Use strict dfa to translate from UTF-8 to code point

[perl5.git] / utf8.c

/*    utf8.c
 *
 *    Copyright (C) 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.
 *
 */

/*
 * 'What a fix!' said Sam.  'That's the one place in all the lands we've ever
 *  heard of that we don't want to see any closer; and that's the one place
 *  we're trying to get to!  And that's just where we can't get, nohow.'
 *
 *     [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
 *
 * 'Well do I understand your speech,' he answered in the same language;
 * 'yet few strangers do so.  Why then do you not speak in the Common Tongue,
 *  as is the custom in the West, if you wish to be answered?'
 *                           --Gandalf, addressing Théoden's door wardens
 *
 *     [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
 *
 * ...the travellers perceived that the floor was paved with stones of many
 * hues; branching runes and strange devices intertwined beneath their feet.
 *
 *     [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
 */

#include "EXTERN.h"
#define PERL_IN_UTF8_C
#include "perl.h"
#include "invlist_inline.h"
#include "uni_keywords.h"

static const char malformed_text[] = "Malformed UTF-8 character";
static const char unees[] =
                        "Malformed UTF-8 character (unexpected end of string)";

/* Be sure to synchronize this message with the similar one in regcomp.c */
static const char cp_above_legal_max[] =
                        "Use of code point 0x%" UVXf " is not allowed; the"
                        " permissible max is 0x%" UVXf;

#define MAX_EXTERNALLY_LEGAL_CP ((UV) (IV_MAX))

/*
=head1 Unicode Support
These are various utility functions for manipulating UTF8-encoded
strings.  For the uninitiated, this is a method of representing arbitrary
Unicode characters as a variable number of bytes, in such a way that
characters in the ASCII range are unmodified, and a zero byte never appears
within non-zero characters.

=cut
*/

void
Perl__force_out_malformed_utf8_message(pTHX_
            const U8 *const p,      /* First byte in UTF-8 sequence */
            const U8 * const e,     /* Final byte in sequence (may include
                                       multiple chars */
            const U32 flags,        /* Flags to pass to utf8n_to_uvchr(),
                                       usually 0, or some DISALLOW flags */
            const bool die_here)    /* If TRUE, this function does not return */
{
    /* This core-only function is to be called when a malformed UTF-8 character
     * is found, in order to output the detailed information about the
     * malformation before dieing.  The reason it exists is for the occasions
     * when such a malformation is fatal, but warnings might be turned off, so
     * that normally they would not be actually output.  This ensures that they
     * do get output.  Because a sequence may be malformed in more than one
     * way, multiple messages may be generated, so we can't make them fatal, as
     * that would cause the first one to die.
     *
     * Instead we pretend -W was passed to perl, then die afterwards.  The
     * flexibility is here to return to the caller so they can finish up and
     * die themselves */
    U32 errors;

    PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;

    ENTER;
    SAVEI8(PL_dowarn);
    SAVESPTR(PL_curcop);

    PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
    if (PL_curcop) {
        PL_curcop->cop_warnings = pWARN_ALL;
    }

    (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);

    LEAVE;

    if (! errors) {
	Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
                         " be called only when there are errors found");
    }

    if (die_here) {
        Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
    }
}

STATIC HV *
S_new_msg_hv(pTHX_ const char * const message, /* The message text */
                   U32 categories,  /* Packed warning categories */
                   U32 flag)        /* Flag associated with this message */
{
    /* Creates, populates, and returns an HV* that describes an error message
     * for the translators between UTF8 and code point */

    SV* msg_sv = newSVpv(message, 0);
    SV* category_sv = newSVuv(categories);
    SV* flag_bit_sv = newSVuv(flag);

    HV* msg_hv = newHV();

    PERL_ARGS_ASSERT_NEW_MSG_HV;

    (void) hv_stores(msg_hv, "text", msg_sv);
    (void) hv_stores(msg_hv, "warn_categories",  category_sv);
    (void) hv_stores(msg_hv, "flag_bit", flag_bit_sv);

    return msg_hv;
}

/*
=for apidoc uvoffuni_to_utf8_flags

THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
Instead, B<Almost all code should use L</uvchr_to_utf8> or
L</uvchr_to_utf8_flags>>.

This function is like them, but the input is a strict Unicode
(as opposed to native) code point.  Only in very rare circumstances should code
not be using the native code point.

For details, see the description for L</uvchr_to_utf8_flags>.

=cut
*/

U8 *
Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
{
    PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;

    return uvoffuni_to_utf8_flags_msgs(d, uv, flags, NULL);
}

/* All these formats take a single UV code point argument */
const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
const char nonchar_cp_format[]   = "Unicode non-character U+%04" UVXf
                                   " is not recommended for open interchange";
const char super_cp_format[]     = "Code point 0x%" UVXf " is not Unicode,"
                                   " may not be portable";
const char perl_extended_cp_format[] = "Code point 0x%" UVXf " is not"        \
                                       " Unicode, requires a Perl extension," \
                                       " and so is not portable";

#define HANDLE_UNICODE_SURROGATE(uv, flags, msgs)                   \
    STMT_START {                                                    \
        if (flags & UNICODE_WARN_SURROGATE) {                       \
            U32 category = packWARN(WARN_SURROGATE);                \
            const char * format = surrogate_cp_format;              \
            if (msgs) {                                             \
                *msgs = new_msg_hv(Perl_form(aTHX_ format, uv),     \
                                   category,                        \
                                   UNICODE_GOT_SURROGATE);          \
            }                                                       \
            else {                                                  \
                Perl_ck_warner_d(aTHX_ category, format, uv);       \
            }                                                       \
        }                                                           \
        if (flags & UNICODE_DISALLOW_SURROGATE) {                   \
            return NULL;                                            \
        }                                                           \
    } STMT_END;

#define HANDLE_UNICODE_NONCHAR(uv, flags, msgs)                     \
    STMT_START {                                                    \
        if (flags & UNICODE_WARN_NONCHAR) {                         \
            U32 category = packWARN(WARN_NONCHAR);                  \
            const char * format = nonchar_cp_format;                \
            if (msgs) {                                             \
                *msgs = new_msg_hv(Perl_form(aTHX_ format, uv),     \
                                   category,                        \
                                   UNICODE_GOT_NONCHAR);            \
            }                                                       \
            else {                                                  \
                Perl_ck_warner_d(aTHX_ category, format, uv);       \
            }                                                       \
        }                                                           \
        if (flags & UNICODE_DISALLOW_NONCHAR) {                     \
            return NULL;                                            \
        }                                                           \
    } STMT_END;

/*  Use shorter names internally in this file */
#define SHIFT   UTF_ACCUMULATION_SHIFT
#undef  MARK
#define MARK    UTF_CONTINUATION_MARK
#define MASK    UTF_CONTINUATION_MASK

/*
=for apidoc uvchr_to_utf8_flags_msgs

THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.

Most code should use C<L</uvchr_to_utf8_flags>()> rather than call this directly.

This function is for code that wants any warning and/or error messages to be
returned to the caller rather than be displayed.  All messages that would have
been displayed if all lexical warnings are enabled will be returned.

It is just like C<L</uvchr_to_utf8_flags>> but it takes an extra parameter
placed after all the others, C<msgs>.  If this parameter is 0, this function
behaves identically to C<L</uvchr_to_utf8_flags>>.  Otherwise, C<msgs> should
be a pointer to an C<HV *> variable, in which this function creates a new HV to
contain any appropriate messages.  The hash has three key-value pairs, as
follows:

=over 4

=item C<text>

The text of the message as a C<SVpv>.

=item C<warn_categories>

The warning category (or categories) packed into a C<SVuv>.

=item C<flag>

A single flag bit associated with this message, in a C<SVuv>.
The bit corresponds to some bit in the C<*errors> return value,
such as C<UNICODE_GOT_SURROGATE>.

=back

It's important to note that specifying this parameter as non-null will cause
any warnings this function would otherwise generate to be suppressed, and
instead be placed in C<*msgs>.  The caller can check the lexical warnings state
(or not) when choosing what to do with the returned messages.

The caller, of course, is responsible for freeing any returned HV.

=cut
*/

/* Undocumented; we don't want people using this.  Instead they should use
 * uvchr_to_utf8_flags_msgs() */
U8 *
Perl_uvoffuni_to_utf8_flags_msgs(pTHX_ U8 *d, UV uv, const UV flags, HV** msgs)
{
    PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS_MSGS;

    if (msgs) {
        *msgs = NULL;
    }

    if (OFFUNI_IS_INVARIANT(uv)) {
	*d++ = LATIN1_TO_NATIVE(uv);
	return d;
    }

    if (uv <= MAX_UTF8_TWO_BYTE) {
        *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
        *d++ = I8_TO_NATIVE_UTF8(( uv           & MASK) |   MARK);
        return d;
    }

    /* Not 2-byte; test for and handle 3-byte result.   In the test immediately
     * below, the 16 is for start bytes E0-EF (which are all the possible ones
     * for 3 byte characters).  The 2 is for 2 continuation bytes; these each
     * contribute SHIFT bits.  This yields 0x4000 on EBCDIC platforms, 0x1_0000
     * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
     * 0x800-0xFFFF on ASCII */
    if (uv < (16 * (1U << (2 * SHIFT)))) {
	*d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
	*d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) |   MARK);
	*d++ = I8_TO_NATIVE_UTF8(( uv  /* (1 - 1) */        & MASK) |   MARK);

#ifndef EBCDIC  /* These problematic code points are 4 bytes on EBCDIC, so
                   aren't tested here */
        /* The most likely code points in this range are below the surrogates.
         * Do an extra test to quickly exclude those. */
        if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
            if (UNLIKELY(   UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
                         || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
            {
                HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
            }
            else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
                HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
            }
        }
#endif
	return d;
    }

    /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
     * platforms, and 0x4000 on EBCDIC.  There are problematic cases that can
     * happen starting with 4-byte characters on ASCII platforms.  We unify the
     * code for these with EBCDIC, even though some of them require 5-bytes on
     * those, because khw believes the code saving is worth the very slight
     * performance hit on these high EBCDIC code points. */

    if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
        if (UNLIKELY(uv > MAX_EXTERNALLY_LEGAL_CP)) {
            Perl_croak(aTHX_ cp_above_legal_max, uv, MAX_EXTERNALLY_LEGAL_CP);
        }
        if (       (flags & UNICODE_WARN_SUPER)
            || (   (flags & UNICODE_WARN_PERL_EXTENDED)
                && UNICODE_IS_PERL_EXTENDED(uv)))
        {
            const char * format = super_cp_format;
            U32 category = packWARN(WARN_NON_UNICODE);
            U32 flag = UNICODE_GOT_SUPER;

            /* Choose the more dire applicable warning */
            if (UNICODE_IS_PERL_EXTENDED(uv)) {
                format = perl_extended_cp_format;
                if (flags & (UNICODE_WARN_PERL_EXTENDED
                            |UNICODE_DISALLOW_PERL_EXTENDED))
                {
                    flag = UNICODE_GOT_PERL_EXTENDED;
                }
            }

            if (msgs) {
                *msgs = new_msg_hv(Perl_form(aTHX_ format, uv),
                                   category, flag);
            }
            else {
                Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE), format, uv);
            }
        }
        if (       (flags & UNICODE_DISALLOW_SUPER)
            || (   (flags & UNICODE_DISALLOW_PERL_EXTENDED)
                &&  UNICODE_IS_PERL_EXTENDED(uv)))
        {
            return NULL;
        }
    }
    else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
        HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
    }

    /* Test for and handle 4-byte result.   In the test immediately below, the
     * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
     * characters).  The 3 is for 3 continuation bytes; these each contribute
     * SHIFT bits.  This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
     * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
     * 0x1_0000-0x1F_FFFF on ASCII */
    if (uv < (8 * (1U << (3 * SHIFT)))) {
	*d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
	*d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) |   MARK);
	*d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) |   MARK);
	*d++ = I8_TO_NATIVE_UTF8(( uv  /* (1 - 1) */        & MASK) |   MARK);

#ifdef EBCDIC   /* These were handled on ASCII platforms in the code for 3-byte
                   characters.  The end-plane non-characters for EBCDIC were
                   handled just above */
        if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
            HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
        }
        else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
            HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
        }
#endif

	return d;
    }

    /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
     * platforms, and 0x4000 on EBCDIC.  At this point we switch to a loop
     * format.  The unrolled version above turns out to not save all that much
     * time, and at these high code points (well above the legal Unicode range
     * on ASCII platforms, and well above anything in common use in EBCDIC),
     * khw believes that less code outweighs slight performance gains. */

    {
	STRLEN len  = OFFUNISKIP(uv);
	U8 *p = d+len-1;
	while (p > d) {
	    *p-- = I8_TO_NATIVE_UTF8((uv & MASK) | MARK);
	    uv >>= SHIFT;
	}
	*p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
	return d+len;
    }
}

/*
=for apidoc uvchr_to_utf8

Adds the UTF-8 representation of the native code point C<uv> to the end
of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
C<UTF8_MAXBYTES+1>) free bytes available.  The return value is the pointer to
the byte after the end of the new character.  In other words,

    d = uvchr_to_utf8(d, uv);

is the recommended wide native character-aware way of saying

    *(d++) = uv;

This function accepts any code point from 0..C<IV_MAX> as input.
C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.

It is possible to forbid or warn on non-Unicode code points, or those that may
be problematic by using L</uvchr_to_utf8_flags>.

=cut
*/

/* This is also a macro */
PERL_CALLCONV U8*       Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);

U8 *
Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
{
    return uvchr_to_utf8(d, uv);
}

/*
=for apidoc uvchr_to_utf8_flags

Adds the UTF-8 representation of the native code point C<uv> to the end
of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
C<UTF8_MAXBYTES+1>) free bytes available.  The return value is the pointer to
the byte after the end of the new character.  In other words,

    d = uvchr_to_utf8_flags(d, uv, flags);

or, in most cases,

    d = uvchr_to_utf8_flags(d, uv, 0);

This is the Unicode-aware way of saying

    *(d++) = uv;

If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as
input.  C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.

Specifying C<flags> can further restrict what is allowed and not warned on, as
follows:

If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
the function will raise a warning, provided UTF8 warnings are enabled.  If
instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
NULL.  If both flags are set, the function will both warn and return NULL.

Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
affect how the function handles a Unicode non-character.

And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
affect the handling of code points that are above the Unicode maximum of
0x10FFFF.  Languages other than Perl may not be able to accept files that
contain these.

The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
three DISALLOW flags.  C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
allowed inputs to the strict UTF-8 traditionally defined by Unicode.
Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
above-Unicode and surrogate flags, but not the non-character ones, as
defined in
L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
See L<perlunicode/Noncharacter code points>.

Extremely high code points were never specified in any standard, and require an
extension to UTF-8 to express, which Perl does.  It is likely that programs
written in something other than Perl would not be able to read files that
contain these; nor would Perl understand files written by something that uses a
different extension.  For these reasons, there is a separate set of flags that
can warn and/or disallow these extremely high code points, even if other
above-Unicode ones are accepted.  They are the C<UNICODE_WARN_PERL_EXTENDED>
and C<UNICODE_DISALLOW_PERL_EXTENDED> flags.  For more information see
L</C<UTF8_GOT_PERL_EXTENDED>>.  Of course C<UNICODE_DISALLOW_SUPER> will
treat all above-Unicode code points, including these, as malformations.  (Note
that the Unicode standard considers anything above 0x10FFFF to be illegal, but
there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1))

A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is
retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>.  Similarly,
C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
C<UNICODE_DISALLOW_PERL_EXTENDED>.  The names are misleading because on EBCDIC
platforms,these flags can apply to code points that actually do fit in 31 bits.
The new names accurately describe the situation in all cases.

=cut
*/

/* This is also a macro */
PERL_CALLCONV U8*       Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);

U8 *
Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
{
    return uvchr_to_utf8_flags(d, uv, flags);
}

#ifndef UV_IS_QUAD

STATIC int
S_is_utf8_cp_above_31_bits(const U8 * const s,
                           const U8 * const e,
                           const bool consider_overlongs)
{
    /* Returns TRUE if the first code point represented by the Perl-extended-
     * UTF-8-encoded string starting at 's', and looking no further than 'e -
     * 1' doesn't fit into 31 bytes.  That is, that if it is >= 2**31.
     *
     * The function handles the case where the input bytes do not include all
     * the ones necessary to represent a full character.  That is, they may be
     * the intial bytes of the representation of a code point, but possibly
     * the final ones necessary for the complete representation may be beyond
     * 'e - 1'.
     *
     * The function also can handle the case where the input is an overlong
     * sequence.  If 'consider_overlongs' is 0, the function assumes the
     * input is not overlong, without checking, and will return based on that
     * assumption.  If this parameter is 1, the function will go to the trouble
     * of figuring out if it actually evaluates to above or below 31 bits.
     *
     * The sequence is otherwise assumed to be well-formed, without checking.
     */

    const STRLEN len = e - s;
    int is_overlong;

    PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;

    assert(! UTF8_IS_INVARIANT(*s) && e > s);

#ifdef EBCDIC

    PERL_UNUSED_ARG(consider_overlongs);

    /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
     * mean a 32-bit or larger code point (0xFF is an invariant).  0xFE can
     * also be the start byte for a 31-bit code point; we need at least 2
     * bytes, and maybe up through 8 bytes, to determine that.  (It can also be
     * the start byte for an overlong sequence, but for 30-bit or smaller code
     * points, so we don't have to worry about overlongs on EBCDIC.) */
    if (*s != 0xFE) {
        return 0;
    }

    if (len == 1) {
        return -1;
    }

#else

    /* On ASCII, FE and FF are the only start bytes that can evaluate to
     * needing more than 31 bits. */
    if (LIKELY(*s < 0xFE)) {
        return 0;
    }

    /* What we have left are FE and FF.  Both of these require more than 31
     * bits unless they are for overlongs. */
    if (! consider_overlongs) {
        return 1;
    }

    /* Here, we have FE or FF.  If the input isn't overlong, it evaluates to
     * above 31 bits.  But we need more than one byte to discern this, so if
     * passed just the start byte, it could be an overlong evaluating to
     * smaller */
    if (len == 1) {
        return -1;
    }

    /* Having excluded len==1, and knowing that FE and FF are both valid start
     * bytes, we can call the function below to see if the sequence is
     * overlong.  (We don't need the full generality of the called function,
     * but for these huge code points, speed shouldn't be a consideration, and
     * the compiler does have enough information, since it's static to this
     * file, to optimize to just the needed parts.) */
    is_overlong = is_utf8_overlong_given_start_byte_ok(s, len);

    /* If it isn't overlong, more than 31 bits are required. */
    if (is_overlong == 0) {
        return 1;
    }

    /* If it is indeterminate if it is overlong, return that */
    if (is_overlong < 0) {
        return -1;
    }

    /* Here is overlong.  Such a sequence starting with FE is below 31 bits, as
     * the max it can be is 2**31 - 1 */
    if (*s == 0xFE) {
        return 0;
    }

#endif

    /* Here, ASCII and EBCDIC rejoin:
    *  On ASCII:   We have an overlong sequence starting with FF
    *  On EBCDIC:  We have a sequence starting with FE. */

    {   /* For C89, use a block so the declaration can be close to its use */

#ifdef EBCDIC

        /* U+7FFFFFFF (2 ** 31 - 1)
         *              [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10  11  12  13
         *   IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
         *    IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
         *   POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
         *         I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
         * U+80000000 (2 ** 31):
         *   IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
         *    IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
         *   POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
         *         I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
         *
         * and since we know that *s = \xfe, any continuation sequcence
         * following it that is gt the below is above 31 bits
                                                [0] [1] [2] [3] [4] [5] [6] */
        const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42";

#else

        /* FF overlong for U+7FFFFFFF (2 ** 31 - 1)
         *      ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF
         * FF overlong for U+80000000 (2 ** 31):
         *      ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80
         * and since we know that *s = \xff, any continuation sequcence
         * following it that is gt the below is above 30 bits
                                                [0] [1] [2] [3] [4] [5] [6] */
        const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81";


#endif
        const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1;
        const STRLEN cmp_len = MIN(conts_len, len - 1);

        /* Now compare the continuation bytes in s with the ones we have
         * compiled in that are for the largest 30 bit code point.  If we have
         * enough bytes available to determine the answer, or the bytes we do
         * have differ from them, we can compare the two to get a definitive
         * answer (Note that in UTF-EBCDIC, the two lowest possible
         * continuation bytes are \x41 and \x42.) */
        if (cmp_len >= conts_len || memNE(s + 1,
                                          conts_for_highest_30_bit,
                                          cmp_len))
        {
            return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len));
        }

        /* Here, all the bytes we have are the same as the highest 30-bit code
         * point, but we are missing so many bytes that we can't make the
         * determination */
        return -1;
    }
}

#endif

PERL_STATIC_INLINE int
S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
{
    /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
     * 's' + 'len' - 1 is an overlong.  It returns 1 if it is an overlong; 0 if
     * it isn't, and -1 if there isn't enough information to tell.  This last
     * return value can happen if the sequence is incomplete, missing some
     * trailing bytes that would form a complete character.  If there are
     * enough bytes to make a definitive decision, this function does so.
     * Usually 2 bytes sufficient.
     *
     * Overlongs can occur whenever the number of continuation bytes changes.
     * That means whenever the number of leading 1 bits in a start byte
     * increases from the next lower start byte.  That happens for start bytes
     * C0, E0, F0, F8, FC, FE, and FF.  On modern perls, the following illegal
     * start bytes have already been excluded, so don't need to be tested here;
     * ASCII platforms: C0, C1
     * EBCDIC platforms C0, C1, C2, C3, C4, E0
     */

    const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
    const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);

    PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
    assert(len > 1 && UTF8_IS_START(*s));

    /* Each platform has overlongs after the start bytes given above (expressed
     * in I8 for EBCDIC).  What constitutes an overlong varies by platform, but
     * the logic is the same, except the E0 overlong has already been excluded
     * on EBCDIC platforms.   The  values below were found by manually
     * inspecting the UTF-8 patterns.  See the tables in utf8.h and
     * utfebcdic.h. */

#       ifdef EBCDIC
#           define F0_ABOVE_OVERLONG 0xB0
#           define F8_ABOVE_OVERLONG 0xA8
#           define FC_ABOVE_OVERLONG 0xA4
#           define FE_ABOVE_OVERLONG 0xA2
#           define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
                                    /* I8(0xfe) is FF */
#       else

    if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
        return 1;
    }

#           define F0_ABOVE_OVERLONG 0x90
#           define F8_ABOVE_OVERLONG 0x88
#           define FC_ABOVE_OVERLONG 0x84
#           define FE_ABOVE_OVERLONG 0x82
#           define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
#       endif


    if (   (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
        || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
        || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
        || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
    {
        return 1;
    }

    /* Check for the FF overlong */
    return isFF_OVERLONG(s, len);
}

PERL_STATIC_INLINE int
S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
{
    /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
     * 'e' - 1 is an overlong beginning with \xFF.  It returns 1 if it is; 0 if
     * it isn't, and -1 if there isn't enough information to tell.  This last
     * return value can happen if the sequence is incomplete, missing some
     * trailing bytes that would form a complete character.  If there are
     * enough bytes to make a definitive decision, this function does so. */

    PERL_ARGS_ASSERT_ISFF_OVERLONG;

    /* To be an FF overlong, all the available bytes must match */
    if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
                     MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
    {
        return 0;
    }

    /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
     * be there; what comes after them doesn't matter.  See tables in utf8.h,
     * utfebcdic.h. */
    if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
        return 1;
    }

    /* The missing bytes could cause the result to go one way or the other, so
     * the result is indeterminate */
    return -1;
}

#if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */
#  ifdef EBCDIC     /* Actually is I8 */
#   define HIGHEST_REPRESENTABLE_UTF8                                       \
                "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
#  else
#   define HIGHEST_REPRESENTABLE_UTF8                                       \
                "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
#  endif
#endif

PERL_STATIC_INLINE int
S_does_utf8_overflow(const U8 * const s,
                     const U8 * e,
                     const bool consider_overlongs)
{
    /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
     * 'e' - 1 would overflow an IV on this platform; that is if it represents
     * a code point larger than the highest representable code point.  It
     * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't
     * enough information to tell.  This last return value can happen if the
     * sequence is incomplete, missing some trailing bytes that would form a
     * complete character.  If there are enough bytes to make a definitive
     * decision, this function does so.
     *
     * If 'consider_overlongs' is TRUE, the function checks for the possibility
     * that the sequence is an overlong that doesn't overflow.  Otherwise, it
     * assumes the sequence is not an overlong.  This can give different
     * results only on ASCII 32-bit platforms.
     *
     * (For ASCII platforms, we could use memcmp() because we don't have to
     * convert each byte to I8, but it's very rare input indeed that would
     * approach overflow, so the loop below will likely only get executed once.)
     *
     * 'e' - 1 must not be beyond a full character. */


    PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
    assert(s <= e && s + UTF8SKIP(s) >= e);

#if ! defined(UV_IS_QUAD)

    return is_utf8_cp_above_31_bits(s, e, consider_overlongs);

#else

    PERL_UNUSED_ARG(consider_overlongs);

    {
        const STRLEN len = e - s;
        const U8 *x;
        const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;

        for (x = s; x < e; x++, y++) {

            if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
                continue;
            }

            /* If this byte is larger than the corresponding highest UTF-8
             * byte, the sequence overflow; otherwise the byte is less than,
             * and so the sequence doesn't overflow */
            return NATIVE_UTF8_TO_I8(*x) > *y;

        }

        /* Got to the end and all bytes are the same.  If the input is a whole
         * character, it doesn't overflow.  And if it is a partial character,
         * there's not enough information to tell */
        if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) {
            return -1;
        }

        return 0;
    }

#endif

}

#if 0

/* This is the portions of the above function that deal with UV_MAX instead of
 * IV_MAX.  They are left here in case we want to combine them so that internal
 * uses can have larger code points.  The only logic difference is that the
 * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has
 * different logic.
 */

/* Anything larger than this will overflow the word if it were converted into a UV */
#if defined(UV_IS_QUAD)
#  ifdef EBCDIC     /* Actually is I8 */
#   define HIGHEST_REPRESENTABLE_UTF8                                       \
                "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
#  else
#   define HIGHEST_REPRESENTABLE_UTF8                                       \
                "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
#  endif
#else   /* 32-bit */
#  ifdef EBCDIC
#   define HIGHEST_REPRESENTABLE_UTF8                                       \
                "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
#  else
#   define HIGHEST_REPRESENTABLE_UTF8  "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
#  endif
#endif

#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)

    /* On 32 bit ASCII machines, many overlongs that start with FF don't
     * overflow */
    if (consider_overlongs && isFF_OVERLONG(s, len) > 0) {

        /* To be such an overlong, the first bytes of 's' must match
         * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80".  If we
         * don't have any additional bytes available, the sequence, when
         * completed might or might not fit in 32 bits.  But if we have that
         * next byte, we can tell for sure.  If it is <= 0x83, then it does
         * fit. */
        if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) {
            return -1;
        }

        return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83;
    }

/* Starting with the #else, the rest of the function is identical except
 *      1.  we need to move the 'len' declaration to be global to the function
 *      2.  the endif move to just after the UNUSED_ARG.
 * An empty endif is given just below to satisfy the preprocessor
 */
#endif

#endif

#undef F0_ABOVE_OVERLONG
#undef F8_ABOVE_OVERLONG
#undef FC_ABOVE_OVERLONG
#undef FE_ABOVE_OVERLONG
#undef FF_OVERLONG_PREFIX

STRLEN
Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
{
    STRLEN len;
    const U8 *x;

    /* A helper function that should not be called directly.
     *
     * This function returns non-zero if the string beginning at 's' and
     * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
     * code point; otherwise it returns 0.  The examination stops after the
     * first code point in 's' is validated, not looking at the rest of the
     * input.  If 'e' is such that there are not enough bytes to represent a
     * complete code point, this function will return non-zero anyway, if the
     * bytes it does have are well-formed UTF-8 as far as they go, and aren't
     * excluded by 'flags'.
     *
     * A non-zero return gives the number of bytes required to represent the
     * code point.  Be aware that if the input is for a partial character, the
     * return will be larger than 'e - s'.
     *
     * This function assumes that the code point represented is UTF-8 variant.
     * The caller should have excluded the possibility of it being invariant
     * before calling this function.
     *
     * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
     * accepted by L</utf8n_to_uvchr>.  If non-zero, this function will return
     * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
     * disallowed by the flags.  If the input is only for a partial character,
     * the function will return non-zero if there is any sequence of
     * well-formed UTF-8 that, when appended to the input sequence, could
     * result in an allowed code point; otherwise it returns 0.  Non characters
     * cannot be determined based on partial character input.  But many  of the
     * other excluded types can be determined with just the first one or two
     * bytes.
     *
     */

    PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;

    assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
                          |UTF8_DISALLOW_PERL_EXTENDED)));
    assert(! UTF8_IS_INVARIANT(*s));

    /* A variant char must begin with a start byte */
    if (UNLIKELY(! UTF8_IS_START(*s))) {
        return 0;
    }

    /* Examine a maximum of a single whole code point */
    if (e - s > UTF8SKIP(s)) {
        e = s + UTF8SKIP(s);
    }

    len = e - s;

    if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
        const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);

        /* Here, we are disallowing some set of largish code points, and the
         * first byte indicates the sequence is for a code point that could be
         * in the excluded set.  We generally don't have to look beyond this or
         * the second byte to see if the sequence is actually for one of the
         * excluded classes.  The code below is derived from this table:
         *
         *              UTF-8            UTF-EBCDIC I8
         *   U+D800: \xED\xA0\x80      \xF1\xB6\xA0\xA0      First surrogate
         *   U+DFFF: \xED\xBF\xBF      \xF1\xB7\xBF\xBF      Final surrogate
         * U+110000: \xF4\x90\x80\x80  \xF9\xA2\xA0\xA0\xA0  First above Unicode
         *
         * Keep in mind that legal continuation bytes range between \x80..\xBF
         * for UTF-8, and \xA0..\xBF for I8.  Anything above those aren't
         * continuation bytes.  Hence, we don't have to test the upper edge
         * because if any of those is encountered, the sequence is malformed,
         * and would fail elsewhere in this function.
         *
         * The code here likewise assumes that there aren't other
         * malformations; again the function should fail elsewhere because of
         * these.  For example, an overlong beginning with FC doesn't actually
         * have to be a super; it could actually represent a small code point,
         * even U+0000.  But, since overlongs (and other malformations) are
         * illegal, the function should return FALSE in either case.
         */

#ifdef EBCDIC   /* On EBCDIC, these are actually I8 bytes */
#  define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER  0xFA
#  define IS_UTF8_2_BYTE_SUPER(s0, s1)           ((s0) == 0xF9 && (s1) >= 0xA2)

#  define IS_UTF8_2_BYTE_SURROGATE(s0, s1)       ((s0) == 0xF1              \
                                                       /* B6 and B7 */      \
                                              && ((s1) & 0xFE ) == 0xB6)
#  define isUTF8_PERL_EXTENDED(s)   (*s == I8_TO_NATIVE_UTF8(0xFF))
#else
#  define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER  0xF5
#  define IS_UTF8_2_BYTE_SUPER(s0, s1)           ((s0) == 0xF4 && (s1) >= 0x90)
#  define IS_UTF8_2_BYTE_SURROGATE(s0, s1)       ((s0) == 0xED && (s1) >= 0xA0)
#  define isUTF8_PERL_EXTENDED(s)   (*s >= 0xFE)
#endif

        if (  (flags & UTF8_DISALLOW_SUPER)
            && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
        {
            return 0;           /* Above Unicode */
        }

        if (   (flags & UTF8_DISALLOW_PERL_EXTENDED)
            &&  UNLIKELY(isUTF8_PERL_EXTENDED(s)))
        {
            return 0;
        }

        if (len > 1) {
            const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);

            if (   (flags & UTF8_DISALLOW_SUPER)
                &&  UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
            {
                return 0;       /* Above Unicode */
            }

            if (   (flags & UTF8_DISALLOW_SURROGATE)
                &&  UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
            {
                return 0;       /* Surrogate */
            }

            if (  (flags & UTF8_DISALLOW_NONCHAR)
                && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
            {
                return 0;       /* Noncharacter code point */
            }
        }
    }

    /* Make sure that all that follows are continuation bytes */
    for (x = s + 1; x < e; x++) {
        if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
            return 0;
        }
    }

    /* Here is syntactically valid.  Next, make sure this isn't the start of an
     * overlong. */
    if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) {
        return 0;
    }

    /* And finally, that the code point represented fits in a word on this
     * platform */
    if (0 < does_utf8_overflow(s, e,
                               0 /* Don't consider overlongs */
                              ))
    {
        return 0;
    }

    return UTF8SKIP(s);
}

char *
Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format)
{
    /* Returns a mortalized C string that is a displayable copy of the 'len'
     * bytes starting at 'start'.  'format' gives how to display each byte.
     * Currently, there are only two formats, so it is currently a bool:
     *      0   \xab
     *      1    ab         (that is a space between two hex digit bytes)
     */

    const STRLEN output_len = 4 * len + 1;  /* 4 bytes per each input, plus a
                                               trailing NUL */
    const U8 * s = start;
    const U8 * const e = start + len;
    char * output;
    char * d;

    PERL_ARGS_ASSERT__BYTE_DUMP_STRING;

    Newx(output, output_len, char);
    SAVEFREEPV(output);

    d = output;
    for (s = start; s < e; s++) {
        const unsigned high_nibble = (*s & 0xF0) >> 4;
        const unsigned low_nibble =  (*s & 0x0F);

        if (format) {
            if (s > start) {
                *d++ = ' ';
            }
        }
        else {
            *d++ = '\\';
            *d++ = 'x';
        }

        if (high_nibble < 10) {
            *d++ = high_nibble + '0';
        }
        else {
            *d++ = high_nibble - 10 + 'a';
        }

        if (low_nibble < 10) {
            *d++ = low_nibble + '0';
        }
        else {
            *d++ = low_nibble - 10 + 'a';
        }
    }

    *d = '\0';
    return output;
}

PERL_STATIC_INLINE char *
S_unexpected_non_continuation_text(pTHX_ const U8 * const s,

                                         /* Max number of bytes to print */
                                         STRLEN print_len,

                                         /* Which one is the non-continuation */
                                         const STRLEN non_cont_byte_pos,

                                         /* How many bytes should there be? */
                                         const STRLEN expect_len)
{
    /* Return the malformation warning text for an unexpected continuation
     * byte. */

    const char * const where = (non_cont_byte_pos == 1)
                               ? "immediately"
                               : Perl_form(aTHX_ "%d bytes",
                                                 (int) non_cont_byte_pos);
    const U8 * x = s + non_cont_byte_pos;
    const U8 * e = s + print_len;

    PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;

    /* We don't need to pass this parameter, but since it has already been
     * calculated, it's likely faster to pass it; verify under DEBUGGING */
    assert(expect_len == UTF8SKIP(s));

    /* As a defensive coding measure, don't output anything past a NUL.  Such
     * bytes shouldn't be in the middle of a malformation, and could mark the
     * end of the allocated string, and what comes after is undefined */
    for (; x < e; x++) {
        if (*x == '\0') {
            x++;            /* Output this particular NUL */
            break;
        }
    }

    return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
                           " %s after start byte 0x%02x; need %d bytes, got %d)",
                           malformed_text,
                           _byte_dump_string(s, x - s, 0),
                           *(s + non_cont_byte_pos),
                           where,
                           *s,
                           (int) expect_len,
                           (int) non_cont_byte_pos);
}

/*

=for apidoc utf8n_to_uvchr

THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.

Bottom level UTF-8 decode routine.
Returns the native code point value of the first character in the string C<s>,
which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
the length, in bytes, of that character.

The value of C<flags> determines the behavior when C<s> does not point to a
well-formed UTF-8 character.  If C<flags> is 0, encountering a malformation
causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
is the next possible position in C<s> that could begin a non-malformed
character.  Also, if UTF-8 warnings haven't been lexically disabled, a warning
is raised.  Some UTF-8 input sequences may contain multiple malformations.
This function tries to find every possible one in each call, so multiple
warnings can be raised for the same sequence.

Various ALLOW flags can be set in C<flags> to allow (and not warn on)
individual types of malformations, such as the sequence being overlong (that
is, when there is a shorter sequence that can express the same code point;
overlong sequences are expressly forbidden in the UTF-8 standard due to
potential security issues).  Another malformation example is the first byte of
a character not being a legal first byte.  See F<utf8.h> for the list of such
flags.  Even if allowed, this function generally returns the Unicode
REPLACEMENT CHARACTER when it encounters a malformation.  There are flags in
F<utf8.h> to override this behavior for the overlong malformations, but don't
do that except for very specialized purposes.

The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
flags) malformation is found.  If this flag is set, the routine assumes that
the caller will raise a warning, and this function will silently just set
C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.

Note that this API requires disambiguation between successful decoding a C<NUL>
character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
be set to 1.  To disambiguate, upon a zero return, see if the first byte of
C<s> is 0 as well.  If so, the input was a C<NUL>; if not, the input had an
error.  Or you can use C<L</utf8n_to_uvchr_error>>.

Certain code points are considered problematic.  These are Unicode surrogates,
Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
By default these are considered regular code points, but certain situations
warrant special handling for them, which can be specified using the C<flags>
parameter.  If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
three classes are treated as malformations and handled as such.  The flags
C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
disallow these categories individually.  C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
restricts the allowed inputs to the strict UTF-8 traditionally defined by
Unicode.  Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
definition given by
L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
The difference between traditional strictness and C9 strictness is that the
latter does not forbid non-character code points.  (They are still discouraged,
however.)  For more discussion see L<perlunicode/Noncharacter code points>.

The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
raised for their respective categories, but otherwise the code points are
considered valid (not malformations).  To get a category to both be treated as
a malformation and raise a warning, specify both the WARN and DISALLOW flags.
(But note that warnings are not raised if lexically disabled nor if
C<UTF8_CHECK_ONLY> is also specified.)

Extremely high code points were never specified in any standard, and require an
extension to UTF-8 to express, which Perl does.  It is likely that programs
written in something other than Perl would not be able to read files that
contain these; nor would Perl understand files written by something that uses a
different extension.  For these reasons, there is a separate set of flags that
can warn and/or disallow these extremely high code points, even if other
above-Unicode ones are accepted.  They are the C<UTF8_WARN_PERL_EXTENDED> and
C<UTF8_DISALLOW_PERL_EXTENDED> flags.  For more information see
L</C<UTF8_GOT_PERL_EXTENDED>>.  Of course C<UTF8_DISALLOW_SUPER> will treat all
above-Unicode code points, including these, as malformations.
(Note that the Unicode standard considers anything above 0x10FFFF to be
illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
(2**31 -1))

A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>.  Similarly,
C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
C<UTF8_DISALLOW_PERL_EXTENDED>.  The names are misleading because these flags
can apply to code points that actually do fit in 31 bits.  This happens on
EBCDIC platforms, and sometimes when the L<overlong
malformation|/C<UTF8_GOT_LONG>> is also present.  The new names accurately
describe the situation in all cases.


All other code points corresponding to Unicode characters, including private
use and those yet to be assigned, are never considered malformed and never
warn.

=cut

Also implemented as a macro in utf8.h
*/

UV
Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
                          STRLEN curlen,
                          STRLEN *retlen,
                          const U32 flags)
{
    PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;

    return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
}

/*

=for apidoc utf8n_to_uvchr_error

THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.

This function is for code that needs to know what the precise malformation(s)
are when an error is found.  If you also need to know the generated warning
messages, use L</utf8n_to_uvchr_msgs>() instead.

It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
all the others, C<errors>.  If this parameter is 0, this function behaves
identically to C<L</utf8n_to_uvchr>>.  Otherwise, C<errors> should be a pointer
to a C<U32> variable, which this function sets to indicate any errors found.
Upon return, if C<*errors> is 0, there were no errors found.  Otherwise,
C<*errors> is the bit-wise C<OR> of the bits described in the list below.  Some
of these bits will be set if a malformation is found, even if the input
C<flags> parameter indicates that the given malformation is allowed; those
exceptions are noted:

=over 4

=item C<UTF8_GOT_PERL_EXTENDED>

The input sequence is not standard UTF-8, but a Perl extension.  This bit is
set only if the input C<flags> parameter contains either the
C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.

Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
and so some extension must be used to express them.  Perl uses a natural
extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
extension to represent even higher ones, so that any code point that fits in a
64-bit word can be represented.  Text using these extensions is not likely to
be portable to non-Perl code.  We lump both of these extensions together and
refer to them as Perl extended UTF-8.  There exist other extensions that people
have invented, incompatible with Perl's.

On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
than on ASCII.  Prior to that, code points 2**31 and higher were simply
unrepresentable, and a different, incompatible method was used to represent
code points between 2**30 and 2**31 - 1.

On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
Perl extended UTF-8 is used.

In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
may use for backward compatibility.  That name is misleading, as this flag may
be set when the code point actually does fit in 31 bits.  This happens on
EBCDIC platforms, and sometimes when the L<overlong
malformation|/C<UTF8_GOT_LONG>> is also present.  The new name accurately
describes the situation in all cases.

=item C<UTF8_GOT_CONTINUATION>

The input sequence was malformed in that the first byte was a a UTF-8
continuation byte.

=item C<UTF8_GOT_EMPTY>

The input C<curlen> parameter was 0.

=item C<UTF8_GOT_LONG>

The input sequence was malformed in that there is some other sequence that
evaluates to the same code point, but that sequence is shorter than this one.

Until Unicode 3.1, it was legal for programs to accept this malformation, but
it was discovered that this created security issues.

=item C<UTF8_GOT_NONCHAR>

The code point represented by the input UTF-8 sequence is for a Unicode
non-character code point.
This bit is set only if the input C<flags> parameter contains either the
C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.

=item C<UTF8_GOT_NON_CONTINUATION>

The input sequence was malformed in that a non-continuation type byte was found
in a position where only a continuation type one should be.

=item C<UTF8_GOT_OVERFLOW>

The input sequence was malformed in that it is for a code point that is not
representable in the number of bits available in an IV on the current platform.

=item C<UTF8_GOT_SHORT>

The input sequence was malformed in that C<curlen> is smaller than required for
a complete sequence.  In other words, the input is for a partial character
sequence.

=item C<UTF8_GOT_SUPER>

The input sequence was malformed in that it is for a non-Unicode code point;
that is, one above the legal Unicode maximum.
This bit is set only if the input C<flags> parameter contains either the
C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.

=item C<UTF8_GOT_SURROGATE>

The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
code point.
This bit is set only if the input C<flags> parameter contains either the
C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.

=back

To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
flag to suppress any warnings, and then examine the C<*errors> return.

=cut

Also implemented as a macro in utf8.h
*/

UV
Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
                          STRLEN curlen,
                          STRLEN *retlen,
                          const U32 flags,
                          U32 * errors)
{
    PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;

    return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL);
}

/*

=for apidoc utf8n_to_uvchr_msgs

THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.

This function is for code that needs to know what the precise malformation(s)
are when an error is found, and wants the corresponding warning and/or error
messages to be returned to the caller rather than be displayed.  All messages
that would have been displayed if all lexcial warnings are enabled will be
returned.

It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter
placed after all the others, C<msgs>.  If this parameter is 0, this function
behaves identically to C<L</utf8n_to_uvchr_error>>.  Otherwise, C<msgs> should
be a pointer to an C<AV *> variable, in which this function creates a new AV to
contain any appropriate messages.  The elements of the array are ordered so
that the first message that would have been displayed is in the 0th element,
and so on.  Each element is a hash with three key-value pairs, as follows:

=over 4

=item C<text>

The text of the message as a C<SVpv>.

=item C<warn_categories>

The warning category (or categories) packed into a C<SVuv>.

=item C<flag>

A single flag bit associated with this message, in a C<SVuv>.
The bit corresponds to some bit in the C<*errors> return value,
such as C<UTF8_GOT_LONG>.

=back

It's important to note that specifying this parameter as non-null will cause
any warnings this function would otherwise generate to be suppressed, and
instead be placed in C<*msgs>.  The caller can check the lexical warnings state
(or not) when choosing what to do with the returned messages.

If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence
no AV is created.

The caller, of course, is responsible for freeing any returned AV.

=cut
*/

UV
Perl_utf8n_to_uvchr_msgs(pTHX_ const U8 *s,
                               STRLEN curlen,
                               STRLEN *retlen,
                               const U32 flags,
                               U32 * errors,
                               AV ** msgs)
{
    const U8 * const s0 = s;
    const U8 * send = s0 + curlen;
    U32 possible_problems = 0;  /* A bit is set here for each potential problem
                                   found as we go along */
    UV uv = (UV) -1;
    STRLEN expectlen   = 0;     /* How long should this sequence be?
                                   (initialized to silence compilers' wrong
                                   warning) */
    STRLEN avail_len   = 0;     /* When input is too short, gives what that is */
    U32 discard_errors = 0;     /* Used to save branches when 'errors' is NULL;
                                   this gets set and discarded */

    /* The below are used only if there is both an overlong malformation and a
     * too short one.  Otherwise the first two are set to 's0' and 'send', and
     * the third not used at all */
    U8 * adjusted_s0 = (U8 *) s0;
    U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
                                            routine; see [perl #130921] */
    UV uv_so_far = 0;   /* (Initialized to silence compilers' wrong warning) */

    UV state = 0;

    PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_MSGS;

    if (errors) {
        *errors = 0;
    }
    else {
        errors = &discard_errors;
    }

    /* The order of malformation tests here is important.  We should consume as
     * few bytes as possible in order to not skip any valid character.  This is
     * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
     * http://unicode.org/reports/tr36 for more discussion as to why.  For
     * example, once we've done a UTF8SKIP, we can tell the expected number of
     * bytes, and could fail right off the bat if the input parameters indicate
     * that there are too few available.  But it could be that just that first
     * byte is garbled, and the intended character occupies fewer bytes.  If we
     * blindly assumed that the first byte is correct, and skipped based on
     * that number, we could skip over a valid input character.  So instead, we
     * always examine the sequence byte-by-byte.
     *
     * We also should not consume too few bytes, otherwise someone could inject
     * things.  For example, an input could be deliberately designed to
     * overflow, and if this code bailed out immediately upon discovering that,
     * returning to the caller C<*retlen> pointing to the very next byte (one
     * which is actually part of of the overflowing sequence), that could look
     * legitimate to the caller, which could discard the initial partial
     * sequence and process the rest, inappropriately.
     *
     * Some possible input sequences are malformed in more than one way.  This
     * function goes to lengths to try to find all of them.  This is necessary
     * for correctness, as the inputs may allow one malformation but not
     * another, and if we abandon searching for others after finding the
     * allowed one, we could allow in something that shouldn't have been.
     */

    if (UNLIKELY(curlen == 0)) {
        possible_problems |= UTF8_GOT_EMPTY;
        curlen = 0;
        uv = UNICODE_REPLACEMENT;
	goto ready_to_handle_errors;
    }

    expectlen = UTF8SKIP(s);

    /* A well-formed UTF-8 character, as the vast majority of calls to this
     * function will be for, has this expected length.  For efficiency, set
     * things up here to return it.  It will be overriden only in those rare
     * cases where a malformation is found */
    if (retlen) {
	*retlen = expectlen;
    }

    /* An invariant is trivially well-formed */
    if (UTF8_IS_INVARIANT(*s0)) {
	return *s0;
    }

    /* Measurements show that this dfa is somewhat faster than the regular code
     * below, so use it first, dropping down for the non-normal cases. */

#define PERL_UTF8_DECODE_REJECT 1

    while (s < send && LIKELY(state != PERL_UTF8_DECODE_REJECT)) {
        UV type = strict_utf8_dfa_tab[*s];

        uv = (state == 0)
             ?  ((0xff >> type) & NATIVE_UTF8_TO_I8(*s))
             : UTF8_ACCUMULATE(uv, *s);
        state = strict_utf8_dfa_tab[256 + state + type];

        if (state == 0) {
            return uv;
        }

        s++;
    }

    /* Here, is one of: a) malformed; b) a problematic code point (surrogate,
     * non-unicode, or nonchar); or c) on ASCII platforms, one of the Hangul
     * syllables that the dfa doesn't properly handle.  Quickly dispose of the
     * final case.
     *
     * Each of the affected Hanguls starts with \xED */

#ifndef EBCDIC

    if (is_HANGUL_ED_utf8_safe(s0, send)) {
        return ((0xED & UTF_START_MASK(3)) << (2 * UTF_ACCUMULATION_SHIFT))
             | ((s0[1] & UTF_CONTINUATION_MASK) << UTF_ACCUMULATION_SHIFT)
             |  (s0[2] & UTF_CONTINUATION_MASK);
    }

#endif

    /* Here is potentially problematic.  Use the full mechanism */
    uv = *s0;

    /* In conjunction with the exhaustive tests that can be enabled in
     * APItest/t/utf8_warn_base.pl, this can make sure the dfa does precisely
     * what it is intended to do, and that no flaws in it are masked by
     * dropping down and executing the code below

    assert(! isUTF8_CHAR(s0, send)
          || UTF8_IS_SURROGATE(s0, send)
          || UTF8_IS_SUPER(s0, send)
          || UTF8_IS_NONCHAR(s0,send));
    */

    /* A continuation character can't start a valid sequence */
    if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
	possible_problems |= UTF8_GOT_CONTINUATION;
        curlen = 1;
        uv = UNICODE_REPLACEMENT;
	goto ready_to_handle_errors;
    }

    /* Here is not a continuation byte, nor an invariant.  The only thing left
     * is a start byte (possibly for an overlong).  (We can't use UTF8_IS_START
     * because it excludes start bytes like \xC0 that always lead to
     * overlongs.) */

    /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
     * that indicate the number of bytes in the character's whole UTF-8
     * sequence, leaving just the bits that are part of the value.  */
    uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);

    /* Setup the loop end point, making sure to not look past the end of the
     * input string, and flag it as too short if the size isn't big enough. */
    if (UNLIKELY(curlen < expectlen)) {
        possible_problems |= UTF8_GOT_SHORT;
        avail_len = curlen;
    }
    else {
        send = (U8*) s0 + expectlen;
    }

    /* Now, loop through the remaining bytes in the character's sequence,
     * accumulating each into the working value as we go. */
    for (s = s0 + 1; s < send; s++) {
	if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
	    uv = UTF8_ACCUMULATE(uv, *s);
            continue;
        }

        /* Here, found a non-continuation before processing all expected bytes.
         * This byte indicates the beginning of a new character, so quit, even
         * if allowing this malformation. */
        possible_problems |= UTF8_GOT_NON_CONTINUATION;
        break;
    } /* End of loop through the character's bytes */

    /* Save how many bytes were actually in the character */
    curlen = s - s0;

    /* Note that there are two types of too-short malformation.  One is when
     * there is actual wrong data before the normal termination of the
     * sequence.  The other is that the sequence wasn't complete before the end
     * of the data we are allowed to look at, based on the input 'curlen'.
     * This means that we were passed data for a partial character, but it is
     * valid as far as we saw.  The other is definitely invalid.  This
     * distinction could be important to a caller, so the two types are kept
     * separate.
     *
     * A convenience macro that matches either of the too-short conditions.  */
#   define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)

    if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
        uv_so_far = uv;
        uv = UNICODE_REPLACEMENT;
    }

    /* Check for overflow.  The algorithm requires us to not look past the end
     * of the current character, even if partial, so the upper limit is 's' */
    if (UNLIKELY(0 < does_utf8_overflow(s0, s,
                                         1 /* Do consider overlongs */
                                        )))
    {
        possible_problems |= UTF8_GOT_OVERFLOW;
        uv = UNICODE_REPLACEMENT;
    }

    /* Check for overlong.  If no problems so far, 'uv' is the correct code
     * point value.  Simply see if it is expressible in fewer bytes.  Otherwise
     * we must look at the UTF-8 byte sequence itself to see if it is for an
     * overlong */
    if (     (   LIKELY(! possible_problems)
              && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
        || (       UNLIKELY(possible_problems)
            && (   UNLIKELY(! UTF8_IS_START(*s0))
                || (   curlen > 1
                    && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
                                                                s - s0))))))
    {
        possible_problems |= UTF8_GOT_LONG;

        if (   UNLIKELY(   possible_problems & UTF8_GOT_TOO_SHORT)

                          /* The calculation in the 'true' branch of this 'if'
                           * below won't work if overflows, and isn't needed
                           * anyway.  Further below we handle all overflow
                           * cases */
            &&   LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
        {
            UV min_uv = uv_so_far;
            STRLEN i;

            /* Here, the input is both overlong and is missing some trailing
             * bytes.  There is no single code point it could be for, but there
             * may be enough information present to determine if what we have
             * so far is for an unallowed code point, such as for a surrogate.
             * The code further below has the intelligence to determine this,
             * but just for non-overlong UTF-8 sequences.  What we do here is
             * calculate the smallest code point the input could represent if
             * there were no too short malformation.  Then we compute and save
             * the UTF-8 for that, which is what the code below looks at
             * instead of the raw input.  It turns out that the smallest such
             * code point is all we need. */
            for (i = curlen; i < expectlen; i++) {
                min_uv = UTF8_ACCUMULATE(min_uv,
                                     I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
            }

            adjusted_s0 = temp_char_buf;
            (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
        }
    }

    /* Here, we have found all the possible problems, except for when the input
     * is for a problematic code point not allowed by the input parameters. */

                                /* uv is valid for overlongs */
    if (   (   (      LIKELY(! (possible_problems & ~UTF8_GOT_LONG))

                      /* isn't problematic if < this */
                   && uv >= UNICODE_SURROGATE_FIRST)
            || (   UNLIKELY(possible_problems)

                          /* if overflow, we know without looking further
                           * precisely which of the problematic types it is,
                           * and we deal with those in the overflow handling
                           * code */
                && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
                && (   isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
                    || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
	&& ((flags & ( UTF8_DISALLOW_NONCHAR
                      |UTF8_DISALLOW_SURROGATE
                      |UTF8_DISALLOW_SUPER
                      |UTF8_DISALLOW_PERL_EXTENDED
	              |UTF8_WARN_NONCHAR
                      |UTF8_WARN_SURROGATE
                      |UTF8_WARN_SUPER
                      |UTF8_WARN_PERL_EXTENDED))))
    {
        /* If there were no malformations, or the only malformation is an
         * overlong, 'uv' is valid */
        if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
            if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
                possible_problems |= UTF8_GOT_SURROGATE;
            }
            else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
                possible_problems |= UTF8_GOT_SUPER;
            }
            else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
                possible_problems |= UTF8_GOT_NONCHAR;
            }
        }
        else {  /* Otherwise, need to look at the source UTF-8, possibly
                   adjusted to be non-overlong */

            if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
                                >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
            {
                possible_problems |= UTF8_GOT_SUPER;
            }
            else if (curlen > 1) {
                if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
                                      NATIVE_UTF8_TO_I8(*adjusted_s0),
                                      NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
                {
                    possible_problems |= UTF8_GOT_SUPER;
                }
                else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
                                      NATIVE_UTF8_TO_I8(*adjusted_s0),
                                      NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
                {
                    possible_problems |= UTF8_GOT_SURROGATE;
                }
            }

            /* We need a complete well-formed UTF-8 character to discern
             * non-characters, so can't look for them here */
        }
    }

  ready_to_handle_errors:

    /* At this point:
     * curlen               contains the number of bytes in the sequence that
     *                      this call should advance the input by.
     * avail_len            gives the available number of bytes passed in, but
     *                      only if this is less than the expected number of
     *                      bytes, based on the code point's start byte.
     * possible_problems'   is 0 if there weren't any problems; otherwise a bit
     *                      is set in it for each potential problem found.
     * uv                   contains the code point the input sequence
     *                      represents; or if there is a problem that prevents
     *                      a well-defined value from being computed, it is
     *                      some subsitute value, typically the REPLACEMENT
     *                      CHARACTER.
     * s0                   points to the first byte of the character
     * s                    points to just after were we left off processing
     *                      the character
     * send                 points to just after where that character should
     *                      end, based on how many bytes the start byte tells
     *                      us should be in it, but no further than s0 +
     *                      avail_len
     */

    if (UNLIKELY(possible_problems)) {
        bool disallowed = FALSE;
        const U32 orig_problems = possible_problems;

        if (msgs) {
            *msgs = NULL;
        }

        while (possible_problems) { /* Handle each possible problem */
            UV pack_warn = 0;
            char * message = NULL;
            U32 this_flag_bit = 0;

            /* Each 'if' clause handles one problem.  They are ordered so that
             * the first ones' messages will be displayed before the later
             * ones; this is kinda in decreasing severity order.  But the
             * overlong must come last, as it changes 'uv' looked at by the
             * others */
            if (possible_problems & UTF8_GOT_OVERFLOW) {

                /* Overflow means also got a super and are using Perl's
                 * extended UTF-8, but we handle all three cases here */
                possible_problems
                  &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
                *errors |= UTF8_GOT_OVERFLOW;

                /* But the API says we flag all errors found */
                if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
                    *errors |= UTF8_GOT_SUPER;
                }
                if (flags
                        & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
                {
                    *errors |= UTF8_GOT_PERL_EXTENDED;
                }

                /* Disallow if any of the three categories say to */
                if ( ! (flags &   UTF8_ALLOW_OVERFLOW)
                    || (flags & ( UTF8_DISALLOW_SUPER
                                 |UTF8_DISALLOW_PERL_EXTENDED)))
                {
                    disallowed = TRUE;
                }

                /* Likewise, warn if any say to */
                if (  ! (flags & UTF8_ALLOW_OVERFLOW)
                    ||  (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
                {

                    /* The warnings code explicitly says it doesn't handle the
                     * case of packWARN2 and two categories which have
                     * parent-child relationship.  Even if it works now to
                     * raise the warning if either is enabled, it wouldn't
                     * necessarily do so in the future.  We output (only) the
                     * most dire warning */
                    if (! (flags & UTF8_CHECK_ONLY)) {
                        if (msgs || ckWARN_d(WARN_UTF8)) {
                            pack_warn = packWARN(WARN_UTF8);
                        }
                        else if (msgs || ckWARN_d(WARN_NON_UNICODE)) {
                            pack_warn = packWARN(WARN_NON_UNICODE);
                        }
                        if (pack_warn) {
                            message = Perl_form(aTHX_ "%s: %s (overflows)",
                                            malformed_text,
                                            _byte_dump_string(s0, curlen, 0));
                            this_flag_bit = UTF8_GOT_OVERFLOW;
                        }
                    }
                }
            }
            else if (possible_problems & UTF8_GOT_EMPTY) {
                possible_problems &= ~UTF8_GOT_EMPTY;
                *errors |= UTF8_GOT_EMPTY;

                if (! (flags & UTF8_ALLOW_EMPTY)) {

                    /* This so-called malformation is now treated as a bug in
                     * the caller.  If you have nothing to decode, skip calling
                     * this function */
                    assert(0);

                    disallowed = TRUE;
                    if (  (msgs
                        || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
                    {
                        pack_warn = packWARN(WARN_UTF8);
                        message = Perl_form(aTHX_ "%s (empty string)",
                                                   malformed_text);
                        this_flag_bit = UTF8_GOT_EMPTY;
                    }
                }
            }
            else if (possible_problems & UTF8_GOT_CONTINUATION) {
                possible_problems &= ~UTF8_GOT_CONTINUATION;
                *errors |= UTF8_GOT_CONTINUATION;

                if (! (flags & UTF8_ALLOW_CONTINUATION)) {
                    disallowed = TRUE;
                    if ((   msgs
                         || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
                    {
                        pack_warn = packWARN(WARN_UTF8);
                        message = Perl_form(aTHX_
                                "%s: %s (unexpected continuation byte 0x%02x,"
                                " with no preceding start byte)",
                                malformed_text,
                                _byte_dump_string(s0, 1, 0), *s0);
                        this_flag_bit = UTF8_GOT_CONTINUATION;
                    }
                }
            }
            else if (possible_problems & UTF8_GOT_SHORT) {
                possible_problems &= ~UTF8_GOT_SHORT;
                *errors |= UTF8_GOT_SHORT;

                if (! (flags & UTF8_ALLOW_SHORT)) {
                    disallowed = TRUE;
                    if ((   msgs
                         || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
                    {
                        pack_warn = packWARN(WARN_UTF8);
                        message = Perl_form(aTHX_
                             "%s: %s (too short; %d byte%s available, need %d)",
                             malformed_text,
                             _byte_dump_string(s0, send - s0, 0),
                             (int)avail_len,
                             avail_len == 1 ? "" : "s",
                             (int)expectlen);
                        this_flag_bit = UTF8_GOT_SHORT;
                    }
                }

            }
            else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
                possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
                *errors |= UTF8_GOT_NON_CONTINUATION;

                if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
                    disallowed = TRUE;
                    if ((   msgs
                         || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
                    {

                        /* If we don't know for sure that the input length is
                         * valid, avoid as much as possible reading past the
                         * end of the buffer */
                        int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
                                       ? s - s0
                                       : send - s0;
                        pack_warn = packWARN(WARN_UTF8);
                        message = Perl_form(aTHX_ "%s",
                            unexpected_non_continuation_text(s0,
                                                            printlen,
                                                            s - s0,
                                                            (int) expectlen));
                        this_flag_bit = UTF8_GOT_NON_CONTINUATION;
                    }
                }
            }
            else if (possible_problems & UTF8_GOT_SURROGATE) {
                possible_problems &= ~UTF8_GOT_SURROGATE;

                if (flags & UTF8_WARN_SURROGATE) {
                    *errors |= UTF8_GOT_SURROGATE;

                    if (   ! (flags & UTF8_CHECK_ONLY)
                        && (msgs || ckWARN_d(WARN_SURROGATE)))
                    {
                        pack_warn = packWARN(WARN_SURROGATE);

                        /* These are the only errors that can occur with a
                        * surrogate when the 'uv' isn't valid */
                        if (orig_problems & UTF8_GOT_TOO_SHORT) {
                            message = Perl_form(aTHX_
                                    "UTF-16 surrogate (any UTF-8 sequence that"
                                    " starts with \"%s\" is for a surrogate)",
                                    _byte_dump_string(s0, curlen, 0));
                        }
                        else {
                            message = Perl_form(aTHX_ surrogate_cp_format, uv);
                        }
                        this_flag_bit = UTF8_GOT_SURROGATE;
                    }
                }

                if (flags & UTF8_DISALLOW_SURROGATE) {
                    disallowed = TRUE;
                    *errors |= UTF8_GOT_SURROGATE;
                }
            }
            else if (possible_problems & UTF8_GOT_SUPER) {
                possible_problems &= ~UTF8_GOT_SUPER;

                if (flags & UTF8_WARN_SUPER) {
                    *errors |= UTF8_GOT_SUPER;

                    if (   ! (flags & UTF8_CHECK_ONLY)
                        && (msgs || ckWARN_d(WARN_NON_UNICODE)))
                    {
                        pack_warn = packWARN(WARN_NON_UNICODE);

                        if (orig_problems & UTF8_GOT_TOO_SHORT) {
                            message = Perl_form(aTHX_
                                    "Any UTF-8 sequence that starts with"
                                    " \"%s\" is for a non-Unicode code point,"
                                    " may not be portable",
                                    _byte_dump_string(s0, curlen, 0));
                        }
                        else {
                            message = Perl_form(aTHX_ super_cp_format, uv);
                        }
                        this_flag_bit = UTF8_GOT_SUPER;
                    }
                }

                /* Test for Perl's extended UTF-8 after the regular SUPER ones,
                 * and before possibly bailing out, so that the more dire
                 * warning will override the regular one. */
                if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
                    if (  ! (flags & UTF8_CHECK_ONLY)
                        &&  (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
                        &&  (msgs || ckWARN_d(WARN_NON_UNICODE)))
                    {
                        pack_warn = packWARN(WARN_NON_UNICODE);

                        /* If it is an overlong that evaluates to a code point
                         * that doesn't have to use the Perl extended UTF-8, it
                         * still used it, and so we output a message that
                         * doesn't refer to the code point.  The same is true
                         * if there was a SHORT malformation where the code
                         * point is not valid.  In that case, 'uv' will have
                         * been set to the REPLACEMENT CHAR, and the message
                         * below without the code point in it will be selected
                         * */
                        if (UNICODE_IS_PERL_EXTENDED(uv)) {
                            message = Perl_form(aTHX_
                                            perl_extended_cp_format, uv);
                        }
                        else {
                            message = Perl_form(aTHX_
                                        "Any UTF-8 sequence that starts with"
                                        " \"%s\" is a Perl extension, and"
                                        " so is not portable",
                                        _byte_dump_string(s0, curlen, 0));
                        }
                        this_flag_bit = UTF8_GOT_PERL_EXTENDED;
                    }

                    if (flags & ( UTF8_WARN_PERL_EXTENDED
                                 |UTF8_DISALLOW_PERL_EXTENDED))
                    {
                        *errors |= UTF8_GOT_PERL_EXTENDED;

                        if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
                            disallowed = TRUE;
                        }
                    }
                }

                if (flags & UTF8_DISALLOW_SUPER) {
                    *errors |= UTF8_GOT_SUPER;
                    disallowed = TRUE;
                }
            }
            else if (possible_problems & UTF8_GOT_NONCHAR) {
                possible_problems &= ~UTF8_GOT_NONCHAR;

                if (flags & UTF8_WARN_NONCHAR) {
                    *errors |= UTF8_GOT_NONCHAR;

                    if (  ! (flags & UTF8_CHECK_ONLY)
                        && (msgs || ckWARN_d(WARN_NONCHAR)))
                    {
                        /* The code above should have guaranteed that we don't
                         * get here with errors other than overlong */
                        assert (! (orig_problems
                                        & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));

                        pack_warn = packWARN(WARN_NONCHAR);
                        message = Perl_form(aTHX_ nonchar_cp_format, uv);
                        this_flag_bit = UTF8_GOT_NONCHAR;
                    }
                }

                if (flags & UTF8_DISALLOW_NONCHAR) {
                    disallowed = TRUE;
                    *errors |= UTF8_GOT_NONCHAR;
                }
            }
            else if (possible_problems & UTF8_GOT_LONG) {
                possible_problems &= ~UTF8_GOT_LONG;
                *errors |= UTF8_GOT_LONG;

                if (flags & UTF8_ALLOW_LONG) {

                    /* We don't allow the actual overlong value, unless the
                     * special extra bit is also set */
                    if (! (flags & (   UTF8_ALLOW_LONG_AND_ITS_VALUE
                                    & ~UTF8_ALLOW_LONG)))
                    {
                        uv = UNICODE_REPLACEMENT;
                    }
                }
                else {
                    disallowed = TRUE;

                    if ((   msgs
                         || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
                    {
                        pack_warn = packWARN(WARN_UTF8);

                        /* These error types cause 'uv' to be something that
                         * isn't what was intended, so can't use it in the
                         * message.  The other error types either can't
                         * generate an overlong, or else the 'uv' is valid */
                        if (orig_problems &
                                        (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
                        {
                            message = Perl_form(aTHX_
                                    "%s: %s (any UTF-8 sequence that starts"
                                    " with \"%s\" is overlong which can and"
                                    " should be represented with a"
                                    " different, shorter sequence)",
                                    malformed_text,
                                    _byte_dump_string(s0, send - s0, 0),
                                    _byte_dump_string(s0, curlen, 0));
                        }
                        else {
                            U8 tmpbuf[UTF8_MAXBYTES+1];
                            const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
                                                                        uv, 0);
                            /* Don't use U+ for non-Unicode code points, which
                             * includes those in the Latin1 range */
                            const char * preface = (    uv > PERL_UNICODE_MAX
#ifdef EBCDIC
                                                     || uv <= 0xFF
#endif
                                                    )
                                                   ? "0x"
                                                   : "U+";
                            message = Perl_form(aTHX_
                                "%s: %s (overlong; instead use %s to represent"
                                " %s%0*" UVXf ")",
                                malformed_text,
                                _byte_dump_string(s0, send - s0, 0),
                                _byte_dump_string(tmpbuf, e - tmpbuf, 0),
                                preface,
                                ((uv < 256) ? 2 : 4), /* Field width of 2 for
                                                         small code points */
                                UNI_TO_NATIVE(uv));
                        }
                        this_flag_bit = UTF8_GOT_LONG;
                    }
                }
            } /* End of looking through the possible flags */

            /* Display the message (if any) for the problem being handled in
             * this iteration of the loop */
            if (message) {
                if (msgs) {
                    assert(this_flag_bit);

                    if (*msgs == NULL) {
                        *msgs = newAV();
                    }

                    av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message,
                                                                pack_warn,
                                                                this_flag_bit)));
                }
                else if (PL_op)
                    Perl_warner(aTHX_ pack_warn, "%s in %s", message,
                                                 OP_DESC(PL_op));
                else
                    Perl_warner(aTHX_ pack_warn, "%s", message);
            }
        }   /* End of 'while (possible_problems)' */

        /* Since there was a possible problem, the returned length may need to
         * be changed from the one stored at the beginning of this function.
         * Instead of trying to figure out if that's needed, just do it. */
        if (retlen) {
            *retlen = curlen;
        }

        if (disallowed) {
            if (flags & UTF8_CHECK_ONLY && retlen) {
                *retlen = ((STRLEN) -1);
            }
            return 0;
        }
    }

    return UNI_TO_NATIVE(uv);
}

/*
=for apidoc utf8_to_uvchr_buf

Returns the native code point of the first character in the string C<s> which
is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
C<*retlen> will be set to the length, in bytes, of that character.

If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
C<NULL>) to -1.  If those warnings are off, the computed value, if well-defined
(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
the next possible position in C<s> that could begin a non-malformed character.
See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
returned.

=cut

Also implemented as a macro in utf8.h

*/


UV
Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
{
    PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;

    assert(s < send);

    return utf8n_to_uvchr(s, send - s, retlen,
                     ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
}

/* This is marked as deprecated
 *
=for apidoc utf8_to_uvuni_buf

Only in very rare circumstances should code need to be dealing in Unicode
(as opposed to native) code points.  In those few cases, use
C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.  If you
are not absolutely sure this is one of those cases, then assume it isn't and
use plain C<utf8_to_uvchr_buf> instead.

Returns the Unicode (not-native) code point of the first character in the
string C<s> which
is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
C<retlen> will be set to the length, in bytes, of that character.

If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
NULL) to -1.  If those warnings are off, the computed value if well-defined (or
the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
next possible position in C<s> that could begin a non-malformed character.
See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.

=cut
*/

UV
Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
{
    PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;

    assert(send > s);

    return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
}

/*
=for apidoc utf8_length

Returns the number of characters in the sequence of UTF-8-encoded bytes starting
at C<s> and ending at the byte just before C<e>.  If <s> and <e> point to the
same place, it returns 0 with no warning raised.

If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning
and returns the number of valid characters.

=cut
*/

STRLEN
Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
{
    STRLEN len = 0;

    PERL_ARGS_ASSERT_UTF8_LENGTH;

    /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
     * the bitops (especially ~) can create illegal UTF-8.
     * In other words: in Perl UTF-8 is not just for Unicode. */

    if (e < s)
	goto warn_and_return;
    while (s < e) {
        s += UTF8SKIP(s);
	len++;
    }

    if (e != s) {
	len--;
        warn_and_return:
	if (PL_op)
	    Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
			     "%s in %s", unees, OP_DESC(PL_op));
	else
	    Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
    }

    return len;
}

/*
=for apidoc bytes_cmp_utf8

Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
sequence of characters (stored as UTF-8)
in C<u>, C<ulen>.  Returns 0 if they are
equal, -1 or -2 if the first string is less than the second string, +1 or +2
if the first string is greater than the second string.

-1 or +1 is returned if the shorter string was identical to the start of the
longer string.  -2 or +2 is returned if
there was a difference between characters
within the strings.

=cut
*/

int
Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
{
    const U8 *const bend = b + blen;
    const U8 *const uend = u + ulen;

    PERL_ARGS_ASSERT_BYTES_CMP_UTF8;

    while (b < bend && u < uend) {
        U8 c = *u++;
	if (!UTF8_IS_INVARIANT(c)) {
	    if (UTF8_IS_DOWNGRADEABLE_START(c)) {
		if (u < uend) {
		    U8 c1 = *u++;
		    if (UTF8_IS_CONTINUATION(c1)) {
			c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
		    } else {
                        /* diag_listed_as: Malformed UTF-8 character%s */
			Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
                              "%s %s%s",
                              unexpected_non_continuation_text(u - 2, 2, 1, 2),
                              PL_op ? " in " : "",
                              PL_op ? OP_DESC(PL_op) : "");
			return -2;
		    }
		} else {
		    if (PL_op)
			Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
					 "%s in %s", unees, OP_DESC(PL_op));
		    else
			Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
		    return -2; /* Really want to return undef :-)  */
		}
	    } else {
		return -2;
	    }
	}
	if (*b != c) {
	    return *b < c ? -2 : +2;
	}
	++b;
    }

    if (b == bend && u == uend)
	return 0;

    return b < bend ? +1 : -1;
}

/*
=for apidoc utf8_to_bytes

Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
Unlike L</bytes_to_utf8>, this over-writes the original string, and
updates C<*lenp> to contain the new length.
Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.

Upon successful return, the number of variants in the string can be computed by
having saved the value of C<*lenp> before the call, and subtracting the
after-call value of C<*lenp> from it.

If you need a copy of the string, see L</bytes_from_utf8>.

=cut
*/

U8 *
Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
{
    U8 * first_variant;

    PERL_ARGS_ASSERT_UTF8_TO_BYTES;
    PERL_UNUSED_CONTEXT;

    /* This is a no-op if no variants at all in the input */
    if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
        return s;
    }

    {
        U8 * const save = s;
        U8 * const send = s + *lenp;
        U8 * d;

        /* Nothing before the first variant needs to be changed, so start the real
         * work there */
        s = first_variant;
        while (s < send) {
            if (! UTF8_IS_INVARIANT(*s)) {
                if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
                    *lenp = ((STRLEN) -1);
                    return 0;
                }
                s++;
            }
            s++;
        }

        /* Is downgradable, so do it */
        d = s = first_variant;
        while (s < send) {
            U8 c = *s++;
            if (! UVCHR_IS_INVARIANT(c)) {
                /* Then it is two-byte encoded */
                c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
                s++;
            }
            *d++ = c;
        }
        *d = '\0';
        *lenp = d - save;

        return save;
    }
}

/*
=for apidoc bytes_from_utf8

Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
byte encoding.  On input, the boolean C<*is_utf8p> gives whether or not C<s> is
actually encoded in UTF-8.

Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
the input string.

Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
not expressible in native byte encoding.  In these cases, C<*is_utf8p> and
C<*lenp> are unchanged, and the return value is the original C<s>.

Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
newly created string containing a downgraded copy of C<s>, and whose length is
returned in C<*lenp>, updated.  The new string is C<NUL>-terminated.  The
caller is responsible for arranging for the memory used by this string to get
freed.

Upon successful return, the number of variants in the string can be computed by
having saved the value of C<*lenp> before the call, and subtracting the
after-call value of C<*lenp> from it.

=cut

There is a macro that avoids this function call, but this is retained for
anyone who calls it with the Perl_ prefix */

U8 *
Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
{
    PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
    PERL_UNUSED_CONTEXT;

    return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
}

/*
No = here because currently externally undocumented
for apidoc bytes_from_utf8_loc

Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
to store the location of the first character in C<"s"> that cannot be
converted to non-UTF8.

If that parameter is C<NULL>, this function behaves identically to
C<bytes_from_utf8>.

Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.

Otherwise, the function returns a newly created C<NUL>-terminated string
containing the non-UTF8 equivalent of the convertible first portion of
C<"s">.  C<*lenp> is set to its length, not including the terminating C<NUL>.
If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
and C<*first_non_downgradable> is set to C<NULL>.

Otherwise, C<*first_non_downgradable> set to point to the first byte of the
first character in the original string that wasn't converted.  C<*is_utf8p> is
unchanged.  Note that the new string may have length 0.

Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
converts as many characters in it as possible stopping at the first one it
finds that can't be converted to non-UTF-8.  C<*first_non_downgradable> is
set to point to that.  The function returns the portion that could be converted
in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
not including the terminating C<NUL>.  If the very first character in the
original could not be converted, C<*lenp> will be 0, and the new string will
contain just a single C<NUL>.  If the entire input string was converted,
C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.

Upon successful return, the number of variants in the converted portion of the
string can be computed by having saved the value of C<*lenp> before the call,
and subtracting the after-call value of C<*lenp> from it.

=cut


*/

U8 *
Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
{
    U8 *d;
    const U8 *original = s;
    U8 *converted_start;
    const U8 *send = s + *lenp;

    PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;

    if (! *is_utf8p) {
        if (first_unconverted) {
            *first_unconverted = NULL;
        }

        return (U8 *) original;
    }

    Newx(d, (*lenp) + 1, U8);

    converted_start = d;
    while (s < send) {
        U8 c = *s++;
        if (! UTF8_IS_INVARIANT(c)) {

            /* Then it is multi-byte encoded.  If the code point is above 0xFF,
             * have to stop now */
            if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
                if (first_unconverted) {
                    *first_unconverted = s - 1;
                    goto finish_and_return;
                }
                else {
                    Safefree(converted_start);
                    return (U8 *) original;
                }
            }

            c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
            s++;
        }
        *d++ = c;
    }

    /* Here, converted the whole of the input */
    *is_utf8p = FALSE;
    if (first_unconverted) {
        *first_unconverted = NULL;
    }

  finish_and_return:
    *d = '\0';
    *lenp = d - converted_start;

    /* Trim unused space */
    Renew(converted_start, *lenp + 1, U8);

    return converted_start;
}

/*
=for apidoc bytes_to_utf8

Converts a string C<s> of length C<*lenp> bytes from the native encoding into
UTF-8.
Returns a pointer to the newly-created string, and sets C<*lenp> to
reflect the new length in bytes.  The caller is responsible for arranging for
the memory used by this string to get freed.

Upon successful return, the number of variants in the string can be computed by
having saved the value of C<*lenp> before the call, and subtracting it from the
after-call value of C<*lenp>.

A C<NUL> character will be written after the end of the string.

If you want to convert to UTF-8 from encodings other than
the native (Latin1 or EBCDIC),
see L</sv_recode_to_utf8>().

=cut
*/

U8*
Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
{
    const U8 * const send = s + (*lenp);
    U8 *d;
    U8 *dst;

    PERL_ARGS_ASSERT_BYTES_TO_UTF8;
    PERL_UNUSED_CONTEXT;

    Newx(d, (*lenp) * 2 + 1, U8);
    dst = d;

    while (s < send) {
        append_utf8_from_native_byte(*s, &d);
        s++;
    }

    *d = '\0';
    *lenp = d-dst;

    /* Trim unused space */
    Renew(dst, *lenp + 1, U8);

    return dst;
}

/*
 * Convert native (big-endian) UTF-16 to UTF-8.  For reversed (little-endian),
 * use utf16_to_utf8_reversed().
 *
 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
 *
 * These functions don't check for overflow.  The worst case is every code
 * point in the input is 2 bytes, and requires 4 bytes on output.  (If the code
 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.)  Therefore the
 * destination must be pre-extended to 2 times the source length.
 *
 * Do not use in-place.  We optimize for native, for obvious reasons. */

U8*
Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
{
    U8* pend;
    U8* dstart = d;

    PERL_ARGS_ASSERT_UTF16_TO_UTF8;

    if (bytelen & 1)
	Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
                                                               (UV)bytelen);

    pend = p + bytelen;

    while (p < pend) {
	UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
	p += 2;
	if (OFFUNI_IS_INVARIANT(uv)) {
	    *d++ = LATIN1_TO_NATIVE((U8) uv);
	    continue;
	}
	if (uv <= MAX_UTF8_TWO_BYTE) {
	    *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
	    *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
	    continue;
	}

#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
#define LAST_HIGH_SURROGATE  0xDBFF
#define FIRST_LOW_SURROGATE  0xDC00
#define LAST_LOW_SURROGATE   UNICODE_SURROGATE_LAST
#define FIRST_IN_PLANE1      0x10000

        /* This assumes that most uses will be in the first Unicode plane, not
         * needing surrogates */
	if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
                  && uv <= UNICODE_SURROGATE_LAST))
        {
            if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
                Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
            }
	    else {
		UV low = (p[0] << 8) + p[1];
		if (   UNLIKELY(low < FIRST_LOW_SURROGATE)
                    || UNLIKELY(low > LAST_LOW_SURROGATE))
                {
		    Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
                }
		p += 2;
		uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
                                + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
	    }
	}
#ifdef EBCDIC
        d = uvoffuni_to_utf8_flags(d, uv, 0);
#else
	if (uv < FIRST_IN_PLANE1) {
	    *d++ = (U8)(( uv >> 12)         | 0xe0);
	    *d++ = (U8)(((uv >>  6) & 0x3f) | 0x80);
	    *d++ = (U8)(( uv        & 0x3f) | 0x80);
	    continue;
	}
	else {
	    *d++ = (U8)(( uv >> 18)         | 0xf0);
	    *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
	    *d++ = (U8)(((uv >>  6) & 0x3f) | 0x80);
	    *d++ = (U8)(( uv        & 0x3f) | 0x80);
	    continue;
	}
#endif
    }
    *newlen = d - dstart;
    return d;
}

/* Note: this one is slightly destructive of the source. */

U8*
Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
{
    U8* s = (U8*)p;
    U8* const send = s + bytelen;

    PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;

    if (bytelen & 1)
	Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
		   (UV)bytelen);

    while (s < send) {
	const U8 tmp = s[0];
	s[0] = s[1];
	s[1] = tmp;
	s += 2;
    }
    return utf16_to_utf8(p, d, bytelen, newlen);
}

bool
Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
{
    return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c);
}

/* Internal function so we can deprecate the external one, and call
   this one from other deprecated functions in this file */

bool
Perl__is_utf8_idstart(pTHX_ const U8 *p)
{
    PERL_ARGS_ASSERT__IS_UTF8_IDSTART;

    if (*p == '_')
	return TRUE;
    return is_utf8_common(p, NULL,
                          "This is buggy if this gets used",
                          PL_utf8_idstart);
}

bool
Perl__is_uni_perl_idcont(pTHX_ UV c)
{
    return _invlist_contains_cp(PL_utf8_perl_idcont, c);
}

bool
Perl__is_uni_perl_idstart(pTHX_ UV c)
{
    return _invlist_contains_cp(PL_utf8_perl_idstart, c);
}

UV
Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
                                  const char S_or_s)
{
    /* We have the latin1-range values compiled into the core, so just use
     * those, converting the result to UTF-8.  The only difference between upper
     * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
     * either "SS" or "Ss".  Which one to use is passed into the routine in
     * 'S_or_s' to avoid a test */

    UV converted = toUPPER_LATIN1_MOD(c);

    PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;

    assert(S_or_s == 'S' || S_or_s == 's');

    if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
					     characters in this range */
	*p = (U8) converted;
	*lenp = 1;
	return converted;
    }

    /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
     * which it maps to one of them, so as to only have to have one check for
     * it in the main case */
    if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
	switch (c) {
	    case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
		converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
		break;
	    case MICRO_SIGN:
		converted = GREEK_CAPITAL_LETTER_MU;
		break;
#if    UNICODE_MAJOR_VERSION > 2                                        \
   || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1           \
                                  && UNICODE_DOT_DOT_VERSION >= 8)
	    case LATIN_SMALL_LETTER_SHARP_S:
		*(p)++ = 'S';
		*p = S_or_s;
		*lenp = 2;
		return 'S';
#endif
	    default:
		Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
                                 " '%c' to map to '%c'",
                                 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
		NOT_REACHED; /* NOTREACHED */
	}
    }

    *(p)++ = UTF8_TWO_BYTE_HI(converted);
    *p = UTF8_TWO_BYTE_LO(converted);
    *lenp = 2;

    return converted;
}

/* If compiled on an early Unicode version, there may not be auxiliary tables
 * */
#ifndef HAS_UC_AUX_TABLES
#  define UC_AUX_TABLE_ptrs     NULL
#  define UC_AUX_TABLE_lengths  NULL
#endif
#ifndef HAS_TC_AUX_TABLES
#  define TC_AUX_TABLE_ptrs     NULL
#  define TC_AUX_TABLE_lengths  NULL
#endif
#ifndef HAS_LC_AUX_TABLES
#  define LC_AUX_TABLE_ptrs     NULL
#  define LC_AUX_TABLE_lengths  NULL
#endif
#ifndef HAS_CF_AUX_TABLES
#  define CF_AUX_TABLE_ptrs     NULL
#  define CF_AUX_TABLE_lengths  NULL
#endif
#ifndef HAS_UC_AUX_TABLES
#  define UC_AUX_TABLE_ptrs     NULL
#  define UC_AUX_TABLE_lengths  NULL
#endif

/* Call the function to convert a UTF-8 encoded character to the specified case.
 * Note that there may be more than one character in the result.
 * 's' is a pointer to the first byte of the input character
 * 'd' will be set to the first byte of the string of changed characters.  It
 *	needs to have space for UTF8_MAXBYTES_CASE+1 bytes
 * 'lenp' will be set to the length in bytes of the string of changed characters
 *
 * The functions return the ordinal of the first character in the string of
 * 'd' */
#define CALL_UPPER_CASE(uv, s, d, lenp)                                     \
                _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper,              \
                                              Uppercase_Mapping_invmap,     \
                                              UC_AUX_TABLE_ptrs,            \
                                              UC_AUX_TABLE_lengths,         \
                                              "uppercase")
#define CALL_TITLE_CASE(uv, s, d, lenp)                                     \
                _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle,              \
                                              Titlecase_Mapping_invmap,     \
                                              TC_AUX_TABLE_ptrs,            \
                                              TC_AUX_TABLE_lengths,         \
                                              "titlecase")
#define CALL_LOWER_CASE(uv, s, d, lenp)                                     \
                _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower,              \
                                              Lowercase_Mapping_invmap,     \
                                              LC_AUX_TABLE_ptrs,            \
                                              LC_AUX_TABLE_lengths,         \
                                              "lowercase")


/* This additionally has the input parameter 'specials', which if non-zero will
 * cause this to use the specials hash for folding (meaning get full case
 * folding); otherwise, when zero, this implies a simple case fold */
#define CALL_FOLD_CASE(uv, s, d, lenp, specials)                            \
        (specials)                                                          \
        ?  _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold,                    \
                                          Case_Folding_invmap,              \
                                          CF_AUX_TABLE_ptrs,                \
                                          CF_AUX_TABLE_lengths,             \
                                          "foldcase")                       \
        : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold,               \
                                         Simple_Case_Folding_invmap,        \
                                         NULL, NULL,                        \
                                         "foldcase")

UV
Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
{
    /* Convert the Unicode character whose ordinal is <c> to its uppercase
     * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
     * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
     * the changed version may be longer than the original character.
     *
     * The ordinal of the first character of the changed version is returned
     * (but note, as explained above, that there may be more.) */

    PERL_ARGS_ASSERT_TO_UNI_UPPER;

    if (c < 256) {
	return _to_upper_title_latin1((U8) c, p, lenp, 'S');
    }

    uvchr_to_utf8(p, c);
    return CALL_UPPER_CASE(c, p, p, lenp);
}

UV
Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
{
    PERL_ARGS_ASSERT_TO_UNI_TITLE;

    if (c < 256) {
	return _to_upper_title_latin1((U8) c, p, lenp, 's');
    }

    uvchr_to_utf8(p, c);
    return CALL_TITLE_CASE(c, p, p, lenp);
}

STATIC U8
S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
{
    /* We have the latin1-range values compiled into the core, so just use
     * those, converting the result to UTF-8.  Since the result is always just
     * one character, we allow <p> to be NULL */

    U8 converted = toLOWER_LATIN1(c);

    PERL_UNUSED_ARG(dummy);

    if (p != NULL) {
	if (NATIVE_BYTE_IS_INVARIANT(converted)) {
	    *p = converted;
	    *lenp = 1;
	}
	else {
            /* Result is known to always be < 256, so can use the EIGHT_BIT
             * macros */
	    *p = UTF8_EIGHT_BIT_HI(converted);
	    *(p+1) = UTF8_EIGHT_BIT_LO(converted);
	    *lenp = 2;
	}
    }
    return converted;
}

UV
Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
{
    PERL_ARGS_ASSERT_TO_UNI_LOWER;

    if (c < 256) {
	return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
    }

    uvchr_to_utf8(p, c);
    return CALL_LOWER_CASE(c, p, p, lenp);
}

UV
Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
{
    /* Corresponds to to_lower_latin1(); <flags> bits meanings:
     *	    FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
     *	    FOLD_FLAGS_FULL  iff full folding is to be used;
     *
     *	Not to be used for locale folds
     */

    UV converted;

    PERL_ARGS_ASSERT__TO_FOLD_LATIN1;

    assert (! (flags & FOLD_FLAGS_LOCALE));

    if (UNLIKELY(c == MICRO_SIGN)) {
	converted = GREEK_SMALL_LETTER_MU;
    }
#if    UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */   \
   || (UNICODE_MAJOR_VERSION == 3 && (   UNICODE_DOT_VERSION > 0)       \
                                      || UNICODE_DOT_DOT_VERSION > 0)
    else if (   (flags & FOLD_FLAGS_FULL)
             && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
    {
        /* If can't cross 127/128 boundary, can't return "ss"; instead return
         * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
         * under those circumstances. */
        if (flags & FOLD_FLAGS_NOMIX_ASCII) {
            *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
            Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
                 p, *lenp, U8);
            return LATIN_SMALL_LETTER_LONG_S;
        }
        else {
            *(p)++ = 's';
            *p = 's';
            *lenp = 2;
            return 's';
        }
    }
#endif
    else { /* In this range the fold of all other characters is their lower
              case */
	converted = toLOWER_LATIN1(c);
    }

    if (UVCHR_IS_INVARIANT(converted)) {
	*p = (U8) converted;
	*lenp = 1;
    }
    else {
	*(p)++ = UTF8_TWO_BYTE_HI(converted);
	*p = UTF8_TWO_BYTE_LO(converted);
	*lenp = 2;
    }

    return converted;
}

UV
Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
{

    /* Not currently externally documented, and subject to change
     *  <flags> bits meanings:
     *	    FOLD_FLAGS_FULL  iff full folding is to be used;
     *	    FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
     *	                      locale are to be used.
     *	    FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
     */

    PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;

    if (flags & FOLD_FLAGS_LOCALE) {
        /* Treat a UTF-8 locale as not being in locale at all, except for
         * potentially warning */
        _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
        if (IN_UTF8_CTYPE_LOCALE) {
            flags &= ~FOLD_FLAGS_LOCALE;
        }
        else {
            goto needs_full_generality;
        }
    }

    if (c < 256) {
        return _to_fold_latin1((U8) c, p, lenp,
			    flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
    }

    /* Here, above 255.  If no special needs, just use the macro */
    if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
	uvchr_to_utf8(p, c);
	return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
    }
    else {  /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
	       the special flags. */
	U8 utf8_c[UTF8_MAXBYTES + 1];

      needs_full_generality:
	uvchr_to_utf8(utf8_c, c);
	return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
                                  p, lenp, flags);
    }
}

PERL_STATIC_INLINE bool
S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
		 const char *const swashname, SV* const invlist)
{
    /* returns a boolean giving whether or not the UTF8-encoded character that
     * starts at <p> is in the swash indicated by <swashname>.  <swash>
     * contains a pointer to where the swash indicated by <swashname>
     * is to be stored; which this routine will do, so that future calls will
     * look at <*swash> and only generate a swash if it is not null.  <invlist>
     * is NULL or an inversion list that defines the swash.  If not null, it
     * saves time during initialization of the swash.
     *
     * Note that it is assumed that the buffer length of <p> is enough to
     * contain all the bytes that comprise the character.  Thus, <*p> should
     * have been checked before this call for mal-formedness enough to assure
     * that. */

    PERL_ARGS_ASSERT_IS_UTF8_COMMON;

    /* The API should have included a length for the UTF-8 character in <p>,
     * but it doesn't.  We therefore assume that p has been validated at least
     * as far as there being enough bytes available in it to accommodate the
     * character without reading beyond the end, and pass that number on to the
     * validating routine */
    if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
        _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
                                          _UTF8_NO_CONFIDENCE_IN_CURLEN,
                                          1 /* Die */ );
        NOT_REACHED; /* NOTREACHED */
    }

    if (invlist) {
        return _invlist_contains_cp(invlist, valid_utf8_to_uvchr(p, NULL));
    }

    assert(swash);

    if (!*swash) {
        U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
        *swash = _core_swash_init("utf8",

                                  /* Only use the name if there is no inversion
                                   * list; otherwise will go out to disk */
                                  (invlist) ? "" : swashname,

                                  &PL_sv_undef, 1, 0, invlist, &flags);
    }

    return swash_fetch(*swash, p, TRUE) != 0;
}

PERL_STATIC_INLINE bool
S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
                          SV **swash, const char *const swashname,
                          SV* const invlist)
{
    /* returns a boolean giving whether or not the UTF8-encoded character that
     * starts at <p>, and extending no further than <e - 1> is in the swash
     * indicated by <swashname>.  <swash> contains a pointer to where the swash
     * indicated by <swashname> is to be stored; which this routine will do, so
     * that future calls will look at <*swash> and only generate a swash if it
     * is not null.  <invlist> is NULL or an inversion list that defines the
     * swash.  If not null, it saves time during initialization of the swash.
     */

    PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;

    if (! isUTF8_CHAR(p, e)) {
        _force_out_malformed_utf8_message(p, e, 0, 1);
        NOT_REACHED; /* NOTREACHED */
    }

    if (invlist) {
        return _invlist_contains_cp(invlist, valid_utf8_to_uvchr(p, NULL));
    }

    assert(swash);

    if (!*swash) {
        U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
        *swash = _core_swash_init("utf8",

                                  /* Only use the name if there is no inversion
                                   * list; otherwise will go out to disk */
                                  (invlist) ? "" : swashname,

                                  &PL_sv_undef, 1, 0, invlist, &flags);
    }

    return swash_fetch(*swash, p, TRUE) != 0;
}

STATIC void
S_warn_on_first_deprecated_use(pTHX_ const char * const name,
                                     const char * const alternative,
                                     const bool use_locale,
                                     const char * const file,
                                     const unsigned line)
{
    const char * key;

    PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;

    if (ckWARN_d(WARN_DEPRECATED)) {

        key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
	if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
            if (! PL_seen_deprecated_macro) {
                PL_seen_deprecated_macro = newHV();
            }
            if (! hv_store(PL_seen_deprecated_macro, key,
                           strlen(key), &PL_sv_undef, 0))
            {
		Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
            }

            if (instr(file, "mathoms.c")) {
                Perl_warner(aTHX_ WARN_DEPRECATED,
                            "In %s, line %d, starting in Perl v5.30, %s()"
                            " will be removed.  Avoid this message by"
                            " converting to use %s().\n",
                            file, line, name, alternative);
            }
            else {
                Perl_warner(aTHX_ WARN_DEPRECATED,
                            "In %s, line %d, starting in Perl v5.30, %s() will"
                            " require an additional parameter.  Avoid this"
                            " message by converting to use %s().\n",
                            file, line, name, alternative);
            }
        }
    }
}

bool
Perl__is_utf8_FOO(pTHX_       U8   classnum,
                        const U8   * const p,
                        const char * const name,
                        const char * const alternative,
                        const bool use_utf8,
                        const bool use_locale,
                        const char * const file,
                        const unsigned line)
{
    PERL_ARGS_ASSERT__IS_UTF8_FOO;

    warn_on_first_deprecated_use(name, alternative, use_locale, file, line);

    if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {

        switch (classnum) {
            case _CC_WORDCHAR:
            case _CC_DIGIT:
            case _CC_ALPHA:
            case _CC_LOWER:
            case _CC_UPPER:
            case _CC_PUNCT:
            case _CC_PRINT:
            case _CC_ALPHANUMERIC:
            case _CC_GRAPH:
            case _CC_CASED:

                return is_utf8_common(p,
                                      NULL,
                                      "This is buggy if this gets used",
                                      PL_XPosix_ptrs[classnum]);

            case _CC_SPACE:
                return is_XPERLSPACE_high(p);
            case _CC_BLANK:
                return is_HORIZWS_high(p);
            case _CC_XDIGIT:
                return is_XDIGIT_high(p);
            case _CC_CNTRL:
                return 0;
            case _CC_ASCII:
                return 0;
            case _CC_VERTSPACE:
                return is_VERTWS_high(p);
            case _CC_IDFIRST:
                return is_utf8_common(p, NULL,
                                      "This is buggy if this gets used",
                                      PL_utf8_perl_idstart);
            case _CC_IDCONT:
                return is_utf8_common(p, NULL,
                                      "This is buggy if this gets used",
                                      PL_utf8_perl_idcont);
        }
    }

    /* idcont is the same as wordchar below 256 */
    if (classnum == _CC_IDCONT) {
        classnum = _CC_WORDCHAR;
    }
    else if (classnum == _CC_IDFIRST) {
        if (*p == '_') {
            return TRUE;
        }
        classnum = _CC_ALPHA;
    }

    if (! use_locale) {
        if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
            return _generic_isCC(*p, classnum);
        }

        return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
    }
    else {
        if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
            return isFOO_lc(classnum, *p);
        }

        return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
    }

    NOT_REACHED; /* NOTREACHED */
}

bool
Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
                                                            const U8 * const e)
{
    PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;

    return is_utf8_common_with_len(p, e, NULL,
                                   "This is buggy if this gets used",
                                   PL_XPosix_ptrs[classnum]);
}

bool
Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
{
    PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;

    return is_utf8_common_with_len(p, e, NULL,
                                   "This is buggy if this gets used",
                                   PL_utf8_perl_idstart);
}

bool
Perl__is_utf8_xidstart(pTHX_ const U8 *p)
{
    PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;

    if (*p == '_')
	return TRUE;
    return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
}

bool
Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
{
    PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;

    return is_utf8_common_with_len(p, e, NULL,
                                   "This is buggy if this gets used",
                                   PL_utf8_perl_idcont);
}

bool
Perl__is_utf8_idcont(pTHX_ const U8 *p)
{
    PERL_ARGS_ASSERT__IS_UTF8_IDCONT;

    return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
}

bool
Perl__is_utf8_xidcont(pTHX_ const U8 *p)
{
    PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;

    return is_utf8_common(p, &PL_utf8_xidcont, "XIdContinue", NULL);
}

bool
Perl__is_utf8_mark(pTHX_ const U8 *p)
{
    PERL_ARGS_ASSERT__IS_UTF8_MARK;

    return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
}

STATIC UV
S__to_utf8_case(pTHX_ const UV uv1, const U8 *p,
                      U8* ustrp, STRLEN *lenp,
                      SV *invlist, const int * const invmap,
                      const unsigned int * const * const aux_tables,
                      const U8 * const aux_table_lengths,
                      const char * const normal)
{
    STRLEN len = 0;

    /* Change the case of code point 'uv1' whose UTF-8 representation (assumed
     * by this routine to be valid) begins at 'p'.  'normal' is a string to use
     * to name the new case in any generated messages, as a fallback if the
     * operation being used is not available.  The new case is given by the
     * data structures in the remaining arguments.
     *
     * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the
     * entire changed case string, and the return value is the first code point
     * in that string */

    PERL_ARGS_ASSERT__TO_UTF8_CASE;

    /* For code points that don't change case, we already know that the output
     * of this function is the unchanged input, so we can skip doing look-ups
     * for them.  Unfortunately the case-changing code points are scattered
     * around.  But there are some long consecutive ranges where there are no
     * case changing code points.  By adding tests, we can eliminate the lookup
     * for all the ones in such ranges.  This is currently done here only for
     * just a few cases where the scripts are in common use in modern commerce
     * (and scripts adjacent to those which can be included without additional
     * tests). */

    if (uv1 >= 0x0590) {
        /* This keeps from needing further processing the code points most
         * likely to be used in the following non-cased scripts: Hebrew,
         * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
         * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
         * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
        if (uv1 < 0x10A0) {
            goto cases_to_self;
        }

        /* The following largish code point ranges also don't have case
         * changes, but khw didn't think they warranted extra tests to speed
         * them up (which would slightly slow down everything else above them):
         * 1100..139F   Hangul Jamo, Ethiopic
         * 1400..1CFF   Unified Canadian Aboriginal Syllabics, Ogham, Runic,
         *              Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
         *              Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
         *              Combining Diacritical Marks Extended, Balinese,
         *              Sundanese, Batak, Lepcha, Ol Chiki
         * 2000..206F   General Punctuation
         */

        if (uv1 >= 0x2D30) {

            /* This keeps the from needing further processing the code points
             * most likely to be used in the following non-cased major scripts:
             * CJK, Katakana, Hiragana, plus some less-likely scripts.
             *
             * (0x2D30 above might have to be changed to 2F00 in the unlikely
             * event that Unicode eventually allocates the unused block as of
             * v8.0 2FE0..2FEF to code points that are cased.  khw has verified
             * that the test suite will start having failures to alert you
             * should that happen) */
            if (uv1 < 0xA640) {
                goto cases_to_self;
            }

            if (uv1 >= 0xAC00) {
                if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
                    if (ckWARN_d(WARN_SURROGATE)) {
                        const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
                        Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
                            "Operation \"%s\" returns its argument for"
                            " UTF-16 surrogate U+%04" UVXf, desc, uv1);
                    }
                    goto cases_to_self;
                }

                /* AC00..FAFF Catches Hangul syllables and private use, plus
                 * some others */
                if (uv1 < 0xFB00) {
                    goto cases_to_self;
                }

                if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
                    if (UNLIKELY(uv1 > MAX_EXTERNALLY_LEGAL_CP)) {
                        Perl_croak(aTHX_ cp_above_legal_max, uv1,
                                         MAX_EXTERNALLY_LEGAL_CP);
                    }
                    if (ckWARN_d(WARN_NON_UNICODE)) {
                        const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
                        Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
                            "Operation \"%s\" returns its argument for"
                            " non-Unicode code point 0x%04" UVXf, desc, uv1);
                    }
                    goto cases_to_self;
                }
#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
                if (UNLIKELY(uv1
                    > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
                {

                    /* As of Unicode 10.0, this means we avoid swash creation
                     * for anything beyond high Plane 1 (below emojis)  */
                    goto cases_to_self;
                }
#endif
            }
        }

	/* Note that non-characters are perfectly legal, so no warning should
         * be given. */
    }

    {
        unsigned int i;
        const unsigned int * cp_list;
        U8 * d;
        SSize_t index = _invlist_search(invlist, uv1);
        IV base = invmap[index];

        /* The data structures are set up so that if 'base' is non-negative,
         * the case change is 1-to-1; and if 0, the change is to itself */
        if (base >= 0) {
            IV lc;

            if (base == 0) {
                goto cases_to_self;
            }

            /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */
            lc = base + uv1 - invlist_array(invlist)[index];
            *lenp = uvchr_to_utf8(ustrp, lc) - ustrp;
            return lc;
        }

        /* Here 'base' is negative.  That means the mapping is 1-to-many, and
         * requires an auxiliary table look up.  abs(base) gives the index into
         * a list of such tables which points to the proper aux table.  And a
         * parallel list gives the length of each corresponding aux table. */
        cp_list = aux_tables[-base];

        /* Create the string of UTF-8 from the mapped-to code points */
        d = ustrp;
        for (i = 0; i < aux_table_lengths[-base]; i++) {
            d = uvchr_to_utf8(d, cp_list[i]);
        }
        *d = '\0';
        *lenp = d - ustrp;

        return cp_list[0];
    }

    /* Here, there was no mapping defined, which means that the code point maps
     * to itself.  Return the inputs */
  cases_to_self:
    len = UTF8SKIP(p);
    if (p != ustrp) {   /* Don't copy onto itself */
        Copy(p, ustrp, len, U8);
    }

    if (lenp)
	 *lenp = len;

    return uv1;

}

Size_t
Perl__inverse_folds(pTHX_ const UV cp, unsigned int * first_folds_to,
                          const unsigned int ** remaining_folds_to)
{
    /* Returns the count of the number of code points that fold to the input
     * 'cp' (besides itself).
     *
     * If the return is 0, there is nothing else that folds to it, and
     * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL.
     *
     * If the return is 1, '*first_folds_to' is set to the single code point,
     * and '*remaining_folds_to' is set to NULL.
     *
     * Otherwise, '*first_folds_to' is set to a code point, and
     * '*remaining_fold_to' is set to an array that contains the others.  The
     * length of this array is the returned count minus 1.
     *
     * The reason for this convolution is to avoid having to deal with
     * allocating and freeing memory.  The lists are already constructed, so
     * the return can point to them, but single code points aren't, so would
     * need to be constructed if we didn't employ something like this API */

    SSize_t index = _invlist_search(PL_utf8_foldclosures, cp);
    int base = _Perl_IVCF_invmap[index];

    PERL_ARGS_ASSERT__INVERSE_FOLDS;

    if (base == 0) {            /* No fold */
        *first_folds_to = 0;
        *remaining_folds_to = NULL;
        return 0;
    }

#ifndef HAS_IVCF_AUX_TABLES     /* This Unicode version only has 1-1 folds */

    assert(base > 0);

#else

    if (UNLIKELY(base < 0)) {   /* Folds to more than one character */

        /* The data structure is set up so that the absolute value of 'base' is
         * an index into a table of pointers to arrays, with the array
         * corresponding to the index being the list of code points that fold
         * to 'cp', and the parallel array containing the length of the list
         * array */
        *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0];
        *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1; /* +1 excludes
                                                                 *first_folds_to
                                                                */
        return IVCF_AUX_TABLE_lengths[-base];
    }

#endif

    /* Only the single code point.  This works like 'fc(G) = G - A + a' */
    *first_folds_to = base + cp - invlist_array(PL_utf8_foldclosures)[index];
    *remaining_folds_to = NULL;
    return 1;
}

STATIC UV
S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
                                       U8* const ustrp, STRLEN *lenp)
{
    /* This is called when changing the case of a UTF-8-encoded character above
     * the Latin1 range, and the operation is in a non-UTF-8 locale.  If the
     * result contains a character that crosses the 255/256 boundary, disallow
     * the change, and return the original code point.  See L<perlfunc/lc> for
     * why;
     *
     * p	points to the original string whose case was changed; assumed
     *          by this routine to be well-formed
     * result	the code point of the first character in the changed-case string
     * ustrp	points to the changed-case string (<result> represents its
     *          first char)
     * lenp	points to the length of <ustrp> */

    UV original;    /* To store the first code point of <p> */

    PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;

    assert(UTF8_IS_ABOVE_LATIN1(*p));

    /* We know immediately if the first character in the string crosses the
     * boundary, so can skip testing */
    if (result > 255) {

	/* Look at every character in the result; if any cross the
	* boundary, the whole thing is disallowed */
	U8* s = ustrp + UTF8SKIP(ustrp);
	U8* e = ustrp + *lenp;
	while (s < e) {
	    if (! UTF8_IS_ABOVE_LATIN1(*s)) {
		goto bad_crossing;
	    }
	    s += UTF8SKIP(s);
	}

        /* Here, no characters crossed, result is ok as-is, but we warn. */
        _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
	return result;
    }

  bad_crossing:

    /* Failed, have to return the original */
    original = valid_utf8_to_uvchr(p, lenp);

    /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
    Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
                           "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
                           " locale; resolved to \"\\x{%" UVXf "}\".",
                           OP_DESC(PL_op),
                           original,
                           original);
    Copy(p, ustrp, *lenp, char);
    return original;
}

STATIC U32
S_check_and_deprecate(pTHX_ const U8 *p,
                            const U8 **e,
                            const unsigned int type,    /* See below */
                            const bool use_locale,      /* Is this a 'LC_'
                                                           macro call? */
                            const char * const file,
                            const unsigned line)
{
    /* This is a temporary function to deprecate the unsafe calls to the case
     * changing macros and functions.  It keeps all the special stuff in just
     * one place.
     *
     * It updates *e with the pointer to the end of the input string.  If using
     * the old-style macros, *e is NULL on input, and so this function assumes
     * the input string is long enough to hold the entire UTF-8 sequence, and
     * sets *e accordingly, but it then returns a flag to pass the
     * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
     * using the full length if possible.
     *
     * It also does the assert that *e > p when *e is not NULL.  This should be
     * migrated to the callers when this function gets deleted.
     *
     * The 'type' parameter is used for the caller to specify which case
     * changing function this is called from: */

#       define DEPRECATE_TO_UPPER 0
#       define DEPRECATE_TO_TITLE 1
#       define DEPRECATE_TO_LOWER 2
#       define DEPRECATE_TO_FOLD  3

    U32 utf8n_flags = 0;
    const char * name;
    const char * alternative;

    PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;

    if (*e == NULL) {
        utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
        *e = p + UTF8SKIP(p);

        /* For mathoms.c calls, we use the function name we know is stored
         * there.  It could be part of a larger path */