This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
perlop: Update to reflect 5.20 changes
[perl5.git] / utf8.c
CommitLineData
a0ed51b3
LW
1/* utf8.c
2 *
1129b882 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
b94e2f88 4 * by Larry Wall and others
a0ed51b3
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550
TC
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
15 *
cdad3b53 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
a0ed51b3
LW
17 *
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
4ac71550 20 * as is the custom in the West, if you wish to be answered?'
cdad3b53 21 * --Gandalf, addressing Théoden's door wardens
4ac71550
TC
22 *
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
24 *
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
4ac71550
TC
27 *
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
29 */
30
31#include "EXTERN.h"
864dbfa3 32#define PERL_IN_UTF8_C
a0ed51b3 33#include "perl.h"
81e983c1 34#include "inline_invlist.c"
b24b43f7 35#include "charclass_invlists.h"
a0ed51b3 36
27da23d5
JH
37static const char unees[] =
38 "Malformed UTF-8 character (unexpected end of string)";
901b21bf 39
48ef279e 40/*
ccfc67b7 41=head1 Unicode Support
7fefc6c1 42These are various utility functions for manipulating UTF8-encoded
72d33970 43strings. For the uninitiated, this is a method of representing arbitrary
61296642 44Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
45characters in the ASCII range are unmodified, and a zero byte never appears
46within non-zero characters.
166f8a29 47
eaf7a4d2
CS
48=cut
49*/
50
51/*
52=for apidoc is_ascii_string
53
a1433954 54Returns true if the first C<len> bytes of the string C<s> are the same whether
970ea3cb
KW
55or not the string is encoded in UTF-8 (or UTF-EBCDIC on EBCDIC machines). That
56is, if they are invariant. On ASCII-ish machines, only ASCII characters
57fit this definition, hence the function's name.
eaf7a4d2 58
75200dff
KW
59If C<len> is 0, it will be calculated using C<strlen(s)>, (which means if you
60use this option, that C<s> can't have embedded C<NUL> characters and has to
61have a terminating C<NUL> byte).
9f7e3d64 62
a1433954 63See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
eaf7a4d2
CS
64
65=cut
66*/
67
68bool
668b6d8d 69Perl_is_ascii_string(const U8 *s, STRLEN len)
eaf7a4d2
CS
70{
71 const U8* const send = s + (len ? len : strlen((const char *)s));
72 const U8* x = s;
73
74 PERL_ARGS_ASSERT_IS_ASCII_STRING;
eaf7a4d2
CS
75
76 for (; x < send; ++x) {
77 if (!UTF8_IS_INVARIANT(*x))
78 break;
79 }
80
81 return x == send;
82}
83
84/*
378516de 85=for apidoc uvoffuni_to_utf8_flags
eebe1485 86
a27992cc 87THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
88Instead, B<Almost all code should use L</uvchr_to_utf8> or
89L</uvchr_to_utf8_flags>>.
a27992cc 90
de69f3af
KW
91This function is like them, but the input is a strict Unicode
92(as opposed to native) code point. Only in very rare circumstances should code
93not be using the native code point.
949cf498 94
de69f3af 95For details, see the description for L</uvchr_to_utf8_flags>>.
949cf498 96
eebe1485
SC
97=cut
98*/
99
dfe13c55 100U8 *
378516de 101Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 102{
378516de 103 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 104
d9432125
KW
105 if (UNI_IS_INVARIANT(uv)) {
106 *d++ = (U8) LATIN1_TO_NATIVE(uv);
107 return d;
108 }
109
979f77b6
KW
110 /* The first problematic code point is the first surrogate */
111 if (uv >= UNICODE_SURROGATE_FIRST
0cfa64bf 112 && ckWARN3_d(WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
979f77b6 113 {
949cf498
KW
114 if (UNICODE_IS_SURROGATE(uv)) {
115 if (flags & UNICODE_WARN_SURROGATE) {
8457b38f 116 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
949cf498
KW
117 "UTF-16 surrogate U+%04"UVXf, uv);
118 }
119 if (flags & UNICODE_DISALLOW_SURROGATE) {
120 return NULL;
121 }
122 }
123 else if (UNICODE_IS_SUPER(uv)) {
124 if (flags & UNICODE_WARN_SUPER
125 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
126 {
8457b38f 127 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
949cf498
KW
128 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
129 }
130 if (flags & UNICODE_DISALLOW_SUPER
131 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
132 {
133 return NULL;
134 }
135 }
136 else if (UNICODE_IS_NONCHAR(uv)) {
137 if (flags & UNICODE_WARN_NONCHAR) {
8457b38f 138 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
949cf498
KW
139 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
140 uv);
141 }
142 if (flags & UNICODE_DISALLOW_NONCHAR) {
143 return NULL;
144 }
145 }
507b9800 146 }
d9432125 147
2d331972 148#if defined(EBCDIC)
d9432125 149 {
5aaebcb3 150 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
151 U8 *p = d+len-1;
152 while (p > d) {
bc3632a8 153 *p-- = (U8) I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
154 uv >>= UTF_ACCUMULATION_SHIFT;
155 }
bc3632a8 156 *p = (U8) I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
157 return d+len;
158 }
159#else /* Non loop style */
a0ed51b3 160 if (uv < 0x800) {
eb160463
GS
161 *d++ = (U8)(( uv >> 6) | 0xc0);
162 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
163 return d;
164 }
165 if (uv < 0x10000) {
eb160463
GS
166 *d++ = (U8)(( uv >> 12) | 0xe0);
167 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
168 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
169 return d;
170 }
171 if (uv < 0x200000) {
eb160463
GS
172 *d++ = (U8)(( uv >> 18) | 0xf0);
173 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
174 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
175 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
176 return d;
177 }
178 if (uv < 0x4000000) {
eb160463
GS
179 *d++ = (U8)(( uv >> 24) | 0xf8);
180 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
181 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
182 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
183 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
184 return d;
185 }
186 if (uv < 0x80000000) {
eb160463
GS
187 *d++ = (U8)(( uv >> 30) | 0xfc);
188 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
189 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
190 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
191 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
192 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
193 return d;
194 }
6588300d 195#ifdef UTF8_QUAD_MAX
d7578b48 196 if (uv < UTF8_QUAD_MAX)
a0ed51b3
LW
197#endif
198 {
eb160463
GS
199 *d++ = 0xfe; /* Can't match U+FEFF! */
200 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
201 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
202 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
203 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
204 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
205 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
206 return d;
207 }
6588300d 208#ifdef UTF8_QUAD_MAX
a0ed51b3 209 {
eb160463
GS
210 *d++ = 0xff; /* Can't match U+FFFE! */
211 *d++ = 0x80; /* 6 Reserved bits */
212 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
213 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
214 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
215 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
217 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
218 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
219 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
220 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
221 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
222 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
223 return d;
224 }
225#endif
537124e4 226#endif /* Non loop style */
a0ed51b3 227}
646ca15d 228/*
07693fe6
KW
229=for apidoc uvchr_to_utf8
230
bcb1a2d4 231Adds the UTF-8 representation of the native code point C<uv> to the end
c749c9fd
KW
232of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
233C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
234the byte after the end of the new character. In other words,
07693fe6
KW
235
236 d = uvchr_to_utf8(d, uv);
237
238is the recommended wide native character-aware way of saying
239
240 *(d++) = uv;
241
de69f3af
KW
242This function accepts any UV as input. To forbid or warn on non-Unicode code
243points, or those that may be problematic, see L</uvchr_to_utf8_flags>.
244
07693fe6
KW
245=cut
246*/
247
de69f3af
KW
248/* This is also a macro */
249PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
250
07693fe6
KW
251U8 *
252Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
253{
de69f3af 254 return uvchr_to_utf8(d, uv);
07693fe6
KW
255}
256
de69f3af
KW
257/*
258=for apidoc uvchr_to_utf8_flags
259
260Adds the UTF-8 representation of the native code point C<uv> to the end
c749c9fd
KW
261of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
262C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
263the byte after the end of the new character. In other words,
de69f3af
KW
264
265 d = uvchr_to_utf8_flags(d, uv, flags);
266
267or, in most cases,
268
269 d = uvchr_to_utf8_flags(d, uv, 0);
270
271This is the Unicode-aware way of saying
272
273 *(d++) = uv;
274
275This function will convert to UTF-8 (and not warn) even code points that aren't
276legal Unicode or are problematic, unless C<flags> contains one or more of the
277following flags:
278
279If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
280the function will raise a warning, provided UTF8 warnings are enabled. If instead
281UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
282If both flags are set, the function will both warn and return NULL.
283
4c3cfd5d 284The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags
de69f3af 285affect how the function handles a Unicode non-character. And likewise, the
4c3cfd5d 286UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags affect the handling of
de69f3af
KW
287code points that are
288above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
289even less portable) can be warned and/or disallowed even if other above-Unicode
290code points are accepted, by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
291flags.
292
293And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
294above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
295DISALLOW flags.
296
297=cut
298*/
299
300/* This is also a macro */
301PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
302
07693fe6
KW
303U8 *
304Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
305{
de69f3af 306 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
307}
308
309/*
492a624f
KW
310=for apidoc is_utf8_char_buf
311
3cedd9d9 312This is identical to the macro L</isUTF8_CHAR>.
492a624f
KW
313
314=cut */
315
316STRLEN
317Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
318{
319
492a624f
KW
320 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
321
6302f837 322 return isUTF8_CHAR(buf, buf_end);
492a624f
KW
323}
324
325/*
87cea99e 326=for apidoc is_utf8_string
6662521e 327
a1433954 328Returns true if the first C<len> bytes of string C<s> form a valid
9f7e3d64 329UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
75200dff
KW
330using C<strlen(s)> (which means if you use this option, that C<s> can't have
331embedded C<NUL> characters and has to have a terminating C<NUL> byte). Note
332that all characters being ASCII constitute 'a valid UTF-8 string'.
6662521e 333
a1433954 334See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
768c67ee 335
6662521e
GS
336=cut
337*/
338
8e84507e 339bool
668b6d8d 340Perl_is_utf8_string(const U8 *s, STRLEN len)
6662521e 341{
35da51f7 342 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 343 const U8* x = s;
067a85ef 344
7918f24d 345 PERL_ARGS_ASSERT_IS_UTF8_STRING;
1aa99e6b 346
6662521e 347 while (x < send) {
6302f837
KW
348 STRLEN len = isUTF8_CHAR(x, send);
349 if (UNLIKELY(! len)) {
350 return FALSE;
351 }
352 x += len;
6662521e 353 }
768c67ee 354
067a85ef 355 return TRUE;
6662521e
GS
356}
357
67e989fb 358/*
814fafa7
NC
359Implemented as a macro in utf8.h
360
87cea99e 361=for apidoc is_utf8_string_loc
814fafa7 362
a1433954
KW
363Like L</is_utf8_string> but stores the location of the failure (in the
364case of "utf8ness failure") or the location C<s>+C<len> (in the case of
814fafa7
NC
365"utf8ness success") in the C<ep>.
366
a1433954 367See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
814fafa7 368
87cea99e 369=for apidoc is_utf8_string_loclen
81cd54e3 370
a1433954
KW
371Like L</is_utf8_string>() but stores the location of the failure (in the
372case of "utf8ness failure") or the location C<s>+C<len> (in the case of
768c67ee
JH
373"utf8ness success") in the C<ep>, and the number of UTF-8
374encoded characters in the C<el>.
375
a1433954 376See also L</is_utf8_string_loc>() and L</is_utf8_string>().
81cd54e3
JH
377
378=cut
379*/
380
381bool
668b6d8d 382Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
81cd54e3 383{
35da51f7 384 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 385 const U8* x = s;
3ebfea28 386 STRLEN outlen = 0;
7918f24d
NC
387
388 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
81cd54e3 389
81cd54e3 390 while (x < send) {
6302f837
KW
391 STRLEN len = isUTF8_CHAR(x, send);
392 if (UNLIKELY(! len)) {
393 goto out;
394 }
395 x += len;
396 outlen++;
81cd54e3 397 }
768c67ee
JH
398
399 out:
3ebfea28
AL
400 if (el)
401 *el = outlen;
402
768c67ee
JH
403 if (ep)
404 *ep = x;
3ebfea28 405 return (x == send);
81cd54e3
JH
406}
407
408/*
768c67ee 409
de69f3af 410=for apidoc utf8n_to_uvchr
378516de
KW
411
412THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 413Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 414
9041c2e3 415Bottom level UTF-8 decode routine.
de69f3af 416Returns the native code point value of the first character in the string C<s>,
746afd53
KW
417which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
418C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
419the length, in bytes, of that character.
949cf498
KW
420
421The value of C<flags> determines the behavior when C<s> does not point to a
422well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
423zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
424next possible position in C<s> that could begin a non-malformed character.
425Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
426
427Various ALLOW flags can be set in C<flags> to allow (and not warn on)
428individual types of malformations, such as the sequence being overlong (that
429is, when there is a shorter sequence that can express the same code point;
430overlong sequences are expressly forbidden in the UTF-8 standard due to
431potential security issues). Another malformation example is the first byte of
432a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
433flags. For allowed 0 length strings, this function returns 0; for allowed
434overlong sequences, the computed code point is returned; for all other allowed
435malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
436determinable reasonable value.
949cf498
KW
437
438The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
439flags) malformation is found. If this flag is set, the routine assumes that
440the caller will raise a warning, and this function will silently just set
d088425d
KW
441C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
442
75200dff 443Note that this API requires disambiguation between successful decoding a C<NUL>
d088425d
KW
444character, and an error return (unless the UTF8_CHECK_ONLY flag is set), as
445in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
75200dff
KW
446first byte of C<s> is 0 as well. If so, the input was a C<NUL>; if not, the
447input had an error.
949cf498
KW
448
449Certain code points are considered problematic. These are Unicode surrogates,
746afd53 450Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 451By default these are considered regular code points, but certain situations
5eafe189 452warrant special handling for them. If C<flags> contains
949cf498
KW
453UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
454malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
455UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
456maximum) can be set to disallow these categories individually.
457
458The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
459UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
460for their respective categories, but otherwise the code points are considered
461valid (not malformations). To get a category to both be treated as a
462malformation and raise a warning, specify both the WARN and DISALLOW flags.
463(But note that warnings are not raised if lexically disabled nor if
464UTF8_CHECK_ONLY is also specified.)
465
466Very large code points (above 0x7FFF_FFFF) are considered more problematic than
467the others that are above the Unicode legal maximum. There are several
eb83ed87
KW
468reasons: they requre at least 32 bits to represent them on ASCII platforms, are
469not representable at all on EBCDIC platforms, and the original UTF-8
470specification never went above this number (the current 0x10FFFF limit was
471imposed later). (The smaller ones, those that fit into 32 bits, are
472representable by a UV on ASCII platforms, but not by an IV, which means that
473the number of operations that can be performed on them is quite restricted.)
474The UTF-8 encoding on ASCII platforms for these large code points begins with a
475byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
476be treated as malformations, while allowing smaller above-Unicode code points.
477(Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
72d33970
FC
478including these, as malformations.)
479Similarly, UTF8_WARN_FE_FF acts just like
eb83ed87 480the other WARN flags, but applies just to these code points.
949cf498
KW
481
482All other code points corresponding to Unicode characters, including private
483use and those yet to be assigned, are never considered malformed and never
484warn.
67e989fb 485
37607a96
PK
486=cut
487*/
67e989fb 488
a0ed51b3 489UV
de69f3af 490Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 491{
d4c19fe8 492 const U8 * const s0 = s;
eb83ed87 493 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 494 U8 * send;
eb83ed87
KW
495 UV uv = *s;
496 STRLEN expectlen;
949cf498 497 SV* sv = NULL;
eb83ed87
KW
498 UV outlier_ret = 0; /* return value when input is in error or problematic
499 */
500 UV pack_warn = 0; /* Save result of packWARN() for later */
501 bool unexpected_non_continuation = FALSE;
502 bool overflowed = FALSE;
2f8f112e 503 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 504
eb83ed87 505 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 506
de69f3af 507 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
a0dbb045 508
eb83ed87
KW
509 /* The order of malformation tests here is important. We should consume as
510 * few bytes as possible in order to not skip any valid character. This is
511 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
512 * http://unicode.org/reports/tr36 for more discussion as to why. For
513 * example, once we've done a UTF8SKIP, we can tell the expected number of
514 * bytes, and could fail right off the bat if the input parameters indicate
515 * that there are too few available. But it could be that just that first
516 * byte is garbled, and the intended character occupies fewer bytes. If we
517 * blindly assumed that the first byte is correct, and skipped based on
518 * that number, we could skip over a valid input character. So instead, we
519 * always examine the sequence byte-by-byte.
520 *
521 * We also should not consume too few bytes, otherwise someone could inject
522 * things. For example, an input could be deliberately designed to
523 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 524 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
525 * which is actually part of of the overflowing sequence), that could look
526 * legitimate to the caller, which could discard the initial partial
527 * sequence and process the rest, inappropriately */
528
529 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 530 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
531 if (retlen) {
532 *retlen = 0;
533 }
a0dbb045 534
eb83ed87
KW
535 if (flags & UTF8_ALLOW_EMPTY) {
536 return 0;
537 }
538 if (! (flags & UTF8_CHECK_ONLY)) {
539 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
540 }
0c443dc2
JH
541 goto malformed;
542 }
543
eb83ed87
KW
544 expectlen = UTF8SKIP(s);
545
546 /* A well-formed UTF-8 character, as the vast majority of calls to this
547 * function will be for, has this expected length. For efficiency, set
548 * things up here to return it. It will be overriden only in those rare
549 * cases where a malformation is found */
550 if (retlen) {
551 *retlen = expectlen;
552 }
553
554 /* An invariant is trivially well-formed */
1d72bdf6 555 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 556 return uv;
a0ed51b3 557 }
67e989fb 558
eb83ed87 559 /* A continuation character can't start a valid sequence */
b5b9af04 560 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
561 if (flags & UTF8_ALLOW_CONTINUATION) {
562 if (retlen) {
563 *retlen = 1;
564 }
565 return UNICODE_REPLACEMENT;
566 }
ba210ebe 567
eb83ed87
KW
568 if (! (flags & UTF8_CHECK_ONLY)) {
569 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
570 }
571 curlen = 1;
ba210ebe
JH
572 goto malformed;
573 }
9041c2e3 574
dcd27b3c
KW
575 /* Here is not a continuation byte, nor an invariant. The only thing left
576 * is a start byte (possibly for an overlong) */
577
1d72bdf6 578#ifdef EBCDIC
bc3632a8 579 uv = NATIVE_UTF8_TO_I8(uv);
1d72bdf6
NIS
580#endif
581
eb83ed87
KW
582 /* Remove the leading bits that indicate the number of bytes in the
583 * character's whole UTF-8 sequence, leaving just the bits that are part of
584 * the value */
585 uv &= UTF_START_MASK(expectlen);
ba210ebe 586
eb83ed87
KW
587 /* Now, loop through the remaining bytes in the character's sequence,
588 * accumulating each into the working value as we go. Be sure to not look
589 * past the end of the input string */
0b8d30e8
KW
590 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
591
eb83ed87 592 for (s = s0 + 1; s < send; s++) {
b5b9af04 593 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
594#ifndef EBCDIC /* Can't overflow in EBCDIC */
595 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
596
597 /* The original implementors viewed this malformation as more
598 * serious than the others (though I, khw, don't understand
599 * why, since other malformations also give very very wrong
600 * results), so there is no way to turn off checking for it.
601 * Set a flag, but keep going in the loop, so that we absorb
602 * the rest of the bytes that comprise the character. */
603 overflowed = TRUE;
604 overflow_byte = *s; /* Save for warning message's use */
605 }
606#endif
8850bf83 607 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
608 }
609 else {
610 /* Here, found a non-continuation before processing all expected
611 * bytes. This byte begins a new character, so quit, even if
612 * allowing this malformation. */
613 unexpected_non_continuation = TRUE;
614 break;
615 }
616 } /* End of loop through the character's bytes */
617
618 /* Save how many bytes were actually in the character */
619 curlen = s - s0;
620
621 /* The loop above finds two types of malformations: non-continuation and/or
622 * overflow. The non-continuation malformation is really a too-short
623 * malformation, as it means that the current character ended before it was
624 * expected to (being terminated prematurely by the beginning of the next
625 * character, whereas in the too-short malformation there just are too few
626 * bytes available to hold the character. In both cases, the check below
627 * that we have found the expected number of bytes would fail if executed.)
628 * Thus the non-continuation malformation is really unnecessary, being a
629 * subset of the too-short malformation. But there may be existing
630 * applications that are expecting the non-continuation type, so we retain
631 * it, and return it in preference to the too-short malformation. (If this
632 * code were being written from scratch, the two types might be collapsed
633 * into one.) I, khw, am also giving priority to returning the
634 * non-continuation and too-short malformations over overflow when multiple
635 * ones are present. I don't know of any real reason to prefer one over
636 * the other, except that it seems to me that multiple-byte errors trumps
637 * errors from a single byte */
b5b9af04 638 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
639 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
640 if (! (flags & UTF8_CHECK_ONLY)) {
641 if (curlen == 1) {
642 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
643 }
644 else {
645 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, %d bytes after start byte 0x%02x, expected %d bytes)", malformed_text, *s, (int) curlen, *s0, (int)expectlen));
a0dbb045
JH
646 }
647 }
eb83ed87
KW
648 goto malformed;
649 }
650 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
651
652 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
653 * as what the original expectations were. */
654 do_overlong_test = FALSE;
eb83ed87
KW
655 if (retlen) {
656 *retlen = curlen;
657 }
658 }
b5b9af04 659 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
660 if (! (flags & UTF8_ALLOW_SHORT)) {
661 if (! (flags & UTF8_CHECK_ONLY)) {
662 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (%d byte%s, need %d, after start byte 0x%02x)", malformed_text, (int)curlen, curlen == 1 ? "" : "s", (int)expectlen, *s0));
a0dbb045 663 }
eb83ed87
KW
664 goto malformed;
665 }
666 uv = UNICODE_REPLACEMENT;
2f8f112e 667 do_overlong_test = FALSE;
eb83ed87
KW
668 if (retlen) {
669 *retlen = curlen;
670 }
671 }
672
ea5ced44 673#ifndef EBCDIC /* EBCDIC can't overflow */
b5b9af04 674 if (UNLIKELY(overflowed)) {
eb83ed87 675 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 676 goto malformed;
eb83ed87
KW
677 }
678#endif
679
2f8f112e 680 if (do_overlong_test
5aaebcb3 681 && expectlen > (STRLEN) OFFUNISKIP(uv)
2f8f112e
KW
682 && ! (flags & UTF8_ALLOW_LONG))
683 {
eb83ed87
KW
684 /* The overlong malformation has lower precedence than the others.
685 * Note that if this malformation is allowed, we return the actual
686 * value, instead of the replacement character. This is because this
687 * value is actually well-defined. */
688 if (! (flags & UTF8_CHECK_ONLY)) {
5aaebcb3 689 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (%d byte%s, need %d, after start byte 0x%02x)", malformed_text, (int)expectlen, expectlen == 1 ? "": "s", OFFUNISKIP(uv), *s0));
eb83ed87
KW
690 }
691 goto malformed;
692 }
693
1a89bb6c 694 /* Here, the input is considered to be well-formed, but it still could be a
eb83ed87
KW
695 * problematic code point that is not allowed by the input parameters. */
696 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
697 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
698 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
699 {
949cf498 700 if (UNICODE_IS_SURROGATE(uv)) {
ea5ced44
KW
701
702 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
703 * generation of the sv, since no warnings are raised under CHECK */
eb83ed87 704 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
54f4afef 705 && ckWARN_d(WARN_SURROGATE))
eb83ed87 706 {
111d382d 707 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
54f4afef 708 pack_warn = packWARN(WARN_SURROGATE);
949cf498
KW
709 }
710 if (flags & UTF8_DISALLOW_SURROGATE) {
711 goto disallowed;
712 }
713 }
949cf498 714 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87 715 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
ea5ced44 716 && ckWARN_d(WARN_NON_UNICODE))
eb83ed87 717 {
111d382d 718 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
54f4afef 719 pack_warn = packWARN(WARN_NON_UNICODE);
949cf498 720 }
ea5ced44
KW
721#ifndef EBCDIC /* EBCDIC always allows FE, FF */
722
723 /* The first byte being 0xFE or 0xFF is a subset of the SUPER code
724 * points. We test for these after the regular SUPER ones, and
725 * before possibly bailing out, so that the more dire warning
726 * overrides the regular one, if applicable */
727 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
728 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
729 {
730 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY))
731 == UTF8_WARN_FE_FF
732 && ckWARN_d(WARN_UTF8))
733 {
734 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%"UVXf" is not Unicode, and not portable", uv));
735 pack_warn = packWARN(WARN_UTF8);
736 }
737 if (flags & UTF8_DISALLOW_FE_FF) {
738 goto disallowed;
739 }
740 }
741#endif
949cf498
KW
742 if (flags & UTF8_DISALLOW_SUPER) {
743 goto disallowed;
744 }
745 }
4190d317
KW
746 else if (UNICODE_IS_NONCHAR(uv)) {
747 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
54f4afef 748 && ckWARN_d(WARN_NONCHAR))
4190d317
KW
749 {
750 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
54f4afef 751 pack_warn = packWARN(WARN_NONCHAR);
4190d317
KW
752 }
753 if (flags & UTF8_DISALLOW_NONCHAR) {
754 goto disallowed;
755 }
756 }
949cf498 757
eb83ed87 758 if (sv) {
de69f3af
KW
759 outlier_ret = uv; /* Note we don't bother to convert to native,
760 as all the outlier code points are the same
761 in both ASCII and EBCDIC */
eb83ed87
KW
762 goto do_warn;
763 }
764
949cf498
KW
765 /* Here, this is not considered a malformed character, so drop through
766 * to return it */
a0ed51b3 767 }
ba210ebe 768
de69f3af 769 return UNI_TO_NATIVE(uv);
ba210ebe 770
eb83ed87
KW
771 /* There are three cases which get to beyond this point. In all 3 cases:
772 * <sv> if not null points to a string to print as a warning.
773 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
774 * set.
775 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
776 * This is done by initializing it to 0, and changing it only
777 * for case 1).
778 * The 3 cases are:
779 * 1) The input is valid but problematic, and to be warned about. The
780 * return value is the resultant code point; <*retlen> is set to
781 * <curlen>, the number of bytes that comprise the code point.
782 * <pack_warn> contains the result of packWARN() for the warning
783 * types. The entry point for this case is the label <do_warn>;
784 * 2) The input is a valid code point but disallowed by the parameters to
785 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
786 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
787 * comprise the code point. <pack_warn> contains the result of
788 * packWARN() for the warning types. The entry point for this case is
789 * the label <disallowed>.
790 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
791 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
792 * bytes that comprise the malformation. All such malformations are
793 * assumed to be warning type <utf8>. The entry point for this case
794 * is the label <malformed>.
795 */
949cf498 796
ba210ebe
JH
797malformed:
798
eb83ed87
KW
799 if (sv && ckWARN_d(WARN_UTF8)) {
800 pack_warn = packWARN(WARN_UTF8);
801 }
802
803disallowed:
804
fcc8fcf6 805 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 806 if (retlen)
10edeb5d 807 *retlen = ((STRLEN) -1);
ba210ebe
JH
808 return 0;
809 }
810
eb83ed87 811do_warn:
5b311467 812
eb83ed87
KW
813 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
814 if warnings are to be raised. */
f555bc63 815 const char * const string = SvPVX_const(sv);
a0dbb045 816
f555bc63
KW
817 if (PL_op)
818 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
819 else
820 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
821 }
822
eb83ed87
KW
823 if (retlen) {
824 *retlen = curlen;
825 }
ba210ebe 826
eb83ed87 827 return outlier_ret;
a0ed51b3
LW
828}
829
8e84507e 830/*
ec5f19d0
KW
831=for apidoc utf8_to_uvchr_buf
832
833Returns the native code point of the first character in the string C<s> which
834is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 835C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 836
524080c4
KW
837If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
838enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
173db420
KW
839NULL) to -1. If those warnings are off, the computed value, if well-defined
840(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
841C<*retlen> is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is
842the next possible position in C<s> that could begin a non-malformed character.
de69f3af 843See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 844returned.
ec5f19d0
KW
845
846=cut
847*/
848
849
850UV
851Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
852{
ec5f19d0
KW
853 assert(s < send);
854
855 return utf8n_to_uvchr(s, send - s, retlen,
856 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
857}
858
27d6c58a 859/* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
3986bb7c 860 * there are no malformations in the input UTF-8 string C<s>. surrogates,
57b0056d 861 * non-character code points, and non-Unicode code points are allowed. */
27d6c58a
KW
862
863UV
864Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
865{
010ab96b
KW
866 UV expectlen = UTF8SKIP(s);
867 const U8* send = s + expectlen;
9ff2f0f7 868 UV uv = *s;
3986bb7c 869
27d6c58a 870 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
81611534 871 PERL_UNUSED_CONTEXT;
27d6c58a 872
010ab96b
KW
873 if (retlen) {
874 *retlen = expectlen;
875 }
876
877 /* An invariant is trivially returned */
878 if (expectlen == 1) {
9ff2f0f7 879 return uv;
010ab96b
KW
880 }
881
9ff2f0f7
KW
882#ifdef EBCDIC
883 uv = NATIVE_UTF8_TO_I8(uv);
884#endif
885
010ab96b
KW
886 /* Remove the leading bits that indicate the number of bytes, leaving just
887 * the bits that are part of the value */
888 uv &= UTF_START_MASK(expectlen);
889
890 /* Now, loop through the remaining bytes, accumulating each into the
891 * working total as we go. (I khw tried unrolling the loop for up to 4
892 * bytes, but there was no performance improvement) */
893 for (++s; s < send; s++) {
894 uv = UTF8_ACCUMULATE(uv, *s);
895 }
896
3986bb7c 897 return UNI_TO_NATIVE(uv);
010ab96b 898
27d6c58a
KW
899}
900
ec5f19d0 901/*
ec5f19d0
KW
902=for apidoc utf8_to_uvuni_buf
903
de69f3af
KW
904Only in very rare circumstances should code need to be dealing in Unicode
905(as opposed to native) code points. In those few cases, use
906C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
907
908Returns the Unicode (not-native) code point of the first character in the
909string C<s> which
ec5f19d0
KW
910is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
911C<retlen> will be set to the length, in bytes, of that character.
912
524080c4
KW
913If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
914enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
915NULL) to -1. If those warnings are off, the computed value if well-defined (or
916the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
917is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
918next possible position in C<s> that could begin a non-malformed character.
de69f3af 919See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
920
921=cut
922*/
923
924UV
925Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
926{
927 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
928
929 assert(send > s);
930
931 /* Call the low level routine asking for checks */
de69f3af
KW
932 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
933 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
ec5f19d0
KW
934}
935
b76347f2 936/*
87cea99e 937=for apidoc utf8_length
b76347f2
JH
938
939Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
940Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
941up past C<e>, croaks.
b76347f2
JH
942
943=cut
944*/
945
946STRLEN
35a4481c 947Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
948{
949 STRLEN len = 0;
950
7918f24d
NC
951 PERL_ARGS_ASSERT_UTF8_LENGTH;
952
8850bf83
JH
953 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
954 * the bitops (especially ~) can create illegal UTF-8.
955 * In other words: in Perl UTF-8 is not just for Unicode. */
956
a3b680e6
AL
957 if (e < s)
958 goto warn_and_return;
b76347f2 959 while (s < e) {
4cbf4130 960 s += UTF8SKIP(s);
8e91ec7f
AV
961 len++;
962 }
963
964 if (e != s) {
965 len--;
966 warn_and_return:
9b387841
NC
967 if (PL_op)
968 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
969 "%s in %s", unees, OP_DESC(PL_op));
970 else
61a12c31 971 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
972 }
973
974 return len;
975}
976
b06226ff 977/*
87cea99e 978=for apidoc utf8_distance
b06226ff 979
1e54db1a 980Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
b06226ff
JH
981and C<b>.
982
983WARNING: use only if you *know* that the pointers point inside the
984same UTF-8 buffer.
985
37607a96
PK
986=cut
987*/
a0ed51b3 988
02eb7b47 989IV
35a4481c 990Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
a0ed51b3 991{
7918f24d
NC
992 PERL_ARGS_ASSERT_UTF8_DISTANCE;
993
bf1665bc 994 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
a0ed51b3
LW
995}
996
b06226ff 997/*
87cea99e 998=for apidoc utf8_hop
b06226ff 999
8850bf83
JH
1000Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1001forward or backward.
b06226ff
JH
1002
1003WARNING: do not use the following unless you *know* C<off> is within
8850bf83
JH
1004the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1005on the first byte of character or just after the last byte of a character.
b06226ff 1006
37607a96
PK
1007=cut
1008*/
a0ed51b3
LW
1009
1010U8 *
ddeaf645 1011Perl_utf8_hop(const U8 *s, I32 off)
a0ed51b3 1012{
7918f24d
NC
1013 PERL_ARGS_ASSERT_UTF8_HOP;
1014
8850bf83
JH
1015 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1016 * the bitops (especially ~) can create illegal UTF-8.
1017 * In other words: in Perl UTF-8 is not just for Unicode. */
1018
a0ed51b3
LW
1019 if (off >= 0) {
1020 while (off--)
1021 s += UTF8SKIP(s);
1022 }
1023 else {
1024 while (off++) {
1025 s--;
8850bf83
JH
1026 while (UTF8_IS_CONTINUATION(*s))
1027 s--;
a0ed51b3
LW
1028 }
1029 }
4373e329 1030 return (U8 *)s;
a0ed51b3
LW
1031}
1032
6940069f 1033/*
fed3ba5d
NC
1034=for apidoc bytes_cmp_utf8
1035
a1433954 1036Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1037sequence of characters (stored as UTF-8)
1038in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1039equal, -1 or -2 if the first string is less than the second string, +1 or +2
1040if the first string is greater than the second string.
1041
1042-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1043longer string. -2 or +2 is returned if
1044there was a difference between characters
fed3ba5d
NC
1045within the strings.
1046
1047=cut
1048*/
1049
1050int
1051Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1052{
1053 const U8 *const bend = b + blen;
1054 const U8 *const uend = u + ulen;
1055
1056 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1057
1058 while (b < bend && u < uend) {
1059 U8 c = *u++;
1060 if (!UTF8_IS_INVARIANT(c)) {
1061 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1062 if (u < uend) {
1063 U8 c1 = *u++;
1064 if (UTF8_IS_CONTINUATION(c1)) {
94bb8c36 1065 c = TWO_BYTE_UTF8_TO_NATIVE(c, c1);
fed3ba5d
NC
1066 } else {
1067 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1068 "Malformed UTF-8 character "
1069 "(unexpected non-continuation byte 0x%02x"
1070 ", immediately after start byte 0x%02x)"
1071 /* Dear diag.t, it's in the pod. */
1072 "%s%s", c1, c,
1073 PL_op ? " in " : "",
1074 PL_op ? OP_DESC(PL_op) : "");
1075 return -2;
1076 }
1077 } else {
1078 if (PL_op)
1079 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1080 "%s in %s", unees, OP_DESC(PL_op));
1081 else
61a12c31 1082 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1083 return -2; /* Really want to return undef :-) */
1084 }
1085 } else {
1086 return -2;
1087 }
1088 }
1089 if (*b != c) {
1090 return *b < c ? -2 : +2;
1091 }
1092 ++b;
1093 }
1094
1095 if (b == bend && u == uend)
1096 return 0;
1097
1098 return b < bend ? +1 : -1;
1099}
1100
1101/*
87cea99e 1102=for apidoc utf8_to_bytes
6940069f 1103
2bbc8d55 1104Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1105Unlike L</bytes_to_utf8>, this over-writes the original string, and
1106updates C<len> to contain the new length.
67e989fb 1107Returns zero on failure, setting C<len> to -1.
6940069f 1108
a1433954 1109If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1110
6940069f
GS
1111=cut
1112*/
1113
1114U8 *
37607a96 1115Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1116{
d4c19fe8
AL
1117 U8 * const save = s;
1118 U8 * const send = s + *len;
6940069f 1119 U8 *d;
246fae53 1120
7918f24d 1121 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 1122 PERL_UNUSED_CONTEXT;
7918f24d 1123
1e54db1a 1124 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1125 while (s < send) {
d59937ca
KW
1126 if (! UTF8_IS_INVARIANT(*s)) {
1127 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1128 *len = ((STRLEN) -1);
1129 return 0;
1130 }
1131 s++;
dcad2880 1132 }
d59937ca 1133 s++;
246fae53 1134 }
dcad2880
JH
1135
1136 d = s = save;
6940069f 1137 while (s < send) {
80e0b38f
KW
1138 U8 c = *s++;
1139 if (! UTF8_IS_INVARIANT(c)) {
1140 /* Then it is two-byte encoded */
1141 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1142 s++;
1143 }
1144 *d++ = c;
6940069f
GS
1145 }
1146 *d = '\0';
246fae53 1147 *len = d - save;
6940069f
GS
1148 return save;
1149}
1150
1151/*
87cea99e 1152=for apidoc bytes_from_utf8
f9a63242 1153
2bbc8d55 1154Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1155Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1156the newly-created string, and updates C<len> to contain the new
1157length. Returns the original string if no conversion occurs, C<len>
72d33970 1158is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55
SP
11590 if C<s> is converted or consisted entirely of characters that are invariant
1160in utf8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1161
37607a96
PK
1162=cut
1163*/
f9a63242
JH
1164
1165U8 *
e1ec3a88 1166Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1167{
f9a63242 1168 U8 *d;
e1ec3a88
AL
1169 const U8 *start = s;
1170 const U8 *send;
f9a63242
JH
1171 I32 count = 0;
1172
7918f24d 1173 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 1174 PERL_UNUSED_CONTEXT;
f9a63242 1175 if (!*is_utf8)
73d840c0 1176 return (U8 *)start;
f9a63242 1177
1e54db1a 1178 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1179 for (send = s + *len; s < send;) {
d59937ca
KW
1180 if (! UTF8_IS_INVARIANT(*s)) {
1181 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1182 return (U8 *)start;
d59937ca
KW
1183 }
1184 count++;
1185 s++;
db42d148 1186 }
d59937ca 1187 s++;
f9a63242
JH
1188 }
1189
35da51f7 1190 *is_utf8 = FALSE;
f9a63242 1191
212542aa 1192 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1193 s = start; start = d;
f9a63242
JH
1194 while (s < send) {
1195 U8 c = *s++;
1a91c45d 1196 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1197 /* Then it is two-byte encoded */
1a91c45d
KW
1198 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1199 s++;
c4d5f83a
NIS
1200 }
1201 *d++ = c;
f9a63242
JH
1202 }
1203 *d = '\0';
1204 *len = d - start;
73d840c0 1205 return (U8 *)start;
f9a63242
JH
1206}
1207
1208/*
87cea99e 1209=for apidoc bytes_to_utf8
6940069f 1210
ff97e5cf
KW
1211Converts a string C<s> of length C<len> bytes from the native encoding into
1212UTF-8.
6662521e 1213Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1214reflect the new length in bytes.
6940069f 1215
75200dff 1216A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
1217
1218If you want to convert to UTF-8 from encodings other than
1219the native (Latin1 or EBCDIC),
a1433954 1220see L</sv_recode_to_utf8>().
c9ada85f 1221
497711e7 1222=cut
6940069f
GS
1223*/
1224
c682ebef
FC
1225/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1226 likewise need duplication. */
1227
6940069f 1228U8*
35a4481c 1229Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1230{
35a4481c 1231 const U8 * const send = s + (*len);
6940069f
GS
1232 U8 *d;
1233 U8 *dst;
7918f24d
NC
1234
1235 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1236 PERL_UNUSED_CONTEXT;
6940069f 1237
212542aa 1238 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1239 dst = d;
1240
1241 while (s < send) {
55d09dc8
KW
1242 append_utf8_from_native_byte(*s, &d);
1243 s++;
6940069f
GS
1244 }
1245 *d = '\0';
6662521e 1246 *len = d-dst;
6940069f
GS
1247 return dst;
1248}
1249
a0ed51b3 1250/*
dea0fc0b 1251 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1252 *
1253 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1254 * We optimize for native, for obvious reasons. */
1255
1256U8*
dea0fc0b 1257Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1258{
dea0fc0b
JH
1259 U8* pend;
1260 U8* dstart = d;
1261
7918f24d
NC
1262 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1263
dea0fc0b 1264 if (bytelen & 1)
f5992bc4 1265 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1266
1267 pend = p + bytelen;
1268
a0ed51b3 1269 while (p < pend) {
dea0fc0b
JH
1270 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1271 p += 2;
56d37426
KW
1272 if (UNI_IS_INVARIANT(uv)) {
1273 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
1274 continue;
1275 }
56d37426
KW
1276 if (uv <= MAX_UTF8_TWO_BYTE) {
1277 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1278 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
1279 continue;
1280 }
46956fad
KW
1281#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1282#define LAST_HIGH_SURROGATE 0xDBFF
1283#define FIRST_LOW_SURROGATE 0xDC00
1284#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
1285 if (uv >= FIRST_HIGH_SURROGATE && uv <= LAST_HIGH_SURROGATE) {
01ea242b 1286 if (p >= pend) {
dea0fc0b 1287 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
01ea242b
NC
1288 } else {
1289 UV low = (p[0] << 8) + p[1];
1290 p += 2;
46956fad 1291 if (low < FIRST_LOW_SURROGATE || low > LAST_LOW_SURROGATE)
01ea242b 1292 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
46956fad
KW
1293 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1294 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 1295 }
46956fad 1296 } else if (uv >= FIRST_LOW_SURROGATE && uv <= LAST_LOW_SURROGATE) {
dbde1951 1297 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
a0ed51b3 1298 }
56d37426
KW
1299#ifdef EBCDIC
1300 d = uvoffuni_to_utf8_flags(d, uv, 0);
1301#else
a0ed51b3 1302 if (uv < 0x10000) {
eb160463
GS
1303 *d++ = (U8)(( uv >> 12) | 0xe0);
1304 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1305 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1306 continue;
1307 }
1308 else {
eb160463
GS
1309 *d++ = (U8)(( uv >> 18) | 0xf0);
1310 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1311 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1312 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1313 continue;
1314 }
56d37426 1315#endif
a0ed51b3 1316 }
dea0fc0b 1317 *newlen = d - dstart;
a0ed51b3
LW
1318 return d;
1319}
1320
1321/* Note: this one is slightly destructive of the source. */
1322
1323U8*
dea0fc0b 1324Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1325{
1326 U8* s = (U8*)p;
d4c19fe8 1327 U8* const send = s + bytelen;
7918f24d
NC
1328
1329 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1330
e0ea5e2d
NC
1331 if (bytelen & 1)
1332 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1333 (UV)bytelen);
1334
a0ed51b3 1335 while (s < send) {
d4c19fe8 1336 const U8 tmp = s[0];
a0ed51b3
LW
1337 s[0] = s[1];
1338 s[1] = tmp;
1339 s += 2;
1340 }
dea0fc0b 1341 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1342}
1343
922e8cb4
KW
1344bool
1345Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1346{
1347 U8 tmpbuf[UTF8_MAXBYTES+1];
1348 uvchr_to_utf8(tmpbuf, c);
1349 return _is_utf8_FOO(classnum, tmpbuf);
1350}
1351
f9ae8fb6
JD
1352/* Internal function so we can deprecate the external one, and call
1353 this one from other deprecated functions in this file */
1354
f2645549
KW
1355bool
1356Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 1357{
f2645549 1358 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
1359
1360 if (*p == '_')
1361 return TRUE;
f25ce844 1362 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
1363}
1364
5092f92a 1365bool
eba68aa0
KW
1366Perl__is_uni_perl_idcont(pTHX_ UV c)
1367{
1368 U8 tmpbuf[UTF8_MAXBYTES+1];
1369 uvchr_to_utf8(tmpbuf, c);
1370 return _is_utf8_perl_idcont(tmpbuf);
1371}
1372
1373bool
f91dcd13
KW
1374Perl__is_uni_perl_idstart(pTHX_ UV c)
1375{
1376 U8 tmpbuf[UTF8_MAXBYTES+1];
1377 uvchr_to_utf8(tmpbuf, c);
1378 return _is_utf8_perl_idstart(tmpbuf);
1379}
1380
3a4c58c9
KW
1381UV
1382Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1383{
1384 /* We have the latin1-range values compiled into the core, so just use
1385 * those, converting the result to utf8. The only difference between upper
1386 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1387 * either "SS" or "Ss". Which one to use is passed into the routine in
1388 * 'S_or_s' to avoid a test */
1389
1390 UV converted = toUPPER_LATIN1_MOD(c);
1391
1392 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1393
1394 assert(S_or_s == 'S' || S_or_s == 's');
1395
6f2d5cbc 1396 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 1397 characters in this range */
3a4c58c9
KW
1398 *p = (U8) converted;
1399 *lenp = 1;
1400 return converted;
1401 }
1402
1403 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1404 * which it maps to one of them, so as to only have to have one check for
1405 * it in the main case */
1406 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1407 switch (c) {
1408 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1409 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1410 break;
1411 case MICRO_SIGN:
1412 converted = GREEK_CAPITAL_LETTER_MU;
1413 break;
1414 case LATIN_SMALL_LETTER_SHARP_S:
1415 *(p)++ = 'S';
1416 *p = S_or_s;
1417 *lenp = 2;
1418 return 'S';
1419 default:
1420 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
a25b5927 1421 assert(0); /* NOTREACHED */
3a4c58c9
KW
1422 }
1423 }
1424
1425 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1426 *p = UTF8_TWO_BYTE_LO(converted);
1427 *lenp = 2;
1428
1429 return converted;
1430}
1431
50bda2c3
KW
1432/* Call the function to convert a UTF-8 encoded character to the specified case.
1433 * Note that there may be more than one character in the result.
1434 * INP is a pointer to the first byte of the input character
1435 * OUTP will be set to the first byte of the string of changed characters. It
1436 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1437 * LENP will be set to the length in bytes of the string of changed characters
1438 *
1439 * The functions return the ordinal of the first character in the string of OUTP */
4a8240a3
KW
1440#define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "")
1441#define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "")
1442#define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "")
50bda2c3
KW
1443
1444/* This additionally has the input parameter SPECIALS, which if non-zero will
1445 * cause this to use the SPECIALS hash for folding (meaning get full case
1446 * folding); otherwise, when zero, this implies a simple case fold */
4a8240a3 1447#define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "" : NULL)
c3fd2246 1448
84afefe6
JH
1449UV
1450Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1451{
a1433954
KW
1452 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1453 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1454 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1455 * the changed version may be longer than the original character.
1456 *
1457 * The ordinal of the first character of the changed version is returned
1458 * (but note, as explained above, that there may be more.) */
1459
7918f24d
NC
1460 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1461
3a4c58c9
KW
1462 if (c < 256) {
1463 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1464 }
1465
0ebc6274 1466 uvchr_to_utf8(p, c);
3a4c58c9 1467 return CALL_UPPER_CASE(p, p, lenp);
a0ed51b3
LW
1468}
1469
84afefe6
JH
1470UV
1471Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1472{
7918f24d
NC
1473 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1474
3a4c58c9
KW
1475 if (c < 256) {
1476 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1477 }
1478
0ebc6274 1479 uvchr_to_utf8(p, c);
3a4c58c9 1480 return CALL_TITLE_CASE(p, p, lenp);
a0ed51b3
LW
1481}
1482
afc16117 1483STATIC U8
81611534 1484S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp)
afc16117
KW
1485{
1486 /* We have the latin1-range values compiled into the core, so just use
1487 * those, converting the result to utf8. Since the result is always just
a1433954 1488 * one character, we allow <p> to be NULL */
afc16117
KW
1489
1490 U8 converted = toLOWER_LATIN1(c);
1491
1492 if (p != NULL) {
6f2d5cbc 1493 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
1494 *p = converted;
1495 *lenp = 1;
1496 }
1497 else {
1498 *p = UTF8_TWO_BYTE_HI(converted);
1499 *(p+1) = UTF8_TWO_BYTE_LO(converted);
1500 *lenp = 2;
1501 }
1502 }
1503 return converted;
1504}
1505
84afefe6
JH
1506UV
1507Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1508{
7918f24d
NC
1509 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1510
afc16117
KW
1511 if (c < 256) {
1512 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1513 }
1514
afc16117 1515 uvchr_to_utf8(p, c);
968c5e6a 1516 return CALL_LOWER_CASE(p, p, lenp);
a0ed51b3
LW
1517}
1518
84afefe6 1519UV
51910141 1520Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 1521{
51910141 1522 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 1523 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 1524 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
1525 *
1526 * Not to be used for locale folds
51910141 1527 */
f673fad4 1528
a1dde8de
KW
1529 UV converted;
1530
1531 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 1532 PERL_UNUSED_CONTEXT;
a1dde8de 1533
1ca267a5
KW
1534 assert (! (flags & FOLD_FLAGS_LOCALE));
1535
a1dde8de
KW
1536 if (c == MICRO_SIGN) {
1537 converted = GREEK_SMALL_LETTER_MU;
1538 }
51910141 1539 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1ca267a5
KW
1540
1541 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1542 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1543 * under those circumstances. */
1544 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1545 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1546 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1547 p, *lenp, U8);
1548 return LATIN_SMALL_LETTER_LONG_S;
1549 }
1550 else {
4f489194
KW
1551 *(p)++ = 's';
1552 *p = 's';
1553 *lenp = 2;
1554 return 's';
1ca267a5 1555 }
a1dde8de
KW
1556 }
1557 else { /* In this range the fold of all other characters is their lower
1558 case */
1559 converted = toLOWER_LATIN1(c);
1560 }
1561
6f2d5cbc 1562 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
1563 *p = (U8) converted;
1564 *lenp = 1;
1565 }
1566 else {
1567 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1568 *p = UTF8_TWO_BYTE_LO(converted);
1569 *lenp = 2;
1570 }
1571
1572 return converted;
1573}
1574
1575UV
31f05a37 1576Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 1577{
4b593389 1578
a0270393
KW
1579 /* Not currently externally documented, and subject to change
1580 * <flags> bits meanings:
1581 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
1582 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1583 * locale are to be used.
a0270393
KW
1584 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1585 */
4b593389 1586
36bb2ab6 1587 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1588
31f05a37
KW
1589 /* Tread a UTF-8 locale as not being in locale at all */
1590 if (IN_UTF8_CTYPE_LOCALE) {
1591 flags &= ~FOLD_FLAGS_LOCALE;
1592 }
1593
a1dde8de 1594 if (c < 256) {
a0270393 1595 UV result = _to_fold_latin1((U8) c, p, lenp,
31f05a37 1596 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a0270393
KW
1597 /* It is illegal for the fold to cross the 255/256 boundary under
1598 * locale; in this case return the original */
1599 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1600 ? c
1601 : result;
a1dde8de
KW
1602 }
1603
a0270393
KW
1604 /* If no special needs, just use the macro */
1605 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1606 uvchr_to_utf8(p, c);
1607 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1608 }
1609 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1610 the special flags. */
1611 U8 utf8_c[UTF8_MAXBYTES + 1];
1612 uvchr_to_utf8(utf8_c, c);
445bf929 1613 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
a0270393 1614 }
84afefe6
JH
1615}
1616
26483009 1617PERL_STATIC_INLINE bool
5141f98e 1618S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 1619 const char *const swashname, SV* const invlist)
bde6a22d 1620{
ea317ccb
KW
1621 /* returns a boolean giving whether or not the UTF8-encoded character that
1622 * starts at <p> is in the swash indicated by <swashname>. <swash>
1623 * contains a pointer to where the swash indicated by <swashname>
1624 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
1625 * look at <*swash> and only generate a swash if it is not null. <invlist>
1626 * is NULL or an inversion list that defines the swash. If not null, it
1627 * saves time during initialization of the swash.
ea317ccb
KW
1628 *
1629 * Note that it is assumed that the buffer length of <p> is enough to
1630 * contain all the bytes that comprise the character. Thus, <*p> should
1631 * have been checked before this call for mal-formedness enough to assure
1632 * that. */
1633
7918f24d
NC
1634 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1635
492a624f 1636 /* The API should have included a length for the UTF-8 character in <p>,
28123549 1637 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
1638 * as far as there being enough bytes available in it to accommodate the
1639 * character without reading beyond the end, and pass that number on to the
1640 * validating routine */
6302f837 1641 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
28123549
KW
1642 if (ckWARN_d(WARN_UTF8)) {
1643 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
9816f121 1644 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
28123549
KW
1645 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1646 what the malformation is */
1647 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1648 }
1649 }
1650 return FALSE;
1651 }
87367d5f
KW
1652 if (!*swash) {
1653 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
1654 *swash = _core_swash_init("utf8",
1655
1656 /* Only use the name if there is no inversion
1657 * list; otherwise will go out to disk */
1658 (invlist) ? "" : swashname,
1659
1660 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 1661 }
28123549 1662
bde6a22d
NC
1663 return swash_fetch(*swash, p, TRUE) != 0;
1664}
1665
1666bool
922e8cb4
KW
1667Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1668{
922e8cb4
KW
1669 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1670
1671 assert(classnum < _FIRST_NON_SWASH_CC);
1672
f25ce844
KW
1673 return is_utf8_common(p,
1674 &PL_utf8_swash_ptrs[classnum],
1675 swash_property_names[classnum],
1676 PL_XPosix_ptrs[classnum]);
922e8cb4
KW
1677}
1678
1679bool
f2645549 1680Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
a0ed51b3 1681{
f2645549 1682 SV* invlist = NULL;
7918f24d 1683
f2645549 1684 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
7918f24d 1685
f2645549
KW
1686 if (! PL_utf8_perl_idstart) {
1687 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1688 }
1689 return is_utf8_common(p, &PL_utf8_perl_idstart, "", invlist);
82686b01
JH
1690}
1691
1692bool
f2645549 1693Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 1694{
f2645549 1695 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
1696
1697 if (*p == '_')
1698 return TRUE;
f25ce844 1699 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
1700}
1701
1702bool
eba68aa0
KW
1703Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1704{
b24b43f7 1705 SV* invlist = NULL;
eba68aa0
KW
1706
1707 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1708
b24b43f7
KW
1709 if (! PL_utf8_perl_idcont) {
1710 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1711 }
1712 return is_utf8_common(p, &PL_utf8_perl_idcont, "", invlist);
eba68aa0
KW
1713}
1714
eba68aa0 1715bool
f2645549 1716Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 1717{
f2645549 1718 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 1719
f25ce844 1720 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
1721}
1722
1723bool
f2645549 1724Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 1725{
f2645549 1726 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 1727
f25ce844 1728 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
1729}
1730
1731bool
7dbf68d2
KW
1732Perl__is_utf8_mark(pTHX_ const U8 *p)
1733{
7dbf68d2
KW
1734 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1735
f25ce844 1736 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
1737}
1738
6b5c0936 1739/*
87cea99e 1740=for apidoc to_utf8_case
6b5c0936 1741
6fae5207 1742C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
1743the character that is being converted. This routine assumes that the character
1744at C<p> is well-formed.
6b5c0936 1745
6fae5207
KW
1746C<ustrp> is a pointer to the character buffer to put the
1747conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
1748of the result.
1749
6fae5207 1750C<swashp> is a pointer to the swash to use.
6b5c0936 1751
a1433954 1752Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
6fae5207 1753and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 1754but not always, a multicharacter mapping), is tried first.
6b5c0936 1755
4a8240a3
KW
1756C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1757any special mappings; C<""> means to use the special mappings. Values other
1758than these two are treated as the name of the hash containing the special
1759mappings, like C<"utf8::ToSpecLower">.
6b5c0936 1760
6fae5207 1761C<normal> is a string like "ToLower" which means the swash
0134edef
JH
1762%utf8::ToLower.
1763
1764=cut */
6b5c0936 1765
2104c8d9 1766UV
9a957fbc
AL
1767Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1768 SV **swashp, const char *normal, const char *special)
a0ed51b3 1769{
0134edef 1770 STRLEN len = 0;
f4cd282c 1771 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
7918f24d
NC
1772
1773 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1774
9ae3ac1a
KW
1775 /* Note that swash_fetch() doesn't output warnings for these because it
1776 * assumes we will */
8457b38f 1777 if (uv1 >= UNICODE_SURROGATE_FIRST) {
9ae3ac1a 1778 if (uv1 <= UNICODE_SURROGATE_LAST) {
8457b38f
KW
1779 if (ckWARN_d(WARN_SURROGATE)) {
1780 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1781 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1782 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1783 }
9ae3ac1a
KW
1784 }
1785 else if (UNICODE_IS_SUPER(uv1)) {
8457b38f
KW
1786 if (ckWARN_d(WARN_NON_UNICODE)) {
1787 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1788 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1789 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1790 }
9ae3ac1a
KW
1791 }
1792
1793 /* Note that non-characters are perfectly legal, so no warning should
1794 * be given */
1795 }
1796
0134edef 1797 if (!*swashp) /* load on-demand */
5ab9d2ef 1798 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 1799
a6f87d8c 1800 if (special) {
0134edef 1801 /* It might be "special" (sometimes, but not always,
2a37f04d 1802 * a multicharacter mapping) */
4a8240a3 1803 HV *hv = NULL;
b08cf34e
JH
1804 SV **svp;
1805
4a8240a3
KW
1806 /* If passed in the specials name, use that; otherwise use any
1807 * given in the swash */
1808 if (*special != '\0') {
1809 hv = get_hv(special, 0);
1810 }
1811 else {
1812 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
1813 if (svp) {
1814 hv = MUTABLE_HV(SvRV(*svp));
1815 }
1816 }
1817
176fe009
KW
1818 if (hv
1819 && (svp = hv_fetch(hv, (const char*)p, UNISKIP(uv1), FALSE))
1820 && (*svp))
1821 {
cfd0369c 1822 const char *s;
47654450 1823
cfd0369c 1824 s = SvPV_const(*svp, len);
47654450 1825 if (len == 1)
f4cd282c 1826 /* EIGHTBIT */
c80e42f3 1827 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 1828 else {
d2dcd0fb 1829 Copy(s, ustrp, len, U8);
29e98929 1830 }
983ffd37 1831 }
0134edef
JH
1832 }
1833
1834 if (!len && *swashp) {
f4cd282c 1835 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is utf8 */);
d4c19fe8 1836
0134edef
JH
1837 if (uv2) {
1838 /* It was "normal" (a single character mapping). */
f4cd282c 1839 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
1840 }
1841 }
1feea2c7 1842
cbe07460
KW
1843 if (len) {
1844 if (lenp) {
1845 *lenp = len;
1846 }
1847 return valid_utf8_to_uvchr(ustrp, 0);
1848 }
1849
1850 /* Here, there was no mapping defined, which means that the code point maps
1851 * to itself. Return the inputs */
bfdf22ec 1852 len = UTF8SKIP(p);
ca9fab46
KW
1853 if (p != ustrp) { /* Don't copy onto itself */
1854 Copy(p, ustrp, len, U8);
1855 }
0134edef 1856
2a37f04d
JH
1857 if (lenp)
1858 *lenp = len;
1859
f4cd282c 1860 return uv1;
cbe07460 1861
a0ed51b3
LW
1862}
1863
051a06d4
KW
1864STATIC UV
1865S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
1866{
1867 /* This is called when changing the case of a utf8-encoded character above
31f05a37
KW
1868 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
1869 * result contains a character that crosses the 255/256 boundary, disallow
1870 * the change, and return the original code point. See L<perlfunc/lc> for
1871 * why;
051a06d4 1872 *
a1433954
KW
1873 * p points to the original string whose case was changed; assumed
1874 * by this routine to be well-formed
051a06d4
KW
1875 * result the code point of the first character in the changed-case string
1876 * ustrp points to the changed-case string (<result> represents its first char)
1877 * lenp points to the length of <ustrp> */
1878
1879 UV original; /* To store the first code point of <p> */
1880
1881 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
1882
a4f12ed7 1883 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
1884
1885 /* We know immediately if the first character in the string crosses the
1886 * boundary, so can skip */
1887 if (result > 255) {
1888
1889 /* Look at every character in the result; if any cross the
1890 * boundary, the whole thing is disallowed */
1891 U8* s = ustrp + UTF8SKIP(ustrp);
1892 U8* e = ustrp + *lenp;
1893 while (s < e) {
a4f12ed7 1894 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
1895 goto bad_crossing;
1896 }
1897 s += UTF8SKIP(s);
1898 }
1899
1900 /* Here, no characters crossed, result is ok as-is */
1901 return result;
1902 }
1903
1904bad_crossing:
1905
1906 /* Failed, have to return the original */
4b88fb76 1907 original = valid_utf8_to_uvchr(p, lenp);
051a06d4
KW
1908 Copy(p, ustrp, *lenp, char);
1909 return original;
1910}
1911
d3e79532 1912/*
87cea99e 1913=for apidoc to_utf8_upper
d3e79532 1914
1f607577 1915Instead use L</toUPPER_utf8>.
a1433954 1916
d3e79532
JH
1917=cut */
1918
051a06d4 1919/* Not currently externally documented, and subject to change:
31f05a37
KW
1920 * <flags> is set iff iff the rules from the current underlying locale are to
1921 * be used. */
051a06d4 1922
2104c8d9 1923UV
31f05a37 1924Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 1925{
051a06d4
KW
1926 UV result;
1927
1928 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 1929
31f05a37
KW
1930 if (flags && IN_UTF8_CTYPE_LOCALE) {
1931 flags = FALSE;
1932 }
1933
3a4c58c9 1934 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
1935 if (flags) {
1936 result = toUPPER_LC(*p);
1937 }
1938 else {
81c6c7ce 1939 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 1940 }
3a4c58c9
KW
1941 }
1942 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 1943 if (flags) {
a6d8b88b 1944 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 1945 result = toUPPER_LC(c);
051a06d4
KW
1946 }
1947 else {
94bb8c36 1948 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 1949 ustrp, lenp, 'S');
051a06d4
KW
1950 }
1951 }
1952 else { /* utf8, ord above 255 */
1953 result = CALL_UPPER_CASE(p, ustrp, lenp);
1954
1955 if (flags) {
1956 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
1957 }
1958 return result;
1959 }
1960
1961 /* Here, used locale rules. Convert back to utf8 */
1962 if (UTF8_IS_INVARIANT(result)) {
1963 *ustrp = (U8) result;
1964 *lenp = 1;
1965 }
1966 else {
62cb07ea
KW
1967 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
1968 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 1969 *lenp = 2;
3a4c58c9 1970 }
baa60164 1971
051a06d4 1972 return result;
983ffd37 1973}
a0ed51b3 1974
d3e79532 1975/*
87cea99e 1976=for apidoc to_utf8_title
d3e79532 1977
1f607577 1978Instead use L</toTITLE_utf8>.
a1433954 1979
d3e79532
JH
1980=cut */
1981
051a06d4 1982/* Not currently externally documented, and subject to change:
31f05a37
KW
1983 * <flags> is set iff the rules from the current underlying locale are to be
1984 * used. Since titlecase is not defined in POSIX, for other than a
1985 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 1986 */
051a06d4 1987
983ffd37 1988UV
31f05a37 1989Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
983ffd37 1990{
051a06d4
KW
1991 UV result;
1992
1993 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 1994
31f05a37
KW
1995 if (flags && IN_UTF8_CTYPE_LOCALE) {
1996 flags = FALSE;
1997 }
1998
3a4c58c9 1999 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2000 if (flags) {
2001 result = toUPPER_LC(*p);
2002 }
2003 else {
81c6c7ce 2004 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2005 }
3a4c58c9
KW
2006 }
2007 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2008 if (flags) {
a6d8b88b 2009 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2010 result = toUPPER_LC(c);
051a06d4
KW
2011 }
2012 else {
94bb8c36 2013 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2014 ustrp, lenp, 's');
051a06d4
KW
2015 }
2016 }
2017 else { /* utf8, ord above 255 */
2018 result = CALL_TITLE_CASE(p, ustrp, lenp);
2019
2020 if (flags) {
2021 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2022 }
2023 return result;
2024 }
2025
2026 /* Here, used locale rules. Convert back to utf8 */
2027 if (UTF8_IS_INVARIANT(result)) {
2028 *ustrp = (U8) result;
2029 *lenp = 1;
2030 }
2031 else {
62cb07ea
KW
2032 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2033 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2034 *lenp = 2;
3a4c58c9
KW
2035 }
2036
051a06d4 2037 return result;
a0ed51b3
LW
2038}
2039
d3e79532 2040/*
87cea99e 2041=for apidoc to_utf8_lower
d3e79532 2042
1f607577 2043Instead use L</toLOWER_utf8>.
a1433954 2044
d3e79532
JH
2045=cut */
2046
051a06d4 2047/* Not currently externally documented, and subject to change:
31f05a37
KW
2048 * <flags> is set iff iff the rules from the current underlying locale are to
2049 * be used.
2050 */
051a06d4 2051
2104c8d9 2052UV
31f05a37 2053Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2054{
051a06d4
KW
2055 UV result;
2056
051a06d4 2057 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2058
31f05a37
KW
2059 if (flags && IN_UTF8_CTYPE_LOCALE) {
2060 flags = FALSE;
2061 }
2062
968c5e6a 2063 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2064 if (flags) {
2065 result = toLOWER_LC(*p);
2066 }
2067 else {
81c6c7ce 2068 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2069 }
968c5e6a
KW
2070 }
2071 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2072 if (flags) {
a6d8b88b 2073 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2074 result = toLOWER_LC(c);
051a06d4
KW
2075 }
2076 else {
94bb8c36 2077 return to_lower_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2078 ustrp, lenp);
051a06d4 2079 }
968c5e6a 2080 }
051a06d4
KW
2081 else { /* utf8, ord above 255 */
2082 result = CALL_LOWER_CASE(p, ustrp, lenp);
2083
2084 if (flags) {
2085 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2086 }
968c5e6a 2087
051a06d4
KW
2088 return result;
2089 }
2090
2091 /* Here, used locale rules. Convert back to utf8 */
2092 if (UTF8_IS_INVARIANT(result)) {
2093 *ustrp = (U8) result;
2094 *lenp = 1;
2095 }
2096 else {
62cb07ea
KW
2097 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2098 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2099 *lenp = 2;
2100 }
2101
051a06d4 2102 return result;
b4e400f9
JH
2103}
2104
d3e79532 2105/*
87cea99e 2106=for apidoc to_utf8_fold
d3e79532 2107
1f607577 2108Instead use L</toFOLD_utf8>.
a1433954 2109
d3e79532
JH
2110=cut */
2111
051a06d4
KW
2112/* Not currently externally documented, and subject to change,
2113 * in <flags>
31f05a37
KW
2114 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2115 * locale are to be used.
051a06d4
KW
2116 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2117 * otherwise simple folds
a0270393
KW
2118 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2119 * prohibited
445bf929 2120 */
36bb2ab6 2121
b4e400f9 2122UV
445bf929 2123Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
b4e400f9 2124{
051a06d4
KW
2125 UV result;
2126
36bb2ab6 2127 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2128
a0270393
KW
2129 /* These are mutually exclusive */
2130 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2131
50ba90ff
KW
2132 assert(p != ustrp); /* Otherwise overwrites */
2133
31f05a37
KW
2134 if (flags & FOLD_FLAGS_LOCALE && IN_UTF8_CTYPE_LOCALE) {
2135 flags &= ~FOLD_FLAGS_LOCALE;
2136 }
2137
a1dde8de 2138 if (UTF8_IS_INVARIANT(*p)) {
051a06d4 2139 if (flags & FOLD_FLAGS_LOCALE) {
d22b930b 2140 result = toFOLD_LC(*p);
051a06d4
KW
2141 }
2142 else {
81c6c7ce 2143 return _to_fold_latin1(*p, ustrp, lenp,
1ca267a5 2144 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2145 }
a1dde8de
KW
2146 }
2147 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2148 if (flags & FOLD_FLAGS_LOCALE) {
a6d8b88b 2149 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2150 result = toFOLD_LC(c);
051a06d4
KW
2151 }
2152 else {
94bb8c36 2153 return _to_fold_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
51910141 2154 ustrp, lenp,
1ca267a5 2155 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2156 }
a1dde8de 2157 }
051a06d4 2158 else { /* utf8, ord above 255 */
a0270393 2159 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2160
1ca267a5
KW
2161 if (flags & FOLD_FLAGS_LOCALE) {
2162
538e84ed
KW
2163 /* Special case these two characters, as what normally gets
2164 * returned under locale doesn't work */
1ca267a5
KW
2165 if (UTF8SKIP(p) == sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1
2166 && memEQ((char *) p, LATIN_CAPITAL_LETTER_SHARP_S_UTF8,
2167 sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1))
2168 {
2169 goto return_long_s;
2170 }
9fc2026f
KW
2171 else if (UTF8SKIP(p) == sizeof(LATIN_SMALL_LIGATURE_LONG_S_T) - 1
2172 && memEQ((char *) p, LATIN_SMALL_LIGATURE_LONG_S_T_UTF8,
2173 sizeof(LATIN_SMALL_LIGATURE_LONG_S_T_UTF8) - 1))
2174 {
2175 goto return_ligature_st;
2176 }
a0270393 2177 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2178 }
a0270393
KW
2179 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2180 return result;
2181 }
2182 else {
2183 /* This is called when changing the case of a utf8-encoded
9fc2026f
KW
2184 * character above the ASCII range, and the result should not
2185 * contain an ASCII character. */
a0270393
KW
2186
2187 UV original; /* To store the first code point of <p> */
2188
2189 /* Look at every character in the result; if any cross the
2190 * boundary, the whole thing is disallowed */
2191 U8* s = ustrp;
2192 U8* e = ustrp + *lenp;
2193 while (s < e) {
2194 if (isASCII(*s)) {
2195 /* Crossed, have to return the original */
2196 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 2197
9fc2026f 2198 /* But in these instances, there is an alternative we can
1ca267a5 2199 * return that is valid */
9fc2026f
KW
2200 if (original == LATIN_CAPITAL_LETTER_SHARP_S
2201 || original == LATIN_SMALL_LETTER_SHARP_S)
2202 {
1ca267a5
KW
2203 goto return_long_s;
2204 }
9fc2026f
KW
2205 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2206 goto return_ligature_st;
2207 }
a0270393
KW
2208 Copy(p, ustrp, *lenp, char);
2209 return original;
2210 }
2211 s += UTF8SKIP(s);
2212 }
051a06d4 2213
a0270393
KW
2214 /* Here, no characters crossed, result is ok as-is */
2215 return result;
2216 }
051a06d4
KW
2217 }
2218
2219 /* Here, used locale rules. Convert back to utf8 */
2220 if (UTF8_IS_INVARIANT(result)) {
2221 *ustrp = (U8) result;
2222 *lenp = 1;
2223 }
2224 else {
62cb07ea
KW
2225 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2226 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2227 *lenp = 2;
2228 }
2229
051a06d4 2230 return result;
1ca267a5
KW
2231
2232 return_long_s:
2233 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2234 * folds to a string of two of these characters. By returning this
2235 * instead, then, e.g.,
2236 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2237 * works. */
2238
2239 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2240 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2241 ustrp, *lenp, U8);
2242 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
2243
2244 return_ligature_st:
2245 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2246 * have the other one fold to it */
2247
2248 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2249 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2250 return LATIN_SMALL_LIGATURE_ST;
a0ed51b3
LW
2251}
2252
711a919c 2253/* Note:
f90a9a02 2254 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2255 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2256 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2257 */
c4a5db0c 2258
a0ed51b3 2259SV*
7fc63493 2260Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2261{
c4a5db0c
KW
2262 PERL_ARGS_ASSERT_SWASH_INIT;
2263
2264 /* Returns a copy of a swash initiated by the called function. This is the
2265 * public interface, and returning a copy prevents others from doing
2266 * mischief on the original */
2267
5d3d13d1 2268 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
2269}
2270
2271SV*
5d3d13d1 2272Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c
KW
2273{
2274 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
2275 * by calling utf8_heavy.pl in the general case. The returned value may be
2276 * the swash's inversion list instead if the input parameters allow it.
2277 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
2278 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2279 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
2280 *
2281 * This interface should only be used by functions that won't destroy or
2282 * adversely change the swash, as doing so affects all other uses of the
2283 * swash in the program; the general public should use 'Perl_swash_init'
2284 * instead.
2285 *
2286 * pkg is the name of the package that <name> should be in.
2287 * name is the name of the swash to find. Typically it is a Unicode
2288 * property name, including user-defined ones
2289 * listsv is a string to initialize the swash with. It must be of the form
2290 * documented as the subroutine return value in
2291 * L<perlunicode/User-Defined Character Properties>
2292 * minbits is the number of bits required to represent each data element.
2293 * It is '1' for binary properties.
2294 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 2295 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
2296 * flags_p if non-NULL is the address of various input and output flag bits
2297 * to the routine, as follows: ('I' means is input to the routine;
2298 * 'O' means output from the routine. Only flags marked O are
2299 * meaningful on return.)
2300 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2301 * came from a user-defined property. (I O)
5d3d13d1
KW
2302 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2303 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
2304 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2305 * return of an inversion list instead of a swash hash if this routine
2306 * thinks that would result in faster execution of swash_fetch() later
2307 * on. (I)
9a53f6cf
KW
2308 *
2309 * Thus there are three possible inputs to find the swash: <name>,
2310 * <listsv>, and <invlist>. At least one must be specified. The result
2311 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
2312 * actions can intersect, etc. what <name> gives. To avoid going out to
2313 * disk at all, <invlist> should specify completely what the swash should
2314 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
2315 *
2316 * <invlist> is only valid for binary properties */
c4a5db0c 2317
c4a5db0c 2318 SV* retval = &PL_sv_undef;
83199d38 2319 HV* swash_hv = NULL;
87367d5f
KW
2320 const int invlist_swash_boundary =
2321 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2322 ? 512 /* Based on some benchmarking, but not extensive, see commit
2323 message */
2324 : -1; /* Never return just an inversion list */
9a53f6cf
KW
2325
2326 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2327 assert(! invlist || minbits == 1);
2328
2329 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2330 * so */
2331 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2332 dSP;
2333 const size_t pkg_len = strlen(pkg);
2334 const size_t name_len = strlen(name);
2335 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2336 SV* errsv_save;
2337 GV *method;
2338
2339 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2340
2341 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2342 ENTER;
69794297
KW
2343 SAVEHINTS();
2344 save_re_context();
650f067c
JL
2345 /* We might get here via a subroutine signature which uses a utf8
2346 * parameter name, at which point PL_subname will have been set
2347 * but not yet used. */
2348 save_item(PL_subname);
69794297
KW
2349 if (PL_parser && PL_parser->error_count)
2350 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2351 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2352 if (!method) { /* demand load utf8 */
2353 ENTER;
db2c6cb3
FC
2354 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2355 GvSV(PL_errgv) = NULL;
69794297
KW
2356 /* It is assumed that callers of this routine are not passing in
2357 * any user derived data. */
2358 /* Need to do this after save_re_context() as it will set
2359 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2360 * in Perl_magic_get). Even line to create errsv_save can turn on
2361 * PL_tainted. */
284167a5
SM
2362#ifndef NO_TAINT_SUPPORT
2363 SAVEBOOL(TAINT_get);
2364 TAINT_NOT;
2365#endif
69794297
KW
2366 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2367 NULL);
eed484f9 2368 {
db2c6cb3
FC
2369 /* Not ERRSV, as there is no need to vivify a scalar we are
2370 about to discard. */
2371 SV * const errsv = GvSV(PL_errgv);
2372 if (!SvTRUE(errsv)) {
2373 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2374 SvREFCNT_dec(errsv);
2375 }
eed484f9 2376 }
69794297
KW
2377 LEAVE;
2378 }
2379 SPAGAIN;
2380 PUSHMARK(SP);
2381 EXTEND(SP,5);
2382 mPUSHp(pkg, pkg_len);
2383 mPUSHp(name, name_len);
2384 PUSHs(listsv);
2385 mPUSHi(minbits);
2386 mPUSHi(none);
2387 PUTBACK;
db2c6cb3
FC
2388 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2389 GvSV(PL_errgv) = NULL;
69794297
KW
2390 /* If we already have a pointer to the method, no need to use
2391 * call_method() to repeat the lookup. */
c41800a8
KW
2392 if (method
2393 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
2394 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2395 {
2396 retval = *PL_stack_sp--;
2397 SvREFCNT_inc(retval);
2398 }
eed484f9 2399 {
db2c6cb3
FC
2400 /* Not ERRSV. See above. */
2401 SV * const errsv = GvSV(PL_errgv);
2402 if (!SvTRUE(errsv)) {
2403 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2404 SvREFCNT_dec(errsv);
2405 }
eed484f9 2406 }
ce3b816e 2407 LEAVE;
69794297
KW
2408 POPSTACK;
2409 if (IN_PERL_COMPILETIME) {
2410 CopHINTS_set(PL_curcop, PL_hints);
2411 }
2412 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2413 if (SvPOK(retval))
2414
2415 /* If caller wants to handle missing properties, let them */
5d3d13d1 2416 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
69794297
KW
2417 return NULL;
2418 }
2419 Perl_croak(aTHX_
2420 "Can't find Unicode property definition \"%"SVf"\"",
2421 SVfARG(retval));
a25b5927 2422 NOT_REACHED; /* NOTREACHED */
69794297 2423 }
9a53f6cf 2424 } /* End of calling the module to find the swash */
36eb48b4 2425
83199d38
KW
2426 /* If this operation fetched a swash, and we will need it later, get it */
2427 if (retval != &PL_sv_undef
2428 && (minbits == 1 || (flags_p
2429 && ! (*flags_p
2430 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2431 {
2432 swash_hv = MUTABLE_HV(SvRV(retval));
2433
2434 /* If we don't already know that there is a user-defined component to
2435 * this swash, and the user has indicated they wish to know if there is
2436 * one (by passing <flags_p>), find out */
2437 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2438 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2439 if (user_defined && SvUV(*user_defined)) {
2440 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2441 }
2442 }
2443 }
2444
36eb48b4
KW
2445 /* Make sure there is an inversion list for binary properties */
2446 if (minbits == 1) {
2447 SV** swash_invlistsvp = NULL;
2448 SV* swash_invlist = NULL;
9a53f6cf 2449 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
2450 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2451 an unclaimed reference count */
36eb48b4 2452
9a53f6cf 2453 /* If this operation fetched a swash, get its already existing
83199d38 2454 * inversion list, or create one for it */
36eb48b4 2455
83199d38 2456 if (swash_hv) {
5c9f4bd2 2457 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
2458 if (swash_invlistsvp) {
2459 swash_invlist = *swash_invlistsvp;
2460 invlist_in_swash_is_valid = TRUE;
2461 }
2462 else {
36eb48b4 2463 swash_invlist = _swash_to_invlist(retval);
02c85471 2464 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
2465 }
2466 }
2467
2468 /* If an inversion list was passed in, have to include it */
2469 if (invlist) {
2470
2471 /* Any fetched swash will by now have an inversion list in it;
2472 * otherwise <swash_invlist> will be NULL, indicating that we
2473 * didn't fetch a swash */
2474 if (swash_invlist) {
2475
2476 /* Add the passed-in inversion list, which invalidates the one
2477 * already stored in the swash */
2478 invlist_in_swash_is_valid = FALSE;
2479 _invlist_union(invlist, swash_invlist, &swash_invlist);
2480 }
2481 else {
2482
87367d5f
KW
2483 /* Here, there is no swash already. Set up a minimal one, if
2484 * we are going to return a swash */
2485 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 2486 swash_hv = newHV();
4aca0fe6 2487 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 2488 }
9a53f6cf
KW
2489 swash_invlist = invlist;
2490 }
9a53f6cf
KW
2491 }
2492
2493 /* Here, we have computed the union of all the passed-in data. It may
2494 * be that there was an inversion list in the swash which didn't get
538e84ed 2495 * touched; otherwise save the computed one */
87367d5f
KW
2496 if (! invlist_in_swash_is_valid
2497 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2498 {
5c9f4bd2 2499 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
2500 {
2501 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2502 }
cc34d8c5
FC
2503 /* We just stole a reference count. */
2504 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2505 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 2506 }
87367d5f 2507
dbfdbd26
KW
2508 SvREADONLY_on(swash_invlist);
2509
c41800a8 2510 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
2511 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2512 SvREFCNT_dec(retval);
02c85471
FC
2513 if (!swash_invlist_unclaimed)
2514 SvREFCNT_inc_simple_void_NN(swash_invlist);
2515 retval = newRV_noinc(swash_invlist);
87367d5f 2516 }
36eb48b4
KW
2517 }
2518
a0ed51b3
LW
2519 return retval;
2520}
2521
035d37be
JH
2522
2523/* This API is wrong for special case conversions since we may need to
2524 * return several Unicode characters for a single Unicode character
2525 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2526 * the lower-level routine, and it is similarly broken for returning
38684baa
KW
2527 * multiple values. --jhi
2528 * For those, you should use to_utf8_case() instead */
b0e3252e 2529/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
2530
2531/* Note:
2532 * Returns the value of property/mapping C<swash> for the first character
2533 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3d0f8846
KW
2534 * assumed to be in well-formed utf8. If C<do_utf8> is false, the string C<ptr>
2535 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
2536 *
2537 * A "swash" is a hash which contains initially the keys/values set up by
2538 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2539 * property for all possible code points. Things are stored in a compact form
2540 * (see utf8_heavy.pl) so that calculation is required to find the actual
2541 * property value for a given code point. As code points are looked up, new
2542 * key/value pairs are added to the hash, so that the calculation doesn't have
2543 * to ever be re-done. Further, each calculation is done, not just for the
2544 * desired one, but for a whole block of code points adjacent to that one.
2545 * For binary properties on ASCII machines, the block is usually for 64 code
2546 * points, starting with a code point evenly divisible by 64. Thus if the
2547 * property value for code point 257 is requested, the code goes out and
2548 * calculates the property values for all 64 code points between 256 and 319,
2549 * and stores these as a single 64-bit long bit vector, called a "swatch",
2550 * under the key for code point 256. The key is the UTF-8 encoding for code
2551 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2552 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2553 * for code point 258 is then requested, this code realizes that it would be
2554 * stored under the key for 256, and would find that value and extract the
2555 * relevant bit, offset from 256.
2556 *
2557 * Non-binary properties are stored in as many bits as necessary to represent
2558 * their values (32 currently, though the code is more general than that), not
2559 * as single bits, but the principal is the same: the value for each key is a
2560 * vector that encompasses the property values for all code points whose UTF-8
2561 * representations are represented by the key. That is, for all code points
2562 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2563 * bytes of that.
680c470c 2564 */
a0ed51b3 2565UV
680c470c 2566Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 2567{
ef8f7699 2568 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
2569 U32 klen;
2570 U32 off;
9b56a019 2571 STRLEN slen = 0;
7d85a32c 2572 STRLEN needents;
cfd0369c 2573 const U8 *tmps = NULL;
979f2922 2574 SV *swatch;
08fb1ac5 2575 const U8 c = *ptr;
3568d838 2576
7918f24d
NC
2577 PERL_ARGS_ASSERT_SWASH_FETCH;
2578
87367d5f
KW
2579 /* If it really isn't a hash, it isn't really swash; must be an inversion
2580 * list */
2581 if (SvTYPE(hv) != SVt_PVHV) {
2582 return _invlist_contains_cp((SV*)hv,
2583 (do_utf8)
2584 ? valid_utf8_to_uvchr(ptr, NULL)
2585 : c);
2586 }
2587
08fb1ac5
KW
2588 /* We store the values in a "swatch" which is a vec() value in a swash
2589 * hash. Code points 0-255 are a single vec() stored with key length
2590 * (klen) 0. All other code points have a UTF-8 representation
2591 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2592 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2593 * length for them is the length of the encoded char - 1. ptr[klen] is the
2594 * final byte in the sequence representing the character */
2595 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2596 klen = 0;
2597 needents = 256;
2598 off = c;
3568d838 2599 }
08fb1ac5
KW
2600 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2601 klen = 0;
2602 needents = 256;
2603 off = TWO_BYTE_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
2604 }
2605 else {
08fb1ac5
KW
2606 klen = UTF8SKIP(ptr) - 1;
2607
2608 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2609 * the vec is the final byte in the sequence. (In EBCDIC this is
2610 * converted to I8 to get consecutive values.) To help you visualize
2611 * all this:
2612 * Straight 1047 After final byte
2613 * UTF-8 UTF-EBCDIC I8 transform
2614 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2615 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2616 * ...
2617 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2618 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2619 * ...
2620 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2621 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2622 * ...
2623 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2624 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2625 * ...
2626 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2627 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2628 *
2629 * (There are no discontinuities in the elided (...) entries.)
2630 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2631 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2632 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2633 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2634 * index into the vec() swatch (after subtracting 0x80, which we
2635 * actually do with an '&').
2636 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2637 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2638 * dicontinuities which go away by transforming it into I8, and we
2639 * effectively subtract 0xA0 to get the index. */
979f2922 2640 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 2641 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 2642 }
7d85a32c 2643
a0ed51b3
LW
2644 /*
2645 * This single-entry cache saves about 1/3 of the utf8 overhead in test
2646 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2647 * it's nothing to sniff at.) Pity we usually come through at least
2648 * two function calls to get here...
2649 *
2650 * NB: this code assumes that swatches are never modified, once generated!
2651 */
2652
3568d838 2653 if (hv == PL_last_swash_hv &&
a0ed51b3 2654 klen == PL_last_swash_klen &&
27da23d5 2655 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
2656 {
2657 tmps = PL_last_swash_tmps;
2658 slen = PL_last_swash_slen;
2659 }
2660 else {
2661 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 2662 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 2663
b0e3252e 2664 /* If not cached, generate it via swatch_get */
979f2922 2665 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
2666 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2667 {
2668 if (klen) {
2669 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2670 swatch = swatch_get(swash,
2671 code_point & ~((UV)needents - 1),
2672 needents);
2673 }
2674 else { /* For the first 256 code points, the swatch has a key of
2675 length 0 */
2676 swatch = swatch_get(swash, 0, needents);
2677 }
979f2922 2678
923e4eb5 2679 if (IN_PERL_COMPILETIME)
623e6609 2680 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 2681
979f2922 2682 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 2683
979f2922
ST
2684 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2685 || (slen << 3) < needents)
5637ef5b
NC
2686 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2687 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2688 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
2689 }
2690
2691 PL_last_swash_hv = hv;
16d8f38a 2692 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 2693 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
2694 /* FIXME change interpvar.h? */
2695 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
2696 PL_last_swash_slen = slen;
2697 if (klen)
2698 Copy(ptr, PL_last_swash_key, klen, U8);
2699 }
2700
9faf8d75 2701 switch ((int)((slen << 3) / needents)) {
a0ed51b3 2702 case 1:
e7aca353 2703 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
a0ed51b3 2704 case 8:
e7aca353 2705 return ((UV) tmps[off]);
a0ed51b3
LW
2706 case 16:
2707 off <<= 1;
e7aca353
JH
2708 return
2709 ((UV) tmps[off ] << 8) +
2710 ((UV) tmps[off + 1]);
a0ed51b3
LW
2711 case 32:
2712 off <<= 2;
e7aca353
JH
2713 return
2714 ((UV) tmps[off ] << 24) +
2715 ((UV) tmps[off + 1] << 16) +
2716 ((UV) tmps[off + 2] << 8) +
2717 ((UV) tmps[off + 3]);
a0ed51b3 2718 }
5637ef5b
NC
2719 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2720 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 2721 NORETURN_FUNCTION_END;
a0ed51b3 2722}
2b9d42f0 2723
319009ee
KW
2724/* Read a single line of the main body of the swash input text. These are of
2725 * the form:
2726 * 0053 0056 0073
2727 * where each number is hex. The first two numbers form the minimum and
2728 * maximum of a range, and the third is the value associated with the range.
2729 * Not all swashes should have a third number
2730 *
2731 * On input: l points to the beginning of the line to be examined; it points
2732 * to somewhere in the string of the whole input text, and is
2733 * terminated by a \n or the null string terminator.
2734 * lend points to the null terminator of that string
2735 * wants_value is non-zero if the swash expects a third number
2736 * typestr is the name of the swash's mapping, like 'ToLower'
2737 * On output: *min, *max, and *val are set to the values read from the line.
2738 * returns a pointer just beyond the line examined. If there was no
2739 * valid min number on the line, returns lend+1
2740 */
2741
2742STATIC U8*
2743S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2744 const bool wants_value, const U8* const typestr)
2745{
2746 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2747 STRLEN numlen; /* Length of the number */
02470786
KW
2748 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2749 | PERL_SCAN_DISALLOW_PREFIX
2750 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2751
2752 /* nl points to the next \n in the scan */
2753 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2754
95543e92
KW
2755 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
2756
319009ee
KW
2757 /* Get the first number on the line: the range minimum */
2758 numlen = lend - l;
2759 *min = grok_hex((char *)l, &numlen, &flags, NULL);
c88850db 2760 *max = *min; /* So can never return without setting max */
319009ee
KW
2761 if (numlen) /* If found a hex number, position past it */
2762 l += numlen;
2763 else if (nl) { /* Else, go handle next line, if any */
2764 return nl + 1; /* 1 is length of "\n" */
2765 }
2766 else { /* Else, no next line */
2767 return lend + 1; /* to LIST's end at which \n is not found */
2768 }
2769
2770 /* The max range value follows, separated by a BLANK */
2771 if (isBLANK(*l)) {
2772 ++l;
02470786
KW
2773 flags = PERL_SCAN_SILENT_ILLDIGIT
2774 | PERL_SCAN_DISALLOW_PREFIX
2775 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2776 numlen = lend - l;
2777 *max = grok_hex((char *)l, &numlen, &flags, NULL);
2778 if (numlen)
2779 l += numlen;
2780 else /* If no value here, it is a single element range */
2781 *max = *min;
2782
2783 /* Non-binary tables have a third entry: what the first element of the
24303724 2784 * range maps to. The map for those currently read here is in hex */
319009ee
KW
2785 if (wants_value) {
2786 if (isBLANK(*l)) {
2787 ++l;
f2a7d0fc
KW
2788 flags = PERL_SCAN_SILENT_ILLDIGIT
2789 | PERL_SCAN_DISALLOW_PREFIX
2790 | PERL_SCAN_SILENT_NON_PORTABLE;
2791 numlen = lend - l;
2792 *val = grok_hex((char *)l, &numlen, &flags, NULL);
2793 if (numlen)
2794 l += numlen;
2795 else
2796 *val = 0;
319009ee
KW
2797 }
2798 else {
2799 *val = 0;
2800 if (typeto) {
dcbac5bb 2801 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
2802 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
2803 typestr, l);
2804 }
2805 }
2806 }
2807 else
2808 *val = 0; /* bits == 1, then any val should be ignored */
2809 }
2810 else { /* Nothing following range min, should be single element with no
2811 mapping expected */
319009ee
KW
2812 if (wants_value) {
2813 *val = 0;
2814 if (typeto) {
dcbac5bb 2815 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
2816 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
2817 }
2818 }
2819 else
2820 *val = 0; /* bits == 1, then val should be ignored */
2821 }
2822
2823 /* Position to next line if any, or EOF */
2824 if (nl)
2825 l = nl + 1;
2826 else
2827 l = lend;
2828
2829 return l;
2830}
2831
979f2922
ST
2832/* Note:
2833 * Returns a swatch (a bit vector string) for a code point sequence
2834 * that starts from the value C<start> and comprises the number C<span>.
2835 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
2836 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
2837 */
2838STATIC SV*
b0e3252e 2839S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
2840{
2841 SV *swatch;
77f9f126 2842 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 2843 STRLEN lcur, xcur, scur;
ef8f7699 2844 HV *const hv = MUTABLE_HV(SvRV(swash));
5c9f4bd2 2845 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
36eb48b4 2846
88d45d28
KW
2847 SV** listsvp = NULL; /* The string containing the main body of the table */
2848 SV** extssvp = NULL;
2849 SV** invert_it_svp = NULL;
2850 U8* typestr = NULL;
786861f5
KW
2851 STRLEN bits;
2852 STRLEN octets; /* if bits == 1, then octets == 0 */
2853 UV none;
2854 UV end = start + span;
972dd592 2855
36eb48b4 2856 if (invlistsvp == NULL) {
786861f5
KW
2857 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
2858 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
2859 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
2860 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
2861 listsvp = hv_fetchs(hv, "LIST", FALSE);
2862 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
2863
2864 bits = SvUV(*bitssvp);
2865 none = SvUV(*nonesvp);
2866 typestr = (U8*)SvPV_nolen(*typesvp);
2867 }
36eb48b4
KW
2868 else {
2869 bits = 1;
2870 none = 0;
2871 }
786861f5 2872 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 2873
b0e3252e 2874 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 2875
979f2922 2876 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 2877 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 2878 (UV)bits);
979f2922
ST
2879 }
2880
84ea5ef6
KW
2881 /* If overflowed, use the max possible */
2882 if (end < start) {
2883 end = UV_MAX;
2884 span = end - start;
2885 }
2886
979f2922 2887 /* create and initialize $swatch */
979f2922 2888 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
2889 swatch = newSV(scur);
2890 SvPOK_on(swatch);
979f2922
ST
2891 s = (U8*)SvPVX(swatch);
2892 if (octets && none) {
0bd48802 2893 const U8* const e = s + scur;
979f2922
ST
2894 while (s < e) {
2895 if (bits == 8)
2896 *s++ = (U8)(none & 0xff);
2897 else if (bits == 16) {
2898 *s++ = (U8)((none >> 8) & 0xff);
2899 *s++ = (U8)( none & 0xff);
2900 }
2901 else if (bits == 32) {
2902 *s++ = (U8)((none >> 24) & 0xff);
2903 *s++ = (U8)((none >> 16) & 0xff);
2904 *s++ = (U8)((none >> 8) & 0xff);
2905 *s++ = (U8)( none & 0xff);
2906 }
2907 }
2908 *s = '\0';
2909 }
2910 else {
2911 (void)memzero((U8*)s, scur + 1);
2912 }
2913 SvCUR_set(swatch, scur);
2914 s = (U8*)SvPVX(swatch);
2915
36eb48b4
KW
2916 if (invlistsvp) { /* If has an inversion list set up use that */
2917 _invlist_populate_swatch(*invlistsvp, start, end, s);
2918 return swatch;
2919 }
2920
2921 /* read $swash->{LIST} */
979f2922
ST
2922 l = (U8*)SvPV(*listsvp, lcur);
2923 lend = l + lcur;
2924 while (l < lend) {
8ed25d53 2925 UV min, max, val, upper;
95543e92
KW
2926 l = swash_scan_list_line(l, lend, &min, &max, &val,
2927 cBOOL(octets), typestr);
319009ee 2928 if (l > lend) {
979f2922
ST
2929 break;
2930 }
2931
972dd592 2932 /* If looking for something beyond this range, go try the next one */
979f2922
ST
2933 if (max < start)
2934 continue;
2935
8ed25d53
KW
2936 /* <end> is generally 1 beyond where we want to set things, but at the
2937 * platform's infinity, where we can't go any higher, we want to
2938 * include the code point at <end> */
2939 upper = (max < end)
2940 ? max
2941 : (max != UV_MAX || end != UV_MAX)
2942 ? end - 1
2943 : end;
2944
979f2922 2945 if (octets) {
35da51f7 2946 UV key;
979f2922
ST
2947 if (min < start) {
2948 if (!none || val < none) {
2949 val += start - min;
2950 }
2951 min = start;
2952 }
8ed25d53 2953 for (key = min; key <= upper; key++) {
979f2922 2954 STRLEN offset;
979f2922
ST
2955 /* offset must be non-negative (start <= min <= key < end) */
2956 offset = octets * (key - start);
2957 if (bits == 8)
2958 s[offset] = (U8)(val & 0xff);
2959 else if (bits == 16) {
2960 s[offset ] = (U8)((val >> 8) & 0xff);
2961 s[offset + 1] = (U8)( val & 0xff);
2962 }
2963 else if (bits == 32) {
2964 s[offset ] = (U8)((val >> 24) & 0xff);
2965 s[offset + 1] = (U8)((val >> 16) & 0xff);
2966 s[offset + 2] = (U8)((val >> 8) & 0xff);
2967 s[offset + 3] = (U8)( val & 0xff);
2968 }
2969
2970 if (!none || val < none)
2971 ++val;
2972 }
2973 }
711a919c 2974 else { /* bits == 1, then val should be ignored */
35da51f7 2975 UV key;
979f2922
ST
2976 if (min < start)
2977 min = start;
6cb05c12 2978
8ed25d53 2979 for (key = min; key <= upper; key++) {
0bd48802 2980 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
2981 s[offset >> 3] |= 1 << (offset & 7);
2982 }
2983 }
2984 } /* while */
979f2922 2985
9479a769 2986 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 2987 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
2988
2989 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
2990 * be 0, and their inversion should also be 0, as we don't succeed any
2991 * Unicode property matches for non-Unicode code points */
2992 if (start <= PERL_UNICODE_MAX) {
2993
2994 /* The code below assumes that we never cross the
2995 * Unicode/above-Unicode boundary in a range, as otherwise we would
2996 * have to figure out where to stop flipping the bits. Since this
2997 * boundary is divisible by a large power of 2, and swatches comes
2998 * in small powers of 2, this should be a valid assumption */
2999 assert(start + span - 1 <= PERL_UNICODE_MAX);
3000
507a8485
KW
3001 send = s + scur;
3002 while (s < send) {
3003 *s = ~(*s);
3004 s++;
3005 }
0bda3001 3006 }
77f9f126
KW
3007 }
3008
d73c39c5
KW
3009 /* read $swash->{EXTRAS}
3010 * This code also copied to swash_to_invlist() below */
979f2922
ST
3011 x = (U8*)SvPV(*extssvp, xcur);
3012 xend = x + xcur;
3013 while (x < xend) {
3014 STRLEN namelen;
3015 U8 *namestr;
3016 SV** othersvp;
3017 HV* otherhv;
3018 STRLEN otherbits;
3019 SV **otherbitssvp, *other;
711a919c 3020 U8 *s, *o, *nl;
979f2922
ST
3021 STRLEN slen, olen;
3022
35da51f7 3023 const U8 opc = *x++;
979f2922
ST
3024 if (opc == '\n')
3025 continue;
3026
3027 nl = (U8*)memchr(x, '\n', xend - x);
3028
3029 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3030 if (nl) {
3031 x = nl + 1; /* 1 is length of "\n" */
3032 continue;
3033 }
3034 else {
3035 x = xend; /* to EXTRAS' end at which \n is not found */
3036 break;
3037 }
3038 }
3039
3040 namestr = x;
3041 if (nl) {
3042 namelen = nl - namestr;
3043 x = nl + 1;
3044 }
3045 else {
3046 namelen = xend - namestr;
3047 x = xend;
3048 }
3049
3050 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3051 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3052 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3053 otherbits = (STRLEN)SvUV(*otherbitssvp);
3054 if (bits < otherbits)
5637ef5b
NC
3055 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3056 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3057
3058 /* The "other" swatch must be destroyed after. */
b0e3252e 3059 other = swatch_get(*othersvp, start, span);
979f2922
ST
3060 o = (U8*)SvPV(other, olen);
3061
3062 if (!olen)
b0e3252e 3063 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3064
3065 s = (U8*)SvPV(swatch, slen);
3066 if (bits == 1 && otherbits == 1) {
3067 if (slen != olen)
5637ef5b
NC
3068 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3069 "mismatch, slen=%"UVuf", olen=%"UVuf,
3070 (UV)slen, (UV)olen);
979f2922
ST
3071
3072 switch (opc) {
3073 case '+':
3074 while (slen--)
3075 *s++ |= *o++;
3076 break;
3077 case '!':
3078 while (slen--)
3079 *s++ |= ~*o++;
3080 break;
3081 case '-':
3082 while (slen--)
3083 *s++ &= ~*o++;
3084 break;
3085 case '&':
3086 while (slen--)
3087 *s++ &= *o++;
3088 break;
3089 default:
3090 break;
3091 }
3092 }
711a919c 3093 else {
979f2922
ST
3094 STRLEN otheroctets = otherbits >> 3;
3095 STRLEN offset = 0;
35da51f7 3096 U8* const send = s + slen;
979f2922
ST
3097
3098 while (s < send) {
3099 UV otherval = 0;
3100
3101 if (otherbits == 1) {
3102 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3103 ++offset;
3104 }
3105 else {
3106 STRLEN vlen = otheroctets;
3107 otherval = *o++;
3108 while (--vlen) {
3109 otherval <<= 8;
3110 otherval |= *o++;
3111 }
3112 }
3113
711a919c 3114 if (opc == '+' && otherval)
6f207bd3 3115 NOOP; /* replace with otherval */
979f2922
ST
3116 else if (opc == '!' && !otherval)
3117 otherval = 1;
3118 else if (opc == '-' && otherval)
3119 otherval = 0;
3120 else if (opc == '&' && !otherval)
3121 otherval = 0;
3122 else {
711a919c 3123 s += octets; /* no replacement */
979f2922
ST
3124 continue;
3125 }
3126
3127 if (bits == 8)
3128 *s++ = (U8)( otherval & 0xff);
3129 else if (bits == 16) {
3130 *s++ = (U8)((otherval >> 8) & 0xff);
3131 *s++ = (U8)( otherval & 0xff);
3132 }
3133 else if (bits == 32) {
3134 *s++ = (U8)((otherval >> 24) & 0xff);
3135 *s++ = (U8)((otherval >> 16) & 0xff);
3136 *s++ = (U8)((otherval >> 8) & 0xff);
3137 *s++ = (U8)( otherval & 0xff);
3138 }
3139 }
3140 }
3141 sv_free(other); /* through with it! */
3142 } /* while */
3143 return swatch;
3144}
3145
064c021d 3146HV*
4c2e1131 3147Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3148{
3149
79a2a0e8 3150 /* Subject to change or removal. For use only in regcomp.c and regexec.c
5662e334
KW
3151 * Can't be used on a property that is subject to user override, as it
3152 * relies on the value of SPECIALS in the swash which would be set by
3153 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3154 * for overridden properties
064c021d
KW
3155 *
3156 * Returns a hash which is the inversion and closure of a swash mapping.
3157 * For example, consider the input lines:
3158 * 004B 006B
3159 * 004C 006C
3160 * 212A 006B
3161 *
3162 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3163 * 006C. The value for each key is an array. For 006C, the array would
17d5c697 3164 * have two elements, the utf8 for itself, and for 004C. For 006B, there
064c021d
KW
3165 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3166 *
538e84ed
KW
3167 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3168 * keys are only code points that are folded-to, so it isn't a full closure.
3169 *
064c021d
KW
3170 * Essentially, for any code point, it gives all the code points that map to
3171 * it, or the list of 'froms' for that point.
3172 *
5662e334
KW
3173 * Currently it ignores any additions or deletions from other swashes,
3174 * looking at just the main body of the swash, and if there are SPECIALS
3175 * in the swash, at that hash
3176 *
3177 * The specials hash can be extra code points, and most likely consists of
3178 * maps from single code points to multiple ones (each expressed as a string
3179 * of utf8 characters). This function currently returns only 1-1 mappings.
3180 * However consider this possible input in the specials hash:
3181 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3182 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3183 *
3184 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3185 * currently handle. But it also means that FB05 and FB06 are equivalent in
3186 * a 1-1 mapping which we should handle, and this relationship may not be in
3187 * the main table. Therefore this function examines all the multi-char
3188 * sequences and adds the 1-1 mappings that come out of that. */
064c021d
KW
3189
3190 U8 *l, *lend;
3191 STRLEN lcur;
3192 HV *const hv = MUTABLE_HV(SvRV(swash));
3193
923b6d4e
KW
3194 /* The string containing the main body of the table. This will have its
3195 * assertion fail if the swash has been converted to its inversion list */
064c021d
KW
3196 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3197
3198 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3199 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3200 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3201 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3202 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3203 const STRLEN bits = SvUV(*bitssvp);
3204 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3205 const UV none = SvUV(*nonesvp);
5662e334 3206 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
064c021d
KW
3207
3208 HV* ret = newHV();
3209
3210 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3211
3212 /* Must have at least 8 bits to get the mappings */
3213 if (bits != 8 && bits != 16 && bits != 32) {
3214 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3215 (UV)bits);
3216 }
3217
5662e334
KW
3218 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3219 mapping to more than one character */
3220
3221 /* Construct an inverse mapping hash for the specials */
3222 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3223 HV * specials_inverse = newHV();
3224 char *char_from; /* the lhs of the map */
3225 I32 from_len; /* its byte length */
3226 char *char_to; /* the rhs of the map */
3227 I32 to_len; /* its byte length */
3228 SV *sv_to; /* and in a sv */
3229 AV* from_list; /* list of things that map to each 'to' */
3230
3231 hv_iterinit(specials_hv);
3232
3233 /* The keys are the characters (in utf8) that map to the corresponding
3234 * utf8 string value. Iterate through the list creating the inverse
3235 * list. */
3236 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3237 SV** listp;
3238 if (! SvPOK(sv_to)) {
5637ef5b
NC
3239 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3240 "unexpectedly is not a string, flags=%lu",
3241 (unsigned long)SvFLAGS(sv_to));
5662e334 3242 }
4b88fb76 3243 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "Found mapping from %"UVXf", First char of to is %"UVXf"\n", valid_utf8_to_uvchr((U8*) char_from, 0), valid_utf8_to_uvchr((U8*) SvPVX(sv_to), 0)));*/
5662e334
KW
3244
3245 /* Each key in the inverse list is a mapped-to value, and the key's
3246 * hash value is a list of the strings (each in utf8) that map to
3247 * it. Those strings are all one character long */
3248 if ((listp = hv_fetch(specials_inverse,
3249 SvPVX(sv_to),
3250 SvCUR(sv_to), 0)))
3251 {
3252 from_list = (AV*) *listp;
3253 }
3254 else { /* No entry yet for it: create one */
3255 from_list = newAV();
3256 if (! hv_store(specials_inverse,
3257 SvPVX(sv_to),
3258 SvCUR(sv_to),
3259 (SV*) from_list, 0))
3260 {
3261 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3262 }
3263 }
3264
3265 /* Here have the list associated with this 'to' (perhaps newly
3266 * created and empty). Just add to it. Note that we ASSUME that
3267 * the input is guaranteed to not have duplications, so we don't
3268 * check for that. Duplications just slow down execution time. */
3269 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3270 }
3271
3272 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3273 * it looking for cases like the FB05/FB06 examples above. There would
3274 * be an entry in the hash like
3275 * 'st' => [ FB05, FB06 ]
3276 * In this example we will create two lists that get stored in the
3277 * returned hash, 'ret':
3278 * FB05 => [ FB05, FB06 ]
3279 * FB06 => [ FB05, FB06 ]
3280 *
3281 * Note that there is nothing to do if the array only has one element.
3282 * (In the normal 1-1 case handled below, we don't have to worry about
3283 * two lists, as everything gets tied to the single list that is
3284 * generated for the single character 'to'. But here, we are omitting
3285 * that list, ('st' in the example), so must have multiple lists.) */
3286 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3287 &char_to, &to_len)))
3288 {
b9f2b683 3289 if (av_tindex(from_list) > 0) {
c70927a6 3290 SSize_t i;
5662e334
KW
3291
3292 /* We iterate over all combinations of i,j to place each code
3293 * point on each list */
b9f2b683 3294 for (i = 0; i <= av_tindex(from_list); i++) {
c70927a6 3295 SSize_t j;
5662e334
KW
3296 AV* i_list = newAV();
3297 SV** entryp = av_fetch(from_list, i, FALSE);
3298 if (entryp == NULL) {
3299 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3300 }
3301 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3302 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3303 }
3304 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3305 (SV*) i_list, FALSE))
3306 {
3307 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3308 }
3309
538e84ed 3310 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
b9f2b683 3311 for (j = 0; j <= av_tindex(from_list); j++) {
5662e334
KW
3312 entryp = av_fetch(from_list, j, FALSE);
3313 if (entryp == NULL) {
3314 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3315 }
3316
3317 /* When i==j this adds itself to the list */
4b88fb76
KW
3318 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3319 (U8*) SvPVX(*entryp),
3320 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3321 0)));
4637d003 3322 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0), u));*/
5662e334
KW
3323 }
3324 }
3325 }
3326 }
3327 SvREFCNT_dec(specials_inverse); /* done with it */
3328 } /* End of specials */
3329
064c021d
KW
3330 /* read $swash->{LIST} */
3331 l = (U8*)SvPV(*listsvp, lcur);
3332 lend = l + lcur;
3333
3334 /* Go through each input line */
3335 while (l < lend) {
3336 UV min, max, val;
3337 UV inverse;
95543e92
KW
3338 l = swash_scan_list_line(l, lend, &min, &max, &val,
3339 cBOOL(octets), typestr);
064c021d
KW
3340 if (l > lend) {
3341 break;
3342 }
3343
3344 /* Each element in the range is to be inverted */
3345 for (inverse = min; inverse <= max; inverse++) {
3346 AV* list;
064c021d
KW
3347 SV** listp;
3348 IV i;
3349 bool found_key = FALSE;
5662e334 3350 bool found_inverse = FALSE;
064c021d
KW
3351
3352 /* The key is the inverse mapping */
3353 char key[UTF8_MAXBYTES+1];
c80e42f3 3354 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
064c021d
KW
3355 STRLEN key_len = key_end - key;
3356
064c021d
KW
3357 /* Get the list for the map */
3358 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3359 list = (AV*) *listp;
3360 }
3361 else { /* No entry yet for it: create one */
3362 list = newAV();
3363 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3364 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3365 }
3366 }
3367
5662e334
KW
3368 /* Look through list to see if this inverse mapping already is
3369 * listed, or if there is a mapping to itself already */
b9f2b683 3370 for (i = 0; i <= av_tindex(list); i++) {
064c021d
KW
3371 SV** entryp = av_fetch(list, i, FALSE);
3372 SV* entry;
414db8a1 3373 UV uv;
064c021d
KW
3374 if (entryp == NULL) {
3375 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3376 }
3377 entry = *entryp;
414db8a1
DM
3378 uv = SvUV(entry);
3379 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, uv));*/
3380 if (uv == val) {
064c021d 3381 found_key = TRUE;
5662e334 3382 }
414db8a1 3383 if (uv == inverse) {
5662e334
KW
3384 found_inverse = TRUE;
3385 }
3386
3387 /* No need to continue searching if found everything we are
3388 * looking for */
3389 if (found_key && found_inverse) {
064c021d
KW
3390 break;
3391 }
3392 }
56ca34ca
KW
3393
3394 /* Make sure there is a mapping to itself on the list */
064c021d 3395 if (! found_key) {
d397ff6a 3396 av_push(list, newSVuv(val));
4637d003 3397 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
064c021d
KW
3398 }
3399
3400
3401 /* Simply add the value to the list */
5662e334
KW
3402 if (! found_inverse) {
3403 av_push(list, newSVuv(inverse));
4637d003 3404 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
5662e334 3405 }
064c021d 3406
b0e3252e 3407 /* swatch_get() increments the value of val for each element in the
064c021d
KW
3408 * range. That makes more compact tables possible. You can
3409 * express the capitalization, for example, of all consecutive
3410 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3411 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
bd3f2f94 3412 * and it's not documented; it appears to be used only in
b0e3252e 3413 * implementing tr//; I copied the semantics from swatch_get(), just
bd3f2f94 3414 * in case */
064c021d
KW
3415 if (!none || val < none) {
3416 ++val;
3417 }
3418 }
3419 }
3420
3421 return ret;
3422}
3423
a25abddc 3424SV*
d764b54e
KW
3425Perl__swash_to_invlist(pTHX_ SV* const swash)
3426{
3427
ed92f1b3
KW
3428 /* Subject to change or removal. For use only in one place in regcomp.c.
3429 * Ownership is given to one reference count in the returned SV* */
d764b54e
KW
3430
3431 U8 *l, *lend;
3432 char *loc;
3433 STRLEN lcur;
3434 HV *const hv = MUTABLE_HV(SvRV(swash));
3435 UV elements = 0; /* Number of elements in the inversion list */
b443038a 3436 U8 empty[] = "";
923b6d4e
KW
3437 SV** listsvp;
3438 SV** typesvp;
3439 SV** bitssvp;
3440 SV** extssvp;
3441 SV** invert_it_svp;
d764b54e 3442
923b6d4e
KW
3443 U8* typestr;
3444 STRLEN bits;
3445 STRLEN octets; /* if bits == 1, then octets == 0 */
d73c39c5
KW
3446 U8 *x, *xend;
3447 STRLEN xcur;
d764b54e 3448
a25abddc 3449 SV* invlist;
d764b54e 3450
b81740c0
KW
3451 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3452
923b6d4e
KW
3453 /* If not a hash, it must be the swash's inversion list instead */
3454 if (SvTYPE(hv) != SVt_PVHV) {
ed92f1b3 3455 return SvREFCNT_inc_simple_NN((SV*) hv);
923b6d4e
KW
3456 }
3457
3458 /* The string containing the main body of the table */
3459 listsvp = hv_fetchs(hv, "LIST", FALSE);
3460 typesvp = hv_fetchs(hv, "TYPE", FALSE);
3461 bitssvp = hv_fetchs(hv, "BITS", FALSE);
3462 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3463 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3464
3465 typestr = (U8*)SvPV_nolen(*typesvp);
3466 bits = SvUV(*bitssvp);
3467 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3468
d764b54e 3469 /* read $swash->{LIST} */
b443038a
KW
3470 if (SvPOK(*listsvp)) {
3471 l = (U8*)SvPV(*listsvp, lcur);
3472 }
3473 else {
3474 /* LIST legitimately doesn't contain a string during compilation phases
3475 * of Perl itself, before the Unicode tables are generated. In this
3476 * case, just fake things up by creating an empty list */
3477 l = empty;
3478 lcur = 0;
3479 }
d764b54e
KW
3480 loc = (char *) l;
3481 lend = l + lcur;
3482
31aa6e0b 3483 if (*l == 'V') { /* Inversion list format */
99944878 3484 const char *after_atou = (char *) lend;
31aa6e0b
KW
3485 UV element0;
3486 UV* other_elements_ptr;
3487
3488 /* The first number is a count of the rest */
3489 l++;
99944878 3490 elements = grok_atou((const char *)l, &after_atou);
eb092534
KW
3491 if (elements == 0) {
3492 invlist = _new_invlist(0);
3493 }
3494 else {
99944878
JH
3495 while (isSPACE(*l)) l++;
3496 l = (U8 *) after_atou;
1f9f7d4c
KW
3497
3498 /* Get the 0th element, which is needed to setup the inversion list */
99944878
JH
3499 while (isSPACE(*l)) l++;
3500 element0 = (UV) grok_atou((const char *)l, &after_atou);
3501 l = (U8 *) after_atou;
1f9f7d4c
KW
3502 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
3503 elements--;
3504
3505 /* Then just populate the rest of the input */
3506 while (elements-- > 0) {
3507 if (l > lend) {
3508 Perl_croak(aTHX_ "panic: Expecting %"UVuf" more elements than available", elements);
3509 }
99944878
JH
3510 while (isSPACE(*l)) l++;
3511 *other_elements_ptr++ = (UV) grok_atou((const char *)l, &after_atou);
3512 l = (U8 *) after_atou;
1f9f7d4c 3513 }
eb092534 3514 }
31aa6e0b
KW
3515 }
3516 else {
3517
1784d2f9
KW
3518 /* Scan the input to count the number of lines to preallocate array
3519 * size based on worst possible case, which is each line in the input
3520 * creates 2 elements in the inversion list: 1) the beginning of a
3521 * range in the list; 2) the beginning of a range not in the list. */
3522 while ((loc = (strchr(loc, '\n'))) != NULL) {
3523 elements += 2;
3524 loc++;
3525 }
d764b54e 3526
1784d2f9
KW
3527 /* If the ending is somehow corrupt and isn't a new line, add another
3528 * element for the final range that isn't in the inversion list */
3529 if (! (*lend == '\n'
3530 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
3531 {
3532 elements++;
3533 }
d764b54e 3534
1784d2f9 3535 invlist = _new_invlist(elements);
d764b54e 3536
1784d2f9
KW
3537 /* Now go through the input again, adding each range to the list */
3538 while (l < lend) {
3539 UV start, end;
3540 UV val; /* Not used by this function */
d764b54e 3541
95543e92
KW
3542 l = swash_scan_list_line(l, lend, &start, &end, &val,
3543 cBOOL(octets), typestr);
d764b54e 3544
1784d2f9
KW
3545 if (l > lend) {
3546 break;
3547 }
3548
3549 invlist = _add_range_to_invlist(invlist, start, end);
3550 }
31aa6e0b 3551 }
d764b54e 3552
77f9f126
KW
3553 /* Invert if the data says it should be */
3554 if (invert_it_svp && SvUV(*invert_it_svp)) {
25151030 3555 _invlist_invert(invlist);
77f9f126
KW
3556 }
3557
b0e3252e 3558 /* This code is copied from swatch_get()
d73c39c5
KW
3559 * read $swash->{EXTRAS} */
3560 x = (U8*)SvPV(*extssvp, xcur);
3561 xend = x + xcur;
3562 while (x < xend) {
3563 STRLEN namelen;
3564 U8 *namestr;
3565 SV** othersvp;
3566 HV* otherhv;
3567 STRLEN otherbits;
3568 SV **otherbitssvp, *other;
3569 U8 *nl;
3570
3571 const U8 opc = *x++;
3572 if (opc == '\n')
3573 continue;
3574
3575 nl = (U8*)memchr(x, '\n', xend - x);
3576
3577 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3578 if (nl) {
3579 x = nl + 1; /* 1 is length of "\n" */
3580 continue;
3581 }
3582 else {
3583 x = xend; /* to EXTRAS' end at which \n is not found */
3584 break;
3585 }
3586 }
3587
3588 namestr = x;
3589 if (nl) {
3590 namelen = nl - namestr;
3591 x = nl + 1;
3592 }
3593 else {
3594 namelen = xend - namestr;
3595 x = xend;
3596 }
3597
3598 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3599 otherhv = MUTABLE_HV(SvRV(*othersvp));
3600 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3601 otherbits = (STRLEN)SvUV(*otherbitssvp);
3602
3603 if (bits != otherbits || bits != 1) {
5637ef5b
NC
3604 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
3605 "properties, bits=%"UVuf", otherbits=%"UVuf,
3606 (UV)bits, (UV)otherbits);
d73c39c5
KW
3607 }
3608
3609 /* The "other" swatch must be destroyed after. */
3610 other = _swash_to_invlist((SV *)*othersvp);
3611
b0e3252e 3612 /* End of code copied from swatch_get() */
d73c39c5
KW
3613 switch (opc) {
3614 case '+':
3615 _invlist_union(invlist, other, &invlist);
3616 break;
3617 case '!':
6c46377d 3618 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
d73c39c5
KW
3619 break;
3620 case '-':
3621 _invlist_subtract(invlist, other, &invlist);
3622 break;
3623 case '&':
3624 _invlist_intersection(invlist, other, &invlist);
3625 break;
3626 default:
3627 break;
3628 }
3629 sv_free(other); /* through with it! */
3630 }
3631
dbfdbd26 3632 SvREADONLY_on(invlist);
d764b54e
KW
3633 return invlist;
3634}
3635
3fdfee00
KW
3636SV*
3637Perl__get_swash_invlist(pTHX_ SV* const swash)
3638{
872dd7e0 3639 SV** ptr;
3fdfee00
KW
3640
3641 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
3642
87367d5f 3643 if (! SvROK(swash)) {
872dd7e0
KW
3644 return NULL;
3645 }
3646
87367d5f
KW
3647 /* If it really isn't a hash, it isn't really swash; must be an inversion
3648 * list */
3649 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
3650 return SvRV(swash);
3651 }
872dd7e0 3652
87367d5f 3653 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
3fdfee00
KW
3654 if (! ptr) {
3655 return NULL;
3656 }
3657
3658 return *ptr;
3659}
3660
0876b9a0 3661bool
5aaab254 3662Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
0876b9a0
KW
3663{
3664 /* May change: warns if surrogates, non-character code points, or
af2af982
KW
3665 * non-Unicode code points are in s which has length len bytes. Returns
3666 * TRUE if none found; FALSE otherwise. The only other validity check is
3667 * to make sure that this won't exceed the string's length */
0876b9a0
KW
3668
3669 const U8* const e = s + len;
3670 bool ok = TRUE;
3671
3672 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
3673
3674 while (s < e) {
3675 if (UTF8SKIP(s) > len) {
3676 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
3677 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
3678 return FALSE;
3679 }
732fbc05 3680 if (UNLIKELY(*s >= UTF8_FIRST_PROBLEMATIC_CODE_POINT_FIRST_BYTE)) {
0876b9a0
KW
3681 STRLEN char_len;
3682 if (UTF8_IS_SUPER(s)) {
8457b38f 3683 if (ckWARN_d(WARN_NON_UNICODE)) {
4b88fb76 3684 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
8457b38f
KW
3685 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
3686 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
3687 ok = FALSE;
3688 }
0876b9a0
KW
3689 }
3690 else if (UTF8_IS_SURROGATE(s)) {
8457b38f 3691 if (ckWARN_d(WARN_SURROGATE)) {
4b88fb76 3692 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
8457b38f
KW
3693 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
3694 "Unicode surrogate U+%04"UVXf" is illegal in UTF-8", uv);
3695 ok = FALSE;
3696 }