APItest/t/utf8.t: Correct to uppercase in print

[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"

static const char malformed_text[] = "Malformed UTF-8 character";
static const char unees[] =
                        "Malformed UTF-8 character (unexpected end of string)";
static const char cp_above_legal_max[] =
 "Use of code point 0x%"UVXf" is deprecated; the permissible max is 0x%"UVXf"";

#define MAX_NON_DEPRECATED_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
*/

/*
=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
*/

#define HANDLE_UNICODE_SURROGATE(uv, flags)                         \
    STMT_START {                                                    \
        if (flags & UNICODE_WARN_SURROGATE) {                       \
            Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),        \
                                "UTF-16 surrogate U+%04"UVXf, uv);  \
        }                                                           \
        if (flags & UNICODE_DISALLOW_SURROGATE) {                   \
            return NULL;                                            \
        }                                                           \
    } STMT_END;

#define HANDLE_UNICODE_NONCHAR(uv, flags)                           \
    STMT_START {                                                    \
        if (flags & UNICODE_WARN_NONCHAR) {                         \
            Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),          \
		 "Unicode non-character U+%04"UVXf" is not "        \
                 "recommended for open interchange", 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

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

    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);
            }
            else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
                HANDLE_UNICODE_SURROGATE(uv, flags);
            }
        }
#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_NON_DEPRECATED_CP)
            && ckWARN_d(WARN_DEPRECATED))
        {
            Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
                        cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP);
        }
        if (   (flags & UNICODE_WARN_SUPER)
            || (   UNICODE_IS_ABOVE_31_BIT(uv)
                && (flags & UNICODE_WARN_ABOVE_31_BIT)))
        {
            Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),

              /* Choose the more dire applicable warning */
              (UNICODE_IS_ABOVE_31_BIT(uv))
              ? "Code point 0x%"UVXf" is not Unicode, and not portable"
              : "Code point 0x%"UVXf" is not Unicode, may not be portable",
             uv);
        }
        if (flags & UNICODE_DISALLOW_SUPER
            || (   UNICODE_IS_ABOVE_31_BIT(uv)
                && (flags & UNICODE_DISALLOW_ABOVE_31_BIT)))
        {
            return NULL;
        }
    }
    else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
        HANDLE_UNICODE_NONCHAR(uv, flags);
    }

    /* 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);
        }
        else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
            HANDLE_UNICODE_SURROGATE(uv, flags);
        }
#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 & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
	    uv >>= UTF_ACCUMULATION_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 UV as input, but very high code points (above
C<IV_MAX> on the platform)  will raise a deprecation warning.  This 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 UV as input, but very high code
points (above C<IV_MAX> for the platform)  will raise a deprecation warning.
This 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>.

Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
so using them is more problematic than other above-Unicode code points.  Perl
invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
likely that non-Perl languages will not be able to read files that contain
these that written by the perl interpreter; 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.  These are the
C<UNICODE_WARN_ABOVE_31_BIT> and C<UNICODE_DISALLOW_ABOVE_31_BIT> flags.  These
are entirely independent from the deprecation warning for code points above
C<IV_MAX>.  On 32-bit machines, it will eventually be forbidden to have any
code point that needs more than 31 bits to represent.  When that happens,
effectively the C<UNICODE_DISALLOW_ABOVE_31_BIT> flag will always be set on
32-bit machines.  (Of course C<UNICODE_DISALLOW_SUPER> will treat all
above-Unicode code points, including these, as malformations; and
C<UNICODE_WARN_SUPER> warns on these.)

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.  The flags C<UNICODE_WARN_ABOVE_31_BIT>
and C<UNICODE_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
platforms, warning and disallowing 2**31 and higher.

=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);
}

PERL_STATIC_INLINE bool
S_is_utf8_cp_above_31_bits(const U8 * const s, const U8 * const e)
{
    /* 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 assumes that the sequence is well-formed UTF-8 as far as it
     * goes, and is for a UTF-8 variant code point.  If the sequence is
     * incomplete, the function returns FALSE if there is any well-formed
     * UTF-8 byte sequence that can complete it in such a way that a code point
     * < 2**31 is produced; otherwise it returns TRUE.
     *
     * Getting this exactly right is slightly tricky, and has to be done in
     * several places in this file, so is centralized here.  It is based on the
     * following table:
     *
     * U+7FFFFFFF (2 ** 31 - 1)
     *      ASCII: \xFD\xBF\xBF\xBF\xBF\xBF
     *   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):
     *      ASCII: \xFE\x82\x80\x80\x80\x80\x80
     *              [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10  11  12  13
     *   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
     */

#ifdef EBCDIC

        /* [0] is start byte           [1] [2] [3] [4] [5] [6] [7] */
    const U8 * const prefix = (U8 *) "\x41\x41\x41\x41\x41\x41\x42";
    const STRLEN prefix_len = sizeof(prefix) - 1;
    const STRLEN len = e - s;
    const STRLEN cmp_len = MIN(prefix_len, len - 1);

#else

    PERL_UNUSED_ARG(e);

#endif

    PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;

    assert(! UTF8_IS_INVARIANT(*s));

#ifndef EBCDIC

    /* Technically, a start byte of FE can be for a code point that fits into
     * 31 bytes, but not for well-formed UTF-8: doing that requires an overlong
     * malformation. */
    return (*s >= 0xFE);

#else

    /* On the EBCDIC code pages we handle, only 0xFE can mean a 32-bit or
     * larger code point (0xFF is an invariant).  For 0xFE, we need at least 2
     * bytes, and maybe up through 8 bytes, to be sure if the value is above 31
     * bits. */
    if (*s != 0xFE || len == 1) {
        return FALSE;
    }

    /* Note that in UTF-EBCDIC, the two lowest possible continuation bytes are
     * \x41 and \x42. */
    return cBOOL(memGT(s + 1, prefix, cmp_len));

#endif

}

PERL_STATIC_INLINE bool
S_does_utf8_overflow(const U8 * const s, const U8 * e)
{
    const U8 *x;
    const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;

    /* Returns a boolean as to if this UTF-8 string would overflow a UV on this
     * platform, that is if it represents a code point larger than the highest
     * representable code point.  (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' must not be beyond a full character.  If it is less than a full
     * character, the function returns FALSE if there is any input beyond 'e'
     * that could result in a non-overflowing code point */

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

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

        /* If this byte is larger than the corresponding highest UTF-8 byte, it
         * overflows */
        if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) > *y)) {
            return TRUE;
        }

        /* If not the same as this byte, it must be smaller, doesn't overflow */
        if (LIKELY(NATIVE_UTF8_TO_I8(*x) != *y)) {
            return FALSE;
        }
    }

    /* 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, so assume doesn't overflow */
    return FALSE;
}

PERL_STATIC_INLINE bool
S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
{
    /* 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
     *
     * At least a second byte is required to determine if other sequences will
     * be an overlong. */

    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 TRUE;
    }

#           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 TRUE;
    }

#   if defined(UV_IS_QUAD) || defined(EBCDIC)

    /* Check for the FF overlong.  This happens only if all these bytes match;
     * what comes after them doesn't matter.  See tables in utf8.h,
     * utfebcdic.h.  (Can't happen on ASCII 32-bit platforms, as overflows
     * instead.) */

    if (   len >= sizeof(FF_OVERLONG_PREFIX) - 1
        && UNLIKELY(memEQ(s, FF_OVERLONG_PREFIX,
                                            sizeof(FF_OVERLONG_PREFIX) - 1)))
    {
        return TRUE;
    }

#else

    PERL_UNUSED_ARG(len);

#endif

    return FALSE;
}

#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 this possibility 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_ABOVE_31_BIT)));
    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]);

        /* The code below is derived from this table.  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
         * are encountered, the sequence is malformed, and will fail elsewhere
         * in this function.
         *              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
         *
         */

#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)
#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)
#endif

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

        if (   (flags & UTF8_DISALLOW_ABOVE_31_BIT)
            &&  UNLIKELY(is_utf8_cp_above_31_bits(s, e)))
        {
            return 0;           /* Above 31 bits */
        }

        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)) {
        return 0;
    }

    /* And finally, that the code point represented fits in a word on this
     * platform */
    if (does_utf8_overflow(s, e)) {
        return 0;
    }

    return UTF8SKIP(s);
}

STATIC char *
S__byte_dump_string(pTHX_ const U8 * s, const STRLEN len)
{
    /* Returns a mortalized C string that is a displayable copy of the 'len'
     * bytes starting at 's', each in a \xXY format. */

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

    PERL_ARGS_ASSERT__BYTE_DUMP_STRING;

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

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

        *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,

                                         /* How many 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);
    unsigned int i;

    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));

    /* It is possible that utf8n_to_uvchr() was called incorrectly, with a
     * length that is larger than is actually available in the buffer.  If we
     * print all the bytes based on that length, we will read past the buffer
     * end.  Often, the strings are NUL terminated, so to lower the chances of
     * this happening, print the malformed bytes only up through any NUL. */
    for (i = 1; i < print_len; i++) {
        if (*(s + i) == '\0') {
            print_len = i + 1;  /* +1 gets the NUL printed */
            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, print_len),
                           *(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 each 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.  For allowed 0 length strings, this function returns 0; for allowed
overlong sequences, the computed code point is returned; for all other allowed
malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
determinable reasonable value.

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.)

It is now deprecated to have very high code points (above C<IV_MAX> on the
platforms) and this function will raise a deprecation warning for these (unless
such warnings are turned off).  This value is typically 0x7FFF_FFFF (2**31 -1)
in a 32-bit word.

Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
so using them is more problematic than other above-Unicode code points.  Perl
invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
likely that non-Perl languages will 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.  These are the
C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags.  These
are entirely independent from the deprecation warning for code points above
C<IV_MAX>.  On 32-bit machines, it will eventually be forbidden to have any
code point that needs more than 31 bits to represent.  When that happens,
effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on
32-bit machines.  (Of course C<UTF8_DISALLOW_SUPER> will treat all
above-Unicode code points, including these, as malformations; and
C<UTF8_WARN_SUPER> warns on these.)

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.  The flags C<UTF8_WARN_ABOVE_31_BIT>
and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
platforms, warning and disallowing 2**31 and higher.

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.

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; the
exceptions are noted:

=over 4

=item C<UTF8_GOT_ABOVE_31_BIT>

The code point represented by the input UTF-8 sequence occupies more than 31
bits.
This bit is set only if the input C<flags> parameter contains either the
C<UTF8_DISALLOW_ABOVE_31_BIT> or the C<UTF8_WARN_ABOVE_31_BIT> flags.

=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.

=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 a UV 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

=cut
*/

UV
Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
                                STRLEN curlen,
                                STRLEN *retlen,
                                const U32 flags,
                                U32 * errors)
{
    const U8 * const s0 = s;
    U8 * send = NULL;           /* (initialized to silence compilers' wrong
                                   warning) */
    U32 possible_problems = 0;  /* A bit is set here for each potential problem
                                   found as we go along */
    UV uv = *s;
    STRLEN expectlen   = 0;     /* How long should this sequence be?
                                   (initialized to silence compilers' wrong
                                   warning) */
    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 * adjusted_send = NULL;  /* (Initialized to silence compilers' wrong
                                   warning) */
    UV uv_so_far = 0;   /* (Initialized to silence compilers' wrong warning) */

    PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;

    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 = 0; /* XXX It could be argued that this should be
                   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(uv)) {
	return uv;
    }

    /* 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) */

    /* 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);

    /* Now, loop through the remaining bytes in the character's sequence,
     * accumulating each into the working value as we go.  Be sure to not look
     * past the end of the input string */
    send = adjusted_send = (U8*) s0 + ((expectlen <= curlen)
                                       ? expectlen
                                       : curlen);
    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. */
        curlen = s - s0;    /* Save how many bytes we actually got */
        possible_problems |= UTF8_GOT_NON_CONTINUATION;
        goto finish_short;
    } /* End of loop through the character's bytes */

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

    /* Did we get all the continuation bytes that were expected?  Note that we
     * know this result even without executing the loop above.  But we had to
     * do the loop to see if there are unexpected non-continuations. */
    if (UNLIKELY(curlen < expectlen)) {
	possible_problems |= UTF8_GOT_SHORT;

      finish_short:
        uv_so_far = uv;
        uv = UNICODE_REPLACEMENT;
    }

    /* 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. */

    /* Check for overflow */
    if (UNLIKELY(does_utf8_overflow(s0, send))) {
        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(is_utf8_overlong_given_start_byte_ok(s0,
                                                                send - s0))))))
    {
        possible_problems |= UTF8_GOT_LONG;

        /* 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 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 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));
            }

            Newx(adjusted_s0, OFFUNISKIP(min_uv) + 1, U8);
            SAVEFREEPV((U8 *) adjusted_s0);    /* Needed because we may not get
                                                  to free it ourselves if
                                                  warnings are made fatal */
            adjusted_send = uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
        }
    }

    /* Now check that the input isn't for a problematic code point not allowed
     * by the input parameters. */
                                              /* isn't problematic if < this */
    if (   (   (   LIKELY(! possible_problems) && uv >= UNICODE_SURROGATE_FIRST)
            || (   UNLIKELY(possible_problems)
                && isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)))
	&& ((flags & ( UTF8_DISALLOW_NONCHAR
                      |UTF8_DISALLOW_SURROGATE
                      |UTF8_DISALLOW_SUPER
                      |UTF8_DISALLOW_ABOVE_31_BIT
	              |UTF8_WARN_NONCHAR
                      |UTF8_WARN_SURROGATE
                      |UTF8_WARN_SUPER
                      |UTF8_WARN_ABOVE_31_BIT))
                   /* In case of a malformation, 'uv' is not valid, and has
                    * been changed to something in the Unicode range.
                    * Currently we don't output a deprecation message if there
                    * is already a malformation, so we don't have to special
                    * case the test immediately below */
            || (   UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
                && ckWARN_d(WARN_DEPRECATED))))
    {
        /* 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.
     * 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
     * send                 points to just after where that (potentially
     *                      partial) character ends
     * adjusted_s0          normally is the same as s0, but in case of an
     *                      overlong for which the UTF-8 matters below, it is
     *                      the first byte of the shortest form representation
     *                      of the input.
     * adjusted_send        normally is the same as 'send', but if adjusted_s0
     *                      is set to something other than s0, this points one
     *                      beyond its end
     */

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

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

            /* 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 */
            if (possible_problems & UTF8_GOT_OVERFLOW) {

                /* Overflow means also got a super and above 31 bits, but we
                 * handle all three cases here */
                possible_problems
                  &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_ABOVE_31_BIT);
                *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_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT)) {
                    *errors |= UTF8_GOT_ABOVE_31_BIT;
                }

                disallowed = TRUE;

                /* 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 (ckWARN_d(WARN_UTF8)) {
                        pack_warn = packWARN(WARN_UTF8);
                    }
                    else if (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, send - s0));
                    }
                }
            }
            else if (possible_problems & UTF8_GOT_EMPTY) {
                possible_problems &= ~UTF8_GOT_EMPTY;
                *errors |= UTF8_GOT_EMPTY;

                if (! (flags & UTF8_ALLOW_EMPTY)) {
                    disallowed = TRUE;
                    if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
                        pack_warn = packWARN(WARN_UTF8);
                        message = Perl_form(aTHX_ "%s (empty string)",
                                                   malformed_text);
                    }
                }
            }
            else if (possible_problems & UTF8_GOT_CONTINUATION) {
                possible_problems &= ~UTF8_GOT_CONTINUATION;
                *errors |= UTF8_GOT_CONTINUATION;

                if (! (flags & UTF8_ALLOW_CONTINUATION)) {
                    disallowed = TRUE;
                    if (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), *s0);
                    }
                }
            }
            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 (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
                        pack_warn = packWARN(WARN_UTF8);
                        message = Perl_form(aTHX_ "%s",
                            unexpected_non_continuation_text(s0,
                                                            send - s0,
                                                            s - s0,
                                                            (int) expectlen));
                    }
                }
            }
            else if (possible_problems & UTF8_GOT_SHORT) {
                possible_problems &= ~UTF8_GOT_SHORT;
                *errors |= UTF8_GOT_SHORT;

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

            }
            else if (possible_problems & UTF8_GOT_LONG) {
                possible_problems &= ~UTF8_GOT_LONG;
                *errors |= UTF8_GOT_LONG;

                if (! (flags & UTF8_ALLOW_LONG)) {
                    disallowed = TRUE;

                    if (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),
                                    _byte_dump_string(s0, curlen));
                        }
                        else {
                            U8 tmpbuf[UTF8_MAXBYTES+1];
                            const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
                                                                        uv, 0);
                            message = Perl_form(aTHX_
                                "%s: %s (overlong; instead use %s to represent"
                                " U+%0*"UVXf")",
                                malformed_text,
                                _byte_dump_string(s0, send - s0),
                                _byte_dump_string(tmpbuf, e - tmpbuf),
                                ((uv < 256) ? 2 : 4), /* Field width of 2 for
                                                         small code points */
                                uv);
                        }
                    }
                }
            }
            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)
                        && 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));
                        }
                        else {
                            message = Perl_form(aTHX_
                                            "UTF-16 surrogate U+%04"UVXf"", uv);
                        }
                    }
                }

                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)
                        && 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));
                        }
                        else {
                            message = Perl_form(aTHX_
                                                "Code point 0x%04"UVXf" is not"
                                                " Unicode, may not be portable",
                                                uv);
                        }
                    }
                }

                /* The maximum code point ever specified by a standard was
                 * 2**31 - 1.  Anything larger than that is a Perl extension
                 * that very well may not be understood by other applications
                 * (including earlier perl versions on EBCDIC platforms).  We
                 * test for these after the regular SUPER ones, and before
                 * possibly bailing out, so that the slightly more dire warning
                 * will override the regular one. */
                if (   (flags & (UTF8_WARN_ABOVE_31_BIT
                                |UTF8_WARN_SUPER
                                |UTF8_DISALLOW_ABOVE_31_BIT))
                    && (   (   UNLIKELY(orig_problems & UTF8_GOT_TOO_SHORT)
                            && UNLIKELY(is_utf8_cp_above_31_bits(
                                                                adjusted_s0,
                                                                adjusted_send)))
                        || (   LIKELY(! (orig_problems & UTF8_GOT_TOO_SHORT))
                            && UNLIKELY(UNICODE_IS_ABOVE_31_BIT(uv)))))
                {
                    if (  ! (flags & UTF8_CHECK_ONLY)
                        &&  (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
                        &&  ckWARN_d(WARN_UTF8))
                    {
                        pack_warn = packWARN(WARN_UTF8);

                        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, and is not portable",
                                        _byte_dump_string(s0, curlen));
                        }
                        else {
                            message = Perl_form(aTHX_
                                        "Code point 0x%"UVXf" is not Unicode,"
                                        " and not portable",
                                         uv);
                        }
                    }

                    if (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT)) {
                        *errors |= UTF8_GOT_ABOVE_31_BIT;

                        if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
                            disallowed = TRUE;
                        }
                    }
                }

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

                /* The deprecated warning overrides any non-deprecated one.  If
                 * there are other problems, a deprecation message is not
                 * really helpful, so don't bother to raise it in that case.
                 * This also keeps the code from having to handle the case
                 * where 'uv' is not valid. */
                if (   ! (orig_problems
                                    & (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
                    && UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
                    && ckWARN_d(WARN_DEPRECATED))
                {
                    message = Perl_form(aTHX_ cp_above_legal_max,
                                              uv, MAX_NON_DEPRECATED_CP);
                    pack_warn = packWARN(WARN_DEPRECATED);
                }
            }
            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)
                        && 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_ "Unicode non-character"
                                                " U+%04"UVXf" is not recommended"
                                                " for open interchange", uv);
                    }
                }

                if (flags & UTF8_DISALLOW_NONCHAR) {
                    disallowed = TRUE;
                    *errors |= UTF8_GOT_NONCHAR;
                }
            } /* 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 (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.

Code points above the platform's C<IV_MAX> will raise a deprecation warning,
unless those are turned off.

=cut

Also implemented as a macro in utf8.h

*/


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

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.

Code points above the platform's C<IV_MAX> will raise a deprecation warning,
unless those are turned off.

=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);

    /* Call the low level routine, asking for checks */
    return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
}

/*
=for apidoc utf8_length

Return the length of the UTF-8 char encoded string C<s> in characters.
Stops at C<e> (inclusive).  If C<e E<lt> s> or if the scan would end
up past C<e>, croaks.

=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 - 1, 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<len> from UTF-8 into native byte encoding.
Unlike L</bytes_to_utf8>, this over-writes the original string, and
updates C<len> to contain the new length.
Returns zero on failure, setting C<len> to -1.

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

=cut
*/

U8 *
Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
{
    U8 * const save = s;
    U8 * const send = s + *len;
    U8 *d;

    PERL_ARGS_ASSERT_UTF8_TO_BYTES;
    PERL_UNUSED_CONTEXT;

    /* ensure valid UTF-8 and chars < 256 before updating string */
    while (s < send) {
        if (! UTF8_IS_INVARIANT(*s)) {
            if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
                *len = ((STRLEN) -1);
                return 0;
            }
            s++;
        }
        s++;
    }

    d = s = save;
    while (s < send) {
	U8 c = *s++;
	if (! UTF8_IS_INVARIANT(c)) {
	    /* Then it is two-byte encoded */
	    c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
            s++;
	}
	*d++ = c;
    }
    *d = '\0';
    *len = d - save;
    return save;
}

/*
=for apidoc bytes_from_utf8

Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
the newly-created string, and updates C<len> to contain the new
length.  Returns the original string if no conversion occurs, C<len>
is unchanged.  Do nothing if C<is_utf8> points to 0.  Sets C<is_utf8> to
0 if C<s> is converted or consisted entirely of characters that are invariant
in UTF-8 (i.e., US-ASCII on non-EBCDIC machines).

=cut
*/

U8 *
Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
{
    U8 *d;
    const U8 *start = s;
    const U8 *send;
    I32 count = 0;

    PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
    PERL_UNUSED_CONTEXT;
    if (!*is_utf8)
        return (U8 *)start;

    /* ensure valid UTF-8 and chars < 256 before converting string */
    for (send = s + *len; s < send;) {
        if (! UTF8_IS_INVARIANT(*s)) {
            if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
                return (U8 *)start;
            }
            count++;
            s++;
	}
        s++;
    }

    *is_utf8 = FALSE;

    Newx(d, (*len) - count + 1, U8);
    s = start; start = d;
    while (s < send) {
	U8 c = *s++;
	if (! UTF8_IS_INVARIANT(c)) {
	    /* Then it is two-byte encoded */
	    c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
            s++;
	}
	*d++ = c;
    }
    *d = '\0';
    *len = d - start;
    return (U8 *)start;
}

/*
=for apidoc bytes_to_utf8

Converts a string C<s> of length C<len> bytes from the native encoding into
UTF-8.
Returns a pointer to the newly-created string, and sets C<len> to
reflect the new length in bytes.

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
*/

/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
   likewise need duplication. */

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

    PERL_ARGS_ASSERT_BYTES_TO_UTF8;
    PERL_UNUSED_CONTEXT;

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

    while (s < send) {
        append_utf8_from_native_byte(*s, &d);
        s++;
    }
    *d = '\0';
    *len = d-dst;
    return dst;
}

/*
 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
 *
 * Destination must be pre-extended to 3/2 source.  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

        /* 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) + 0x10000;
	    }
	}
#ifdef EBCDIC
        d = uvoffuni_to_utf8_flags(d, uv, 0);
#else
	if (uv < 0x10000) {
	    *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)
{
    U8 tmpbuf[UTF8_MAXBYTES+1];
    uvchr_to_utf8(tmpbuf, c);
    return _is_utf8_FOO(classnum, tmpbuf);
}

/* 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, &PL_utf8_idstart, "IdStart", NULL);
}

bool
Perl__is_uni_perl_idcont(pTHX_ UV c)
{
    U8 tmpbuf[UTF8_MAXBYTES+1];
    uvchr_to_utf8(tmpbuf, c);
    return _is_utf8_perl_idcont(tmpbuf);
}

bool
Perl__is_uni_perl_idstart(pTHX_ UV c)
{
    U8 tmpbuf[UTF8_MAXBYTES+1];
    uvchr_to_utf8(tmpbuf, c);
    return _is_utf8_perl_idstart(tmpbuf);
}

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;
}

/* 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.
 * INP is a pointer to the first byte of the input character
 * OUTP 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 OUTP */
#define CALL_UPPER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
#define CALL_TITLE_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
#define CALL_LOWER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")

/* 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) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)

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)
{
    /* 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);

    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);
    }

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

UV
Perl__to_fold_latin1(pTHX_ 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;
    PERL_UNUSED_CONTEXT;

    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 */
        if (IN_UTF8_CTYPE_LOCALE) {
            flags &= ~FOLD_FLAGS_LOCALE;
        }
        else {
            _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
            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, _to_utf8_fold_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 _to_utf8_fold_flags(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))) {
        if (ckWARN_d(WARN_UTF8)) {
            Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
		    "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
            if (ckWARN(WARN_UTF8)) {    /* This will output details as to the
                                           what the malformation is */
                utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
            }
        }
        return FALSE;
    }
    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;
}

bool
Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
{
    PERL_ARGS_ASSERT__IS_UTF8_FOO;

    assert(classnum < _FIRST_NON_SWASH_CC);

    return is_utf8_common(p,
                          &PL_utf8_swash_ptrs[classnum],
                          swash_property_names[classnum],
                          PL_XPosix_ptrs[classnum]);
}

bool
Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
{
    SV* invlist = NULL;

    PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;

    if (! PL_utf8_perl_idstart) {
        invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
    }
    return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
}

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(pTHX_ const U8 *p)
{
    SV* invlist = NULL;

    PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;

    if (! PL_utf8_perl_idcont) {
        invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
    }
    return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
}

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_idcont, "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);
}

/*
=for apidoc to_utf8_case

Instead use the appropriate one of L</toUPPER_utf8>,
L</toTITLE_utf8>,
L</toLOWER_utf8>,
or L</toFOLD_utf8>.

C<p> contains the pointer to the UTF-8 string encoding
the character that is being converted.  This routine assumes that the character
at C<p> is well-formed.

C<ustrp> is a pointer to the character buffer to put the
conversion result to.  C<lenp> is a pointer to the length
of the result.

C<swashp> is a pointer to the swash to use.

Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>.  C<special> (usually,
but not always, a multicharacter mapping), is tried first.

C<special> is a string, normally C<NULL> or C<"">.  C<NULL> means to not use
any special mappings; C<""> means to use the special mappings.  Values other
than these two are treated as the name of the hash containing the special
mappings, like C<"utf8::ToSpecLower">.

C<normal> is a string like C<"ToLower"> which means the swash
C<%utf8::ToLower>.

Code points above the platform's C<IV_MAX> will raise a deprecation warning,
unless those are turned off.

=cut */

UV
Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
			SV **swashp, const char *normal, const char *special)
{
    PERL_ARGS_ASSERT_TO_UTF8_CASE;

    return _to_utf8_case(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, swashp, normal, special);
}

    /* change namve uv1 to 'from' */
STATIC UV
S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
		SV **swashp, const char *normal, const char *special)
{
    STRLEN len = 0;

    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_NON_DEPRECATED_CP)
                        && ckWARN_d(WARN_DEPRECATED))
                    {
                        Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
                                cp_above_legal_max, uv1, MAX_NON_DEPRECATED_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 this writing, this means we avoid swash creation
                     * for anything beyond low Plane 1 */
                    goto cases_to_self;
                }
#endif
            }
        }

	/* Note that non-characters are perfectly legal, so no warning should
         * be given.  There are so few of them, that it isn't worth the extra
         * tests to avoid swash creation */
    }

    if (!*swashp) /* load on-demand */
         *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);

    if (special) {
         /* It might be "special" (sometimes, but not always,
	  * a multicharacter mapping) */
         HV *hv = NULL;
	 SV **svp;

	 /* If passed in the specials name, use that; otherwise use any
	  * given in the swash */
         if (*special != '\0') {
            hv = get_hv(special, 0);
        }
        else {
            svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
            if (svp) {
                hv = MUTABLE_HV(SvRV(*svp));
            }
        }

	 if (hv
             && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
             && (*svp))
         {
	     const char *s;

	      s = SvPV_const(*svp, len);
	      if (len == 1)
                  /* EIGHTBIT */
		   len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
	      else {
		   Copy(s, ustrp, len, U8);
	      }
	 }
    }

    if (!len && *swashp) {
	const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);

	 if (uv2) {
	      /* It was "normal" (a single character mapping). */
	      len = uvchr_to_utf8(ustrp, uv2) - ustrp;
	 }
    }

    if (len) {
        if (lenp) {
            *lenp = len;
        }
        return valid_utf8_to_uvchr(ustrp, 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;

}

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 */
    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;
}

/*
=for apidoc to_utf8_upper

Instead use L</toUPPER_utf8>.

=cut */

/* Not currently externally documented, and subject to change:
 * <flags> is set iff iff the rules from the current underlying locale are to
 *         be used. */

UV
Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
{
    UV result;

    PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;

    if (flags) {
        /* Treat a UTF-8 locale as not being in locale at all */
        if (IN_UTF8_CTYPE_LOCALE) {
            flags = FALSE;
        }
        else {
            _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
        }
    }

    if (UTF8_IS_INVARIANT(*p)) {
	if (flags) {
	    result = toUPPER_LC(*p);
	}
	else {
	    return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
	}
    }
    else if UTF8_IS_DOWNGRADEABLE_START(*p) {
	if (flags) {
            U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
	    result = toUPPER_LC(c);
	}
	else {
	    return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
				          ustrp, lenp, 'S');
	}
    }
    else {  /* UTF-8, ord above 255 */
	result = CALL_UPPER_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp);

	if (flags) {
	    result = check_locale_boundary_crossing(p, result, ustrp, lenp);
	}
	return result;
    }

    /* Here, used locale rules.  Convert back to UTF-8 */
    if (UTF8_IS_INVARIANT(result)) {
	*ustrp = (U8) result;
	*lenp = 1;
    }
    else {
	*ustrp = UTF8_EIGHT_BIT_HI((U8) result);
	*(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
	*lenp = 2;
    }

    return result;
}

/*
=for apidoc to_utf8_title

Instead use L</toTITLE_utf8>.

=cut */

/* Not currently externally documented, and subject to change:
 * <flags> is set iff the rules from the current underlying locale are to be
 *         used.  Since titlecase is not defined in POSIX, for other than a
 *         UTF-8 locale, uppercase is used instead for code points < 256.
 */

UV
Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
{
    UV result;

    PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;

    if (flags) {
        /* Treat a UTF-8 locale as not being in locale at all */
        if (IN_UTF8_CTYPE_LOCALE) {
            flags = FALSE;
        }
        else {
            _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
        }
    }

    if (UTF8_IS_INVARIANT(*p)) {
	if (flags) {
	    result = toUPPER_LC(*p);
	}
	else {
	    return _to_upper_title_latin1(*p, ustrp, lenp, 's');
	}
    }
    else if UTF8_IS_DOWNGRADEABLE_START(*p) {
	if (flags) {
            U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
	    result = toUPPER_LC(c);
	}
	else {
	    return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
				          ustrp, lenp, 's');
	}
    }
    else {  /* UTF-8, ord above 255 */
	result = CALL_TITLE_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp);

	if (flags) {
	    result = check_locale_boundary_crossing(p, result, ustrp, lenp);
	}
	return result;
    }

    /* Here, used locale rules.  Convert back to UTF-8 */
    if (UTF8_IS_INVARIANT(result)) {
	*ustrp = (U8) result;
	*lenp = 1;
    }
    else {
	*ustrp = UTF8_EIGHT_BIT_HI((U8) result);
	*(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
	*lenp = 2;
    }

    return result;
}

/*
=for apidoc to_utf8_lower

Instead use L</toLOWER_utf8>.

=cut */

/* Not currently externally documented, and subject to change:
 * <flags> is set iff iff the rules from the current underlying locale are to
 *         be used.
 */

UV
Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
{
    UV result;

    PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;

    if (flags) {
        /* Treat a UTF-8 locale as not being in locale at all */
        if (IN_UTF8_CTYPE_LOCALE) {
            flags = FALSE;
        }
        else {
            _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
        }
    }

    if (UTF8_IS_INVARIANT(*p)) {
	if (flags) {
	    result = toLOWER_LC(*p);
	}
	else {
	    return to_lower_latin1(*p, ustrp, lenp);
	}
    }
    else if UTF8_IS_DOWNGRADEABLE_START(*p) {
	if (flags) {
            U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
	    result = toLOWER_LC(c);
	}
	else {
	    return to_lower_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
		                   ustrp, lenp);
	}
    }
    else {  /* UTF-8, ord above 255 */
	result = CALL_LOWER_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp);

	if (flags) {
	    result = check_locale_boundary_crossing(p, result, ustrp, lenp);
	}

	return result;
    }

    /* Here, used locale rules.  Convert back to UTF-8 */
    if (UTF8_IS_INVARIANT(result)) {
	*ustrp = (U8) result;
	*lenp = 1;
    }
    else {
	*ustrp = UTF8_EIGHT_BIT_HI((U8) result);
	*(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
	*lenp = 2;
    }

    return result;
}

/*
=for apidoc to_utf8_fold

Instead use L</toFOLD_utf8>.

=cut */

/* Not currently externally documented, and subject to change,
 * in <flags>
 *	bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
 *	                      locale are to be used.
 *      bit FOLD_FLAGS_FULL   is set iff full case folds are to be used;
 *			      otherwise simple folds
 *      bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
 *			      prohibited
 */

UV
Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
{
    UV result;

    PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;

    /* These are mutually exclusive */
    assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));

    assert(p != ustrp); /* Otherwise overwrites */

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

    if (UTF8_IS_INVARIANT(*p)) {
	if (flags & FOLD_FLAGS_LOCALE) {
	    result = toFOLD_LC(*p);
	}
	else {
	    return _to_fold_latin1(*p, ustrp, lenp,
                            flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
	}
    }
    else if UTF8_IS_DOWNGRADEABLE_START(*p) {
	if (flags & FOLD_FLAGS_LOCALE) {
            U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
	    result = toFOLD_LC(c);
	}
	else {
	    return _to_fold_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
                            ustrp, lenp,
                            flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
	}
    }
    else {  /* UTF-8, ord above 255 */
	result = CALL_FOLD_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, flags & FOLD_FLAGS_FULL);

	if (flags & FOLD_FLAGS_LOCALE) {

#           define LONG_S_T      LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
            const unsigned int long_s_t_len    = sizeof(LONG_S_T) - 1;

#         ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
#           define CAP_SHARP_S   LATIN_CAPITAL_LETTER_SHARP_S_UTF8

            const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;

            /* Special case these two characters, as what normally gets
             * returned under locale doesn't work */
            if (UTF8SKIP(p) == cap_sharp_s_len
                && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
            {
                /* 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 fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
                              "resolved to \"\\x{17F}\\x{17F}\".");
                goto return_long_s;
            }
            else
#endif
                 if (UTF8SKIP(p) == long_s_t_len
                     && memEQ((char *) p, LONG_S_T, long_s_t_len))
            {
                /* 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 fc(\"\\x{FB05}\") on non-UTF-8 locale; "
                              "resolved to \"\\x{FB06}\".");
                goto return_ligature_st;
            }

#if    UNICODE_MAJOR_VERSION   == 3         \
    && UNICODE_DOT_VERSION     == 0         \
    && UNICODE_DOT_DOT_VERSION == 1
#           define DOTTED_I   LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8

            /* And special case this on this Unicode version only, for the same
             * reaons the other two are special cased.  They would cross the
             * 255/256 boundary which is forbidden under /l, and so the code
             * wouldn't catch that they are equivalent (which they are only in
             * this release) */
            else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
                     && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
            {
                /* 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 fc(\"\\x{0130}\") on non-UTF-8 locale; "
                              "resolved to \"\\x{0131}\".");
                goto return_dotless_i;
            }
#endif

	    return check_locale_boundary_crossing(p, result, ustrp, lenp);
	}
	else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
	    return result;
	}
	else {
	    /* This is called when changing the case of a UTF-8-encoded
             * character above the ASCII range, and the result should not
             * contain an ASCII character. */

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

	    /* Look at every character in the result; if any cross the
	    * boundary, the whole thing is disallowed */
	    U8* s = ustrp;
	    U8* e = ustrp + *lenp;
	    while (s < e) {
		if (isASCII(*s)) {
		    /* Crossed, have to return the original */
		    original = valid_utf8_to_uvchr(p, lenp);

                    /* But in these instances, there is an alternative we can
                     * return that is valid */
                    if (original == LATIN_SMALL_LETTER_SHARP_S
#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
                        || original == LATIN_CAPITAL_LETTER_SHARP_S
#endif
                    ) {
                        goto return_long_s;
                    }
                    else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
                        goto return_ligature_st;
                    }
#if    UNICODE_MAJOR_VERSION   == 3         \
    && UNICODE_DOT_VERSION     == 0         \
    && UNICODE_DOT_DOT_VERSION == 1

                    else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
                        goto return_dotless_i;
                    }
#endif
		    Copy(p, ustrp, *lenp, char);
		    return original;
		}
		s += UTF8SKIP(s);
	    }

	    /* Here, no characters crossed, result is ok as-is */
	    return result;
	}
    }

    /* Here, used locale rules.  Convert back to UTF-8 */
    if (UTF8_IS_INVARIANT(result)) {
	*ustrp = (U8) result;
	*lenp = 1;
    }
    else {
	*ustrp = UTF8_EIGHT_BIT_HI((U8) result);
	*(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
	*lenp = 2;
    }

    return result;

  return_long_s:
    /* Certain folds to 'ss' are prohibited by the options, but they do allow
     * folds to a string of two of these characters.  By returning this
     * instead, then, e.g.,
     *      fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
     * works. */

    *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
    Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
        ustrp, *lenp, U8);
    return LATIN_SMALL_LETTER_LONG_S;

  return_ligature_st:
    /* Two folds to 'st' are prohibited by the options; instead we pick one and
     * have the other one fold to it */

    *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
    Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
    return LATIN_SMALL_LIGATURE_ST;

#if    UNICODE_MAJOR_VERSION   == 3         \
    && UNICODE_DOT_VERSION     == 0         \
    && UNICODE_DOT_DOT_VERSION == 1

  return_dotless_i:
    *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
    Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
    return LATIN_SMALL_LETTER_DOTLESS_I;

#endif

}

/* Note:
 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
 */

SV*
Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
{
    PERL_ARGS_ASSERT_SWASH_INIT;

    /* Returns a copy of a swash initiated by the called function.  This is the
     * public interface, and returning a copy prevents others from doing
     * mischief on the original */

    return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
}

SV*
Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
{

    /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
     * use the following define */

#define CORE_SWASH_INIT_RETURN(x)   \
    PL_curpm= old_PL_curpm;         \
    return x

    /* Initialize and return a swash, creating it if necessary.  It does this
     * by calling utf8_heavy.pl in the general case.  The returned value may be
     * the swash's inversion list instead if the input parameters allow it.
     * Which is returned should be immaterial to callers, as the only
     * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
     * and swash_to_invlist() handle both these transparently.
     *
     * This interface should only be used by functions that won't destroy or
     * adversely change the swash, as doing so affects all other uses of the
     * swash in the program; the general public should use 'Perl_swash_init'
     * instead.
     *
     * pkg  is the name of the package that <name> should be in.
     * name is the name of the swash to find.  Typically it is a Unicode
     *	    property name, including user-defined ones
     * listsv is a string to initialize the swash with.  It must be of the form
     *	    documented as the subroutine return value in
     *	    L<perlunicode/User-Defined Character Properties>
     * minbits is the number of bits required to represent each data element.
     *	    It is '1' for binary properties.
     * none I (khw) do not understand this one, but it is used only in tr///.
     * invlist is an inversion list to initialize the swash with (or NULL)
     * flags_p if non-NULL is the address of various input and output flag bits
     *      to the routine, as follows:  ('I' means is input to the routine;
     *      'O' means output from the routine.  Only flags marked O are
     *      meaningful on return.)
     *  _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
     *      came from a user-defined property.  (I O)
     *  _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
     *      when the swash cannot be located, to simply return NULL. (I)
     *  _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
     *      return of an inversion list instead of a swash hash if this routine
     *      thinks that would result in faster execution of swash_fetch() later
     *      on. (I)
     *
     * Thus there are three possible inputs to find the swash: <name>,
     * <listsv>, and <invlist>.  At least one must be specified.  The result
     * will be the union of the specified ones, although <listsv>'s various
     * actions can intersect, etc. what <name> gives.  To avoid going out to
     * disk at all, <invlist> should specify completely what the swash should
     * have, and <listsv> should be &PL_sv_undef and <name> should be "".
     *
     * <invlist> is only valid for binary properties */

    PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */

    SV* retval = &PL_sv_undef;
    HV* swash_hv = NULL;
    const int invlist_swash_boundary =
        (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
        ? 512    /* Based on some benchmarking, but not extensive, see commit
                    message */
        : -1;   /* Never return just an inversion list */

    assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
    assert(! invlist || minbits == 1);

    PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
                       that triggered the swash init and the swash init perl logic itself.
                       See perl #122747 */

    /* If data was passed in to go out to utf8_heavy to find the swash of, do
     * so */
    if (listsv != &PL_sv_undef || strNE(name, "")) {
	dSP;
	const size_t pkg_len = strlen(pkg);
	const size_t name_len = strlen(name);
	HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
	SV* errsv_save;
	GV *method;

	PERL_ARGS_ASSERT__CORE_SWASH_INIT;

	PUSHSTACKi(PERLSI_MAGIC);
	ENTER;
	SAVEHINTS();
	save_re_context();
	/* We might get here via a subroutine signature which uses a utf8
	 * parameter name, at which point PL_subname will have been set
	 * but not yet used. */
	save_item(PL_subname);
	if (PL_parser && PL_parser->error_count)
	    SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
	method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
	if (!method) {	/* demand load UTF-8 */
	    ENTER;
	    if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
	    GvSV(PL_errgv) = NULL;
#ifndef NO_TAINT_SUPPORT
	    /* It is assumed that callers of this routine are not passing in
	     * any user derived data.  */
	    /* Need to do this after save_re_context() as it will set
	     * PL_tainted to 1 while saving $1 etc (see the code after getrx:
	     * in Perl_magic_get).  Even line to create errsv_save can turn on
	     * PL_tainted.  */
	    SAVEBOOL(TAINT_get);
	    TAINT_NOT;
#endif
	    Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
			     NULL);
	    {
		/* Not ERRSV, as there is no need to vivify a scalar we are
		   about to discard. */
		SV * const errsv = GvSV(PL_errgv);
		if (!SvTRUE(errsv)) {
		    GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
		    SvREFCNT_dec(errsv);
		}
	    }
	    LEAVE;
	}
	SPAGAIN;
	PUSHMARK(SP);
	EXTEND(SP,5);
	mPUSHp(pkg, pkg_len);
	mPUSHp(name, name_len);
	PUSHs(listsv);
	mPUSHi(minbits);
	mPUSHi(none);
	PUTBACK;
	if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
	GvSV(PL_errgv) = NULL;
	/* If we already have a pointer to the method, no need to use
	 * call_method() to repeat the lookup.  */
	if (method
            ? call_sv(MUTABLE_SV(method), G_SCALAR)
	    : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
	{
	    retval = *PL_stack_sp--;
	    SvREFCNT_inc(retval);
	}
	{
	    /* Not ERRSV.  See above. */
	    SV * const errsv = GvSV(PL_errgv);
	    if (!SvTRUE(errsv)) {
		GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
		SvREFCNT_dec(errsv);
	    }
	}
	LEAVE;
	POPSTACK;
	if (IN_PERL_COMPILETIME) {
	    CopHINTS_set(PL_curcop, PL_hints);
	}
	if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
	    if (SvPOK(retval)) {

		/* If caller wants to handle missing properties, let them */
		if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
                    CORE_SWASH_INIT_RETURN(NULL);
		}
		Perl_croak(aTHX_
			   "Can't find Unicode property definition \"%"SVf"\"",
			   SVfARG(retval));
                NOT_REACHED; /* NOTREACHED */
            }
	}
    } /* End of calling the module to find the swash */

    /* If this operation fetched a swash, and we will need it later, get it */
    if (retval != &PL_sv_undef
        && (minbits == 1 || (flags_p
                            && ! (*flags_p
                                  & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
    {
        swash_hv = MUTABLE_HV(SvRV(retval));

        /* If we don't already know that there is a user-defined component to
         * this swash, and the user has indicated they wish to know if there is
         * one (by passing <flags_p>), find out */
        if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
            SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
            if (user_defined && SvUV(*user_defined)) {
                *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
            }
        }
    }

    /* Make sure there is an inversion list for binary properties */
    if (minbits == 1) {
	SV** swash_invlistsvp = NULL;
	SV* swash_invlist = NULL;
	bool invlist_in_swash_is_valid = FALSE;
	bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
					    an unclaimed reference count */

        /* If this operation fetched a swash, get its already existing
         * inversion list, or create one for it */

        if (swash_hv) {
	    swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
	    if (swash_invlistsvp) {
		swash_invlist = *swash_invlistsvp;
		invlist_in_swash_is_valid = TRUE;
	    }
	    else {
		swash_invlist = _swash_to_invlist(retval);
		swash_invlist_unclaimed = TRUE;
	    }
	}

	/* If an inversion list was passed in, have to include it */
	if (invlist) {

            /* Any fetched swash will by now have an inversion list in it;
             * otherwise <swash_invlist>  will be NULL, indicating that we
             * didn't fetch a swash */
	    if (swash_invlist) {

		/* Add the passed-in inversion list, which invalidates the one
		 * already stored in the swash */
		invlist_in_swash_is_valid = FALSE;
		_invlist_union(invlist, swash_invlist, &swash_invlist);
	    }
	    else {

                /* Here, there is no swash already.  Set up a minimal one, if
                 * we are going to return a swash */
                if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
                    swash_hv = newHV();
                    retval = newRV_noinc(MUTABLE_SV(swash_hv));
                }
		swash_invlist = invlist;
	    }
	}

        /* Here, we have computed the union of all the passed-in data.  It may
         * be that there was an inversion list in the swash which didn't get
         * touched; otherwise save the computed one */
	if (! invlist_in_swash_is_valid
            && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
        {
	    if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
            {
		Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
	    }
	    /* We just stole a reference count. */
	    if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
	    else SvREFCNT_inc_simple_void_NN(swash_invlist);
	}

        SvREADONLY_on(swash_invlist);

        /* Use the inversion list stand-alone if small enough */
        if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
	    SvREFCNT_dec(retval);
	    if (!swash_invlist_unclaimed)
		SvREFCNT_inc_simple_void_NN(swash_invlist);
            retval = newRV_noinc(swash_invlist);
        }
    }

    CORE_SWASH_INIT_RETURN(retval);
#undef CORE_SWASH_INIT_RETURN
}


/* This API is wrong for special case conversions since we may need to
 * return several Unicode characters for a single Unicode character
 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
 * the lower-level routine, and it is similarly broken for returning
 * multiple values.  --jhi
 * For those, you should use S__to_utf8_case() instead */
/* Now SWASHGET is recasted into S_swatch_get in this file. */

/* Note:
 * Returns the value of property/mapping C<swash> for the first character
 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
 *
 * A "swash" is a hash which contains initially the keys/values set up by
 * SWASHNEW.  The purpose is to be able to completely represent a Unicode
 * property for all possible code points.  Things are stored in a compact form
 * (see utf8_heavy.pl) so that calculation is required to find the actual
 * property value for a given code point.  As code points are looked up, new
 * key/value pairs are added to the hash, so that the calculation doesn't have
 * to ever be re-done.  Further, each calculation is done, not just for the
 * desired one, but for a whole block of code points adjacent to that one.
 * For binary properties on ASCII machines, the block is usually for 64 code
 * points, starting with a code point evenly divisible by 64.  Thus if the
 * property value for code point 257 is requested, the code goes out and
 * calculates the property values for all 64 code points between 256 and 319,
 * and stores these as a single 64-bit long bit vector, called a "swatch",
 * under the key for code point 256.  The key is the UTF-8 encoding for code
 * point 256, minus the final byte.  Thus, if the length of the UTF-8 encoding
 * for a code point is 13 bytes, the key will be 12 bytes long.  If the value
 * for code point 258 is then requested, this code realizes that it would be
 * stored under the key for 256, and would find that value and extract the
 * relevant bit, offset from 256.
 *
 * Non-binary properties are stored in as many bits as necessary to represent
 * their values (32 currently, though the code is more general than that), not
 * as single bits, but the principle is the same: the value for each key is a
 * vector that encompasses the property values for all code points whose UTF-8
 * representations are represented by the key.  That is, for all code points
 * whose UTF-8 representations are length N bytes, and the key is the first N-1
 * bytes of that.
 */
UV
Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
{
    HV *const hv = MUTABLE_HV(SvRV(swash));
    U32 klen;
    U32 off;
    STRLEN slen = 0;
    STRLEN needents;
    const U8 *tmps = NULL;
    SV *swatch;
    const U8 c = *ptr;

    PERL_ARGS_ASSERT_SWASH_FETCH;

    /* If it really isn't a hash, it isn't really swash; must be an inversion
     * list */
    if (SvTYPE(hv) != SVt_PVHV) {
        return _invlist_contains_cp((SV*)hv,
                                    (do_utf8)
                                     ? valid_utf8_to_uvchr(ptr, NULL)
                                     : c);
    }

    /* We store the values in a "swatch" which is a vec() value in a swash
     * hash.  Code points 0-255 are a single vec() stored with key length
     * (klen) 0.  All other code points have a UTF-8 representation
     * 0xAA..0xYY,0xZZ.  A vec() is constructed containing all of them which
     * share 0xAA..0xYY, which is the key in the hash to that vec.  So the key
     * length for them is the length of the encoded char - 1.  ptr[klen] is the
     * final byte in the sequence representing the character */
    if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
        klen = 0;
	needents = 256;
        off = c;
    }
    else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
        klen = 0;
	needents = 256;
        off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
    }
    else {
        klen = UTF8SKIP(ptr) - 1;

        /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values.  The offset into
         * the vec is the final byte in the sequence.  (In EBCDIC this is
         * converted to I8 to get consecutive values.)  To help you visualize
         * all this:
         *                       Straight 1047   After final byte
         *             UTF-8      UTF-EBCDIC     I8 transform
         *  U+0400:  \xD0\x80    \xB8\x41\x41    \xB8\x41\xA0
         *  U+0401:  \xD0\x81    \xB8\x41\x42    \xB8\x41\xA1
         *    ...
         *  U+0409:  \xD0\x89    \xB8\x41\x4A    \xB8\x41\xA9
         *  U+040A:  \xD0\x8A    \xB8\x41\x51    \xB8\x41\xAA
         *    ...
         *  U+0412:  \xD0\x92    \xB8\x41\x59    \xB8\x41\xB2
         *  U+0413:  \xD0\x93    \xB8\x41\x62    \xB8\x41\xB3
         *    ...
         *  U+041B:  \xD0\x9B    \xB8\x41\x6A    \xB8\x41\xBB
         *  U+041C:  \xD0\x9C    \xB8\x41\x70    \xB8\x41\xBC
         *    ...
         *  U+041F:  \xD0\x9F    \xB8\x41\x73    \xB8\x41\xBF
         *  U+0420:  \xD0\xA0    \xB8\x42\x41    \xB8\x42\x41
         *
         * (There are no discontinuities in the elided (...) entries.)
         * The UTF-8 key for these 33 code points is '\xD0' (which also is the
         * key for the next 31, up through U+043F, whose UTF-8 final byte is
         * \xBF).  Thus in UTF-8, each key is for a vec() for 64 code points.
         * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
         * index into the vec() swatch (after subtracting 0x80, which we
         * actually do with an '&').
         * In UTF-EBCDIC, each key is for a 32 code point vec().  The first 32
         * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
         * dicontinuities which go away by transforming it into I8, and we
         * effectively subtract 0xA0 to get the index. */
	needents = (1 << UTF_ACCUMULATION_SHIFT);
	off      = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
    }

    /*
     * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
     * suite.  (That is, only 7-8% overall over just a hash cache.  Still,
     * it's nothing to sniff at.)  Pity we usually come through at least
     * two function calls to get here...
     *
     * NB: this code assumes that swatches are never modified, once generated!
     */

    if (hv   == PL_last_swash_hv &&
	klen == PL_last_swash_klen &&
	(!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
    {
	tmps = PL_last_swash_tmps;
	slen = PL_last_swash_slen;
    }
    else {
	/* Try our second-level swatch cache, kept in a hash. */
	SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);

	/* If not cached, generate it via swatch_get */
	if (!svp || !SvPOK(*svp)
		 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
        {
            if (klen) {
                const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
                swatch = swatch_get(swash,
                                    code_point & ~((UV)needents - 1),
				    needents);
            }
            else {  /* For the first 256 code points, the swatch has a key of
                       length 0 */
                swatch = swatch_get(swash, 0, needents);
            }

	    if (IN_PERL_COMPILETIME)
		CopHINTS_set(PL_curcop, PL_hints);

	    svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);

	    if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
		     || (slen << 3) < needents)
		Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
			   "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
			   svp, tmps, (UV)slen, (UV)needents);
	}

	PL_last_swash_hv = hv;
	assert(klen <= sizeof(PL_last_swash_key));
	PL_last_swash_klen = (U8)klen;
	/* FIXME change interpvar.h?  */
	PL_last_swash_tmps = (U8 *) tmps;
	PL_last_swash_slen = slen;
	if (klen)
	    Copy(ptr, PL_last_swash_key, klen, U8);
    }

    switch ((int)((slen << 3) / needents)) {
    case 1:
	return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
    case 8:
	return ((UV) tmps[off]);
    case 16:
	off <<= 1;
	return
            ((UV) tmps[off    ] << 8) +
            ((UV) tmps[off + 1]);
    case 32:
	off <<= 2;
	return
            ((UV) tmps[off    ] << 24) +
            ((UV) tmps[off + 1] << 16) +
            ((UV) tmps[off + 2] <<  8) +
            ((UV) tmps[off + 3]);
    }
    Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
	       "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
    NORETURN_FUNCTION_END;
}

/* Read a single line of the main body of the swash input text.  These are of
 * the form:
 * 0053	0056	0073
 * where each number is hex.  The first two numbers form the minimum and
 * maximum of a range, and the third is the value associated with the range.
 * Not all swashes should have a third number
 *
 * On input: l	  points to the beginning of the line to be examined; it points
 *		  to somewhere in the string of the whole input text, and is
 *		  terminated by a \n or the null string terminator.
 *	     lend   points to the null terminator of that string
 *	     wants_value    is non-zero if the swash expects a third number
 *	     typestr is the name of the swash's mapping, like 'ToLower'
 * On output: *min, *max, and *val are set to the values read from the line.
 *	      returns a pointer just beyond the line examined.  If there was no
 *	      valid min number on the line, returns lend+1
 */

STATIC U8*
S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
			     const bool wants_value, const U8* const typestr)
{
    const int  typeto  = typestr[0] == 'T' && typestr[1] == 'o';
    STRLEN numlen;	    /* Length of the number */
    I32 flags = PERL_SCAN_SILENT_ILLDIGIT
		| PERL_SCAN_DISALLOW_PREFIX
		| PERL_SCAN_SILENT_NON_PORTABLE;

    /* nl points to the next \n in the scan */
    U8* const nl = (U8*)memchr(l, '\n', lend - l);

    PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;

    /* Get the first number on the line: the range minimum */
    numlen = lend - l;
    *min = grok_hex((char *)l, &numlen, &flags, NULL);
    *max = *min;    /* So can never return without setting max */
    if (numlen)	    /* If found a hex number, position past it */
	l += numlen;
    else if (nl) {	    /* Else, go handle next line, if any */
	return nl + 1;	/* 1 is length of "\n" */
    }
    else {		/* Else, no next line */
	return lend + 1;	/* to LIST's end at which \n is not found */
    }

    /* The max range value follows, separated by a BLANK */
    if (isBLANK(*l)) {
	++l;
	flags = PERL_SCAN_SILENT_ILLDIGIT
		| PERL_SCAN_DISALLOW_PREFIX
		| PERL_SCAN_SILENT_NON_PORTABLE;
	numlen = lend - l;
	*max = grok_hex((char *)l, &numlen, &flags, NULL);
	if (numlen)
	    l += numlen;
	else    /* If no value here, it is a single element range */
	    *max = *min;

	/* Non-binary tables have a third entry: what the first element of the
	 * range maps to.  The map for those currently read here is in hex */
	if (wants_value) {
	    if (isBLANK(*l)) {
		++l;
                flags = PERL_SCAN_SILENT_ILLDIGIT
                    | PERL_SCAN_DISALLOW_PREFIX
                    | PERL_SCAN_SILENT_NON_PORTABLE;
                numlen = lend - l;
                *val = grok_hex((char *)l, &numlen, &flags, NULL);
                if (numlen)
                    l += numlen;
                else
                    *val = 0;
	    }
	    else {
		*val = 0;
		if (typeto) {
		    /* diag_listed_as: To%s: illegal mapping '%s' */
		    Perl_croak(aTHX_ "%s: illegal mapping '%s'",
				     typestr, l);
		}
	    }
	}
	else
	    *val = 0; /* bits == 1, then any val should be ignored */
    }
    else { /* Nothing following range min, should be single element with no
	      mapping expected */
	if (wants_value) {
	    *val = 0;
	    if (typeto) {
		/* diag_listed_as