****    Alterations to Henry's code are...
 ****
 ****    Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
 ****    2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
 ****    by Larry Wall and others
 ****
 ****    You may distribute under the terms of either the GNU General Public
 ****    License or the Artistic License, as specified in the README file.
 *
 * Beware that some of this code is subtly aware of the way operator
 * precedence is structured in regular expressions.  Serious changes in
 * regular-expression syntax might require a total rethink.
 */
#include "EXTERN.h"
#define PERL_IN_REGEXEC_C
#include "perl.h"

#ifdef PERL_IN_XSUB_RE
#  include "re_comp.h"
#else
#  include "regcomp.h"
#endif

#include "invlist_inline.h"
#include "unicode_constants.h"

#define B_ON_NON_UTF8_LOCALE_IS_WRONG            \
 "Use of \\b{} or \\B{} for non-UTF-8 locale is wrong.  Assuming a UTF-8 locale"

static const char utf8_locale_required[] =
      "Use of (?[ ]) for non-UTF-8 locale is wrong.  Assuming a UTF-8 locale";

#ifdef DEBUGGING
/* At least one required character in the target string is expressible only in
 * UTF-8. */
static const char non_utf8_target_but_utf8_required[]
                = "Can't match, because target string needs to be in UTF-8\n";
#endif

#define NON_UTF8_TARGET_BUT_UTF8_REQUIRED(target) STMT_START {           \
    DEBUG_EXECUTE_r(Perl_re_printf( aTHX_  "%s", non_utf8_target_but_utf8_required));\
    goto target;                                                         \
} STMT_END

#define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)

#ifndef STATIC
#define STATIC  static
#endif

/*
 * Forwards.
 */

#define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))

#define HOPc(pos,off) \
        (char *)(reginfo->is_utf8_target \
            ? reghop3((U8*)pos, off, \
                    (U8*)(off >= 0 ? reginfo->strend : reginfo->strbeg)) \
            : (U8*)(pos + off))

/* like HOPMAYBE3 but backwards. lim must be +ve. Returns NULL on overshoot */
#define HOPBACK3(pos, off, lim) \
        (reginfo->is_utf8_target                          \
            ? reghopmaybe3((U8*)pos, (SSize_t)0-off, (U8*)(lim)) \
            : (pos - off >= lim)                                 \
                ? (U8*)pos - off                                 \
                : NULL)

#define HOPBACKc(pos, off) ((char*)HOPBACK3(pos, off, reginfo->strbeg))

#define HOP3(pos,off,lim) (reginfo->is_utf8_target  ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
#define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))

/* lim must be +ve. Returns NULL on overshoot */
#define HOPMAYBE3(pos,off,lim) \
        (reginfo->is_utf8_target                        \
            ? reghopmaybe3((U8*)pos, off, (U8*)(lim))   \
            : ((U8*)pos + off <= lim)                   \
                ? (U8*)pos + off                        \
                : NULL)

/* like HOP3, but limits the result to <= lim even for the non-utf8 case.
 * off must be >=0; args should be vars rather than expressions */
#define HOP3lim(pos,off,lim) (reginfo->is_utf8_target \
    ? reghop3((U8*)(pos), off, (U8*)(lim)) \
    : (U8*)((pos + off) > lim ? lim : (pos + off)))
#define HOP3clim(pos,off,lim) ((char*)HOP3lim(pos,off,lim))

#define HOP4(pos,off,llim, rlim) (reginfo->is_utf8_target \
    ? reghop4((U8*)(pos), off, (U8*)(llim), (U8*)(rlim)) \
    : (U8*)(pos + off))
#define HOP4c(pos,off,llim, rlim) ((char*)HOP4(pos,off,llim, rlim))

#define PLACEHOLDER     /* Something for the preprocessor to grab onto */
/* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */

/* for use after a quantifier and before an EXACT-like node -- japhy */
/* it would be nice to rework regcomp.sym to generate this stuff. sigh
 *
 * NOTE that *nothing* that affects backtracking should be in here, specifically
 * VERBS must NOT be included. JUMPABLE is used to determine  if we can ignore a
 * node that is in between two EXACT like nodes when ascertaining what the required
 * "follow" character is. This should probably be moved to regex compile time
 * although it may be done at run time beause of the REF possibility - more
 * investigation required. -- demerphq
*/
#define JUMPABLE(rn) (                                                             \
    OP(rn) == OPEN ||                                                              \
    (OP(rn) == CLOSE &&                                                            \
     !EVAL_CLOSE_PAREN_IS(cur_eval,ARG(rn)) ) ||                                   \
    OP(rn) == EVAL ||                                                              \
    OP(rn) == SUSPEND || OP(rn) == IFMATCH ||                                      \
    OP(rn) == PLUS || OP(rn) == MINMOD ||                                          \
    OP(rn) == KEEPS ||                                                             \
    (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0)                                  \
)
#define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)

#define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )

/*
  Search for mandatory following text node; for lookahead, the text must
  follow but for lookbehind (rn->flags != 0) we skip to the next step.
*/
#define FIND_NEXT_IMPT(rn) STMT_START {                                   \
    while (JUMPABLE(rn)) { \
        const OPCODE type = OP(rn); \
        if (type == SUSPEND || PL_regkind[type] == CURLY) \
            rn = NEXTOPER(NEXTOPER(rn)); \
        else if (type == PLUS) \
            rn = NEXTOPER(rn); \
        else if (type == IFMATCH) \
            rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
        else rn += NEXT_OFF(rn); \
    } \
} STMT_END 

#define SLAB_FIRST(s) (&(s)->states[0])
#define SLAB_LAST(s)  (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])

static void S_setup_eval_state(pTHX_ regmatch_info *const reginfo);
static void S_cleanup_regmatch_info_aux(pTHX_ void *arg);
static regmatch_state * S_push_slab(pTHX);

#define REGCP_PAREN_ELEMS 3
#define REGCP_OTHER_ELEMS 3
#define REGCP_FRAME_ELEMS 1
/* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
 * are needed for the regexp context stack bookkeeping. */

STATIC CHECKPOINT
S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen _pDEPTH)
{
    const int retval = PL_savestack_ix;
    const int paren_elems_to_push =
                (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS;
    const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
    const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
    I32 p;
    GET_RE_DEBUG_FLAGS_DECL;

    PERL_ARGS_ASSERT_REGCPPUSH;

    if (paren_elems_to_push < 0)
        Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0, maxopenparen: %i parenfloor: %i REGCP_PAREN_ELEMS: %u",
                   (int)paren_elems_to_push, (int)maxopenparen,
                   (int)parenfloor, (unsigned)REGCP_PAREN_ELEMS);

    if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
        Perl_croak(aTHX_ "panic: paren_elems_to_push offset %" UVuf
                   " out of range (%lu-%ld)",
                   total_elems,
                   (unsigned long)maxopenparen,
                   (long)parenfloor);

    SSGROW(total_elems + REGCP_FRAME_ELEMS);
    
    DEBUG_BUFFERS_r(
        if ((int)maxopenparen > (int)parenfloor)
            Perl_re_exec_indentf( aTHX_
                "rex=0x%" UVxf " offs=0x%" UVxf ": saving capture indices:\n",
                depth,
                PTR2UV(rex),
                PTR2UV(rex->offs)
            );
    );
    for (p = parenfloor+1; p <= (I32)maxopenparen;  p++) {
/* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
        SSPUSHIV(rex->offs[p].end);
        SSPUSHIV(rex->offs[p].start);
        SSPUSHINT(rex->offs[p].start_tmp);
        DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
            "    \\%" UVuf ": %" IVdf "(%" IVdf ")..%" IVdf "\n",
            depth,
            (UV)p,
            (IV)rex->offs[p].start,
            (IV)rex->offs[p].start_tmp,
            (IV)rex->offs[p].end
        ));
    }
/* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
    SSPUSHINT(maxopenparen);
    SSPUSHINT(rex->lastparen);
    SSPUSHINT(rex->lastcloseparen);
    SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */

    return retval;
}

/* These are needed since we do not localize EVAL nodes: */
#define REGCP_SET(cp)                                           \
    DEBUG_STATE_r(                                              \
        Perl_re_exec_indentf( aTHX_                             \
            "Setting an EVAL scope, savestack=%" IVdf ",\n",    \
            depth, (IV)PL_savestack_ix                          \
        )                                                       \
    );                                                          \
    cp = PL_savestack_ix

#define REGCP_UNWIND(cp)                                        \
    DEBUG_STATE_r(                                              \
        if (cp != PL_savestack_ix)                              \
            Perl_re_exec_indentf( aTHX_                         \
                "Clearing an EVAL scope, savestack=%"           \
                IVdf "..%" IVdf "\n",                           \
                depth, (IV)(cp), (IV)PL_savestack_ix            \
            )                                                   \
    );                                                          \
    regcpblow(cp)

/* set the start and end positions of capture ix */
#define CLOSE_CAPTURE(ix, s, e)                                            \
    rex->offs[ix].start = s;                                               \
    rex->offs[ix].end = e;                                                 \
    if (ix > rex->lastparen)                                               \
        rex->lastparen = ix;                                               \
    rex->lastcloseparen = ix;                                              \
    DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_                            \
        "CLOSE: rex=0x%" UVxf " offs=0x%" UVxf ": \\%" UVuf ": set %" IVdf "..%" IVdf " max: %" UVuf "\n", \
        depth,                                                             \
        PTR2UV(rex),                                                       \
        PTR2UV(rex->offs),                                                 \
        (UV)ix,                                                            \
        (IV)rex->offs[ix].start,                                           \
        (IV)rex->offs[ix].end,                                             \
        (UV)rex->lastparen                                                 \
    ))

#define UNWIND_PAREN(lp, lcp)               \
    DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_  \
        "UNWIND_PAREN: rex=0x%" UVxf " offs=0x%" UVxf ": invalidate (%" UVuf "..%" UVuf "] set lcp: %" UVuf "\n", \
        depth,                              \
        PTR2UV(rex),                        \
        PTR2UV(rex->offs),                  \
        (UV)(lp),                           \
        (UV)(rex->lastparen),               \
        (UV)(lcp)                           \
    ));                                     \
    for (n = rex->lastparen; n > lp; n--)   \
        rex->offs[n].end = -1;              \
    rex->lastparen = n;                     \
    rex->lastcloseparen = lcp;


STATIC void
S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p _pDEPTH)
{
    UV i;
    U32 paren;
    GET_RE_DEBUG_FLAGS_DECL;

    PERL_ARGS_ASSERT_REGCPPOP;

    /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
    i = SSPOPUV;
    assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
    i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
    rex->lastcloseparen = SSPOPINT;
    rex->lastparen = SSPOPINT;
    *maxopenparen_p = SSPOPINT;

    i -= REGCP_OTHER_ELEMS;
    /* Now restore the parentheses context. */
    DEBUG_BUFFERS_r(
        if (i || rex->lastparen + 1 <= rex->nparens)
            Perl_re_exec_indentf( aTHX_
                "rex=0x%" UVxf " offs=0x%" UVxf ": restoring capture indices to:\n",
                depth,
                PTR2UV(rex),
                PTR2UV(rex->offs)
            );
    );
    paren = *maxopenparen_p;
    for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
        SSize_t tmps;
        rex->offs[paren].start_tmp = SSPOPINT;
        rex->offs[paren].start = SSPOPIV;
        tmps = SSPOPIV;
        if (paren <= rex->lastparen)
            rex->offs[paren].end = tmps;
        DEBUG_BUFFERS_r( Perl_re_exec_indentf( aTHX_
            "    \\%" UVuf ": %" IVdf "(%" IVdf ")..%" IVdf "%s\n",
            depth,
            (UV)paren,
            (IV)rex->offs[paren].start,
            (IV)rex->offs[paren].start_tmp,
            (IV)rex->offs[paren].end,
            (paren > rex->lastparen ? "(skipped)" : ""));
        );
        paren--;
    }
#if 1
    /* It would seem that the similar code in regtry()
     * already takes care of this, and in fact it is in
     * a better location to since this code can #if 0-ed out
     * but the code in regtry() is needed or otherwise tests
     * requiring null fields (pat.t#187 and split.t#{13,14}
     * (as of patchlevel 7877)  will fail.  Then again,
     * this code seems to be necessary or otherwise
     * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
     * --jhi updated by dapm */
    for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
        if (i > *maxopenparen_p)
            rex->offs[i].start = -1;
        rex->offs[i].end = -1;
        DEBUG_BUFFERS_r( Perl_re_exec_indentf( aTHX_
            "    \\%" UVuf ": %s   ..-1 undeffing\n",
            depth,
            (UV)i,
            (i > *maxopenparen_p) ? "-1" : "  "
        ));
    }
#endif
}

/* restore the parens and associated vars at savestack position ix,
 * but without popping the stack */

STATIC void
S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p _pDEPTH)
{
    I32 tmpix = PL_savestack_ix;
    PERL_ARGS_ASSERT_REGCP_RESTORE;

    PL_savestack_ix = ix;
    regcppop(rex, maxopenparen_p);
    PL_savestack_ix = tmpix;
}

#define regcpblow(cp) LEAVE_SCOPE(cp)   /* Ignores regcppush()ed data. */

#ifndef PERL_IN_XSUB_RE

bool
Perl_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
{
    /* Returns a boolean as to whether or not 'character' is a member of the
     * Posix character class given by 'classnum' that should be equivalent to a
     * value in the typedef '_char_class_number'.
     *
     * Ideally this could be replaced by a just an array of function pointers
     * to the C library functions that implement the macros this calls.
     * However, to compile, the precise function signatures are required, and
     * these may vary from platform to to platform.  To avoid having to figure
     * out what those all are on each platform, I (khw) am using this method,
     * which adds an extra layer of function call overhead (unless the C
     * optimizer strips it away).  But we don't particularly care about
     * performance with locales anyway. */

    switch ((_char_class_number) classnum) {
        case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character);
        case _CC_ENUM_ALPHA:     return isALPHA_LC(character);
        case _CC_ENUM_ASCII:     return isASCII_LC(character);
        case _CC_ENUM_BLANK:     return isBLANK_LC(character);
        case _CC_ENUM_CASED:     return    isLOWER_LC(character)
                                        || isUPPER_LC(character);
        case _CC_ENUM_CNTRL:     return isCNTRL_LC(character);
        case _CC_ENUM_DIGIT:     return isDIGIT_LC(character);
        case _CC_ENUM_GRAPH:     return isGRAPH_LC(character);
        case _CC_ENUM_LOWER:     return isLOWER_LC(character);
        case _CC_ENUM_PRINT:     return isPRINT_LC(character);
        case _CC_ENUM_PUNCT:     return isPUNCT_LC(character);
        case _CC_ENUM_SPACE:     return isSPACE_LC(character);
        case _CC_ENUM_UPPER:     return isUPPER_LC(character);
        case _CC_ENUM_WORDCHAR:  return isWORDCHAR_LC(character);
        case _CC_ENUM_XDIGIT:    return isXDIGIT_LC(character);
        default:    /* VERTSPACE should never occur in locales */
            Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
    }

    NOT_REACHED; /* NOTREACHED */
    return FALSE;
}

#endif

PERL_STATIC_INLINE I32
S_foldEQ_latin1_s2_folded(const char *s1, const char *s2, I32 len)
{
    /* Compare non-UTF-8 using Unicode (Latin1) semantics.  s2 must already be
     * folded.  Works on all folds representable without UTF-8, except for
     * LATIN_SMALL_LETTER_SHARP_S, and does not check for this.  Nor does it
     * check that the strings each have at least 'len' characters.
     *
     * There is almost an identical API function where s2 need not be folded:
     * Perl_foldEQ_latin1() */

    const U8 *a = (const U8 *)s1;
    const U8 *b = (const U8 *)s2;

    PERL_ARGS_ASSERT_FOLDEQ_LATIN1_S2_FOLDED;

    assert(len >= 0);

    while (len--) {
        assert(! isUPPER_L1(*b));
        if (toLOWER_L1(*a) != *b) {
            return 0;
        }
        a++, b++;
    }
    return 1;
}

STATIC bool
S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character, const U8* e)
{
    /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded
     * 'character' is a member of the Posix character class given by 'classnum'
     * that should be equivalent to a value in the typedef
     * '_char_class_number'.
     *
     * This just calls isFOO_lc on the code point for the character if it is in
     * the range 0-255.  Outside that range, all characters use Unicode
     * rules, ignoring any locale.  So use the Unicode function if this class
     * requires an inversion list, and use the Unicode macro otherwise. */

    dVAR;

    PERL_ARGS_ASSERT_ISFOO_UTF8_LC;

    if (UTF8_IS_INVARIANT(*character)) {
        return isFOO_lc(classnum, *character);
    }
    else if (UTF8_IS_DOWNGRADEABLE_START(*character)) {
        return isFOO_lc(classnum,
                        EIGHT_BIT_UTF8_TO_NATIVE(*character, *(character + 1)));
    }

    _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(character, e);

    switch ((_char_class_number) classnum) {
        case _CC_ENUM_SPACE:     return is_XPERLSPACE_high(character);
        case _CC_ENUM_BLANK:     return is_HORIZWS_high(character);
        case _CC_ENUM_XDIGIT:    return is_XDIGIT_high(character);
        case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character);
        default:
            return _invlist_contains_cp(PL_XPosix_ptrs[classnum],
                                        utf8_to_uvchr_buf(character, e, NULL));
    }

    return FALSE; /* Things like CNTRL are always below 256 */
}

STATIC U8 *
S_find_span_end(U8 * s, const U8 * send, const U8 span_byte)
{
    /* Returns the position of the first byte in the sequence between 's' and
     * 'send-1' inclusive that isn't 'span_byte'; returns 'send' if none found.
     * */

    PERL_ARGS_ASSERT_FIND_SPAN_END;

    assert(send >= s);

    if ((STRLEN) (send - s) >= PERL_WORDSIZE
                          + PERL_WORDSIZE * PERL_IS_SUBWORD_ADDR(s)
                          - (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK))
    {
        PERL_UINTMAX_T span_word;

        /* Process per-byte until reach word boundary.  XXX This loop could be
         * eliminated if we knew that this platform had fast unaligned reads */
        while (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK) {
            if (*s != span_byte) {
                return s;
            }
            s++;
        }

        /* Create a word filled with the bytes we are spanning */
        span_word = PERL_COUNT_MULTIPLIER * span_byte;

        /* Process per-word as long as we have at least a full word left */
        do {

            /* Keep going if the whole word is composed of 'span_byte's */
            if ((* (PERL_UINTMAX_T *) s) == span_word)  {
                s += PERL_WORDSIZE;
                continue;
            }

            /* Here, at least one byte in the word isn't 'span_byte'. */

#ifdef EBCDIC

            break;

#else

            /* This xor leaves 1 bits only in those non-matching bytes */
            span_word ^= * (PERL_UINTMAX_T *) s;

            /* Make sure the upper bit of each non-matching byte is set.  This
             * makes each such byte look like an ASCII platform variant byte */
            span_word |= span_word << 1;
            span_word |= span_word << 2;
            span_word |= span_word << 4;

            /* That reduces the problem to what this function solves */
            return s + _variant_byte_number(span_word);

#endif

        } while (s + PERL_WORDSIZE <= send);
    }

    /* Process the straggler bytes beyond the final word boundary */
    while (s < send) {
        if (*s != span_byte) {
            return s;
        }
        s++;
    }

    return s;
}

STATIC U8 *
S_find_next_masked(U8 * s, const U8 * send, const U8 byte, const U8 mask)
{
    /* Returns the position of the first byte in the sequence between 's'
     * and 'send-1' inclusive that when ANDed with 'mask' yields 'byte';
     * returns 'send' if none found.  It uses word-level operations instead of
     * byte to speed up the process */

    PERL_ARGS_ASSERT_FIND_NEXT_MASKED;

    assert(send >= s);
    assert((byte & mask) == byte);

#ifndef EBCDIC

    if ((STRLEN) (send - s) >= PERL_WORDSIZE
                          + PERL_WORDSIZE * PERL_IS_SUBWORD_ADDR(s)
                          - (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK))
    {
        PERL_UINTMAX_T word, mask_word;

        while (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK) {
            if (((*s) & mask) == byte) {
                return s;
            }
            s++;
        }

        word      = PERL_COUNT_MULTIPLIER * byte;
        mask_word = PERL_COUNT_MULTIPLIER * mask;

        do {
            PERL_UINTMAX_T masked = (* (PERL_UINTMAX_T *) s) & mask_word;

            /* If 'masked' contains bytes with the bit pattern of 'byte' within
             * it, xoring with 'word' will leave each of the 8 bits in such
             * bytes be 0, and no byte containing any other bit pattern will be
             * 0. */
            masked ^= word;

            /* This causes the most significant bit to be set to 1 for any
             * bytes in the word that aren't completely 0 */
            masked |= masked << 1;
            masked |= masked << 2;
            masked |= masked << 4;

            /* The msbits are the same as what marks a byte as variant, so we
             * can use this mask.  If all msbits are 1, the word doesn't
             * contain 'byte' */
            if ((masked & PERL_VARIANTS_WORD_MASK) == PERL_VARIANTS_WORD_MASK) {
                s += PERL_WORDSIZE;
                continue;
            }

            /* Here, the msbit of bytes in the word that aren't 'byte' are 1,
             * and any that are, are 0.  Complement and re-AND to swap that */
            masked = ~ masked;
            masked &= PERL_VARIANTS_WORD_MASK;

            /* This reduces the problem to that solved by this function */
            s += _variant_byte_number(masked);
            return s;

        } while (s + PERL_WORDSIZE <= send);
    }

#endif

    while (s < send) {
        if (((*s) & mask) == byte) {
            return s;
        }
        s++;
    }

    return s;
}

STATIC U8 *
S_find_span_end_mask(U8 * s, const U8 * send, const U8 span_byte, const U8 mask)
{
    /* Returns the position of the first byte in the sequence between 's' and
     * 'send-1' inclusive that when ANDed with 'mask' isn't 'span_byte'.
     * 'span_byte' should have been ANDed with 'mask' in the call of this
     * function.  Returns 'send' if none found.  Works like find_span_end(),
     * except for the AND */

    PERL_ARGS_ASSERT_FIND_SPAN_END_MASK;

    assert(send >= s);
    assert((span_byte & mask) == span_byte);

    if ((STRLEN) (send - s) >= PERL_WORDSIZE
                          + PERL_WORDSIZE * PERL_IS_SUBWORD_ADDR(s)
                          - (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK))
    {
        PERL_UINTMAX_T span_word, mask_word;

        while (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK) {
            if (((*s) & mask) != span_byte) {
                return s;
            }
            s++;
        }

        span_word = PERL_COUNT_MULTIPLIER * span_byte;
        mask_word = PERL_COUNT_MULTIPLIER * mask;

        do {
            PERL_UINTMAX_T masked = (* (PERL_UINTMAX_T *) s) & mask_word;

            if (masked == span_word) {
                s += PERL_WORDSIZE;
                continue;
            }

#ifdef EBCDIC

            break;

#else

            masked ^= span_word;
            masked |= masked << 1;
            masked |= masked << 2;
            masked |= masked << 4;
            return s + _variant_byte_number(masked);

#endif

        } while (s + PERL_WORDSIZE <= send);
    }

    while (s < send) {
        if (((*s) & mask) != span_byte) {
            return s;
        }
        s++;
    }

    return s;
}

/*
 * pregexec and friends
 */

#ifndef PERL_IN_XSUB_RE
/*
 - pregexec - match a regexp against a string
 */
I32
Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
         char *strbeg, SSize_t minend, SV *screamer, U32 nosave)
/* stringarg: the point in the string at which to begin matching */
/* strend:    pointer to null at end of string */
/* strbeg:    real beginning of string */
/* minend:    end of match must be >= minend bytes after stringarg. */
/* screamer:  SV being matched: only used for utf8 flag, pos() etc; string
 *            itself is accessed via the pointers above */
/* nosave:    For optimizations. */
{
    PERL_ARGS_ASSERT_PREGEXEC;

    return
        regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
                      nosave ? 0 : REXEC_COPY_STR);
}
#endif



/* re_intuit_start():
 *
 * Based on some optimiser hints, try to find the earliest position in the
 * string where the regex could match.
 *
 *   rx:     the regex to match against
 *   sv:     the SV being matched: only used for utf8 flag; the string
 *           itself is accessed via the pointers below. Note that on
 *           something like an overloaded SV, SvPOK(sv) may be false
 *           and the string pointers may point to something unrelated to
 *           the SV itself.
 *   strbeg: real beginning of string
 *   strpos: the point in the string at which to begin matching
 *   strend: pointer to the byte following the last char of the string
 *   flags   currently unused; set to 0
 *   data:   currently unused; set to NULL
 *
 * The basic idea of re_intuit_start() is to use some known information
 * about the pattern, namely:
 *
 *   a) the longest known anchored substring (i.e. one that's at a
 *      constant offset from the beginning of the pattern; but not
 *      necessarily at a fixed offset from the beginning of the
 *      string);
 *   b) the longest floating substring (i.e. one that's not at a constant
 *      offset from the beginning of the pattern);
 *   c) Whether the pattern is anchored to the string; either
 *      an absolute anchor: /^../, or anchored to \n: /^.../m,
 *      or anchored to pos(): /\G/;
 *   d) A start class: a real or synthetic character class which
 *      represents which characters are legal at the start of the pattern;
 *
 * to either quickly reject the match, or to find the earliest position
 * within the string at which the pattern might match, thus avoiding
 * running the full NFA engine at those earlier locations, only to
 * eventually fail and retry further along.
 *
 * Returns NULL if the pattern can't match, or returns the address within
 * the string which is the earliest place the match could occur.
 *
 * The longest of the anchored and floating substrings is called 'check'
 * and is checked first. The other is called 'other' and is checked
 * second. The 'other' substring may not be present.  For example,
 *
 *    /(abc|xyz)ABC\d{0,3}DEFG/
 *
 * will have
 *
 *   check substr (float)    = "DEFG", offset 6..9 chars
 *   other substr (anchored) = "ABC",  offset 3..3 chars
 *   stclass = [ax]
 *
 * Be aware that during the course of this function, sometimes 'anchored'
 * refers to a substring being anchored relative to the start of the
 * pattern, and sometimes to the pattern itself being anchored relative to
 * the string. For example:
 *
 *   /\dabc/:   "abc" is anchored to the pattern;
 *   /^\dabc/:  "abc" is anchored to the pattern and the string;
 *   /\d+abc/:  "abc" is anchored to neither the pattern nor the string;
 *   /^\d+abc/: "abc" is anchored to neither the pattern nor the string,
 *                    but the pattern is anchored to the string.
 */

char *
Perl_re_intuit_start(pTHX_
                    REGEXP * const rx,
                    SV *sv,
                    const char * const strbeg,
                    char *strpos,
                    char *strend,
                    const U32 flags,
                    re_scream_pos_data *data)
{
    struct regexp *const prog = ReANY(rx);
    SSize_t start_shift = prog->check_offset_min;
    /* Should be nonnegative! */
    SSize_t end_shift   = 0;
    /* current lowest pos in string where the regex can start matching */
    char *rx_origin = strpos;
    SV *check;
    const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
    U8   other_ix = 1 - prog->substrs->check_ix;
    bool ml_anch = 0;
    char *other_last = strpos;/* latest pos 'other' substr already checked to */
    char *check_at = NULL;              /* check substr found at this pos */
    const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
    RXi_GET_DECL(prog,progi);
    regmatch_info reginfo_buf;  /* create some info to pass to find_byclass */
    regmatch_info *const reginfo = &reginfo_buf;
    GET_RE_DEBUG_FLAGS_DECL;

    PERL_ARGS_ASSERT_RE_INTUIT_START;
    PERL_UNUSED_ARG(flags);
    PERL_UNUSED_ARG(data);

    DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
                "Intuit: trying to determine minimum start position...\n"));

    /* for now, assume that all substr offsets are positive. If at some point
     * in the future someone wants to do clever things with lookbehind and
     * -ve offsets, they'll need to fix up any code in this function
     * which uses these offsets. See the thread beginning
     * <20140113145929.GF27210@iabyn.com>
     */
    assert(prog->substrs->data[0].min_offset >= 0);
    assert(prog->substrs->data[0].max_offset >= 0);
    assert(prog->substrs->data[1].min_offset >= 0);
    assert(prog->substrs->data[1].max_offset >= 0);
    assert(prog->substrs->data[2].min_offset >= 0);
    assert(prog->substrs->data[2].max_offset >= 0);

    /* for now, assume that if both present, that the floating substring
     * doesn't start before the anchored substring.
     * If you break this assumption (e.g. doing better optimisations
     * with lookahead/behind), then you'll need to audit the code in this
     * function carefully first
     */
    assert(
            ! (  (prog->anchored_utf8 || prog->anchored_substr)
              && (prog->float_utf8    || prog->float_substr))
           || (prog->float_min_offset >= prog->anchored_offset));

    /* byte rather than char calculation for efficiency. It fails
     * to quickly reject some cases that can't match, but will reject
     * them later after doing full char arithmetic */
    if (prog->minlen > strend - strpos) {
        DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
                              "  String too short...\n"));
        goto fail;
    }

    RXp_MATCH_UTF8_set(prog, utf8_target);
    reginfo->is_utf8_target = cBOOL(utf8_target);
    reginfo->info_aux = NULL;
    reginfo->strbeg = strbeg;
    reginfo->strend = strend;
    reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
    reginfo->intuit = 1;
    /* not actually used within intuit, but zero for safety anyway */
    reginfo->poscache_maxiter = 0;

    if (utf8_target) {
        if ((!prog->anchored_utf8 && prog->anchored_substr)
                || (!prog->float_utf8 && prog->float_substr))
            to_utf8_substr(prog);
        check = prog->check_utf8;
    } else {
        if (!prog->check_substr && prog->check_utf8) {
            if (! to_byte_substr(prog)) {
                NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
            }
        }
        check = prog->check_substr;
    }

    /* dump the various substring data */
    DEBUG_OPTIMISE_MORE_r({
        int i;
        for (i=0; i<=2; i++) {
            SV *sv = (utf8_target ? prog->substrs->data[i].utf8_substr
                                  : prog->substrs->data[i].substr);
            if (!sv)
                continue;

            Perl_re_printf( aTHX_
                "  substrs[%d]: min=%" IVdf " max=%" IVdf " end shift=%" IVdf
                " useful=%" IVdf " utf8=%d [%s]\n",
                i,
                (IV)prog->substrs->data[i].min_offset,
                (IV)prog->substrs->data[i].max_offset,
                (IV)prog->substrs->data[i].end_shift,
                BmUSEFUL(sv),
                utf8_target ? 1 : 0,
                SvPEEK(sv));
        }
    });

    if (prog->intflags & PREGf_ANCH) { /* Match at \G, beg-of-str or after \n */

        /* ml_anch: check after \n?
         *
         * A note about PREGf_IMPLICIT: on an un-anchored pattern beginning
         * with /.*.../, these flags will have been added by the
         * compiler:
         *   /.*abc/, /.*abc/m:  PREGf_IMPLICIT | PREGf_ANCH_MBOL
         *   /.*abc/s:           PREGf_IMPLICIT | PREGf_ANCH_SBOL
         */
        ml_anch =      (prog->intflags & PREGf_ANCH_MBOL)
                   && !(prog->intflags & PREGf_IMPLICIT);

        if (!ml_anch && !(prog->intflags & PREGf_IMPLICIT)) {
            /* we are only allowed to match at BOS or \G */

            /* trivially reject if there's a BOS anchor and we're not at BOS.
             *
             * Note that we don't try to do a similar quick reject for
             * \G, since generally the caller will have calculated strpos
             * based on pos() and gofs, so the string is already correctly
             * anchored by definition; and handling the exceptions would
             * be too fiddly (e.g. REXEC_IGNOREPOS).
             */
            if (   strpos != strbeg
                && (prog->intflags & PREGf_ANCH_SBOL))
            {
                DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
                                "  Not at start...\n"));
                goto fail;
            }

            /* in the presence of an anchor, the anchored (relative to the
             * start of the regex) substr must also be anchored relative
             * to strpos. So quickly reject if substr isn't found there.
             * This works for \G too, because the caller will already have
             * subtracted gofs from pos, and gofs is the offset from the
             * \G to the start of the regex. For example, in /.abc\Gdef/,
             * where substr="abcdef", pos()=3, gofs=4, offset_min=1:
             * caller will have set strpos=pos()-4; we look for the substr
             * at position pos()-4+1, which lines up with the "a" */

            if (prog->check_offset_min == prog->check_offset_max) {
                /* Substring at constant offset from beg-of-str... */
                SSize_t slen = SvCUR(check);
                char *s = HOP3c(strpos, prog->check_offset_min, strend);
            
                DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
                    "  Looking for check substr at fixed offset %" IVdf "...\n",
                    (IV)prog->check_offset_min));

                if (SvTAIL(check)) {
                    /* In this case, the regex is anchored at the end too.
                     * Unless it's a multiline match, the lengths must match
                     * exactly, give or take a \n.  NB: slen >= 1 since
                     * the last char of check is \n */
                    if (!multiline
                        && (   strend - s > slen
                            || strend - s < slen - 1
                            || (strend - s == slen && strend[-1] != '\n')))
                    {
                        DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
                                            "  String too long...\n"));
                        goto fail_finish;
                    }
                    /* Now should match s[0..slen-2] */
                    slen--;
                }
                if (slen && (strend - s < slen
                    || *SvPVX_const(check) != *s
                    || (slen > 1 && (memNE(SvPVX_const(check), s, slen)))))
                {
                    DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
                                    "  String not equal...\n"));
                    goto fail_finish;
                }

                check_at = s;
                goto success_at_start;
            }
        }
    }

    end_shift = prog->check_end_shift;

#ifdef DEBUGGING        /* 7/99: reports of failure (with the older version) */
    if (end_shift < 0)
        Perl_croak(aTHX_ "panic: end_shift: %" IVdf " pattern:\n%s\n ",
                   (IV)end_shift, RX_PRECOMP(rx));
#endif

  restart:
    
    /* This is the (re)entry point of the main loop in this function.
     * The goal of this loop is to:
     * 1) find the "check" substring in the region rx_origin..strend
     *    (adjusted by start_shift / end_shift). If not found, reject
     *    immediately.
     * 2) If it exists, look for the "other" substr too if defined; for
     *    example, if the check substr maps to the anchored substr, then
     *    check the floating substr, and vice-versa. If not found, go
     *    back to (1) with rx_origin suitably incremented.
     * 3) If we find an rx_origin position that doesn't contradict
     *    either of the substrings, then check the possible additional
     *    constraints on rx_origin of /^.../m or a known start class.
     *    If these fail, then depending on which constraints fail, jump
     *    back to here, or to various other re-entry points further along
     *    that skip some of the first steps.
     * 4) If we pass all those tests, update the BmUSEFUL() count on the
     *    substring. If the start position was determined to be at the
     *    beginning of the string  - so, not rejected, but not optimised,
     *    since we have to run regmatch from position 0 - decrement the
     *    BmUSEFUL() count. Otherwise increment it.
     */


    /* first, look for the 'check' substring */

    {
        U8* start_point;
        U8* end_point;

        DEBUG_OPTIMISE_MORE_r({
            Perl_re_printf( aTHX_
                "  At restart: rx_origin=%" IVdf " Check offset min: %" IVdf
                " Start shift: %" IVdf " End shift %" IVdf
                " Real end Shift: %" IVdf "\n",
                (IV)(rx_origin - strbeg),
                (IV)prog->check_offset_min,
                (IV)start_shift,
                (IV)end_shift,
                (IV)prog->check_end_shift);
        });
        
        end_point = HOPBACK3(strend, end_shift, rx_origin);
        if (!end_point)
            goto fail_finish;
        start_point = HOPMAYBE3(rx_origin, start_shift, end_point);
        if (!start_point)
            goto fail_finish;


        /* If the regex is absolutely anchored to either the start of the
         * string (SBOL) or to pos() (ANCH_GPOS), then
         * check_offset_max represents an upper bound on the string where
         * the substr could start. For the ANCH_GPOS case, we assume that
         * the caller of intuit will have already set strpos to
         * pos()-gofs, so in this case strpos + offset_max will still be
         * an upper bound on the substr.
         */
        if (!ml_anch
            && prog->intflags & PREGf_ANCH
            && prog->check_offset_max != SSize_t_MAX)
        {
            SSize_t check_len = SvCUR(check) - !!SvTAIL(check);
            const char * const anchor =
                        (prog->intflags & PREGf_ANCH_GPOS ? strpos : strbeg);
            SSize_t targ_len = (char*)end_point - anchor;

            if (check_len > targ_len) {
                DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
                              "Target string too short to match required substring...\n"));
                goto fail_finish;
            }

            /* do a bytes rather than chars comparison. It's conservative;
             * so it skips doing the HOP if the result can't possibly end
             * up earlier than the old value of end_point.
             */
            assert(anchor + check_len <= (char *)end_point);
            if (prog->check_offset_max + check_len < targ_len) {
                end_point = HOP3lim((U8*)anchor,
                                prog->check_offset_max,
                                end_point - check_len
                            )
                            + check_len;
                if (end_point < start_point)
                    goto fail_finish;
            }
        }

        check_at = fbm_instr( start_point, end_point,
                      check, multiline ? FBMrf_MULTILINE : 0);

        DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
            "  doing 'check' fbm scan, [%" IVdf "..%" IVdf "] gave %" IVdf "\n",
            (IV)((char*)start_point - strbeg),
            (IV)((char*)end_point   - strbeg),
            (IV)(check_at ? check_at - strbeg : -1)
        ));

        /* Update the count-of-usability, remove useless subpatterns,
            unshift s.  */

        DEBUG_EXECUTE_r({
            RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
                SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
            Perl_re_printf( aTHX_  "  %s %s substr %s%s%s",
                              (check_at ? "Found" : "Did not find"),
                (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
                    ? "anchored" : "floating"),
                quoted,
                RE_SV_TAIL(check),
                (check_at ? " at offset " : "...\n") );
        });

        if (!check_at)
            goto fail_finish;
        /* set rx_origin to the minimum position where the regex could start
         * matching, given the constraint of the just-matched check substring.
         * But don't set it lower than previously.
         */

        if (check_at - rx_origin > prog->check_offset_max)
            rx_origin = HOP3c(check_at, -prog->check_offset_max, rx_origin);
        /* Finish the diagnostic message */
        DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
            "%ld (rx_origin now %" IVdf ")...\n",
            (long)(check_at - strbeg),
            (IV)(rx_origin - strbeg)
        ));
    }


    /* now look for the 'other' substring if defined */

    if (prog->substrs->data[other_ix].utf8_substr
        || prog->substrs->data[other_ix].substr)
    {
        /* Take into account the "other" substring. */
        char *last, *last1;
        char *s;
        SV* must;
        struct reg_substr_datum *other;

      do_other_substr:
        other = &prog->substrs->data[other_ix];
        if (!utf8_target && !other->substr) {
            if (!to_byte_substr(prog)) {
                NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
            }
        }

        /* if "other" is anchored:
         * we've previously found a floating substr starting at check_at.
         * This means that the regex origin must lie somewhere
         * between min (rx_origin): HOP3(check_at, -check_offset_max)
         * and max:                 HOP3(check_at, -check_offset_min)
         * (except that min will be >= strpos)
         * So the fixed  substr must lie somewhere between
         *  HOP3(min, anchored_offset)
         *  HOP3(max, anchored_offset) + SvCUR(substr)
         */

        /* if "other" is floating
         * Calculate last1, the absolute latest point where the
         * floating substr could start in the string, ignoring any
         * constraints from the earlier fixed match. It is calculated
         * as follows:
         *
         * strend - prog->minlen (in chars) is the absolute latest
         * position within the string where the origin of the regex
         * could appear. The latest start point for the floating
         * substr is float_min_offset(*) on from the start of the
         * regex.  last1 simply combines thee two offsets.
         *
         * (*) You might think the latest start point should be
         * float_max_offset from the regex origin, and technically
         * you'd be correct. However, consider
         *    /a\d{2,4}bcd\w/
         * Here, float min, max are 3,5 and minlen is 7.
         * This can match either
         *    /a\d\dbcd\w/
         *    /a\d\d\dbcd\w/
         *    /a\d\d\d\dbcd\w/
         * In the first case, the regex matches minlen chars; in the
         * second, minlen+1, in the third, minlen+2.
         * In the first case, the floating offset is 3 (which equals
         * float_min), in the second, 4, and in the third, 5 (which
         * equals float_max). In all cases, the floating string bcd
         * can never start more than 4 chars from the end of the
         * string, which equals minlen - float_min. As the substring
         * starts to match more than float_min from the start of the
         * regex, it makes the regex match more than minlen chars,
         * and the two cancel each other out. So we can always use
         * float_min - minlen, rather than float_max - minlen for the
         * latest position in the string.
         *
         * Note that -minlen + float_min_offset is equivalent (AFAIKT)
         * to CHR_SVLEN(must) - !!SvTAIL(must) + prog->float_end_shift
         */

        assert(prog->minlen >= other->min_offset);
        last1 = HOP3c(strend,
                        other->min_offset - prog->minlen, strbeg);

        if (other_ix) {/* i.e. if (other-is-float) */
            /* last is the latest point where the floating substr could
             * start, *given* any constraints from the earlier fixed
             * match. This constraint is that the floating string starts
             * <= float_max_offset chars from the regex origin (rx_origin).
             * If this value is less than last1, use it instead.
             */
            assert(rx_origin <= last1);
            last =
                /* this condition handles the offset==infinity case, and
                 * is a short-cut otherwise. Although it's comparing a
                 * byte offset to a char length, it does so in a safe way,
                 * since 1 char always occupies 1 or more bytes,
                 * so if a string range is  (last1 - rx_origin) bytes,
                 * it will be less than or equal to  (last1 - rx_origin)
                 * chars; meaning it errs towards doing the accurate HOP3
                 * rather than just using last1 as a short-cut */
                (last1 - rx_origin) < other->max_offset
                    ? last1
                    : (char*)HOP3lim(rx_origin, other->max_offset, last1);
        }
        else {
            assert(strpos + start_shift <= check_at);
            last = HOP4c(check_at, other->min_offset - start_shift,
                        strbeg, strend);
        }

        s = HOP3c(rx_origin, other->min_offset, strend);
        if (s < other_last)     /* These positions already checked */
            s = other_last;

        must = utf8_target ? other->utf8_substr : other->substr;
        assert(SvPOK(must));
        {
            char *from = s;
            char *to   = last + SvCUR(must) - (SvTAIL(must)!=0);

            if (to > strend)
                to = strend;
            if (from > to) {
                s = NULL;
                DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
                    "  skipping 'other' fbm scan: %" IVdf " > %" IVdf "\n",
                    (IV)(from - strbeg),
                    (IV)(to   - strbeg)
                ));
            }
            else {
                s = fbm_instr(
                    (unsigned char*)from,
                    (unsigned char*)to,
                    must,
                    multiline ? FBMrf_MULTILINE : 0
                );
                DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
                    "  doing 'other' fbm scan, [%" IVdf "..%" IVdf "] gave %" IVdf "\n",
                    (IV)(from - strbeg),
                    (IV)(to   - strbeg),
                    (IV)(s ? s - strbeg : -1)
                ));
            }
        }

        DEBUG_EXECUTE_r({
            RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
                SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
            Perl_re_printf( aTHX_  "  %s %s substr %s%s",
                s ? "Found" : "Contradicts",
                other_ix ? "floating" : "anchored",
                quoted, RE_SV_TAIL(must));
        });


        if (!s) {
            /* last1 is latest possible substr location. If we didn't
             * find it before there, we never will */
            if (last >= last1) {
                DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
                                        "; giving up...\n"));
                goto fail_finish;
            }

            /* try to find the check substr again at a later
             * position. Maybe next time we'll find the "other" substr
             * in range too */
            other_last = HOP3c(last, 1, strend) /* highest failure */;
            rx_origin =
                other_ix /* i.e. if other-is-float */
                    ? HOP3c(rx_origin, 1, strend)
                    : HOP4c(last, 1 - other->min_offset, strbeg, strend);
            DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
                "; about to retry %s at offset %ld (rx_origin now %" IVdf ")...\n",
                (other_ix ? "floating" : "anchored"),
                (long)(HOP3c(check_at, 1, strend) - strbeg),
                (IV)(rx_origin - strbeg)
            ));
            goto restart;
        }
        else {
            if (other_ix) { /* if (other-is-float) */
                /* other_last is set to s, not s+1, since its possible for
                 * a floating substr to fail first time, then succeed
                 * second time at the same floating position; e.g.:
                 *     "-AB--AABZ" =~ /\wAB\d*Z/
                 * The first time round, anchored and float match at
                 * "-(AB)--AAB(Z)" then fail on the initial \w character
                 * class. Second time round, they match at "-AB--A(AB)(Z)".
                 */
                other_last = s;
            }
            else {
                rx_origin = HOP3c(s, -other->min_offset, strbeg);
                other_last = HOP3c(s, 1, strend);
            }
            DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
                " at offset %ld (rx_origin now %" IVdf ")...\n",
                  (long)(s - strbeg),
                (IV)(rx_origin - strbeg)
              ));

        }
    }
    else {
        DEBUG_OPTIMISE_MORE_r(
            Perl_re_printf( aTHX_
                "  Check-only match: offset min:%" IVdf " max:%" IVdf
                " check_at:%" IVdf " rx_origin:%" IVdf " rx_origin-check_at:%" IVdf
                " strend:%" IVdf "\n",
                (IV)prog->check_offset_min,
                (IV)prog->check_offset_max,
                (IV)(check_at-strbeg),
                (IV)(rx_origin-strbeg),
                (IV)(rx_origin-check_at),
                (IV)(strend-strbeg)
            )
        );
    }

  postprocess_substr_matches:

    /* handle the extra constraint of /^.../m if present */

    if (ml_anch && rx_origin != strbeg && rx_origin[-1] != '\n') {
        char *s;

        DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
                        "  looking for /^/m anchor"));

        /* we have failed the constraint of a \n before rx_origin.
         * Find the next \n, if any, even if it's beyond the current
         * anchored and/or floating substrings. Whether we should be
         * scanning ahead for the next \n or the next substr is debatable.
         * On the one hand you'd expect rare substrings to appear less
         * often than \n's. On the other hand, searching for \n means
         * we're effectively flipping between check_substr and "\n" on each
         * iteration as the current "rarest" string candidate, which
         * means for example that we'll quickly reject the whole string if
         * hasn't got a \n, rather than trying every substr position
         * first
         */

        s = HOP3c(strend, - prog->minlen, strpos);
        if (s <= rx_origin ||
            ! ( rx_origin = (char *)memchr(rx_origin, '\n', s - rx_origin)))
        {
            DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
                            "  Did not find /%s^%s/m...\n",
                            PL_colors[0], PL_colors[1]));
            goto fail_finish;
        }

        /* earliest possible origin is 1 char after the \n.
         * (since *rx_origin == '\n', it's safe to ++ here rather than
         * HOP(rx_origin, 1)) */
        rx_origin++;

        if (prog->substrs->check_ix == 0  /* check is anchored */
            || rx_origin >= HOP3c(check_at,  - prog->check_offset_min, strpos))
        {
            /* Position contradicts check-string; either because
             * check was anchored (and thus has no wiggle room),
             * or check was float and rx_origin is above the float range */
            DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
                "  Found /%s^%s/m, about to restart lookup for check-string with rx_origin %ld...\n",
                PL_colors[0], PL_colors[1], (long)(rx_origin - strbeg)));
            goto restart;
        }

        /* if we get here, the check substr must have been float,
         * is in range, and we may or may not have had an anchored
         * "other" substr which still contradicts */
        assert(prog->substrs->check_ix); /* check is float */

        if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
            /* whoops, the anchored "other" substr exists, so we still
             * contradict. On the other hand, the float "check" substr
             * didn't contradict, so just retry the anchored "other"
             * substr */
            DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
                "  Found /%s^%s/m, rescanning for anchored from offset %" IVdf " (rx_origin now %" IVdf ")...\n",
                PL_colors[0], PL_colors[1],
                (IV)(rx_origin - strbeg + prog->anchored_offset),
                (IV)(rx_origin - strbeg)
            ));
            goto do_other_substr;
        }

        /* success: we don't contradict the found floating substring
         * (and there's no anchored substr). */
        DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
            "  Found /%s^%s/m with rx_origin %ld...\n",
            PL_colors[0], PL_colors[1], (long)(rx_origin - strbeg)));
    }
    else {
        DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
            "  (multiline anchor test skipped)\n"));
    }

  success_at_start:


    /* if we have a starting character class, then test that extra constraint.
     * (trie stclasses are too expensive to use here, we are better off to
     * leave it to regmatch itself) */

    if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
        const U8* const str = (U8*)STRING(progi->regstclass);

        /* XXX this value could be pre-computed */
        const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
                    ?  (reginfo->is_utf8_pat
                        ? utf8_distance(str + STR_LEN(progi->regstclass), str)
                        : STR_LEN(progi->regstclass))
                    : 1);
        char * endpos;
        char *s;
        /* latest pos that a matching float substr constrains rx start to */
        char *rx_max_float = NULL;

        /* if the current rx_origin is anchored, either by satisfying an
         * anchored substring constraint, or a /^.../m constraint, then we
         * can reject the current origin if the start class isn't found
         * at the current position. If we have a float-only match, then
         * rx_origin is constrained to a range; so look for the start class
         * in that range. if neither, then look for the start class in the
         * whole rest of the string */

        /* XXX DAPM it's not clear what the minlen test is for, and why
         * it's not used in the floating case. Nothing in the test suite
         * causes minlen == 0 here. See <20140313134639.GS12844@iabyn.com>.
         * Here are some old comments, which may or may not be correct:
         *
         *   minlen == 0 is possible if regstclass is \b or \B,
         *   and the fixed substr is ''$.
         *   Since minlen is already taken into account, rx_origin+1 is
         *   before strend; accidentally, minlen >= 1 guaranties no false
         *   positives at rx_origin + 1 even for \b or \B.  But (minlen? 1 :
         *   0) below assumes that regstclass does not come from lookahead...
         *   If regstclass takes bytelength more than 1: If charlength==1, OK.
         *   This leaves EXACTF-ish only, which are dealt with in
         *   find_byclass().
         */

        if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
            endpos = HOP3clim(rx_origin, (prog->minlen ? cl_l : 0), strend);
        else if (prog->float_substr || prog->float_utf8) {
            rx_max_float = HOP3c(check_at, -start_shift, strbeg);
            endpos = HOP3clim(rx_max_float, cl_l, strend);
        }
        else 
            endpos= strend;
                    
        DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
            "  looking for class: start_shift: %" IVdf " check_at: %" IVdf
            " rx_origin: %" IVdf " endpos: %" IVdf "\n",
              (IV)start_shift, (IV)(check_at - strbeg),
              (IV)(rx_origin - strbeg), (IV)(endpos - strbeg)));

        s = find_byclass(prog, progi->regstclass, rx_origin, endpos,
                            reginfo);
        if (!s) {
            if (endpos == strend) {
                DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
                                "  Could not match STCLASS...\n") );
                goto fail;
            }
            DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
                               "  This position contradicts STCLASS...\n") );
            if ((prog->intflags & PREGf_ANCH) && !ml_anch
                        && !(prog->intflags & PREGf_IMPLICIT))
                goto fail;

            /* Contradict one of substrings */
            if (prog->anchored_substr || prog->anchored_utf8) {
                if (prog->substrs->check_ix == 1) { /* check is float */
                    /* Have both, check_string is floating */
                    assert(rx_origin + start_shift <= check_at);
                    if (rx_origin + start_shift != check_at) {
                        /* not at latest position float substr could match:
                         * Recheck anchored substring, but not floating.
                         * The condition above is in bytes rather than
                         * chars for efficiency. It's conservative, in
                         * that it errs on the side of doing 'goto
                         * do_other_substr'. In this case, at worst,
                         * an extra anchored search may get done, but in
                         * practice the extra fbm_instr() is likely to
                         * get skipped anyway. */
                        DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
                            "  about to retry anchored at offset %ld (rx_origin now %" IVdf ")...\n",
                            (long)(other_last - strbeg),
                            (IV)(rx_origin - strbeg)
                        ));
                        goto do_other_substr;
                    }
                }
            }
            else {
                /* float-only */

                if (ml_anch) {
                    /* In the presence of ml_anch, we might be able to
                     * find another \n without breaking the current float
                     * constraint. */

                    /* strictly speaking this should be HOP3c(..., 1, ...),
                     * but since we goto a block of code that's going to
                     * search for the next \n if any, its safe here */
                    rx_origin++;
                    DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
                              "  about to look for /%s^%s/m starting at rx_origin %ld...\n",
                              PL_colors[0], PL_colors[1],
                              (long)(rx_origin - strbeg)) );
                    goto postprocess_substr_matches;
                }

                /* strictly speaking this can never be true; but might
                 * be if we ever allow intuit without substrings */
                if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
                    goto fail;

                rx_origin = rx_max_float;
            }

            /* at this point, any matching substrings have been
             * contradicted. Start again... */

            rx_origin = HOP3c(rx_origin, 1, strend);

            /* uses bytes rather than char calculations for efficiency.
             * It's conservative: it errs on the side of doing 'goto restart',
             * where there is code that does a proper char-based test */
            if (rx_origin + start_shift + end_shift > strend) {
                DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
                                       "  Could not match STCLASS...\n") );
                goto fail;
            }
            DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
                "  about to look for %s substr starting at offset %ld (rx_origin now %" IVdf ")...\n",
                (prog->substrs->check_ix ? "floating" : "anchored"),
                (long)(rx_origin + start_shift - strbeg),
                (IV)(rx_origin - strbeg)
            ));
            goto restart;
        }

        /* Success !!! */

        if (rx_origin != s) {
            DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
                        "  By STCLASS: moving %ld --> %ld\n",
                                  (long)(rx_origin - strbeg), (long)(s - strbeg))
                   );
        }
        else {
            DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
                                  "  Does not contradict STCLASS...\n");
                   );
        }
    }

    /* Decide whether using the substrings helped */

    if (rx_origin != strpos) {
        /* Fixed substring is found far enough so that the match
           cannot start at strpos. */

        DEBUG_EXECUTE_r(Perl_re_printf( aTHX_  "  try at offset...\n"));
        ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr);        /* hooray/5 */
    }
    else {
        /* The found rx_origin position does not prohibit matching at
         * strpos, so calling intuit didn't gain us anything. Decrement
         * the BmUSEFUL() count on the check substring, and if we reach
         * zero, free it.  */
        if (!(prog->intflags & PREGf_NAUGHTY)
            && (utf8_target ? (
                prog->check_utf8                /* Could be deleted already */
                && --BmUSEFUL(prog->check_utf8) < 0
                && (prog->check_utf8 == prog->float_utf8)
            ) : (
                prog->check_substr              /* Could be deleted already */
                && --BmUSEFUL(prog->check_substr) < 0
                && (prog->check_substr == prog->float_substr)
            )))
        {
            /* If flags & SOMETHING - do not do it many times on the same match */
            DEBUG_EXECUTE_r(Perl_re_printf( aTHX_  "  ... Disabling check substring...\n"));
            /* XXX Does the destruction order has to change with utf8_target? */
            SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
            SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
            prog->check_substr = prog->check_utf8 = NULL;       /* disable */
            prog->float_substr = prog->float_utf8 = NULL;       /* clear */
            check = NULL;                       /* abort */
            /* XXXX This is a remnant of the old implementation.  It
                    looks wasteful, since now INTUIT can use many
                    other heuristics. */
            prog->extflags &= ~RXf_USE_INTUIT;
        }
    }

    DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
            "Intuit: %sSuccessfully guessed:%s match at offset %ld\n",
             PL_colors[4], PL_colors[5], (long)(rx_origin - strbeg)) );

    return rx_origin;

  fail_finish:                          /* Substring not found */
    if (prog->check_substr || prog->check_utf8)         /* could be removed already */
        BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
  fail:
    DEBUG_EXECUTE_r(Perl_re_printf( aTHX_  "%sMatch rejected by optimizer%s\n",
                          PL_colors[4], PL_colors[5]));
    return NULL;
}


#define DECL_TRIE_TYPE(scan) \
    const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold,       \
                 trie_utf8_exactfa_fold, trie_latin_utf8_exactfa_fold,              \
                 trie_utf8l, trie_flu8, trie_flu8_latin }                           \
                    trie_type = ((scan->flags == EXACT)                             \
                                 ? (utf8_target ? trie_utf8 : trie_plain)           \
                                 : (scan->flags == EXACTL)                          \
                                    ? (utf8_target ? trie_utf8l : trie_plain)       \
                                    : (scan->flags == EXACTFAA)                     \
                                      ? (utf8_target                                \
                                         ? trie_utf8_exactfa_fold                   \
                                         : trie_latin_utf8_exactfa_fold)            \
                                      : (scan->flags == EXACTFLU8                   \
                                         ? (utf8_target                             \
                                           ? trie_flu8                              \
                                           : trie_flu8_latin)                       \
                                         : (utf8_target                             \
                                           ? trie_utf8_fold                         \
                                           : trie_latin_utf8_fold)))

/* 'uscan' is set to foldbuf, and incremented, so below the end of uscan is
 * 'foldbuf+sizeof(foldbuf)' */
#define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uc_end, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \
STMT_START {                                                                        \
    STRLEN skiplen;                                                                 \
    U8 flags = FOLD_FLAGS_FULL;                                                     \
    switch (trie_type) {                                                            \
    case trie_flu8:                                                                 \
        _CHECK_AND_WARN_PROBLEMATIC_LOCALE;                                         \
        if (UTF8_IS_ABOVE_LATIN1(*uc)) {                                            \
            _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc_end);                     \
        }                                                                           \
        goto do_trie_utf8_fold;                                                     \
    case trie_utf8_exactfa_fold:                                                    \
        flags |= FOLD_FLAGS_NOMIX_ASCII;                                            \
        /* FALLTHROUGH */                                                           \
    case trie_utf8_fold:                                                            \
      do_trie_utf8_fold:                                                            \
        if ( foldlen>0 ) {                                                          \
            uvc = utf8n_to_uvchr( (const U8*) uscan, foldlen, &len, uniflags );     \
            foldlen -= len;                                                         \
            uscan += len;                                                           \
            len=0;                                                                  \
        } else {                                                                    \
            uvc = _toFOLD_utf8_flags( (const U8*) uc, uc_end, foldbuf, &foldlen,    \
                                                                            flags); \
            len = UTF8_SAFE_SKIP(uc, uc_end);                                       \
            skiplen = UVCHR_SKIP( uvc );                                            \
            foldlen -= skiplen;                                                     \
            uscan = foldbuf + skiplen;                                              \
        }                                                                           \
        break;                                                                      \
    case trie_flu8_latin:                                                           \
        _CHECK_AND_WARN_PROBLEMATIC_LOCALE;                                         \
        goto do_trie_latin_utf8_fold;                                               \
    case trie_latin_utf8_exactfa_fold:                                              \
        flags |= FOLD_FLAGS_NOMIX_ASCII;                                            \
        /* FALLTHROUGH */                                                           \
    case trie_latin_utf8_fold:                                                      \
      do_trie_latin_utf8_fold:                                                      \
        if ( foldlen>0 ) {                                                          \
            uvc = utf8n_to_uvchr( (const U8*) uscan, foldlen, &len, uniflags );     \
            foldlen -= len;                                                         \
            uscan += len;                                                           \
            len=0;                                                                  \
        } else {                                                                    \
            len = 1;                                                                \
            uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags);             \
            skiplen = UVCHR_SKIP( uvc );                                            \
            foldlen -= skiplen;                                                     \
            uscan = foldbuf + skiplen;                                              \
        }                                                                           \
        break;                                                                      \
    case trie_utf8l:                                                                \
        _CHECK_AND_WARN_PROBLEMATIC_LOCALE;                                         \
        if (utf8_target && UTF8_IS_ABOVE_LATIN1(*uc)) {                             \
            _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc_end);                     \
        }                                                                           \
        /* FALLTHROUGH */                                                           \
    case trie_utf8:                                                                 \
        uvc = utf8n_to_uvchr( (const U8*) uc, uc_end - uc, &len, uniflags );        \
        break;                                                                      \
    case trie_plain:                                                                \
        uvc = (UV)*uc;                                                              \
        len = 1;                                                                    \
    }                                                                               \
    if (uvc < 256) {                                                                \
        charid = trie->charmap[ uvc ];                                              \
    }                                                                               \
    else {                                                                          \
        charid = 0;                                                                 \
        if (widecharmap) {                                                          \
            SV** const svpp = hv_fetch(widecharmap,                                 \
                        (char*)&uvc, sizeof(UV), 0);                                \
            if (svpp)                                                               \
                charid = (U16)SvIV(*svpp);                                          \
        }                                                                           \
    }                                                                               \
} STMT_END

#define DUMP_EXEC_POS(li,s,doutf8,depth)                    \
    dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \
                startpos, doutf8, depth)

#define REXEC_FBC_SCAN(UTF8, CODE)                          \
    STMT_START {                                            \
        while (s < strend) {                                \
            CODE                                            \
            s += ((UTF8)                                    \
                  ? UTF8_SAFE_SKIP(s, reginfo->strend)      \
                  : 1);                                     \
        }                                                   \
    } STMT_END

#define REXEC_FBC_CLASS_SCAN(UTF8, COND)                    \
    STMT_START {                                            \
        while (s < strend) {                                \
            REXEC_FBC_CLASS_SCAN_GUTS(UTF8, COND)           \
        }                                                   \
    } STMT_END

#define REXEC_FBC_CLASS_SCAN_GUTS(UTF8, COND)                  \
    if (COND) {                                                \
        FBC_CHECK_AND_TRY                                      \
        s += ((UTF8) ? UTF8_SAFE_SKIP(s, reginfo->strend) : 1);\
        previous_occurrence_end = s;                           \
    }                                                          \
    else {                                                     \
        s += ((UTF8) ? UTF8SKIP(s) : 1);                       \
    }

#define REXEC_FBC_CSCAN(CONDUTF8,COND)                         \
    if (utf8_target) {                                         \
        REXEC_FBC_CLASS_SCAN(1, CONDUTF8);                     \
    }                                                          \
    else {                                                     \
        REXEC_FBC_CLASS_SCAN(0, COND);                         \
    }

/* We keep track of where the next character should start after an occurrence
 * of the one we're looking for.  Knowing that, we can see right away if the
 * next occurrence is adjacent to the previous.  When 'doevery' is FALSE, we
 * don't accept the 2nd and succeeding adjacent occurrences */
#define FBC_CHECK_AND_TRY                                           \
        if (   (   doevery                                          \
                || s != previous_occurrence_end)                    \
            && (   reginfo->intuit                                  \
                || (s <= reginfo->strend && regtry(reginfo, &s))))  \
        {                                                           \
            goto got_it;                                            \
        }


/* This differs from the above macros in that it calls a function which returns
 * the next occurrence of the thing being looked for in 's'; and 'strend' if
 * there is no such occurrence. */
#define REXEC_FBC_FIND_NEXT_SCAN(UTF8, f)                   \
    while (s < strend) {                                    \
        s = (f);                                            \
        if (s >= strend) {                                  \
            break;                                          \
        }                                                   \
                                                            \
        FBC_CHECK_AND_TRY                                   \
        s += (UTF8) ? UTF8SKIP(s) : 1;                      \
        previous_occurrence_end = s;                        \
    }

/* This differs from the above macros in that it is passed a single byte that
 * is known to begin the next occurrence of the thing being looked for in 's'.
 * It does a memchr to find the next occurrence of 'byte', before trying 'COND'
 * at that position. */
#define REXEC_FBC_FIND_NEXT_UTF8_BYTE_SCAN(byte, COND)      \
    while (s < strend) {                                    \
        s = (char *) memchr(s, byte, strend -s);            \
        if (s == NULL) {                                    \
            s = (char *) strend;                            \
            break;                                          \
        }                                                   \
                                                            \
        if (COND) {                                         \
            FBC_CHECK_AND_TRY                               \
            s += UTF8_SAFE_SKIP(s, reginfo->strend);        \
            previous_occurrence_end = s;                    \
        }                                                   \
        else {                                              \
            s += UTF8SKIP(s);                               \
        }                                                   \
    }

/* The three macros below are slightly different versions of the same logic.
 *
 * The first is for /a and /aa when the target string is UTF-8.  This can only
 * match ascii, but it must advance based on UTF-8.   The other two handle the
 * non-UTF-8 and the more generic UTF-8 cases.   In all three, we are looking
 * for the boundary (or non-boundary) between a word and non-word character.
 * The utf8 and non-utf8 cases have the same logic, but the details must be
 * different.  Find the "wordness" of the character just prior to this one, and
 * compare it with the wordness of this one.  If they differ, we have a
 * boundary.  At the beginning of the string, pretend that the previous
 * character was a new-line.
 *
 * All these macros uncleanly have side-effects with each other and outside
 * variables.  So far it's been too much trouble to clean-up
 *
 * TEST_NON_UTF8 is the macro or function to call to test if its byte input is
 *               a word character or not.
 * IF_SUCCESS    is code to do if it finds that we are at a boundary between
 *               word/non-word
 * IF_FAIL       is code to do if we aren't at a boundary between word/non-word
 *
 * Exactly one of the two IF_FOO parameters is a no-op, depending on whether we
 * are looking for a boundary or for a non-boundary.  If we are looking for a
 * boundary, we want IF_FAIL to be the no-op, and for IF_SUCCESS to go out and
 * see if this tentative match actually works, and if so, to quit the loop
 * here.  And vice-versa if we are looking for a non-boundary.
 *
 * 'tmp' below in the next three macros in the REXEC_FBC_SCAN and
 * REXEC_FBC_SCAN loops is a loop invariant, a bool giving the return of
 * TEST_NON_UTF8(s-1).  To see this, note that that's what it is defined to be
 * at entry to the loop, and to get to the IF_FAIL branch, tmp must equal
 * TEST_NON_UTF8(s), and in the opposite branch, IF_SUCCESS, tmp is that
 * complement.  But in that branch we complement tmp, meaning that at the
 * bottom of the loop tmp is always going to be equal to TEST_NON_UTF8(s),
 * which means at the top of the loop in the next iteration, it is
 * TEST_NON_UTF8(s-1) */
#define FBC_UTF8_A(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL)                         \
    tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n';                      \
    tmp = TEST_NON_UTF8(tmp);                                                  \
    REXEC_FBC_SCAN(1,  /* 1=>is-utf8; advances s while s < strend */           \
        if (tmp == ! TEST_NON_UTF8((U8) *s)) {                                 \
            tmp = !tmp;                                                        \
            IF_SUCCESS; /* Is a boundary if values for s-1 and s differ */     \
        }                                                                      \
        else {                                                                 \
            IF_FAIL;                                                           \
        }                                                                      \
    );                                                                         \

/* Like FBC_UTF8_A, but TEST_UV is a macro which takes a UV as its input, and
 * TEST_UTF8 is a macro that for the same input code points returns identically
 * to TEST_UV, but takes a pointer to a UTF-8 encoded string instead */
#define FBC_UTF8(TEST_UV, TEST_UTF8, IF_SUCCESS, IF_FAIL)                      \
    if (s == reginfo->strbeg) {                                                \
        tmp = '\n';                                                            \
    }                                                                          \
    else { /* Back-up to the start of the previous character */                \
        U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg);              \
        tmp = utf8n_to_uvchr(r, (U8*) reginfo->strend - r,                     \
                                                       0, UTF8_ALLOW_DEFAULT); \
    }                                                                          \
    tmp = TEST_UV(tmp);                                                        \
    REXEC_FBC_SCAN(1,  /* 1=>is-utf8; advances s while s < strend */           \
        if (tmp == ! (TEST_UTF8((U8 *) s, (U8 *) reginfo->strend))) {          \
            tmp = !tmp;                                                        \
            IF_SUCCESS;                                                        \
        }                                                                      \
        else {                                                                 \
            IF_FAIL;                                                           \
        }                                                                      \
    );

/* Like the above two macros.  UTF8_CODE is the complete code for handling
 * UTF-8.  Common to the BOUND and NBOUND cases, set-up by the FBC_BOUND, etc
 * macros below */
#define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL)        \
    if (utf8_target) {                                                         \
        UTF8_CODE                                                              \
    }                                                                          \
    else {  /* Not utf8 */                                                     \
        tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n';                  \
        tmp = TEST_NON_UTF8(tmp);                                              \
        REXEC_FBC_SCAN(0, /* 0=>not-utf8; advances s while s < strend */       \
            if (tmp == ! TEST_NON_UTF8((U8) *s)) {                             \
                IF_SUCCESS;                                                    \
                tmp = !tmp;                                                    \
            }                                                                  \
            else {                                                             \
                IF_FAIL;                                                       \
            }                                                                  \
        );                                                                     \
    }                                                                          \
    /* Here, things have been set up by the previous code so that tmp is the   \
     * return of TEST_NON_UTF(s-1) or TEST_UTF8(s-1) (depending on the         \
     * utf8ness of the target).  We also have to check if this matches against \
     * the EOS, which we treat as a \n (which is the same value in both UTF-8  \
     * or non-UTF8, so can use the non-utf8 test condition even for a UTF-8    \
     * string */                                                               \
    if (tmp == ! TEST_NON_UTF8('\n')) {                                        \
        IF_SUCCESS;                                                            \
    }                                                                          \
    else {                                                                     \
        IF_FAIL;                                                               \
    }

/* This is the macro to use when we want to see if something that looks like it
 * could match, actually does, and if so exits the loop.  It needs to be used
 * only for bounds checking macros, as it allows for matching beyond the end of
 * string (which should be zero length without having to look at the string
 * contents) */
#define REXEC_FBC_TRYIT                                                     \
    if (reginfo->intuit || (s <= reginfo->strend && regtry(reginfo, &s)))   \
        goto got_it

/* The only difference between the BOUND and NBOUND cases is that
 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
 * NBOUND.  This is accomplished by passing it as either the if or else clause,
 * with the other one being empty (PLACEHOLDER is defined as empty).
 *
 * The TEST_FOO parameters are for operating on different forms of input, but
 * all should be ones that return identically for the same underlying code
 * points */
#define FBC_BOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8)                           \
    FBC_BOUND_COMMON(                                                          \
          FBC_UTF8(TEST_UV, TEST_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER),          \
          TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)

#define FBC_BOUND_A(TEST_NON_UTF8)                                             \
    FBC_BOUND_COMMON(                                                          \
            FBC_UTF8_A(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER),           \
            TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)

#define FBC_NBOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8)                          \
    FBC_BOUND_COMMON(                                                          \
          FBC_UTF8(TEST_UV, TEST_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT),          \
          TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)

#define FBC_NBOUND_A(TEST_NON_UTF8)                                            \
    FBC_BOUND_COMMON(                                                          \
            FBC_UTF8_A(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT),           \
            TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)

#ifdef DEBUGGING
static IV
S_get_break_val_cp_checked(SV* const invlist, const UV cp_in) {
  IV cp_out = _invlist_search(invlist, cp_in);
  assert(cp_out >= 0);
  return cp_out;
}
#  define _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp) \
        invmap[S_get_break_val_cp_checked(invlist, cp)]
#else
#  define _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp) \
        invmap[_invlist_search(invlist, cp)]
#endif

/* Takes a pointer to an inversion list, a pointer to its corresponding
 * inversion map, and a code point, and returns the code point's value
 * according to the two arrays.  It assumes that all code points have a value.
 * This is used as the base macro for macros for particular properties */
#define _generic_GET_BREAK_VAL_CP(invlist, invmap, cp)              \
        _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp)

/* Same as above, but takes begin, end ptrs to a UTF-8 encoded string instead
 * of a code point, returning the value for the first code point in the string.
 * And it takes the particular macro name that finds the desired value given a
 * code point.  Merely convert the UTF-8 to code point and call the cp macro */
#define _generic_GET_BREAK_VAL_UTF8(cp_macro, pos, strend)                     \
             (__ASSERT_(pos < strend)                                          \
                 /* Note assumes is valid UTF-8 */                             \
             (cp_macro(utf8_to_uvchr_buf((pos), (strend), NULL))))

/* Returns the GCB value for the input code point */
#define getGCB_VAL_CP(cp)                                                      \
          _generic_GET_BREAK_VAL_CP(                                           \
                                    PL_GCB_invlist,                            \
                                    _Perl_GCB_invmap,                          \
                                    (cp))

/* Returns the GCB value for the first code point in the UTF-8 encoded string
 * bounded by pos and strend */
#define getGCB_VAL_UTF8(pos, strend)                                           \
    _generic_GET_BREAK_VAL_UTF8(getGCB_VAL_CP, pos, strend)

/* Returns the LB value for the input code point */
#define getLB_VAL_CP(cp)                                                       \
          _generic_GET_BREAK_VAL_CP(                                           \
                                    PL_LB_invlist,                             \
                                    _Perl_LB_invmap,                           \
                                    (cp))

/* Returns the LB value for the first code point in the UTF-8 encoded string
 * bounded by pos and strend */
#define getLB_VAL_UTF8(pos, strend)                                            \
    _generic_GET_BREAK_VAL_UTF8(getLB_VAL_CP, pos, strend)


/* Returns the SB value for the input code point */
#define getSB_VAL_CP(cp)                                                       \
          _generic_GET_BREAK_VAL_CP(                                           \
                                    PL_SB_invlist,                             \
                                    _Perl_SB_invmap,                     \
                                    (cp))

/* Returns the SB value for the first code point in the UTF-8 encoded string
 * bounded by pos and strend */
#define getSB_VAL_UTF8(pos, strend)                                            \
    _generic_GET_BREAK_VAL_UTF8(getSB_VAL_CP, pos, strend)

/* Returns the WB value for the input code point */
#define getWB_VAL_CP(cp)                                                       \
          _generic_GET_BREAK_VAL_CP(                                           \
                                    PL_WB_invlist,                             \
                                    _Perl_WB_invmap,                         \
                                    (cp))

/* Returns the WB value for the first code point in the UTF-8 encoded string
 * bounded by pos and strend */
#define getWB_VAL_UTF8(pos, strend)                                            \
    _generic_GET_BREAK_VAL_UTF8(getWB_VAL_CP, pos, strend)

/* We know what class REx starts with.  Try to find this position... */
/* if reginfo->intuit, its a dryrun */
/* annoyingly all the vars in this routine have different names from their counterparts
   in regmatch. /grrr */
STATIC char *
S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, 
    const char *strend, regmatch_info *reginfo)
{
    dVAR;

    /* TRUE if x+ need not match at just the 1st pos of run of x's */
    const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;

    char *pat_string;   /* The pattern's exactish string */
    char *pat_end;          /* ptr to end char of pat_string */
    re_fold_t folder;   /* Function for computing non-utf8 folds */
    const U8 *fold_array;   /* array for folding ords < 256 */
    STRLEN ln;
    STRLEN lnc;
    U8 c1;
    U8 c2;
    char *e = NULL;

    /* In some cases we accept only the first occurence of 'x' in a sequence of
     * them.  This variable points to just beyond the end of the previous
     * occurrence of 'x', hence we can tell if we are in a sequence.  (Having
     * it point to beyond the 'x' allows us to work for UTF-8 without having to
     * hop back.) */
    char * previous_occurrence_end = 0;

    I32 tmp;            /* Scratch variable */
    const bool utf8_target = reginfo->is_utf8_target;
    UV utf8_fold_flags = 0;
    const bool is_utf8_pat = reginfo->is_utf8_pat;
    bool to_complement = FALSE; /* Invert the result?  Taking the xor of this
                                   with a result inverts that result, as 0^1 =
                                   1 and 1^1 = 0 */
    _char_class_number classnum;

    RXi_GET_DECL(prog,progi);

    PERL_ARGS_ASSERT_FIND_BYCLASS;

    /* We know what class it must start with. */
    switch (OP(c)) {
    case ANYOFPOSIXL:
    case ANYOFL:
        _CHECK_AND_WARN_PROBLEMATIC_LOCALE;

        if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(c)) && ! IN_UTF8_CTYPE_LOCALE) {
            Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required);
        }

        /* FALLTHROUGH */
    case ANYOFD:
    case ANYOF:
        if (utf8_target) {
            REXEC_FBC_CLASS_SCAN(1, /* 1=>is-utf8 */
                      reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
        }
        else if (ANYOF_FLAGS(c) & ~ ANYOF_MATCHES_ALL_ABOVE_BITMAP) {
            /* We know that s is in the bitmap range since the target isn't
             * UTF-8, so what happens for out-of-range values is not relevant,
             * so exclude that from the flags */
            REXEC_FBC_CLASS_SCAN(0, reginclass(prog,c, (U8*)s, (U8*)s+1, 0));
        }
        else {
            REXEC_FBC_CLASS_SCAN(0, ANYOF_BITMAP_TEST(c, *((U8*)s)));
        }
        break;

    case ANYOFM:    /* ARG() is the base byte; FLAGS() the mask byte */
        /* UTF-8ness doesn't matter because only matches UTF-8 invariants, so
         * use 0 */
        REXEC_FBC_FIND_NEXT_SCAN(0,
         (char *) find_next_masked((U8 *) s, (U8 *) strend,
                                   (U8) ARG(c), FLAGS(c)));
        break;

    case NANYOFM:   /* UTF-8ness does matter because can match UTF-8 variants.
                     */
        REXEC_FBC_FIND_NEXT_SCAN(utf8_target,
         (char *) find_span_end_mask((U8 *) s, (U8 *) strend,
                                   (U8) ARG(c), FLAGS(c)));
        break;

    case ANYOFH:
        if (utf8_target) {  /* Can't possibly match a non-UTF-8 target */
            U8 first_byte = FLAGS(c);

            if (first_byte) {   /* We know what the first byte of any matched
                                   string should be */
                REXEC_FBC_FIND_NEXT_UTF8_BYTE_SCAN(first_byte,
                      reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
            }
            else {
                REXEC_FBC_CLASS_SCAN(TRUE,
                      reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
            }
        }
        break;

    case EXACTFAA_NO_TRIE: /* This node only generated for non-utf8 patterns */
        assert(! is_utf8_pat);
        /* FALLTHROUGH */
    case EXACTFAA:
        if (is_utf8_pat) {
            utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII
                             |FOLDEQ_S2_ALREADY_FOLDED|FOLDEQ_S2_FOLDS_SANE;
            goto do_exactf_utf8;
        }
        else if (utf8_target) {

            /* Here, and elsewhere in this file, the reason we can't consider a
             * non-UTF-8 pattern already folded in the presence of a UTF-8
             * target is because any MICRO SIGN in the pattern won't be folded.
             * Since the fold of the MICRO SIGN requires UTF-8 to represent, we
             * can consider a non-UTF-8 pattern folded when matching a
             * non-UTF-8 target */
            utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
            goto do_exactf_utf8;
        }

        /* Latin1 folds are not affected by /a, except it excludes the sharp s,
         * which these functions don't handle anyway */
        fold_array = PL_fold_latin1;
        folder = foldEQ_latin1_s2_folded;
        goto do_exactf_non_utf8;

    case EXACTF:   /* This node only generated for non-utf8 patterns */
        assert(! is_utf8_pat);
        if (utf8_target) {
            goto do_exactf_utf8;
        }
        fold_array = PL_fold;
        folder = foldEQ;
        goto do_exactf_non_utf8;

    case EXACTFL:
        _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
        if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) {
            utf8_fold_flags = FOLDEQ_LOCALE;
            goto do_exactf_utf8;
        }
        fold_array = PL_fold_locale;
        folder = foldEQ_locale;
        goto do_exactf_non_utf8;

    case EXACTFUP:      /* Problematic even though pattern isn't UTF-8.  Use
                           full functionality normally not done except for
                           UTF-8 */
        assert(! is_utf8_pat);
        goto do_exactf_utf8;

    case EXACTFLU8:
            if (! utf8_target) {    /* All code points in this node require
                                       UTF-8 to express.  */
                break;
            }
            utf8_fold_flags =  FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED
                                             | FOLDEQ_S2_FOLDS_SANE;
            goto do_exactf_utf8;

    case EXACTFU_ONLY8:
        if (! utf8_target) {
            break;
        }
        assert(is_utf8_pat);
        utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
        goto do_exactf_utf8;

    case EXACTFU:
        if (is_utf8_pat || utf8_target) {
            utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
            goto do_exactf_utf8;
        }

        /* Any 'ss' in the pattern should have been replaced by regcomp,
         * so we don't have to worry here about this single special case
         * in the Latin1 range */
        fold_array = PL_fold_latin1;
        folder = foldEQ_latin1_s2_folded;

        /* FALLTHROUGH */

      do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
                           are no glitches with fold-length differences
                           between the target string and pattern */

        /* The idea in the non-utf8 EXACTF* cases is to first find the
         * first character of the EXACTF* node and then, if necessary,
         * case-insensitively compare the full text of the node.  c1 is the
         * first character.  c2 is its fold.  This logic will not work for
         * Unicode semantics and the german sharp ss, which hence should
         * not be compiled into a node that gets here. */
        pat_string = STRING(c);
        ln  = STR_LEN(c);       /* length to match in octets/bytes */

        /* We know that we have to match at least 'ln' bytes (which is the
         * same as characters, since not utf8).  If we have to match 3
         * characters, and there are only 2 availabe, we know without
         * trying that it will fail; so don't start a match past the
         * required minimum number from the far end */
        e = HOP3c(strend, -((SSize_t)ln), s);
        if (e < s)
            break;

        c1 = *pat_string;
        c2 = fold_array[c1];
        if (c1 == c2) { /* If char and fold are the same */
            while (s <= e) {
                s = (char *) memchr(s, c1, e + 1 - s);
                if (s == NULL) {
                    break;
                }

                /* Check that the rest of the node matches */
                if (   (ln == 1 || folder(s + 1, pat_string + 1, ln - 1))
                    && (reginfo->intuit || regtry(reginfo, &s)) )
                {
                    goto got_it;
                }
                s++;
            }
        }
        else {
            U8 bits_differing = c1 ^ c2;

            /* If the folds differ in one bit position only, we can mask to
             * match either of them, and can use this faster find method.  Both
             * ASCII and EBCDIC tend to have their case folds differ in only
             * one position, so this is very likely */
            if (LIKELY(PL_bitcount[bits_differing] == 1)) {
                bits_differing = ~ bits_differing;
                while (s <= e) {
                    s = (char *) find_next_masked((U8 *) s, (U8 *) e + 1,
                                        (c1 & bits_differing), bits_differing);
                    if (s > e) {
                        break;
                    }

                    if (   (ln == 1 || folder(s + 1, pat_string + 1, ln - 1))
                        && (reginfo->intuit || regtry(reginfo, &s)) )
                    {
                        goto got_it;
                    }
                    s++;
                }
            }
            else {  /* Otherwise, stuck with looking byte-at-a-time.  This
                       should actually happen only in EXACTFL nodes */
                while (s <= e) {
                    if (    (*(U8*)s == c1 || *(U8*)s == c2)
                        && (ln == 1 || folder(s + 1, pat_string + 1, ln - 1))
                        && (reginfo->intuit || regtry(reginfo, &s)) )
                    {
                        goto got_it;
                    }
                    s++;
                }
            }
        }
        break;

      do_exactf_utf8:
      {
        unsigned expansion;

        /* If one of the operands is in utf8, we can't use the simpler folding
         * above, due to the fact that many different characters can have the
         * same fold, or portion of a fold, or different- length fold */
        pat_string = STRING(c);
        ln  = STR_LEN(c);       /* length to match in octets/bytes */
        pat_end = pat_string + ln;
        lnc = is_utf8_pat       /* length to match in characters */
                ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
                : ln;

        /* We have 'lnc' characters to match in the pattern, but because of
         * multi-character folding, each character in the target can match
         * up to 3 characters (Unicode guarantees it will never exceed
         * this) if it is utf8-encoded; and up to 2 if not (based on the
         * fact that the Latin 1 folds are already determined, and the
         * only multi-char fold in that range is the sharp-s folding to
         * 'ss'.  Thus, a pattern character can match as little as 1/3 of a
         * string character.  Adjust lnc accordingly, rounding up, so that
         * if we need to match at least 4+1/3 chars, that really is 5. */
        expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
        lnc = (lnc + expansion - 1) / expansion;

        /* As in the non-UTF8 case, if we have to match 3 characters, and
         * only 2 are left, it's guaranteed to fail, so don't start a
         * match that would require us to go beyond the end of the string
         */
        e = HOP3c(strend, -((SSize_t)lnc), s);

        /* XXX Note that we could recalculate e to stop the loop earlier,
         * as the worst case expansion above will rarely be met, and as we
         * go along we would usually find that e moves further to the left.
         * This would happen only after we reached the point in the loop
         * where if there were no expansion we should fail.  Unclear if
         * worth the expense */

        while (s <= e) {
            char *my_strend= (char *)strend;
            if (foldEQ_utf8_flags(s, &my_strend, 0,  utf8_target,
                  pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
                && (reginfo->intuit || regtry(reginfo, &s)) )
            {
                goto got_it;
            }
            s += (utf8_target) ? UTF8_SAFE_SKIP(s, reginfo->strend) : 1;
        }
        break;
    }

    case BOUNDL:
        _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
        if (FLAGS(c) != TRADITIONAL_BOUND) {
            if (! IN_UTF8_CTYPE_LOCALE) {
                Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
                                                B_ON_NON_UTF8_LOCALE_IS_WRONG);
            }
            goto do_boundu;
        }

        FBC_BOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8_safe);
        break;

    case NBOUNDL:
        _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
        if (FLAGS(c) != TRADITIONAL_BOUND) {
            if (! IN_UTF8_CTYPE_LOCALE) {
                Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
                                                B_ON_NON_UTF8_LOCALE_IS_WRONG);
            }
            goto do_nboundu;
        }

        FBC_NBOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8_safe);
        break;

    case BOUND: /* regcomp.c makes sure that this only has the traditional \b
                   meaning */
        assert(FLAGS(c) == TRADITIONAL_BOUND);

        FBC_BOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
        break;

    case BOUNDA: /* regcomp.c makes sure that this only has the traditional \b
                   meaning */
        assert(FLAGS(c) == TRADITIONAL_BOUND);

        FBC_BOUND_A(isWORDCHAR_A);
        break;

    case NBOUND: /* regcomp.c makes sure that this only has the traditional \b
                   meaning */
        assert(FLAGS(c) == TRADITIONAL_BOUND);

        FBC_NBOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
        break;

    case NBOUNDA: /* regcomp.c makes sure that this only has the traditional \b
                   meaning */
        assert(FLAGS(c) == TRADITIONAL_BOUND);

        FBC_NBOUND_A(isWORDCHAR_A);
        break;

    case NBOUNDU:
        if ((bound_type) FLAGS(c) == TRADITIONAL_BOUND) {
            FBC_NBOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
            break;
        }

      do_nboundu:

        to_complement = 1;
        /* FALLTHROUGH */

    case BOUNDU:
      do_boundu:
        switch((bound_type) FLAGS(c)) {
            case TRADITIONAL_BOUND:
                FBC_BOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
                break;
            case GCB_BOUND:
                if (s == reginfo->strbeg) {
                    if (reginfo->intuit || regtry(reginfo, &s))
                    {
                        goto got_it;
                    }

                    /* Didn't match.  Try at the next position (if there is one) */
                    s += (utf8_target) ? UTF8_SAFE_SKIP(s, reginfo->strend) : 1;
                    if (UNLIKELY(s >= reginfo->strend)) {
                        break;
                    }
                }

                if (utf8_target) {
                    GCB_enum before = getGCB_VAL_UTF8(
                                               reghop3((U8*)s, -1,
                                                       (U8*)(reginfo->strbeg)),
                                               (U8*) reginfo->strend);
                    while (s < strend) {
                        GCB_enum after = getGCB_VAL_UTF8((U8*) s,
                                                        (U8*) reginfo->strend);
                        if (   (to_complement ^ isGCB(before,
                                                      after,
                                                      (U8*) reginfo->strbeg,
                                                      (U8*) s,
                                                      utf8_target))
                            && (reginfo->intuit || regtry(reginfo, &s)))
                        {
                            goto got_it;
                        }
                        before = after;
                        s += UTF8_SAFE_SKIP(s, reginfo->strend);
                    }
                }
                else {  /* Not utf8.  Everything is a GCB except between CR and
                           LF */
                    while (s < strend) {
                        if ((to_complement ^ (   UCHARAT(s - 1) != '\r'
                                              || UCHARAT(s) != '\n'))
                            && (reginfo->intuit || regtry(reginfo, &s)))
                        {
                            goto got_it;
                        }
                        s++;
                    }
                }

                /* And, since this is a bound, it can match after the final
                 * character in the string */
                if (   reginfo->intuit
                    || (s <= reginfo->strend && regtry(reginfo, &s)))
                {
                    goto got_it;
                }
                break;

            case LB_BOUND:
                if (s == reginfo->strbeg) {
                    if (reginfo->intuit || regtry(reginfo, &s)) {
                        goto got_it;
                    }
                    s += (utf8_target) ? UTF8_SAFE_SKIP(s, reginfo->strend) : 1;
                    if (UNLIKELY(s >= reginfo->strend)) {
                        break;
                    }
                }

                if (utf8_target) {
                    LB_enum before = getLB_VAL_UTF8(reghop3((U8*)s,
                                                               -1,
                                                               (U8*)(reginfo->strbeg)),
                                                       (U8*) reginfo->strend);
                    while (s < strend) {
                        LB_enum after = getLB_VAL_UTF8((U8*) s, (U8*) reginfo->strend);
                        if (to_complement ^ isLB(before,
                                                 after,
                                                 (U8*) reginfo->strbeg,
                                                 (U8*) s,
                                                 (U8*) reginfo->strend,
                                                 utf8_target)
                            && (reginfo->intuit || regtry(reginfo, &s)))
                        {
                            goto got_it;
                        }
                        before = after;
                        s += UTF8_SAFE_SKIP(s, reginfo->strend);
                    }
                }
                else {  /* Not utf8. */
                    LB_enum before = getLB_VAL_CP((U8) *(s -1));
                    while (s < strend) {
                        LB_enum after = getLB_VAL_CP((U8) *s);
                        if (to_complement ^ isLB(before,
                                                 after,
                                                 (U8*) reginfo->strbeg,
                                                 (U8*) s,
                                                 (U8*) reginfo->strend,
                                                 utf8_target)
                            && (reginfo->intuit || regtry(reginfo, &s)))
                        {
                            goto got_it;
                        }
                        before = after;
                        s++;
                    }
                }

                if (   reginfo->intuit
                    || (s <= reginfo->strend && regtry(reginfo, &s)))
                {
                    goto got_it;
                }

                break;

            case SB_BOUND:
                if (s == reginfo->strbeg) {
                    if (reginfo->intuit || regtry(reginfo, &s)) {
                        goto got_it;
                    }
                    s += (utf8_target) ? UTF8_SAFE_SKIP(s, reginfo->strend) : 1;
                    if (UNLIKELY(s >= reginfo->strend)) {
                        break;
                    }
                }

                if (utf8_target) {
                    SB_enum before = getSB_VAL_UTF8(reghop3((U8*)s,
                                                        -1,
                                                        (U8*)(reginfo->strbeg)),
                                                      (U8*) reginfo->strend);
                    while (s < strend) {
                        SB_enum after = getSB_VAL_UTF8((U8*) s,
                                                         (U8*) reginfo->strend);
                        if ((to_complement ^ isSB(before,
                                                  after,
                                                  (U8*) reginfo->strbeg,
                                                  (U8*) s,
                                                  (U8*) reginfo->strend,
                                                  utf8_target))
                            && (reginfo->intuit || regtry(reginfo, &s)))
                        {
                            goto got_it;
                        }
                        before = after;
                        s += UTF8_SAFE_SKIP(s, reginfo->strend);
                    }
                }
                else {  /* Not utf8. */
                    SB_enum before = getSB_VAL_CP((U8) *(s -1));
                    while (s < strend) {
                        SB_enum after = getSB_VAL_CP((U8) *s);
                        if ((to_complement ^ isSB(before,
                                                  after,
                                                  (U8*) reginfo->strbeg,
                                                  (U8*) s,
                                                  (U8*) reginfo->strend,
                                                  utf8_target))
                            && (reginfo->intuit || regtry(reginfo, &s)))
                        {
                            goto got_it;
                        }
                        before = after;
                        s++;
                    }
                }

                /* Here are at the final position in the target string.  The SB
                 * value is always true here, so matches, depending on other
                 * constraints */
                if (   reginfo->intuit
                    || (s <= reginfo->strend && regtry(reginfo, &s)))
                {
                    goto got_it;
                }

                break;

            case WB_BOUND:
                if (s == reginfo->strbeg) {
                    if (reginfo->intuit || regtry(reginfo, &s)) {
                        goto got_it;
                    }
                    s += (utf8_target) ? UTF8_SAFE_SKIP(s, reginfo->strend) : 1;
                    if (UNLIKELY(s >= reginfo->strend)) {
                        break;
                    }
                }

                if (utf8_target) {
                    /* We are at a boundary between char_sub_0 and char_sub_1.
                     * We also keep track of the value for char_sub_-1 as we
                     * loop through the line.   Context may be needed to make a
                     * determination, and if so, this can save having to
                     * recalculate it */
                    WB_enum previous = WB_UNKNOWN;
                    WB_enum before = getWB_VAL_UTF8(
                                              reghop3((U8*)s,
                                                      -1,
                                                      (U8*)(reginfo->strbeg)),
                                              (U8*) reginfo->strend);
                    while (s < strend) {
                        WB_enum after = getWB_VAL_UTF8((U8*) s,
                                                        (U8*) reginfo->strend);
                        if ((to_complement ^ isWB(previous,
                                                  before,
                                                  after,
                                                  (U8*) reginfo->strbeg,
                                                  (U8*) s,
                                                  (U8*) reginfo->strend,
                                                  utf8_target))
                            && (reginfo->intuit || regtry(reginfo, &s)))
                        {
                            goto got_it;
                        }
                        previous = before;
                        before = after;
                        s += UTF8_SAFE_SKIP(s, reginfo->strend);
                    }
                }
                else {  /* Not utf8. */
                    WB_enum previous = WB_UNKNOWN;
                    WB_enum before = getWB_VAL_CP((U8) *(s -1));
                    while (s < strend) {
                        WB_enum after = getWB_VAL_CP((U8) *s);
                        if ((to_complement ^ isWB(previous,
                                                  before,
                                                  after,
                                                  (U8*) reginfo->strbeg,
                                                  (U8*) s,
                                                  (U8*) reginfo->strend,
                                                  utf8_target))
                            && (reginfo->intuit || regtry(reginfo, &s)))
                        {
                            goto got_it;
                        }
                        previous = before;
                        before = after;
                        s++;
                    }
                }

                if (   reginfo->intuit
                    || (s <= reginfo->strend && regtry(reginfo, &s)))
                {
                    goto got_it;
                }
        }
        break;

    case LNBREAK:
        REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
                        is_LNBREAK_latin1_safe(s, strend)
        );
        break;

    /* The argument to all the POSIX node types is the class number to pass to
     * _generic_isCC() to build a mask for searching in PL_charclass[] */

    case NPOSIXL:
        to_complement = 1;
        /* FALLTHROUGH */

    case POSIXL:
        _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
        REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s, (U8 *) strend)),
                        to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
        break;

    case NPOSIXD:
        to_complement = 1;
        /* FALLTHROUGH */

    case POSIXD:
        if (utf8_target) {
            goto posix_utf8;
        }
        goto posixa;

    case NPOSIXA:
        if (utf8_target) {
            /* The complement of something that matches only ASCII matches all
             * non-ASCII, plus everything in ASCII that isn't in the class. */
            REXEC_FBC_CLASS_SCAN(1,   ! isASCII_utf8_safe(s, strend)
                                   || ! _generic_isCC_A(*s, FLAGS(c)));
            break;
        }

        to_complement = 1;
        goto posixa;

    case POSIXA:
        /* Don't need to worry about utf8, as it can match only a single
         * byte invariant character.  But we do anyway for performance reasons,
         * as otherwise we would have to examine all the continuation
         * characters */
        if (utf8_target) {
            REXEC_FBC_CLASS_SCAN(1, _generic_isCC_A(*s, FLAGS(c)));
            break;
        }

      posixa:
        REXEC_FBC_CLASS_SCAN(0, /* 0=>not-utf8 */
                        to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
        break;

    case NPOSIXU:
        to_complement = 1;
        /* FALLTHROUGH */

    case POSIXU:
        if (! utf8_target) {
            REXEC_FBC_CLASS_SCAN(0, /* 0=>not-utf8 */
                                 to_complement ^ cBOOL(_generic_isCC(*s,
                                                                    FLAGS(c))));
        }
        else {

          posix_utf8:
            classnum = (_char_class_number) FLAGS(c);
            switch (classnum) {
                default:
                    REXEC_FBC_CLASS_SCAN(1, /* 1=>is-utf8 */
                        to_complement ^ cBOOL(_invlist_contains_cp(
                                              PL_XPosix_ptrs[classnum],
                                              utf8_to_uvchr_buf((U8 *) s,
                                                                (U8 *) strend,
                                                                NULL))));
                    break;
                case _CC_ENUM_SPACE:
                    REXEC_FBC_CLASS_SCAN(1, /* 1=>is-utf8 */
                        to_complement ^ cBOOL(isSPACE_utf8_safe(s, strend)));
                    break;

                case _CC_ENUM_BLANK:
                    REXEC_FBC_CLASS_SCAN(1,
                        to_complement ^ cBOOL(isBLANK_utf8_safe(s, strend)));
                    break;

                case _CC_ENUM_XDIGIT:
                    REXEC_FBC_CLASS_SCAN(1,
                       to_complement ^ cBOOL(isXDIGIT_utf8_safe(s, strend)));
                    break;

                case _CC_ENUM_VERTSPACE:
                    REXEC_FBC_CLASS_SCAN(1,
                       to_complement ^ cBOOL(isVERTWS_utf8_safe(s, strend)));
                    break;

                case _CC_ENUM_CNTRL:
                    REXEC_FBC_CLASS_SCAN(1,
                        to_complement ^ cBOOL(isCNTRL_utf8_safe(s, strend)));
                    break;
            }
        }
        break;

    case AHOCORASICKC:
    case AHOCORASICK:
        {
            DECL_TRIE_TYPE(c);
            /* what trie are we using right now */
            reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
            reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
            HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);

            const char *last_start = strend - trie->minlen;
#ifdef DEBUGGING
            const char *real_start = s;
#endif
            STRLEN maxlen = trie->maxlen;
            SV *sv_points;
            U8 **points; /* map of where we were in the input string
                            when reading a given char. For ASCII this
                            is unnecessary overhead as the relationship
                            is always 1:1, but for Unicode, especially
                            case folded Unicode this is not true. */
            U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
            U8 *bitmap=NULL;


            GET_RE_DEBUG_FLAGS_DECL;

            /* We can't just allocate points here. We need to wrap it in
             * an SV so it gets freed properly if there is a croak while
             * running the match */
            ENTER;
            SAVETMPS;
            sv_points=newSV(maxlen * sizeof(U8 *));
            SvCUR_set(sv_points,
                maxlen * sizeof(U8 *));
            SvPOK_on(sv_points);
            sv_2mortal(sv_points);
            points=(U8**)SvPV_nolen(sv_points );
            if ( trie_type != trie_utf8_fold
                 && (trie->bitmap || OP(c)==AHOCORASICKC) )
            {
                if (trie->bitmap)
                    bitmap=(U8*)trie->bitmap;
                else
                    bitmap=(U8*)ANYOF_BITMAP(c);
            }
            /* this is the Aho-Corasick algorithm modified a touch
               to include special handling for long "unknown char" sequences.
               The basic idea being that we use AC as long as we are dealing
               with a possible matching char, when we encounter an unknown char
               (and we have not encountered an accepting state) we scan forward
               until we find a legal starting char.
               AC matching is basically that of trie matching, except that when
               we encounter a failing transition, we fall back to the current
               states "fail state", and try the current char again, a process
               we repeat until we reach the root state, state 1, or a legal
               transition. If we fail on the root state then we can either
               terminate if we have reached an accepting state previously, or
               restart the entire process from the beginning if we have not.

             */
            while (s <= last_start) {
                const U32 uniflags = UTF8_ALLOW_DEFAULT;
                U8 *uc = (U8*)s;
                U16 charid = 0;
                U32 base = 1;
                U32 state = 1;
                UV uvc = 0;
                STRLEN len = 0;
                STRLEN foldlen = 0;
                U8 *uscan = (U8*)NULL;
                U8 *leftmost = NULL;
#ifdef DEBUGGING
                U32 accepted_word= 0;
#endif
                U32 pointpos = 0;

                while ( state && uc <= (U8*)strend ) {
                    int failed=0;
                    U32 word = aho->states[ state ].wordnum;

                    if( state==1 ) {
                        if ( bitmap ) {
                            DEBUG_TRIE_EXECUTE_r(
                                if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
                                    dump_exec_pos( (char *)uc, c, strend, real_start,
                                        (char *)uc, utf8_target, 0 );
                                    Perl_re_printf( aTHX_
                                        " Scanning for legal start char...\n");
                                }
                            );
                            if (utf8_target) {
                                while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
                                    uc += UTF8SKIP(uc);
                                }
                            } else {
                                while ( uc <= (U8*)last_start  && !BITMAP_TEST(bitmap,*uc) ) {
                                    uc++;
                                }
                            }
                            s= (char *)uc;
                        }
                        if (uc >(U8*)last_start) break;
                    }

                    if ( word ) {
                        U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
                        if (!leftmost || lpos < leftmost) {
                            DEBUG_r(accepted_word=word);
                            leftmost= lpos;
                        }
                        if (base==0) break;

                    }
                    points[pointpos++ % maxlen]= uc;
                    if (foldlen || uc < (U8*)strend) {
                        REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
                                             (U8 *) strend, uscan, len, uvc,
                                             charid, foldlen, foldbuf,
                                             uniflags);
                        DEBUG_TRIE_EXECUTE_r({
                            dump_exec_pos( (char *)uc, c, strend,
                                        real_start, s, utf8_target, 0);
                            Perl_re_printf( aTHX_
                                " Charid:%3u CP:%4" UVxf " ",
                                 charid, uvc);
                        });
                    }
                    else {
                        len = 0;
                        charid = 0;
                    }


                    do {
#ifdef DEBUGGING
                        word = aho->states[ state ].wordnum;
#endif
                        base = aho->states[ state ].trans.base;

                        DEBUG_TRIE_EXECUTE_r({
                            if (failed)
                                dump_exec_pos( (char *)uc, c, strend, real_start,
                                    s,   utf8_target, 0 );
                            Perl_re_printf( aTHX_
                                "%sState: %4" UVxf ", word=%" UVxf,
                                failed ? " Fail transition to " : "",
                                (UV)state, (UV)word);
                        });
                        if ( base ) {
                            U32 tmp;
                            I32 offset;
                            if (charid &&
                                 ( ((offset = base + charid
                                    - 1 - trie->uniquecharcount)) >= 0)
                                 && ((U32)offset < trie->lasttrans)
                                 && trie->trans[offset].check == state
                                 && (tmp=trie->trans[offset].next))
                            {
                                DEBUG_TRIE_EXECUTE_r(
                                    Perl_re_printf( aTHX_ " - legal\n"));
                                state = tmp;
                                break;
                            }
                            else {
                                DEBUG_TRIE_EXECUTE_r(
                                    Perl_re_printf( aTHX_ " - fail\n"));
                                failed = 1;
                                state = aho->fail[state];
                            }
                        }
                        else {
                            /* we must be accepting here */
                            DEBUG_TRIE_EXECUTE_r(
                                    Perl_re_printf( aTHX_ " - accepting\n"));
                            failed = 1;
                            break;
                        }
                    } while(state);
                    uc += len;
                    if (failed) {
                        if (leftmost)
                            break;
                        if (!state) state = 1;
                    }
                }
                if ( aho->states[ state ].wordnum ) {
                    U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
                    if (!leftmost || lpos < leftmost) {
                        DEBUG_r(accepted_word=aho->states[ state ].wordnum);
                        leftmost = lpos;
                    }
                }
                if (leftmost) {
                    s = (char*)leftmost;
                    DEBUG_TRIE_EXECUTE_r({
                        Perl_re_printf( aTHX_  "Matches word #%" UVxf " at position %" IVdf ". Trying full pattern...\n",
                            (UV)accepted_word, (IV)(s - real_start)
                        );
                    });
                    if (reginfo->intuit || regtry(reginfo, &s)) {
                        FREETMPS;
                        LEAVE;
                        goto got_it;
                    }
                    if (s < reginfo->strend) {
                        s = HOPc(s,1);
                    }
                    DEBUG_TRIE_EXECUTE_r({
                        Perl_re_printf( aTHX_ "Pattern failed. Looking for new start point...\n");
                    });
                } else {
                    DEBUG_TRIE_EXECUTE_r(
                        Perl_re_printf( aTHX_ "No match.\n"));
                    break;
                }
            }
            FREETMPS;
            LEAVE;
        }
        break;
    default:
        Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
    }
    return 0;
  got_it:
    return s;
}

/* set RX_SAVED_COPY, RX_SUBBEG etc.
 * flags have same meanings as with regexec_flags() */

static void
S_reg_set_capture_string(pTHX_ REGEXP * const rx,
                            char *strbeg,
                            char *strend,
                            SV *sv,
                            U32 flags,
                            bool utf8_target)
{
    struct regexp *const prog = ReANY(rx);

    if (flags & REXEC_COPY_STR) {
#ifdef PERL_ANY_COW
        if (SvCANCOW(sv)) {
            DEBUG_C(Perl_re_printf( aTHX_
                              "Copy on write: regexp capture, type %d\n",
                                    (int) SvTYPE(sv)));
            /* Create a new COW SV to share the match string and store
             * in saved_copy, unless the current COW SV in saved_copy
             * is valid and suitable for our purpose */
            if ((   prog->saved_copy
                 && SvIsCOW(prog->saved_copy)
                 && SvPOKp(prog->saved_copy)
                 && SvIsCOW(sv)
                 && SvPOKp(sv)
                 && SvPVX(sv) == SvPVX(prog->saved_copy)))
            {
                /* just reuse saved_copy SV */
                if (RXp_MATCH_COPIED(prog)) {
                    Safefree(prog->subbeg);
                    RXp_MATCH_COPIED_off(prog);
                }
            }
            else {
                /* create new COW SV to share string */
                RXp_MATCH_COPY_FREE(prog);
                prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
            }
            prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
            assert (SvPOKp(prog->saved_copy));
            prog->sublen  = strend - strbeg;
            prog->suboffset = 0;
            prog->subcoffset = 0;
        } else
#endif
        {
            SSize_t min = 0;
            SSize_t max = strend - strbeg;
            SSize_t sublen;

            if (    (flags & REXEC_COPY_SKIP_POST)
                && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
                && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
            ) { /* don't copy $' part of string */
                U32 n = 0;
                max = -1;
                /* calculate the right-most part of the string covered
                 * by a capture. Due to lookahead, this may be to
                 * the right of $&, so we have to scan all captures */
                while (n <= prog->lastparen) {
                    if (prog->offs[n].end > max)
                        max = prog->offs[n].end;
                    n++;
                }
                if (max == -1)
                    max = (PL_sawampersand & SAWAMPERSAND_LEFT)
                            ? prog->offs[0].start
                            : 0;
                assert(max >= 0 && max <= strend - strbeg);
            }

            if (    (flags & REXEC_COPY_SKIP_PRE)
                && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
                && !(PL_sawampersand & SAWAMPERSAND_LEFT)
            ) { /* don't copy $` part of string */
                U32 n = 0;
                min = max;
                /* calculate the left-most part of the string covered
                 * by a capture. Due to lookbehind, this may be to
                 * the left of $&, so we have to scan all captures */
                while (min && n <= prog->lastparen) {
                    if (   prog->offs[n].start != -1
                        && prog->offs[n].start < min)
                    {
                        min = prog->offs[n].start;
                    }
                    n++;
                }
                if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
                    && min >  prog->offs[0].end
                )
                    min = prog->offs[0].end;

            }

            assert(min >= 0 && min <= max && min <= strend - strbeg);
            sublen = max - min;

            if (RXp_MATCH_COPIED(prog)) {
                if (sublen > prog->sublen)
                    prog->subbeg =
                            (char*)saferealloc(prog->subbeg, sublen+1);
            }
            else
                prog->subbeg = (char*)safemalloc(sublen+1);
            Copy(strbeg + min, prog->subbeg, sublen, char);
            prog->subbeg[sublen] = '\0';
            prog->suboffset = min;
            prog->sublen = sublen;
            RXp_MATCH_COPIED_on(prog);
        }
        prog->subcoffset = prog->suboffset;
        if (prog->suboffset && utf8_target) {
            /* Convert byte offset to chars.
             * XXX ideally should only compute this if @-/@+
             * has been seen, a la PL_sawampersand ??? */

            /* If there's a direct correspondence between the
             * string which we're matching and the original SV,
             * then we can use the utf8 len cache associated with
             * the SV. In particular, it means that under //g,
             * sv_pos_b2u() will use the previously cached
             * position to speed up working out the new length of
             * subcoffset, rather than counting from the start of
             * the string each time. This stops
             *   $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
             * from going quadratic */
            if (SvPOKp(sv) && SvPVX(sv) == strbeg)
                prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset,
                                                SV_GMAGIC|SV_CONST_RETURN);
            else
                prog->subcoffset = utf8_length((U8*)strbeg,
                                    (U8*)(strbeg+prog->suboffset));
        }
    }
    else {
        RXp_MATCH_COPY_FREE(prog);
        prog->subbeg = strbeg;
        prog->suboffset = 0;
        prog->subcoffset = 0;
        prog->sublen = strend - strbeg;
    }
}




/*
 - regexec_flags - match a regexp against a string
 */
I32
Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
              char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags)
/* stringarg: the point in the string at which to begin matching */
/* strend:    pointer to null at end of string */
/* strbeg:    real beginning of string */
/* minend:    end of match must be >= minend bytes after stringarg. */
/* sv:        SV being matched: only used for utf8 flag, pos() etc; string
 *            itself is accessed via the pointers above */
/* data:      May be used for some additional optimizations.
              Currently unused. */
/* flags:     For optimizations. See REXEC_* in regexp.h */

{
    struct regexp *const prog = ReANY(rx);
    char *s;
    regnode *c;
    char *startpos;
    SSize_t minlen;             /* must match at least this many chars */
    SSize_t dontbother = 0;     /* how many characters not to try at end */
    const bool utf8_target = cBOOL(DO_UTF8(sv));
    I32 multiline;
    RXi_GET_DECL(prog,progi);
    regmatch_info reginfo_buf;  /* create some info to pass to regtry etc */
    regmatch_info *const reginfo = &reginfo_buf;
    regexp_paren_pair *swap = NULL;
    I32 oldsave;
    GET_RE_DEBUG_FLAGS_DECL;

    PERL_ARGS_ASSERT_REGEXEC_FLAGS;
    PERL_UNUSED_ARG(data);

    /* Be paranoid... */
    if (prog == NULL) {
        Perl_croak(aTHX_ "NULL regexp parameter");
    }

    DEBUG_EXECUTE_r(
        debug_start_match(rx, utf8_target, stringarg, strend,
        "Matching");
    );

    startpos = stringarg;

    /* set these early as they may be used by the HOP macros below */
    reginfo->strbeg = strbeg;
    reginfo->strend = strend;
    reginfo->is_utf8_target = cBOOL(utf8_target);

    if (prog->intflags & PREGf_GPOS_SEEN) {
        MAGIC *mg;

        /* set reginfo->ganch, the position where \G can match */

        reginfo->ganch =
            (flags & REXEC_IGNOREPOS)
            ? stringarg /* use start pos rather than pos() */
            : ((mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
              /* Defined pos(): */
            ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg)
            : strbeg; /* pos() not defined; use start of string */

        DEBUG_GPOS_r(Perl_re_printf( aTHX_
            "GPOS ganch set to strbeg[%" IVdf "]\n", (IV)(reginfo->ganch - strbeg)));

        /* in the presence of \G, we may need to start looking earlier in
         * the string than the suggested start point of stringarg:
         * if prog->gofs is set, then that's a known, fixed minimum
         * offset, such as
         * /..\G/:   gofs = 2
         * /ab|c\G/: gofs = 1
         * or if the minimum offset isn't known, then we have to go back
         * to the start of the string, e.g. /w+\G/
         */

        if (prog->intflags & PREGf_ANCH_GPOS) {
            if (prog->gofs) {
                startpos = HOPBACKc(reginfo->ganch, prog->gofs);
                if (!startpos ||
                    ((flags & REXEC_FAIL_ON_UNDERFLOW) && startpos < stringarg))
                {
                    DEBUG_r(Perl_re_printf( aTHX_
                            "fail: ganch-gofs before earliest possible start\n"));
                    return 0;
                }
            }
            else
                startpos = reginfo->ganch;
        }
        else if (prog->gofs) {
            startpos = HOPBACKc(startpos, prog->gofs);
            if (!startpos)
                startpos = strbeg;
        }
        else if (prog->intflags & PREGf_GPOS_FLOAT)
            startpos = strbeg;
    }

    minlen = prog->minlen;
    if ((startpos + minlen) > strend || startpos < strbeg) {
        DEBUG_r(Perl_re_printf( aTHX_
                    "Regex match can't succeed, so not even tried\n"));
        return 0;
    }

    /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
     * which will call destuctors to reset PL_regmatch_state, free higher
     * PL_regmatch_slabs, and clean up regmatch_info_aux and
     * regmatch_info_aux_eval */

    oldsave = PL_savestack_ix;

    s = startpos;

    if ((prog->extflags & RXf_USE_INTUIT)
        && !(flags & REXEC_CHECKED))
    {
        s = re_intuit_start(rx, sv, strbeg, startpos, strend,
                                    flags, NULL);
        if (!s)
            return 0;

        if (prog->extflags & RXf_CHECK_ALL) {
            /* we can match based purely on the result of INTUIT.
             * Set up captures etc just for $& and $-[0]
             * (an intuit-only match wont have $1,$2,..) */
            assert(!prog->nparens);

            /* s/// doesn't like it if $& is earlier than where we asked it to
             * start searching (which can happen on something like /.\G/) */
            if (       (flags & REXEC_FAIL_ON_UNDERFLOW)
                    && (s < stringarg))
            {
                /* this should only be possible under \G */
                assert(prog->intflags & PREGf_GPOS_SEEN);
                DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
                    "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
                goto phooey;
            }

            /* match via INTUIT shouldn't have any captures.
             * Let @-, @+, $^N know */
            prog->lastparen = prog->lastcloseparen = 0;
            RXp_MATCH_UTF8_set(prog, utf8_target);
            prog->offs[0].start = s - strbeg;
            prog->offs[0].end = utf8_target
                ? (char*)utf8_hop_forward((U8*)s, prog->minlenret, (U8 *) strend) - strbeg
                : s - strbeg + prog->minlenret;
            if ( !(flags & REXEC_NOT_FIRST) )
                S_reg_set_capture_string(aTHX_ rx,
                                        strbeg, strend,
                                        sv, flags, utf8_target);

            return 1;
        }
    }

    multiline = prog->extflags & RXf_PMf_MULTILINE;
    
    if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
        DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
                              "String too short [regexec_flags]...\n"));
        goto phooey;
    }
    
    /* Check validity of program. */
    if (UCHARAT(progi->program) != REG_MAGIC) {
        Perl_croak(aTHX_ "corrupted regexp program");
    }

    RXp_MATCH_TAINTED_off(prog);
    RXp_MATCH_UTF8_set(prog, utf8_target);

    reginfo->prog = rx;  /* Yes, sorry that this is confusing.  */
    reginfo->intuit = 0;
    reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
    reginfo->warned = FALSE;
    reginfo->sv = sv;
    reginfo->poscache_maxiter = 0; /* not yet started a countdown */
    /* see how far we have to get to not match where we matched before */
    reginfo->till = stringarg + minend;

    if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) {
        /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
           S_cleanup_regmatch_info_aux has executed (registered by
           SAVEDESTRUCTOR_X below).  S_cleanup_regmatch_info_aux modifies
           magic belonging to this SV.
           Not newSVsv, either, as it does not COW.
        */
        reginfo->sv = newSV(0);
        SvSetSV_nosteal(reginfo->sv, sv);
        SAVEFREESV(reginfo->sv);
    }

    /* reserve next 2 or 3 slots in PL_regmatch_state:
     * slot N+0: may currently be in use: skip it
     * slot N+1: use for regmatch_info_aux struct
     * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
     * slot N+3: ready for use by regmatch()
     */

    {
        regmatch_state *old_regmatch_state;
        regmatch_slab  *old_regmatch_slab;
        int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;

        /* on first ever match, allocate first slab */
        if (!PL_regmatch_slab) {
            Newx(PL_regmatch_slab, 1, regmatch_slab);
            PL_regmatch_slab->prev = NULL;
            PL_regmatch_slab->next = NULL;
            PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
        }

        old_regmatch_state = PL_regmatch_state;
        old_regmatch_slab  = PL_regmatch_slab;

        for (i=0; i <= max; i++) {
            if (i == 1)
                reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
            else if (i ==2)
                reginfo->info_aux_eval =
                reginfo->info_aux->info_aux_eval =
                            &(PL_regmatch_state->u.info_aux_eval);

            if (++PL_regmatch_state >  SLAB_LAST(PL_regmatch_slab))
                PL_regmatch_state = S_push_slab(aTHX);
        }

        /* note initial PL_regmatch_state position; at end of match we'll
         * pop back to there and free any higher slabs */

        reginfo->info_aux->old_regmatch_state = old_regmatch_state;
        reginfo->info_aux->old_regmatch_slab  = old_regmatch_slab;
        reginfo->info_aux->poscache = NULL;

        SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);

        if ((prog->extflags & RXf_EVAL_SEEN))
            S_setup_eval_state(aTHX_ reginfo);
        else
            reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
    }

    /* If there is a "must appear" string, look for it. */

    if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
        /* We have to be careful. If the previous successful match
           was from this regex we don't want a subsequent partially
           successful match to clobber the old results.
           So when we detect this possibility we add a swap buffer
           to the re, and switch the buffer each match. If we fail,
           we switch it back; otherwise we leave it swapped.
        */
        swap = prog->offs;
        /* avoid leak if we die, or clean up anyway if match completes */
        SAVEFREEPV(swap);
        Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
        DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
            "rex=0x%" UVxf " saving  offs: orig=0x%" UVxf " new=0x%" UVxf "\n",
            0,
            PTR2UV(prog),
            PTR2UV(swap),
            PTR2UV(prog->offs)
        ));
    }

    if (prog->recurse_locinput)
        Zero(prog->recurse_locinput,prog->nparens + 1, char *);

    /* Simplest case: anchored match need be tried only once, or with
     * MBOL, only at the beginning of each line.
     *
     * Note that /.*.../ sets PREGf_IMPLICIT|MBOL, while /.*.../s sets
     * PREGf_IMPLICIT|SBOL. The idea is that with /.*.../s, if it doesn't
     * match at the start of the string then it won't match anywhere else
     * either; while with /.*.../, if it doesn't match at the beginning,
     * the earliest it could match is at the start of the next line */

    if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) {
        char *end;

        if (regtry(reginfo, &s))
            goto got_it;

        if (!(prog->intflags & PREGf_ANCH_MBOL))
            goto phooey;

        /* didn't match at start, try at other newline positions */

        if (minlen)
            dontbother = minlen - 1;
        end = HOP3c(strend, -dontbother, strbeg) - 1;

        /* skip to next newline */

        while (s <= end) { /* note it could be possible to match at the end of the string */
            /* NB: newlines are the same in unicode as they are in latin */
            if (*s++ != '\n')
                continue;
            if (prog->check_substr || prog->check_utf8) {
            /* note that with PREGf_IMPLICIT, intuit can only fail
             * or return the start position, so it's of limited utility.
             * Nevertheless, I made the decision that the potential for
             * quick fail was still worth it - DAPM */
                s = re_intuit_start(rx, sv, strbeg, s, strend, flags, NULL);
                if (!s)
                    goto phooey;
            }
            if (regtry(reginfo, &s))
                goto got_it;
        }
        goto phooey;
    } /* end anchored search */

    if (prog->intflags & PREGf_ANCH_GPOS)
    {
        /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */
        assert(prog->intflags & PREGf_GPOS_SEEN);
        /* For anchored \G, the only position it can match from is
         * (ganch-gofs); we already set startpos to this above; if intuit
         * moved us on from there, we can't possibly succeed */
        assert(startpos == HOPBACKc(reginfo->ganch, prog->gofs));
        if (s == startpos && regtry(reginfo, &s))
            goto got_it;
        goto phooey;
    }

    /* Messy cases:  unanchored match. */
    if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
        /* we have /x+whatever/ */
        /* it must be a one character string (XXXX Except is_utf8_pat?) */
        char ch;
#ifdef DEBUGGING
        int did_match = 0;
#endif
        if (utf8_target) {
            if (! prog->anchored_utf8) {
                to_utf8_substr(prog);
            }
            ch = SvPVX_const(prog->anchored_utf8)[0];
            REXEC_FBC_SCAN(0,   /* 0=>not-utf8 */
                if (*s == ch) {
                    DEBUG_EXECUTE_r( did_match = 1 );
                    if (regtry(reginfo, &s)) goto got_it;
                    s += UTF8_SAFE_SKIP(s, strend);
                    while (s < strend && *s == ch)
                        s += UTF8SKIP(s);
                }
            );

        }
        else {
            if (! prog->anchored_substr) {
                if (! to_byte_substr(prog)) {
                    NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
                }
            }
            ch = SvPVX_const(prog->anchored_substr)[0];
            REXEC_FBC_SCAN(0,   /* 0=>not-utf8 */
                if (*s == ch) {
                    DEBUG_EXECUTE_r( did_match = 1 );
                    if (regtry(reginfo, &s)) goto got_it;
                    s++;
                    while (s < strend && *s == ch)
                        s++;
                }
            );
        }
        DEBUG_EXECUTE_r(if (!did_match)
                Perl_re_printf( aTHX_
                                  "Did not find anchored character...\n")
               );
    }
    else if (prog->anchored_substr != NULL
              || prog->anchored_utf8 != NULL
              || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
                  && prog->float_max_offset < strend - s)) {
        SV *must;
        SSize_t back_max;
        SSize_t back_min;
        char *last;
        char *last1;            /* Last position checked before */
#ifdef DEBUGGING
        int did_match = 0;
#endif
        if (prog->anchored_substr || prog->anchored_utf8) {
            if (utf8_target) {
                if (! prog->anchored_utf8) {
                    to_utf8_substr(prog);
                }
                must = prog->anchored_utf8;
            }
            else {
                if (! prog->anchored_substr) {
                    if (! to_byte_substr(prog)) {
                        NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
                    }
                }
                must = prog->anchored_substr;
            }
            back_max = back_min = prog->anchored_offset;
        } else {
            if (utf8_target) {
                if (! prog->float_utf8) {
                    to_utf8_substr(prog);
                }
                must = prog->float_utf8;
            }
            else {
                if (! prog->float_substr) {
                    if (! to_byte_substr(prog)) {
                        NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
                    }
                }
                must = prog->float_substr;
            }
            back_max = prog->float_max_offset;
            back_min = prog->float_min_offset;
        }
            
        if (back_min<0) {
            last = strend;
        } else {
            last = HOP3c(strend,        /* Cannot start after this */
                  -(SSize_t)(CHR_SVLEN(must)
                         - (SvTAIL(must) != 0) + back_min), strbeg);
        }
        if (s > reginfo->strbeg)
            last1 = HOPc(s, -1);
        else
            last1 = s - 1;      /* bogus */

        /* XXXX check_substr already used to find "s", can optimize if
           check_substr==must. */
        dontbother = 0;
        strend = HOPc(strend, -dontbother);
        while ( (s <= last) &&
                (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg,  strend),
                                  (unsigned char*)strend, must,
                                  multiline ? FBMrf_MULTILINE : 0)) ) {
            DEBUG_EXECUTE_r( did_match = 1 );
            if (HOPc(s, -back_max) > last1) {
                last1 = HOPc(s, -back_min);
                s = HOPc(s, -back_max);
            }
            else {
                char * const t = (last1 >= reginfo->strbeg)
                                    ? HOPc(last1, 1) : last1 + 1;

                last1 = HOPc(s, -back_min);
                s = t;
            }
            if (utf8_target) {
                while (s <= last1) {
                    if (regtry(reginfo, &s))
                        goto got_it;
                    if (s >= last1) {
                        s++; /* to break out of outer loop */
                        break;
                    }
                    s += UTF8SKIP(s);
                }
            }
            else {
                while (s <= last1) {
                    if (regtry(reginfo, &s))
                        goto got_it;
                    s++;
                }
            }
        }
        DEBUG_EXECUTE_r(if (!did_match) {
            RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
                SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
            Perl_re_printf( aTHX_  "Did not find %s substr %s%s...\n",
                              ((must == prog->anchored_substr || must == prog->anchored_utf8)
                               ? "anchored" : "floating"),
                quoted, RE_SV_TAIL(must));
        });                 
        goto phooey;
    }
    else if ( (c = progi->regstclass) ) {
        if (minlen) {
            const OPCODE op = OP(progi->regstclass);
            /* don't bother with what can't match */
            if (PL_regkind[op] != EXACT && PL_regkind[op] != TRIE)
                strend = HOPc(strend, -(minlen - 1));
        }
        DEBUG_EXECUTE_r({
            SV * const prop = sv_newmortal();
            regprop(prog, prop, c, reginfo, NULL);
            {
                RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
                    s,strend-s,PL_dump_re_max_len);
                Perl_re_printf( aTHX_
                    "Matching stclass %.*s against %s (%d bytes)\n",
                    (int)SvCUR(prop), SvPVX_const(prop),
                     quoted, (int)(strend - s));
            }
        });
        if (find_byclass(prog, c, s, strend, reginfo))
            goto got_it;
        DEBUG_EXECUTE_r(Perl_re_printf( aTHX_  "Contradicts stclass... [regexec_flags]\n"));
    }
    else {
        dontbother = 0;
        if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
            /* Trim the end. */
            char *last= NULL;
            SV* float_real;
            STRLEN len;
            const char *little;

            if (utf8_target) {
                if (! prog->float_utf8) {
                    to_utf8_substr(prog);
                }
                float_real = prog->float_utf8;
            }
            else {
                if (! prog->float_substr) {
                    if (! to_byte_substr(prog)) {
                        NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
                    }
                }
                float_real = prog->float_substr;
            }

            little = SvPV_const(float_real, len);
            if (SvTAIL(float_real)) {
                    /* This means that float_real contains an artificial \n on
                     * the end due to the presence of something like this:
                     * /foo$/ where we can match both "foo" and "foo\n" at the
                     * end of the string.  So we have to compare the end of the
                     * string first against the float_real without the \n and
                     * then against the full float_real with the string.  We
                     * have to watch out for cases where the string might be
                     * smaller than the float_real or the float_real without
                     * the \n. */
                    char *checkpos= strend - len;
                    DEBUG_OPTIMISE_r(
                        Perl_re_printf( aTHX_
                            "%sChecking for float_real.%s\n",
                            PL_colors[4], PL_colors[5]));
                    if (checkpos + 1 < strbeg) {
                        /* can't match, even if we remove the trailing \n
                         * string is too short to match */
                        DEBUG_EXECUTE_r(
                            Perl_re_printf( aTHX_
                                "%sString shorter than required trailing substring, cannot match.%s\n",
                                PL_colors[4], PL_colors[5]));
                        goto phooey;
                    } else if (memEQ(checkpos + 1, little, len - 1)) {
                        /* can match, the end of the string matches without the
                         * "\n" */
                        last = checkpos + 1;
                    } else if (checkpos < strbeg) {
                        /* cant match, string is too short when the "\n" is
                         * included */
                        DEBUG_EXECUTE_r(
                            Perl_re_printf( aTHX_
                                "%sString does not contain required trailing substring, cannot match.%s\n",
                                PL_colors[4], PL_colors[5]));
                        goto phooey;
                    } else if (!multiline) {
                        /* non multiline match, so compare with the "\n" at the
                         * end of the string */
                        if (memEQ(checkpos, little, len)) {
                            last= checkpos;
                        } else {
                            DEBUG_EXECUTE_r(
                                Perl_re_printf( aTHX_
                                    "%sString does not contain required trailing substring, cannot match.%s\n",
                                    PL_colors[4], PL_colors[5]));
                            goto phooey;
                        }
                    } else {
                        /* multiline match, so we have to search for a place
                         * where the full string is located */
                        goto find_last;
                    }
            } else {
                  find_last:
                    if (len)
                        last = rninstr(s, strend, little, little + len);
                    else
                        last = strend;  /* matching "$" */
            }
            if (!last) {
                /* at one point this block contained a comment which was
                 * probably incorrect, which said that this was a "should not
                 * happen" case.  Even if it was true when it was written I am