This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
toke.c: Remove soon-to-be invalid t assumption
[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"
b992490d 34#include "invlist_inline.h"
a0ed51b3 35
27da23d5
JH
36static const char unees[] =
37 "Malformed UTF-8 character (unexpected end of string)";
901b21bf 38
48ef279e 39/*
ccfc67b7 40=head1 Unicode Support
7fefc6c1 41These are various utility functions for manipulating UTF8-encoded
72d33970 42strings. For the uninitiated, this is a method of representing arbitrary
61296642 43Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
44characters in the ASCII range are unmodified, and a zero byte never appears
45within non-zero characters.
166f8a29 46
eaf7a4d2
CS
47=cut
48*/
49
50/*
7bbfa158 51=for apidoc is_invariant_string
eaf7a4d2 52
b8ed1ca9
KW
53Returns true iff the first C<len> bytes of the string C<s> are the same
54regardless of the UTF-8 encoding of the string (or UTF-EBCDIC encoding on
55EBCDIC machines). That is, if they are UTF-8 invariant. On ASCII-ish
56machines, all the ASCII characters and only the ASCII characters fit this
7bbfa158 57definition. On EBCDIC machines, the ASCII-range characters are invariant, but
b8ed1ca9
KW
58so also are the C1 controls and C<\c?> (which isn't in the ASCII range on
59EBCDIC).
eaf7a4d2 60
75200dff
KW
61If C<len> is 0, it will be calculated using C<strlen(s)>, (which means if you
62use this option, that C<s> can't have embedded C<NUL> characters and has to
63have a terminating C<NUL> byte).
9f7e3d64 64
a1433954 65See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
eaf7a4d2
CS
66
67=cut
68*/
69
70bool
7bbfa158 71Perl_is_invariant_string(const U8 *s, STRLEN len)
eaf7a4d2
CS
72{
73 const U8* const send = s + (len ? len : strlen((const char *)s));
74 const U8* x = s;
75
7bbfa158 76 PERL_ARGS_ASSERT_IS_INVARIANT_STRING;
eaf7a4d2
CS
77
78 for (; x < send; ++x) {
79 if (!UTF8_IS_INVARIANT(*x))
80 break;
81 }
82
83 return x == send;
84}
85
86/*
378516de 87=for apidoc uvoffuni_to_utf8_flags
eebe1485 88
a27992cc 89THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
90Instead, B<Almost all code should use L</uvchr_to_utf8> or
91L</uvchr_to_utf8_flags>>.
a27992cc 92
de69f3af
KW
93This function is like them, but the input is a strict Unicode
94(as opposed to native) code point. Only in very rare circumstances should code
95not be using the native code point.
949cf498 96
efa9cd84 97For details, see the description for L</uvchr_to_utf8_flags>.
949cf498 98
eebe1485
SC
99=cut
100*/
101
dfe13c55 102U8 *
378516de 103Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 104{
378516de 105 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 106
2d1545e5 107 if (OFFUNI_IS_INVARIANT(uv)) {
d9432125
KW
108 *d++ = (U8) LATIN1_TO_NATIVE(uv);
109 return d;
110 }
111
9415f659
KW
112#ifdef EBCDIC
113 /* Not representable in UTF-EBCDIC */
114 flags |= UNICODE_DISALLOW_FE_FF;
115#endif
116
979f77b6 117 /* The first problematic code point is the first surrogate */
57162cb6 118 if ( flags /* It's common to turn off all these */
80670d37 119 && uv >= UNICODE_SURROGATE_FIRST)
979f77b6 120 {
949cf498
KW
121 if (UNICODE_IS_SURROGATE(uv)) {
122 if (flags & UNICODE_WARN_SURROGATE) {
8457b38f 123 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
949cf498
KW
124 "UTF-16 surrogate U+%04"UVXf, uv);
125 }
126 if (flags & UNICODE_DISALLOW_SURROGATE) {
127 return NULL;
128 }
129 }
130 else if (UNICODE_IS_SUPER(uv)) {
0bcdd8f6 131 if ( (flags & UNICODE_WARN_SUPER)
949cf498 132 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
0bcdd8f6
KW
133 {
134 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
135
136 /* Choose the more dire applicable warning */
137 (UNICODE_IS_FE_FF(uv))
138 ? "Code point 0x%"UVXf" is not Unicode, and not portable"
139 : "Code point 0x%"UVXf" is not Unicode, may not be portable",
140 uv);
949cf498
KW
141 }
142 if (flags & UNICODE_DISALLOW_SUPER
143 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
144 {
9415f659
KW
145#ifdef EBCDIC
146 Perl_die(aTHX_ "Can't represent character for Ox%"UVXf" on this platform", uv);
661d43c4 147 NOT_REACHED; /* NOTREACHED */
9415f659 148#endif
949cf498
KW
149 return NULL;
150 }
151 }
152 else if (UNICODE_IS_NONCHAR(uv)) {
153 if (flags & UNICODE_WARN_NONCHAR) {
8457b38f 154 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
ba707cdc 155 "Unicode non-character U+%04"UVXf" is not recommended for open interchange",
949cf498
KW
156 uv);
157 }
158 if (flags & UNICODE_DISALLOW_NONCHAR) {
159 return NULL;
160 }
161 }
507b9800 162 }
d9432125 163
2d331972 164#if defined(EBCDIC)
d9432125 165 {
5aaebcb3 166 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
167 U8 *p = d+len-1;
168 while (p > d) {
bc3632a8 169 *p-- = (U8) I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
170 uv >>= UTF_ACCUMULATION_SHIFT;
171 }
bc3632a8 172 *p = (U8) I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
173 return d+len;
174 }
175#else /* Non loop style */
a0ed51b3 176 if (uv < 0x800) {
eb160463
GS
177 *d++ = (U8)(( uv >> 6) | 0xc0);
178 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
179 return d;
180 }
181 if (uv < 0x10000) {
eb160463
GS
182 *d++ = (U8)(( uv >> 12) | 0xe0);
183 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
184 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
185 return d;
186 }
187 if (uv < 0x200000) {
eb160463
GS
188 *d++ = (U8)(( uv >> 18) | 0xf0);
189 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
190 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
191 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
192 return d;
193 }
194 if (uv < 0x4000000) {
eb160463
GS
195 *d++ = (U8)(( uv >> 24) | 0xf8);
196 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
197 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
198 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
199 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
200 return d;
201 }
202 if (uv < 0x80000000) {
eb160463
GS
203 *d++ = (U8)(( uv >> 30) | 0xfc);
204 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
205 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
206 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
207 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
208 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
209 return d;
210 }
6588300d 211#ifdef UTF8_QUAD_MAX
d7578b48 212 if (uv < UTF8_QUAD_MAX)
a0ed51b3
LW
213#endif
214 {
eb160463
GS
215 *d++ = 0xfe; /* Can't match U+FEFF! */
216 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
217 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
218 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
219 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
220 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
221 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
222 return d;
223 }
6588300d 224#ifdef UTF8_QUAD_MAX
a0ed51b3 225 {
eb160463
GS
226 *d++ = 0xff; /* Can't match U+FFFE! */
227 *d++ = 0x80; /* 6 Reserved bits */
228 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
229 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
230 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
231 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
232 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
233 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
234 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
235 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
236 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
237 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
238 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
239 return d;
240 }
241#endif
537124e4 242#endif /* Non loop style */
a0ed51b3 243}
646ca15d 244/*
07693fe6
KW
245=for apidoc uvchr_to_utf8
246
bcb1a2d4 247Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 248of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
249C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
250the byte after the end of the new character. In other words,
07693fe6
KW
251
252 d = uvchr_to_utf8(d, uv);
253
254is the recommended wide native character-aware way of saying
255
256 *(d++) = uv;
257
de69f3af
KW
258This function accepts any UV as input. To forbid or warn on non-Unicode code
259points, or those that may be problematic, see L</uvchr_to_utf8_flags>.
260
07693fe6
KW
261=cut
262*/
263
de69f3af
KW
264/* This is also a macro */
265PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
266
07693fe6
KW
267U8 *
268Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
269{
de69f3af 270 return uvchr_to_utf8(d, uv);
07693fe6
KW
271}
272
de69f3af
KW
273/*
274=for apidoc uvchr_to_utf8_flags
275
276Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 277of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
278C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
279the byte after the end of the new character. In other words,
de69f3af
KW
280
281 d = uvchr_to_utf8_flags(d, uv, flags);
282
283or, in most cases,
284
285 d = uvchr_to_utf8_flags(d, uv, 0);
286
287This is the Unicode-aware way of saying
288
289 *(d++) = uv;
290
291This function will convert to UTF-8 (and not warn) even code points that aren't
292legal Unicode or are problematic, unless C<flags> contains one or more of the
293following flags:
294
796b6530 295If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
de69f3af 296the function will raise a warning, provided UTF8 warnings are enabled. If instead
796b6530 297C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return NULL.
de69f3af
KW
298If both flags are set, the function will both warn and return NULL.
299
796b6530 300The C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
de69f3af 301affect how the function handles a Unicode non-character. And likewise, the
796b6530 302C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags affect the handling of
de69f3af
KW
303code points that are
304above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
305even less portable) can be warned and/or disallowed even if other above-Unicode
796b6530
KW
306code points are accepted, by the C<UNICODE_WARN_FE_FF> and
307C<UNICODE_DISALLOW_FE_FF> flags.
de69f3af 308
796b6530
KW
309And finally, the flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all four of
310the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
311four DISALLOW flags.
de69f3af
KW
312
313=cut
314*/
315
316/* This is also a macro */
317PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
318
07693fe6
KW
319U8 *
320Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
321{
de69f3af 322 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
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
7bbfa158 334See also L</is_invariant_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 437
796b6530 438The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
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>
796b6530 444character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
d088425d 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
796b6530
KW
453C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all three classes are treated as
454malformations and handled as such. The flags C<UTF8_DISALLOW_SURROGATE>,
455C<UTF8_DISALLOW_NONCHAR>, and C<UTF8_DISALLOW_SUPER> (meaning above the legal
456Unicode maximum) can be set to disallow these categories individually.
457
458The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
459C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
460raised for their respective categories, but otherwise the code points are
461considered valid (not malformations). To get a category to both be treated as
462a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 463(But note that warnings are not raised if lexically disabled nor if
796b6530 464C<UTF8_CHECK_ONLY> is also specified.)
949cf498
KW
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
796b6530 475byte containing 0xFE or 0xFF. The C<UTF8_DISALLOW_FE_FF> flag will cause them to
eb83ed87 476be treated as malformations, while allowing smaller above-Unicode code points.
796b6530 477(Of course C<UTF8_DISALLOW_SUPER> will treat all above-Unicode code points,
72d33970 478including these, as malformations.)
796b6530 479Similarly, C<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 {
0bcdd8f6
KW
718 sv = sv_2mortal(Perl_newSVpvf(aTHX_
719 "Code point 0x%04"UVXf" is not Unicode, may not be portable",
720 uv));
54f4afef 721 pack_warn = packWARN(WARN_NON_UNICODE);
949cf498 722 }
e543b0e5
KW
723#ifndef EBCDIC /* Can never have the equivalent of FE nor FF on EBCDIC, since
724 not representable in UTF-EBCDIC */
ea5ced44
KW
725
726 /* The first byte being 0xFE or 0xFF is a subset of the SUPER code
727 * points. We test for these after the regular SUPER ones, and
728 * before possibly bailing out, so that the more dire warning
729 * overrides the regular one, if applicable */
730 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
0bcdd8f6 731 && (flags & (UTF8_WARN_FE_FF|UTF8_WARN_SUPER|UTF8_DISALLOW_FE_FF)))
ea5ced44 732 {
0bcdd8f6
KW
733 if ( ! (flags & UTF8_CHECK_ONLY)
734 && (flags & (UTF8_WARN_FE_FF|UTF8_WARN_SUPER))
735 && ckWARN_d(WARN_UTF8))
ea5ced44 736 {
0bcdd8f6
KW
737 sv = sv_2mortal(Perl_newSVpvf(aTHX_
738 "Code point 0x%"UVXf" is not Unicode, and not portable",
739 uv));
ea5ced44
KW
740 pack_warn = packWARN(WARN_UTF8);
741 }
742 if (flags & UTF8_DISALLOW_FE_FF) {
743 goto disallowed;
744 }
745 }
746#endif
949cf498
KW
747 if (flags & UTF8_DISALLOW_SUPER) {
748 goto disallowed;
749 }
750 }
4190d317
KW
751 else if (UNICODE_IS_NONCHAR(uv)) {
752 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
54f4afef 753 && ckWARN_d(WARN_NONCHAR))
4190d317 754 {
ba707cdc 755 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is not recommended for open interchange", uv));
54f4afef 756 pack_warn = packWARN(WARN_NONCHAR);
4190d317
KW
757 }
758 if (flags & UTF8_DISALLOW_NONCHAR) {
759 goto disallowed;
760 }
761 }
949cf498 762
eb83ed87 763 if (sv) {
de69f3af
KW
764 outlier_ret = uv; /* Note we don't bother to convert to native,
765 as all the outlier code points are the same
766 in both ASCII and EBCDIC */
eb83ed87
KW
767 goto do_warn;
768 }
769
949cf498
KW
770 /* Here, this is not considered a malformed character, so drop through
771 * to return it */
a0ed51b3 772 }
ba210ebe 773
de69f3af 774 return UNI_TO_NATIVE(uv);
ba210ebe 775
eb83ed87
KW
776 /* There are three cases which get to beyond this point. In all 3 cases:
777 * <sv> if not null points to a string to print as a warning.
778 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
779 * set.
780 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
781 * This is done by initializing it to 0, and changing it only
782 * for case 1).
783 * The 3 cases are:
784 * 1) The input is valid but problematic, and to be warned about. The
785 * return value is the resultant code point; <*retlen> is set to
786 * <curlen>, the number of bytes that comprise the code point.
787 * <pack_warn> contains the result of packWARN() for the warning
788 * types. The entry point for this case is the label <do_warn>;
789 * 2) The input is a valid code point but disallowed by the parameters to
790 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
791 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
792 * comprise the code point. <pack_warn> contains the result of
793 * packWARN() for the warning types. The entry point for this case is
794 * the label <disallowed>.
795 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
796 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
797 * bytes that comprise the malformation. All such malformations are
798 * assumed to be warning type <utf8>. The entry point for this case
799 * is the label <malformed>.
800 */
949cf498 801
7b52d656 802 malformed:
ba210ebe 803
eb83ed87
KW
804 if (sv && ckWARN_d(WARN_UTF8)) {
805 pack_warn = packWARN(WARN_UTF8);
806 }
807
7b52d656 808 disallowed:
eb83ed87 809
fcc8fcf6 810 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 811 if (retlen)
10edeb5d 812 *retlen = ((STRLEN) -1);
ba210ebe
JH
813 return 0;
814 }
815
7b52d656 816 do_warn:
5b311467 817
eb83ed87
KW
818 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
819 if warnings are to be raised. */
f555bc63 820 const char * const string = SvPVX_const(sv);
a0dbb045 821
f555bc63
KW
822 if (PL_op)
823 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
824 else
825 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
826 }
827
eb83ed87
KW
828 if (retlen) {
829 *retlen = curlen;
830 }
ba210ebe 831
eb83ed87 832 return outlier_ret;
a0ed51b3
LW
833}
834
8e84507e 835/*
ec5f19d0
KW
836=for apidoc utf8_to_uvchr_buf
837
838Returns the native code point of the first character in the string C<s> which
839is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 840C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 841
524080c4
KW
842If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
843enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 844C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 845(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 846C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 847the next possible position in C<s> that could begin a non-malformed character.
de69f3af 848See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 849returned.
ec5f19d0
KW
850
851=cut
852*/
853
854
855UV
856Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
857{
ec5f19d0
KW
858 assert(s < send);
859
860 return utf8n_to_uvchr(s, send - s, retlen,
861 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
862}
863
27d6c58a 864/* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
3986bb7c 865 * there are no malformations in the input UTF-8 string C<s>. surrogates,
57b0056d 866 * non-character code points, and non-Unicode code points are allowed. */
27d6c58a
KW
867
868UV
869Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
870{
010ab96b
KW
871 UV expectlen = UTF8SKIP(s);
872 const U8* send = s + expectlen;
9ff2f0f7 873 UV uv = *s;
3986bb7c 874
27d6c58a 875 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
81611534 876 PERL_UNUSED_CONTEXT;
27d6c58a 877
010ab96b
KW
878 if (retlen) {
879 *retlen = expectlen;
880 }
881
882 /* An invariant is trivially returned */
883 if (expectlen == 1) {
9ff2f0f7 884 return uv;
010ab96b
KW
885 }
886
9ff2f0f7
KW
887#ifdef EBCDIC
888 uv = NATIVE_UTF8_TO_I8(uv);
889#endif
890
010ab96b
KW
891 /* Remove the leading bits that indicate the number of bytes, leaving just
892 * the bits that are part of the value */
893 uv &= UTF_START_MASK(expectlen);
894
895 /* Now, loop through the remaining bytes, accumulating each into the
896 * working total as we go. (I khw tried unrolling the loop for up to 4
897 * bytes, but there was no performance improvement) */
898 for (++s; s < send; s++) {
899 uv = UTF8_ACCUMULATE(uv, *s);
900 }
901
3986bb7c 902 return UNI_TO_NATIVE(uv);
010ab96b 903
27d6c58a
KW
904}
905
ec5f19d0 906/*
ec5f19d0
KW
907=for apidoc utf8_to_uvuni_buf
908
de69f3af
KW
909Only in very rare circumstances should code need to be dealing in Unicode
910(as opposed to native) code points. In those few cases, use
911C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
912
913Returns the Unicode (not-native) code point of the first character in the
914string C<s> which
ec5f19d0
KW
915is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
916C<retlen> will be set to the length, in bytes, of that character.
917
524080c4
KW
918If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
919enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
920NULL) to -1. If those warnings are off, the computed value if well-defined (or
921the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
922is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
923next possible position in C<s> that could begin a non-malformed character.
de69f3af 924See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
925
926=cut
927*/
928
929UV
930Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
931{
932 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
933
934 assert(send > s);
935
936 /* Call the low level routine asking for checks */
de69f3af
KW
937 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
938 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
ec5f19d0
KW
939}
940
b76347f2 941/*
87cea99e 942=for apidoc utf8_length
b76347f2
JH
943
944Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
945Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
946up past C<e>, croaks.
b76347f2
JH
947
948=cut
949*/
950
951STRLEN
35a4481c 952Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
953{
954 STRLEN len = 0;
955
7918f24d
NC
956 PERL_ARGS_ASSERT_UTF8_LENGTH;
957
8850bf83
JH
958 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
959 * the bitops (especially ~) can create illegal UTF-8.
960 * In other words: in Perl UTF-8 is not just for Unicode. */
961
a3b680e6
AL
962 if (e < s)
963 goto warn_and_return;
b76347f2 964 while (s < e) {
4cbf4130 965 s += UTF8SKIP(s);
8e91ec7f
AV
966 len++;
967 }
968
969 if (e != s) {
970 len--;
971 warn_and_return:
9b387841
NC
972 if (PL_op)
973 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
974 "%s in %s", unees, OP_DESC(PL_op));
975 else
61a12c31 976 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
977 }
978
979 return len;
980}
981
b06226ff 982/*
87cea99e 983=for apidoc utf8_distance
b06226ff 984
1e54db1a 985Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
b06226ff
JH
986and C<b>.
987
988WARNING: use only if you *know* that the pointers point inside the
989same UTF-8 buffer.
990
37607a96
PK
991=cut
992*/
a0ed51b3 993
02eb7b47 994IV
35a4481c 995Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
a0ed51b3 996{
7918f24d
NC
997 PERL_ARGS_ASSERT_UTF8_DISTANCE;
998
bf1665bc 999 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
a0ed51b3
LW
1000}
1001
b06226ff 1002/*
87cea99e 1003=for apidoc utf8_hop
b06226ff 1004
8850bf83
JH
1005Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1006forward or backward.
b06226ff
JH
1007
1008WARNING: do not use the following unless you *know* C<off> is within
8850bf83
JH
1009the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1010on the first byte of character or just after the last byte of a character.
b06226ff 1011
37607a96
PK
1012=cut
1013*/
a0ed51b3
LW
1014
1015U8 *
ddeaf645 1016Perl_utf8_hop(const U8 *s, I32 off)
a0ed51b3 1017{
7918f24d
NC
1018 PERL_ARGS_ASSERT_UTF8_HOP;
1019
8850bf83
JH
1020 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1021 * the bitops (especially ~) can create illegal UTF-8.
1022 * In other words: in Perl UTF-8 is not just for Unicode. */
1023
a0ed51b3
LW
1024 if (off >= 0) {
1025 while (off--)
1026 s += UTF8SKIP(s);
1027 }
1028 else {
1029 while (off++) {
1030 s--;
8850bf83
JH
1031 while (UTF8_IS_CONTINUATION(*s))
1032 s--;
a0ed51b3
LW
1033 }
1034 }
4373e329 1035 return (U8 *)s;
a0ed51b3
LW
1036}
1037
6940069f 1038/*
fed3ba5d
NC
1039=for apidoc bytes_cmp_utf8
1040
a1433954 1041Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1042sequence of characters (stored as UTF-8)
1043in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1044equal, -1 or -2 if the first string is less than the second string, +1 or +2
1045if the first string is greater than the second string.
1046
1047-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1048longer string. -2 or +2 is returned if
1049there was a difference between characters
fed3ba5d
NC
1050within the strings.
1051
1052=cut
1053*/
1054
1055int
1056Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1057{
1058 const U8 *const bend = b + blen;
1059 const U8 *const uend = u + ulen;
1060
1061 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1062
1063 while (b < bend && u < uend) {
1064 U8 c = *u++;
1065 if (!UTF8_IS_INVARIANT(c)) {
1066 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1067 if (u < uend) {
1068 U8 c1 = *u++;
1069 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 1070 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d
NC
1071 } else {
1072 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1073 "Malformed UTF-8 character "
1074 "(unexpected non-continuation byte 0x%02x"
1075 ", immediately after start byte 0x%02x)"
1076 /* Dear diag.t, it's in the pod. */
1077 "%s%s", c1, c,
1078 PL_op ? " in " : "",
1079 PL_op ? OP_DESC(PL_op) : "");
1080 return -2;
1081 }
1082 } else {
1083 if (PL_op)
1084 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1085 "%s in %s", unees, OP_DESC(PL_op));
1086 else
61a12c31 1087 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1088 return -2; /* Really want to return undef :-) */
1089 }
1090 } else {
1091 return -2;
1092 }
1093 }
1094 if (*b != c) {
1095 return *b < c ? -2 : +2;
1096 }
1097 ++b;
1098 }
1099
1100 if (b == bend && u == uend)
1101 return 0;
1102
1103 return b < bend ? +1 : -1;
1104}
1105
1106/*
87cea99e 1107=for apidoc utf8_to_bytes
6940069f 1108
2bbc8d55 1109Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1110Unlike L</bytes_to_utf8>, this over-writes the original string, and
1111updates C<len> to contain the new length.
67e989fb 1112Returns zero on failure, setting C<len> to -1.
6940069f 1113
a1433954 1114If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1115
6940069f
GS
1116=cut
1117*/
1118
1119U8 *
37607a96 1120Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1121{
d4c19fe8
AL
1122 U8 * const save = s;
1123 U8 * const send = s + *len;
6940069f 1124 U8 *d;
246fae53 1125
7918f24d 1126 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 1127 PERL_UNUSED_CONTEXT;
7918f24d 1128
1e54db1a 1129 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1130 while (s < send) {
d59937ca
KW
1131 if (! UTF8_IS_INVARIANT(*s)) {
1132 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1133 *len = ((STRLEN) -1);
1134 return 0;
1135 }
1136 s++;
dcad2880 1137 }
d59937ca 1138 s++;
246fae53 1139 }
dcad2880
JH
1140
1141 d = s = save;
6940069f 1142 while (s < send) {
80e0b38f
KW
1143 U8 c = *s++;
1144 if (! UTF8_IS_INVARIANT(c)) {
1145 /* Then it is two-byte encoded */
a62b247b 1146 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
80e0b38f
KW
1147 s++;
1148 }
1149 *d++ = c;
6940069f
GS
1150 }
1151 *d = '\0';
246fae53 1152 *len = d - save;
6940069f
GS
1153 return save;
1154}
1155
1156/*
87cea99e 1157=for apidoc bytes_from_utf8
f9a63242 1158
2bbc8d55 1159Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1160Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1161the newly-created string, and updates C<len> to contain the new
1162length. Returns the original string if no conversion occurs, C<len>
72d33970 1163is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55 11640 if C<s> is converted or consisted entirely of characters that are invariant
4a4088c4 1165in UTF-8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1166
37607a96
PK
1167=cut
1168*/
f9a63242
JH
1169
1170U8 *
e1ec3a88 1171Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1172{
f9a63242 1173 U8 *d;
e1ec3a88
AL
1174 const U8 *start = s;
1175 const U8 *send;
f9a63242
JH
1176 I32 count = 0;
1177
7918f24d 1178 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 1179 PERL_UNUSED_CONTEXT;
f9a63242 1180 if (!*is_utf8)
73d840c0 1181 return (U8 *)start;
f9a63242 1182
1e54db1a 1183 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1184 for (send = s + *len; s < send;) {
d59937ca
KW
1185 if (! UTF8_IS_INVARIANT(*s)) {
1186 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1187 return (U8 *)start;
d59937ca
KW
1188 }
1189 count++;
1190 s++;
db42d148 1191 }
d59937ca 1192 s++;
f9a63242
JH
1193 }
1194
35da51f7 1195 *is_utf8 = FALSE;
f9a63242 1196
212542aa 1197 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1198 s = start; start = d;
f9a63242
JH
1199 while (s < send) {
1200 U8 c = *s++;
1a91c45d 1201 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1202 /* Then it is two-byte encoded */
a62b247b 1203 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1a91c45d 1204 s++;
c4d5f83a
NIS
1205 }
1206 *d++ = c;
f9a63242
JH
1207 }
1208 *d = '\0';
1209 *len = d - start;
73d840c0 1210 return (U8 *)start;
f9a63242
JH
1211}
1212
1213/*
87cea99e 1214=for apidoc bytes_to_utf8
6940069f 1215
ff97e5cf
KW
1216Converts a string C<s> of length C<len> bytes from the native encoding into
1217UTF-8.
6662521e 1218Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1219reflect the new length in bytes.
6940069f 1220
75200dff 1221A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
1222
1223If you want to convert to UTF-8 from encodings other than
1224the native (Latin1 or EBCDIC),
a1433954 1225see L</sv_recode_to_utf8>().
c9ada85f 1226
497711e7 1227=cut
6940069f
GS
1228*/
1229
c682ebef
FC
1230/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1231 likewise need duplication. */
1232
6940069f 1233U8*
35a4481c 1234Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1235{
35a4481c 1236 const U8 * const send = s + (*len);
6940069f
GS
1237 U8 *d;
1238 U8 *dst;
7918f24d
NC
1239
1240 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1241 PERL_UNUSED_CONTEXT;
6940069f 1242
212542aa 1243 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1244 dst = d;
1245
1246 while (s < send) {
55d09dc8
KW
1247 append_utf8_from_native_byte(*s, &d);
1248 s++;
6940069f
GS
1249 }
1250 *d = '\0';
6662521e 1251 *len = d-dst;
6940069f
GS
1252 return dst;
1253}
1254
a0ed51b3 1255/*
dea0fc0b 1256 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1257 *
1258 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1259 * We optimize for native, for obvious reasons. */
1260
1261U8*
dea0fc0b 1262Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1263{
dea0fc0b
JH
1264 U8* pend;
1265 U8* dstart = d;
1266
7918f24d
NC
1267 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1268
dea0fc0b 1269 if (bytelen & 1)
f5992bc4 1270 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1271
1272 pend = p + bytelen;
1273
a0ed51b3 1274 while (p < pend) {
dea0fc0b
JH
1275 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1276 p += 2;
2d1545e5 1277 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 1278 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
1279 continue;
1280 }
56d37426
KW
1281 if (uv <= MAX_UTF8_TWO_BYTE) {
1282 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1283 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
1284 continue;
1285 }
46956fad
KW
1286#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1287#define LAST_HIGH_SURROGATE 0xDBFF
1288#define FIRST_LOW_SURROGATE 0xDC00
1289#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
1290
1291 /* This assumes that most uses will be in the first Unicode plane, not
1292 * needing surrogates */
1293 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
1294 && uv <= UNICODE_SURROGATE_LAST))
1295 {
1296 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
1297 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1298 }
1299 else {
01ea242b 1300 UV low = (p[0] << 8) + p[1];
e23c50db
KW
1301 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
1302 || UNLIKELY(low > LAST_LOW_SURROGATE))
1303 {
01ea242b 1304 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
1305 }
1306 p += 2;
46956fad
KW
1307 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1308 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 1309 }
a0ed51b3 1310 }
56d37426
KW
1311#ifdef EBCDIC
1312 d = uvoffuni_to_utf8_flags(d, uv, 0);
1313#else
a0ed51b3 1314 if (uv < 0x10000) {
eb160463
GS
1315 *d++ = (U8)(( uv >> 12) | 0xe0);
1316 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1317 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1318 continue;
1319 }
1320 else {
eb160463
GS
1321 *d++ = (U8)(( uv >> 18) | 0xf0);
1322 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1323 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1324 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1325 continue;
1326 }
56d37426 1327#endif
a0ed51b3 1328 }
dea0fc0b 1329 *newlen = d - dstart;
a0ed51b3
LW
1330 return d;
1331}
1332
1333/* Note: this one is slightly destructive of the source. */
1334
1335U8*
dea0fc0b 1336Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1337{
1338 U8* s = (U8*)p;
d4c19fe8 1339 U8* const send = s + bytelen;
7918f24d
NC
1340
1341 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1342
e0ea5e2d
NC
1343 if (bytelen & 1)
1344 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1345 (UV)bytelen);
1346
a0ed51b3 1347 while (s < send) {
d4c19fe8 1348 const U8 tmp = s[0];
a0ed51b3
LW
1349 s[0] = s[1];
1350 s[1] = tmp;
1351 s += 2;
1352 }
dea0fc0b 1353 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1354}
1355
922e8cb4
KW
1356bool
1357Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1358{
1359 U8 tmpbuf[UTF8_MAXBYTES+1];
1360 uvchr_to_utf8(tmpbuf, c);
1361 return _is_utf8_FOO(classnum, tmpbuf);
1362}
1363
f9ae8fb6
JD
1364/* Internal function so we can deprecate the external one, and call
1365 this one from other deprecated functions in this file */
1366
f2645549
KW
1367bool
1368Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 1369{
f2645549 1370 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
1371
1372 if (*p == '_')
1373 return TRUE;
f25ce844 1374 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
1375}
1376
5092f92a 1377bool
eba68aa0
KW
1378Perl__is_uni_perl_idcont(pTHX_ UV c)
1379{
1380 U8 tmpbuf[UTF8_MAXBYTES+1];
1381 uvchr_to_utf8(tmpbuf, c);
1382 return _is_utf8_perl_idcont(tmpbuf);
1383}
1384
1385bool
f91dcd13
KW
1386Perl__is_uni_perl_idstart(pTHX_ UV c)
1387{
1388 U8 tmpbuf[UTF8_MAXBYTES+1];
1389 uvchr_to_utf8(tmpbuf, c);
1390 return _is_utf8_perl_idstart(tmpbuf);
1391}
1392
3a4c58c9
KW
1393UV
1394Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1395{
1396 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 1397 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
1398 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1399 * either "SS" or "Ss". Which one to use is passed into the routine in
1400 * 'S_or_s' to avoid a test */
1401
1402 UV converted = toUPPER_LATIN1_MOD(c);
1403
1404 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1405
1406 assert(S_or_s == 'S' || S_or_s == 's');
1407
6f2d5cbc 1408 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 1409 characters in this range */
3a4c58c9
KW
1410 *p = (U8) converted;
1411 *lenp = 1;
1412 return converted;
1413 }
1414
1415 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1416 * which it maps to one of them, so as to only have to have one check for
1417 * it in the main case */
1418 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1419 switch (c) {
1420 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1421 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1422 break;
1423 case MICRO_SIGN:
1424 converted = GREEK_CAPITAL_LETTER_MU;
1425 break;
79e064b9
KW
1426#if UNICODE_MAJOR_VERSION > 2 \
1427 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
1428 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
1429 case LATIN_SMALL_LETTER_SHARP_S:
1430 *(p)++ = 'S';
1431 *p = S_or_s;
1432 *lenp = 2;
1433 return 'S';
79e064b9 1434#endif
3a4c58c9
KW
1435 default:
1436 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 1437 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
1438 }
1439 }
1440
1441 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1442 *p = UTF8_TWO_BYTE_LO(converted);
1443 *lenp = 2;
1444
1445 return converted;
1446}
1447
50bda2c3
KW
1448/* Call the function to convert a UTF-8 encoded character to the specified case.
1449 * Note that there may be more than one character in the result.
1450 * INP is a pointer to the first byte of the input character
1451 * OUTP will be set to the first byte of the string of changed characters. It
1452 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1453 * LENP will be set to the length in bytes of the string of changed characters
1454 *
1455 * The functions return the ordinal of the first character in the string of OUTP */
4a8240a3
KW
1456#define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "")
1457#define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "")
1458#define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "")
50bda2c3
KW
1459
1460/* This additionally has the input parameter SPECIALS, which if non-zero will
1461 * cause this to use the SPECIALS hash for folding (meaning get full case
1462 * folding); otherwise, when zero, this implies a simple case fold */
4a8240a3 1463#define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "" : NULL)
c3fd2246 1464
84afefe6
JH
1465UV
1466Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1467{
a1433954
KW
1468 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1469 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1470 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1471 * the changed version may be longer than the original character.
1472 *
1473 * The ordinal of the first character of the changed version is returned
1474 * (but note, as explained above, that there may be more.) */
1475
7918f24d
NC
1476 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1477
3a4c58c9
KW
1478 if (c < 256) {
1479 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1480 }
1481
0ebc6274 1482 uvchr_to_utf8(p, c);
3a4c58c9 1483 return CALL_UPPER_CASE(p, p, lenp);
a0ed51b3
LW
1484}
1485
84afefe6
JH
1486UV
1487Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1488{
7918f24d
NC
1489 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1490
3a4c58c9
KW
1491 if (c < 256) {
1492 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1493 }
1494
0ebc6274 1495 uvchr_to_utf8(p, c);
3a4c58c9 1496 return CALL_TITLE_CASE(p, p, lenp);
a0ed51b3
LW
1497}
1498
afc16117 1499STATIC U8
81611534 1500S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp)
afc16117
KW
1501{
1502 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 1503 * those, converting the result to UTF-8. Since the result is always just
a1433954 1504 * one character, we allow <p> to be NULL */
afc16117
KW
1505
1506 U8 converted = toLOWER_LATIN1(c);
1507
1508 if (p != NULL) {
6f2d5cbc 1509 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
1510 *p = converted;
1511 *lenp = 1;
1512 }
1513 else {
430c9760
KW
1514 /* Result is known to always be < 256, so can use the EIGHT_BIT
1515 * macros */
1516 *p = UTF8_EIGHT_BIT_HI(converted);
1517 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
1518 *lenp = 2;
1519 }
1520 }
1521 return converted;
1522}
1523
84afefe6
JH
1524UV
1525Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1526{
7918f24d
NC
1527 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1528
afc16117
KW
1529 if (c < 256) {
1530 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1531 }
1532
afc16117 1533 uvchr_to_utf8(p, c);
968c5e6a 1534 return CALL_LOWER_CASE(p, p, lenp);
a0ed51b3
LW
1535}
1536
84afefe6 1537UV
51910141 1538Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 1539{
51910141 1540 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 1541 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 1542 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
1543 *
1544 * Not to be used for locale folds
51910141 1545 */
f673fad4 1546
a1dde8de
KW
1547 UV converted;
1548
1549 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 1550 PERL_UNUSED_CONTEXT;
a1dde8de 1551
1ca267a5
KW
1552 assert (! (flags & FOLD_FLAGS_LOCALE));
1553
a1dde8de
KW
1554 if (c == MICRO_SIGN) {
1555 converted = GREEK_SMALL_LETTER_MU;
1556 }
9b63e895
KW
1557#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
1558 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
1559 || UNICODE_DOT_DOT_VERSION > 0)
51910141 1560 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1ca267a5
KW
1561
1562 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1563 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1564 * under those circumstances. */
1565 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1566 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1567 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1568 p, *lenp, U8);
1569 return LATIN_SMALL_LETTER_LONG_S;
1570 }
1571 else {
4f489194
KW
1572 *(p)++ = 's';
1573 *p = 's';
1574 *lenp = 2;
1575 return 's';
1ca267a5 1576 }
a1dde8de 1577 }
9b63e895 1578#endif
a1dde8de
KW
1579 else { /* In this range the fold of all other characters is their lower
1580 case */
1581 converted = toLOWER_LATIN1(c);
1582 }
1583
6f2d5cbc 1584 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
1585 *p = (U8) converted;
1586 *lenp = 1;
1587 }
1588 else {
1589 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1590 *p = UTF8_TWO_BYTE_LO(converted);
1591 *lenp = 2;
1592 }
1593
1594 return converted;
1595}
1596
1597UV
31f05a37 1598Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 1599{
4b593389 1600
a0270393
KW
1601 /* Not currently externally documented, and subject to change
1602 * <flags> bits meanings:
1603 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
1604 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1605 * locale are to be used.
a0270393
KW
1606 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1607 */
4b593389 1608
36bb2ab6 1609 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1610
780fcc9f
KW
1611 if (flags & FOLD_FLAGS_LOCALE) {
1612 /* Treat a UTF-8 locale as not being in locale at all */
1613 if (IN_UTF8_CTYPE_LOCALE) {
1614 flags &= ~FOLD_FLAGS_LOCALE;
1615 }
1616 else {
1617 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 1618 goto needs_full_generality;
780fcc9f 1619 }
31f05a37
KW
1620 }
1621
a1dde8de 1622 if (c < 256) {
e7b7ac46 1623 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 1624 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
1625 }
1626
2f306ab9 1627 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
1628 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1629 uvchr_to_utf8(p, c);
1630 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1631 }
1632 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1633 the special flags. */
1634 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
1635
1636 needs_full_generality:
a0270393 1637 uvchr_to_utf8(utf8_c, c);
445bf929 1638 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
a0270393 1639 }
84afefe6
JH
1640}
1641
26483009 1642PERL_STATIC_INLINE bool
5141f98e 1643S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 1644 const char *const swashname, SV* const invlist)
bde6a22d 1645{
ea317ccb
KW
1646 /* returns a boolean giving whether or not the UTF8-encoded character that
1647 * starts at <p> is in the swash indicated by <swashname>. <swash>
1648 * contains a pointer to where the swash indicated by <swashname>
1649 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
1650 * look at <*swash> and only generate a swash if it is not null. <invlist>
1651 * is NULL or an inversion list that defines the swash. If not null, it
1652 * saves time during initialization of the swash.
ea317ccb
KW
1653 *
1654 * Note that it is assumed that the buffer length of <p> is enough to
1655 * contain all the bytes that comprise the character. Thus, <*p> should
1656 * have been checked before this call for mal-formedness enough to assure
1657 * that. */
1658
7918f24d
NC
1659 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1660
492a624f 1661 /* The API should have included a length for the UTF-8 character in <p>,
28123549 1662 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
1663 * as far as there being enough bytes available in it to accommodate the
1664 * character without reading beyond the end, and pass that number on to the
1665 * validating routine */
6302f837 1666 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
28123549
KW
1667 if (ckWARN_d(WARN_UTF8)) {
1668 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
9816f121 1669 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
28123549
KW
1670 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1671 what the malformation is */
1672 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1673 }
1674 }
1675 return FALSE;
1676 }
87367d5f
KW
1677 if (!*swash) {
1678 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
1679 *swash = _core_swash_init("utf8",
1680
1681 /* Only use the name if there is no inversion
1682 * list; otherwise will go out to disk */
1683 (invlist) ? "" : swashname,
1684
1685 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 1686 }
28123549 1687
bde6a22d
NC
1688 return swash_fetch(*swash, p, TRUE) != 0;
1689}
1690
1691bool
922e8cb4
KW
1692Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1693{
922e8cb4
KW
1694 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1695
1696 assert(classnum < _FIRST_NON_SWASH_CC);
1697
f25ce844
KW
1698 return is_utf8_common(p,
1699 &PL_utf8_swash_ptrs[classnum],
1700 swash_property_names[classnum],
1701 PL_XPosix_ptrs[classnum]);
922e8cb4
KW
1702}
1703
1704bool
f2645549 1705Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
a0ed51b3 1706{
f2645549 1707 SV* invlist = NULL;
7918f24d 1708
f2645549 1709 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
7918f24d 1710
f2645549
KW
1711 if (! PL_utf8_perl_idstart) {
1712 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1713 }
60071a22 1714 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
82686b01
JH
1715}
1716
1717bool
f2645549 1718Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 1719{
f2645549 1720 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
1721
1722 if (*p == '_')
1723 return TRUE;
f25ce844 1724 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
1725}
1726
1727bool
eba68aa0
KW
1728Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1729{
b24b43f7 1730 SV* invlist = NULL;
eba68aa0
KW
1731
1732 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1733
b24b43f7
KW
1734 if (! PL_utf8_perl_idcont) {
1735 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1736 }
60071a22 1737 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
eba68aa0
KW
1738}
1739
eba68aa0 1740bool
f2645549 1741Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 1742{
f2645549 1743 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 1744
f25ce844 1745 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
1746}
1747
1748bool
f2645549 1749Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 1750{
f2645549 1751 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 1752
f25ce844 1753 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
1754}
1755
1756bool
7dbf68d2
KW
1757Perl__is_utf8_mark(pTHX_ const U8 *p)
1758{
7dbf68d2
KW
1759 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1760
f25ce844 1761 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
1762}
1763
6b5c0936 1764/*
87cea99e 1765=for apidoc to_utf8_case
6b5c0936 1766
6fae5207 1767C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
1768the character that is being converted. This routine assumes that the character
1769at C<p> is well-formed.
6b5c0936 1770
6fae5207
KW
1771C<ustrp> is a pointer to the character buffer to put the
1772conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
1773of the result.
1774
6fae5207 1775C<swashp> is a pointer to the swash to use.
6b5c0936 1776
a1433954 1777Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
796b6530 1778and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 1779but not always, a multicharacter mapping), is tried first.
6b5c0936 1780
4a8240a3
KW
1781C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1782any special mappings; C<""> means to use the special mappings. Values other
1783than these two are treated as the name of the hash containing the special
1784mappings, like C<"utf8::ToSpecLower">.
6b5c0936 1785
796b6530
KW
1786C<normal> is a string like C<"ToLower"> which means the swash
1787C<%utf8::ToLower>.
0134edef
JH
1788
1789=cut */
6b5c0936 1790
2104c8d9 1791UV
9a957fbc
AL
1792Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1793 SV **swashp, const char *normal, const char *special)
a0ed51b3 1794{
0134edef 1795 STRLEN len = 0;
f4cd282c 1796 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
7918f24d
NC
1797
1798 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1799
9ae3ac1a
KW
1800 /* Note that swash_fetch() doesn't output warnings for these because it
1801 * assumes we will */
8457b38f 1802 if (uv1 >= UNICODE_SURROGATE_FIRST) {
9ae3ac1a 1803 if (uv1 <= UNICODE_SURROGATE_LAST) {
8457b38f
KW
1804 if (ckWARN_d(WARN_SURROGATE)) {
1805 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1806 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1807 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1808 }
9ae3ac1a
KW
1809 }
1810 else if (UNICODE_IS_SUPER(uv1)) {
8457b38f
KW
1811 if (ckWARN_d(WARN_NON_UNICODE)) {
1812 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1813 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1814 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1815 }
9ae3ac1a
KW
1816 }
1817
1818 /* Note that non-characters are perfectly legal, so no warning should
1819 * be given */
1820 }
1821
0134edef 1822 if (!*swashp) /* load on-demand */
5ab9d2ef 1823 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 1824
a6f87d8c 1825 if (special) {
0134edef 1826 /* It might be "special" (sometimes, but not always,
2a37f04d 1827 * a multicharacter mapping) */
4a8240a3 1828 HV *hv = NULL;
b08cf34e
JH
1829 SV **svp;
1830
4a8240a3
KW
1831 /* If passed in the specials name, use that; otherwise use any
1832 * given in the swash */
1833 if (*special != '\0') {
1834 hv = get_hv(special, 0);
1835 }
1836 else {
1837 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
1838 if (svp) {
1839 hv = MUTABLE_HV(SvRV(*svp));
1840 }
1841 }
1842
176fe009 1843 if (hv
5f560d8a 1844 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
1845 && (*svp))
1846 {
cfd0369c 1847 const char *s;
47654450 1848
cfd0369c 1849 s = SvPV_const(*svp, len);
47654450 1850 if (len == 1)
f4cd282c 1851 /* EIGHTBIT */
c80e42f3 1852 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 1853 else {
d2dcd0fb 1854 Copy(s, ustrp, len, U8);
29e98929 1855 }
983ffd37 1856 }
0134edef
JH
1857 }
1858
1859 if (!len && *swashp) {
4a4088c4 1860 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 1861
0134edef
JH
1862 if (uv2) {
1863 /* It was "normal" (a single character mapping). */
f4cd282c 1864 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
1865 }
1866 }
1feea2c7 1867
cbe07460
KW
1868 if (len) {
1869 if (lenp) {
1870 *lenp = len;
1871 }
1872 return valid_utf8_to_uvchr(ustrp, 0);
1873 }
1874
1875 /* Here, there was no mapping defined, which means that the code point maps
1876 * to itself. Return the inputs */
bfdf22ec 1877 len = UTF8SKIP(p);
ca9fab46
KW
1878 if (p != ustrp) { /* Don't copy onto itself */
1879 Copy(p, ustrp, len, U8);
1880 }
0134edef 1881
2a37f04d
JH
1882 if (lenp)
1883 *lenp = len;
1884
f4cd282c 1885 return uv1;
cbe07460 1886
a0ed51b3
LW
1887}
1888
051a06d4 1889STATIC UV
357aadde 1890S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
051a06d4 1891{
4a4088c4 1892 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
1893 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
1894 * result contains a character that crosses the 255/256 boundary, disallow
1895 * the change, and return the original code point. See L<perlfunc/lc> for
1896 * why;
051a06d4 1897 *
a1433954
KW
1898 * p points to the original string whose case was changed; assumed
1899 * by this routine to be well-formed
051a06d4
KW
1900 * result the code point of the first character in the changed-case string
1901 * ustrp points to the changed-case string (<result> represents its first char)
1902 * lenp points to the length of <ustrp> */
1903
1904 UV original; /* To store the first code point of <p> */
1905
1906 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
1907
a4f12ed7 1908 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
1909
1910 /* We know immediately if the first character in the string crosses the
1911 * boundary, so can skip */
1912 if (result > 255) {
1913
1914 /* Look at every character in the result; if any cross the
1915 * boundary, the whole thing is disallowed */
1916 U8* s = ustrp + UTF8SKIP(ustrp);
1917 U8* e = ustrp + *lenp;
1918 while (s < e) {
a4f12ed7 1919 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
1920 goto bad_crossing;
1921 }
1922 s += UTF8SKIP(s);
1923 }
1924
613abc6d
KW
1925 /* Here, no characters crossed, result is ok as-is, but we warn. */
1926 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
1927 return result;
1928 }
1929
7b52d656 1930 bad_crossing:
051a06d4
KW
1931
1932 /* Failed, have to return the original */
4b88fb76 1933 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
1934
1935 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
1936 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1937 "Can't do %s(\"\\x{%"UVXf"}\") on non-UTF-8 locale; "
1938 "resolved to \"\\x{%"UVXf"}\".",
357aadde 1939 OP_DESC(PL_op),
ab0b796c
KW
1940 original,
1941 original);
051a06d4
KW
1942 Copy(p, ustrp, *lenp, char);
1943 return original;
1944}
1945
d3e79532 1946/*
87cea99e 1947=for apidoc to_utf8_upper
d3e79532 1948
1f607577 1949Instead use L</toUPPER_utf8>.
a1433954 1950
d3e79532
JH
1951=cut */
1952
051a06d4 1953/* Not currently externally documented, and subject to change:
31f05a37
KW
1954 * <flags> is set iff iff the rules from the current underlying locale are to
1955 * be used. */
051a06d4 1956
2104c8d9 1957UV
31f05a37 1958Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 1959{
051a06d4
KW
1960 UV result;
1961
1962 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 1963
780fcc9f
KW
1964 if (flags) {
1965 /* Treat a UTF-8 locale as not being in locale at all */
1966 if (IN_UTF8_CTYPE_LOCALE) {
1967 flags = FALSE;
1968 }
1969 else {
1970 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
1971 }
31f05a37
KW
1972 }
1973
3a4c58c9 1974 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
1975 if (flags) {
1976 result = toUPPER_LC(*p);
1977 }
1978 else {
81c6c7ce 1979 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 1980 }
3a4c58c9
KW
1981 }
1982 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 1983 if (flags) {
a62b247b 1984 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 1985 result = toUPPER_LC(c);
051a06d4
KW
1986 }
1987 else {
a62b247b 1988 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 1989 ustrp, lenp, 'S');
051a06d4
KW
1990 }
1991 }
4a4088c4 1992 else { /* UTF-8, ord above 255 */
051a06d4
KW
1993 result = CALL_UPPER_CASE(p, ustrp, lenp);
1994
1995 if (flags) {
357aadde 1996 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4
KW
1997 }
1998 return result;
1999 }
2000
4a4088c4 2001 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2002 if (UTF8_IS_INVARIANT(result)) {
2003 *ustrp = (U8) result;
2004 *lenp = 1;
2005 }
2006 else {
62cb07ea
KW
2007 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2008 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2009 *lenp = 2;
3a4c58c9 2010 }
baa60164 2011
051a06d4 2012 return result;
983ffd37 2013}
a0ed51b3 2014
d3e79532 2015/*
87cea99e 2016=for apidoc to_utf8_title
d3e79532 2017
1f607577 2018Instead use L</toTITLE_utf8>.
a1433954 2019
d3e79532
JH
2020=cut */
2021
051a06d4 2022/* Not currently externally documented, and subject to change:
31f05a37
KW
2023 * <flags> is set iff the rules from the current underlying locale are to be
2024 * used. Since titlecase is not defined in POSIX, for other than a
2025 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 2026 */
051a06d4 2027
983ffd37 2028UV
31f05a37 2029Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
983ffd37 2030{
051a06d4
KW
2031 UV result;
2032
2033 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2034
780fcc9f
KW
2035 if (flags) {
2036 /* Treat a UTF-8 locale as not being in locale at all */
2037 if (IN_UTF8_CTYPE_LOCALE) {
2038 flags = FALSE;
2039 }
2040 else {
2041 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2042 }
31f05a37
KW
2043 }
2044
3a4c58c9 2045 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2046 if (flags) {
2047 result = toUPPER_LC(*p);
2048 }
2049 else {
81c6c7ce 2050 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2051 }
3a4c58c9
KW
2052 }
2053 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2054 if (flags) {
a62b247b 2055 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2056 result = toUPPER_LC(c);
051a06d4
KW
2057 }
2058 else {
a62b247b 2059 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2060 ustrp, lenp, 's');
051a06d4
KW
2061 }
2062 }
4a4088c4 2063 else { /* UTF-8, ord above 255 */
051a06d4
KW
2064 result = CALL_TITLE_CASE(p, ustrp, lenp);
2065
2066 if (flags) {
357aadde 2067 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4
KW
2068 }
2069 return result;
2070 }
2071
4a4088c4 2072 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2073 if (UTF8_IS_INVARIANT(result)) {
2074 *ustrp = (U8) result;
2075 *lenp = 1;
2076 }
2077 else {
62cb07ea
KW
2078 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2079 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2080 *lenp = 2;
3a4c58c9
KW
2081 }
2082
051a06d4 2083 return result;
a0ed51b3
LW
2084}
2085
d3e79532 2086/*
87cea99e 2087=for apidoc to_utf8_lower
d3e79532 2088
1f607577 2089Instead use L</toLOWER_utf8>.
a1433954 2090
d3e79532
JH
2091=cut */
2092
051a06d4 2093/* Not currently externally documented, and subject to change:
31f05a37
KW
2094 * <flags> is set iff iff the rules from the current underlying locale are to
2095 * be used.
2096 */
051a06d4 2097
2104c8d9 2098UV
31f05a37 2099Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2100{
051a06d4
KW
2101 UV result;
2102
051a06d4 2103 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2104
780fcc9f
KW
2105 if (flags) {
2106 /* Treat a UTF-8 locale as not being in locale at all */
2107 if (IN_UTF8_CTYPE_LOCALE) {
2108 flags = FALSE;
2109 }
2110 else {
2111 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2112 }
31f05a37
KW
2113 }
2114
968c5e6a 2115 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2116 if (flags) {
2117 result = toLOWER_LC(*p);
2118 }
2119 else {
81c6c7ce 2120 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2121 }
968c5e6a
KW
2122 }
2123 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2124 if (flags) {
a62b247b 2125 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2126 result = toLOWER_LC(c);
051a06d4
KW
2127 }
2128 else {
a62b247b 2129 return to_lower_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2130 ustrp, lenp);
051a06d4 2131 }
968c5e6a 2132 }
4a4088c4 2133 else { /* UTF-8, ord above 255 */
051a06d4
KW
2134 result = CALL_LOWER_CASE(p, ustrp, lenp);
2135
2136 if (flags) {
357aadde 2137 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2138 }
968c5e6a 2139
051a06d4
KW
2140 return result;
2141 }
2142
4a4088c4 2143 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2144 if (UTF8_IS_INVARIANT(result)) {
2145 *ustrp = (U8) result;
2146 *lenp = 1;
2147 }
2148 else {
62cb07ea
KW
2149 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2150 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2151 *lenp = 2;
2152 }
2153
051a06d4 2154 return result;
b4e400f9
JH
2155}
2156
d3e79532 2157/*
87cea99e 2158=for apidoc to_utf8_fold
d3e79532 2159
1f607577 2160Instead use L</toFOLD_utf8>.
a1433954 2161
d3e79532
JH
2162=cut */
2163
051a06d4
KW
2164/* Not currently externally documented, and subject to change,
2165 * in <flags>
31f05a37
KW
2166 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2167 * locale are to be used.
051a06d4
KW
2168 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2169 * otherwise simple folds
a0270393
KW
2170 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2171 * prohibited
445bf929 2172 */
36bb2ab6 2173
b4e400f9 2174UV
445bf929 2175Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
b4e400f9 2176{
051a06d4
KW
2177 UV result;
2178
36bb2ab6 2179 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2180
a0270393
KW
2181 /* These are mutually exclusive */
2182 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2183
50ba90ff
KW
2184 assert(p != ustrp); /* Otherwise overwrites */
2185
780fcc9f
KW
2186 if (flags & FOLD_FLAGS_LOCALE) {
2187 /* Treat a UTF-8 locale as not being in locale at all */
2188 if (IN_UTF8_CTYPE_LOCALE) {
2189 flags &= ~FOLD_FLAGS_LOCALE;
2190 }
2191 else {
2192 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2193 }
31f05a37
KW
2194 }
2195
a1dde8de 2196 if (UTF8_IS_INVARIANT(*p)) {
051a06d4 2197 if (flags & FOLD_FLAGS_LOCALE) {
d22b930b 2198 result = toFOLD_LC(*p);
051a06d4
KW
2199 }
2200 else {
81c6c7ce 2201 return _to_fold_latin1(*p, ustrp, lenp,
1ca267a5 2202 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2203 }
a1dde8de
KW
2204 }
2205 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2206 if (flags & FOLD_FLAGS_LOCALE) {
a62b247b 2207 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2208 result = toFOLD_LC(c);
051a06d4
KW
2209 }
2210 else {
a62b247b 2211 return _to_fold_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
51910141 2212 ustrp, lenp,
1ca267a5 2213 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2214 }
a1dde8de 2215 }
4a4088c4 2216 else { /* UTF-8, ord above 255 */
a0270393 2217 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2218
1ca267a5
KW
2219 if (flags & FOLD_FLAGS_LOCALE) {
2220
76f2ffcd 2221# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
2222 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
2223
2224# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
2225# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd
KW
2226
2227 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
76f2ffcd 2228
538e84ed
KW
2229 /* Special case these two characters, as what normally gets
2230 * returned under locale doesn't work */
76f2ffcd
KW
2231 if (UTF8SKIP(p) == cap_sharp_s_len
2232 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 2233 {
ab0b796c
KW
2234 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2235 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2236 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
2237 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
2238 goto return_long_s;
2239 }
0766489e
KW
2240 else
2241#endif
2242 if (UTF8SKIP(p) == long_s_t_len
76f2ffcd 2243 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 2244 {
ab0b796c
KW
2245 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2246 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2247 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
2248 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
2249 goto return_ligature_st;
2250 }
74894415
KW
2251
2252#if UNICODE_MAJOR_VERSION == 3 \
2253 && UNICODE_DOT_VERSION == 0 \
2254 && UNICODE_DOT_DOT_VERSION == 1
2255# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
2256
2257 /* And special case this on this Unicode version only, for the same
2258 * reaons the other two are special cased. They would cross the
2259 * 255/256 boundary which is forbidden under /l, and so the code
2260 * wouldn't catch that they are equivalent (which they are only in
2261 * this release) */
2262 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
2263 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
2264 {
2265 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2266 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2267 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
2268 "resolved to \"\\x{0131}\".");
2269 goto return_dotless_i;
2270 }
2271#endif
2272
357aadde 2273 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2274 }
a0270393
KW
2275 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2276 return result;
2277 }
2278 else {
4a4088c4 2279 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
2280 * character above the ASCII range, and the result should not
2281 * contain an ASCII character. */
a0270393
KW
2282
2283 UV original; /* To store the first code point of <p> */
2284
2285 /* Look at every character in the result; if any cross the
2286 * boundary, the whole thing is disallowed */
2287 U8* s = ustrp;
2288 U8* e = ustrp + *lenp;
2289 while (s < e) {
2290 if (isASCII(*s)) {
2291 /* Crossed, have to return the original */
2292 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 2293
9fc2026f 2294 /* But in these instances, there is an alternative we can
1ca267a5 2295 * return that is valid */
0766489e
KW
2296 if (original == LATIN_SMALL_LETTER_SHARP_S
2297#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
2298 || original == LATIN_CAPITAL_LETTER_SHARP_S
2299#endif
2300 ) {
1ca267a5
KW
2301 goto return_long_s;
2302 }
9fc2026f
KW
2303 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2304 goto return_ligature_st;
2305 }
74894415
KW
2306#if UNICODE_MAJOR_VERSION == 3 \
2307 && UNICODE_DOT_VERSION == 0 \
2308 && UNICODE_DOT_DOT_VERSION == 1
2309
2310 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
2311 goto return_dotless_i;
2312 }
2313#endif
a0270393
KW
2314 Copy(p, ustrp, *lenp, char);
2315 return original;
2316 }
2317 s += UTF8SKIP(s);
2318 }
051a06d4 2319
a0270393
KW
2320 /* Here, no characters crossed, result is ok as-is */
2321 return result;
2322 }
051a06d4
KW
2323 }
2324
4a4088c4 2325 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2326 if (UTF8_IS_INVARIANT(result)) {
2327 *ustrp = (U8) result;
2328 *lenp = 1;
2329 }
2330 else {
62cb07ea
KW
2331 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2332 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2333 *lenp = 2;
2334 }
2335
051a06d4 2336 return result;
1ca267a5
KW
2337
2338 return_long_s:
2339 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2340 * folds to a string of two of these characters. By returning this
2341 * instead, then, e.g.,
2342 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2343 * works. */
2344
2345 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2346 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2347 ustrp, *lenp, U8);
2348 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
2349
2350 return_ligature_st:
2351 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2352 * have the other one fold to it */
2353
2354 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2355 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2356 return LATIN_SMALL_LIGATURE_ST;
74894415
KW
2357
2358#if UNICODE_MAJOR_VERSION == 3 \
2359 && UNICODE_DOT_VERSION == 0 \
2360 && UNICODE_DOT_DOT_VERSION == 1
2361
2362 return_dotless_i:
2363 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
2364 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
2365 return LATIN_SMALL_LETTER_DOTLESS_I;
2366
2367#endif
2368
a0ed51b3
LW
2369}
2370
711a919c 2371/* Note:
f90a9a02 2372 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2373 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2374 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2375 */
c4a5db0c 2376
a0ed51b3 2377SV*
7fc63493 2378Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2379{
c4a5db0c
KW
2380 PERL_ARGS_ASSERT_SWASH_INIT;
2381
2382 /* Returns a copy of a swash initiated by the called function. This is the
2383 * public interface, and returning a copy prevents others from doing
2384 * mischief on the original */
2385
5d3d13d1 2386 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
2387}
2388
2389SV*
5d3d13d1 2390Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c 2391{
2c1f00b9
YO
2392
2393 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
2394 * use the following define */
2395
2396#define CORE_SWASH_INIT_RETURN(x) \
2397 PL_curpm= old_PL_curpm; \
2398 return x
2399
c4a5db0c 2400 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
2401 * by calling utf8_heavy.pl in the general case. The returned value may be
2402 * the swash's inversion list instead if the input parameters allow it.
2403 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
2404 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2405 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
2406 *
2407 * This interface should only be used by functions that won't destroy or
2408 * adversely change the swash, as doing so affects all other uses of the
2409 * swash in the program; the general public should use 'Perl_swash_init'
2410 * instead.
2411 *
2412 * pkg is the name of the package that <name> should be in.
2413 * name is the name of the swash to find. Typically it is a Unicode
2414 * property name, including user-defined ones
2415 * listsv is a string to initialize the swash with. It must be of the form
2416 * documented as the subroutine return value in
2417 * L<perlunicode/User-Defined Character Properties>
2418 * minbits is the number of bits required to represent each data element.
2419 * It is '1' for binary properties.
2420 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 2421 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
2422 * flags_p if non-NULL is the address of various input and output flag bits
2423 * to the routine, as follows: ('I' means is input to the routine;
2424 * 'O' means output from the routine. Only flags marked O are
2425 * meaningful on return.)
2426 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2427 * came from a user-defined property. (I O)
5d3d13d1
KW
2428 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2429 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
2430 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2431 * return of an inversion list instead of a swash hash if this routine
2432 * thinks that would result in faster execution of swash_fetch() later
2433 * on. (I)
9a53f6cf
KW
2434 *
2435 * Thus there are three possible inputs to find the swash: <name>,
2436 * <listsv>, and <invlist>. At least one must be specified. The result
2437 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
2438 * actions can intersect, etc. what <name> gives. To avoid going out to
2439 * disk at all, <invlist> should specify completely what the swash should
2440 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
2441 *
2442 * <invlist> is only valid for binary properties */
c4a5db0c 2443
2c1f00b9
YO
2444 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
2445
c4a5db0c 2446 SV* retval = &PL_sv_undef;
83199d38 2447 HV* swash_hv = NULL;
87367d5f
KW
2448 const int invlist_swash_boundary =
2449 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2450 ? 512 /* Based on some benchmarking, but not extensive, see commit
2451 message */
2452 : -1; /* Never return just an inversion list */
9a53f6cf
KW
2453
2454 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2455 assert(! invlist || minbits == 1);
2456
2c1f00b9
YO
2457 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
2458 that triggered the swash init and the swash init perl logic itself.
2459 See perl #122747 */
2460
9a53f6cf
KW
2461 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2462 * so */
2463 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2464 dSP;
2465 const size_t pkg_len = strlen(pkg);
2466 const size_t name_len = strlen(name);
2467 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2468 SV* errsv_save;
2469 GV *method;
2470
2471 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2472
2473 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2474 ENTER;
69794297 2475 SAVEHINTS();
2782061f 2476 save_re_context();
650f067c
JL
2477 /* We might get here via a subroutine signature which uses a utf8
2478 * parameter name, at which point PL_subname will have been set
2479 * but not yet used. */
2480 save_item(PL_subname);
69794297
KW
2481 if (PL_parser && PL_parser->error_count)
2482 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2483 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
4a4088c4 2484 if (!method) { /* demand load UTF-8 */
69794297 2485 ENTER;
db2c6cb3
FC
2486 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2487 GvSV(PL_errgv) = NULL;
1a419e6b 2488#ifndef NO_TAINT_SUPPORT
69794297
KW
2489 /* It is assumed that callers of this routine are not passing in
2490 * any user derived data. */
2782061f
DM
2491 /* Need to do this after save_re_context() as it will set
2492 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2493 * in Perl_magic_get). Even line to create errsv_save can turn on
2494 * PL_tainted. */
284167a5
SM
2495 SAVEBOOL(TAINT_get);
2496 TAINT_NOT;
2497#endif
69794297
KW
2498 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2499 NULL);
eed484f9 2500 {
db2c6cb3
FC
2501 /* Not ERRSV, as there is no need to vivify a scalar we are
2502 about to discard. */
2503 SV * const errsv = GvSV(PL_errgv);
2504 if (!SvTRUE(errsv)) {
2505 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2506 SvREFCNT_dec(errsv);
2507 }
eed484f9 2508 }
69794297
KW
2509 LEAVE;
2510 }
2511 SPAGAIN;
2512 PUSHMARK(SP);
2513 EXTEND(SP,5);
2514 mPUSHp(pkg, pkg_len);
2515 mPUSHp(name, name_len);
2516 PUSHs(listsv);
2517 mPUSHi(minbits);
2518 mPUSHi(none);
2519 PUTBACK;
db2c6cb3
FC
2520 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2521 GvSV(PL_errgv) = NULL;
69794297
KW
2522 /* If we already have a pointer to the method, no need to use
2523 * call_method() to repeat the lookup. */
c41800a8
KW
2524 if (method
2525 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
2526 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2527 {
2528 retval = *PL_stack_sp--;
2529 SvREFCNT_inc(retval);
2530 }
eed484f9 2531 {
db2c6cb3
FC
2532 /* Not ERRSV. See above. */
2533 SV * const errsv = GvSV(PL_errgv);
2534 if (!SvTRUE(errsv)) {
2535 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2536 SvREFCNT_dec(errsv);
2537 }
eed484f9 2538 }
ce3b816e 2539 LEAVE;
69794297
KW
2540 POPSTACK;
2541 if (IN_PERL_COMPILETIME) {
2542 CopHINTS_set(PL_curcop, PL_hints);
2543 }
2544 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2545 if (SvPOK(retval))
2546
2547 /* If caller wants to handle missing properties, let them */
5d3d13d1 2548 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2c1f00b9 2549 CORE_SWASH_INIT_RETURN(NULL);
69794297
KW
2550 }
2551 Perl_croak(aTHX_
2552 "Can't find Unicode property definition \"%"SVf"\"",
2553 SVfARG(retval));
a25b5927 2554 NOT_REACHED; /* NOTREACHED */
69794297 2555 }
9a53f6cf 2556 } /* End of calling the module to find the swash */
36eb48b4 2557
83199d38
KW
2558 /* If this operation fetched a swash, and we will need it later, get it */
2559 if (retval != &PL_sv_undef
2560 && (minbits == 1 || (flags_p
2561 && ! (*flags_p
2562 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2563 {
2564 swash_hv = MUTABLE_HV(SvRV(retval));
2565
2566 /* If we don't already know that there is a user-defined component to
2567 * this swash, and the user has indicated they wish to know if there is
2568 * one (by passing <flags_p>), find out */
2569 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2570 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2571 if (user_defined && SvUV(*user_defined)) {
2572 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2573 }
2574 }
2575 }
2576
36eb48b4
KW
2577 /* Make sure there is an inversion list for binary properties */
2578 if (minbits == 1) {
2579 SV** swash_invlistsvp = NULL;
2580 SV* swash_invlist = NULL;
9a53f6cf 2581 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
2582 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2583 an unclaimed reference count */
36eb48b4 2584
9a53f6cf 2585 /* If this operation fetched a swash, get its already existing
83199d38 2586 * inversion list, or create one for it */
36eb48b4 2587
83199d38 2588 if (swash_hv) {
5c9f4bd2 2589 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
2590 if (swash_invlistsvp) {
2591 swash_invlist = *swash_invlistsvp;
2592 invlist_in_swash_is_valid = TRUE;
2593 }
2594 else {
36eb48b4 2595 swash_invlist = _swash_to_invlist(retval);
02c85471 2596 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
2597 }
2598 }
2599
2600 /* If an inversion list was passed in, have to include it */
2601 if (invlist) {
2602
2603 /* Any fetched swash will by now have an inversion list in it;
2604 * otherwise <swash_invlist> will be NULL, indicating that we
2605 * didn't fetch a swash */
2606 if (swash_invlist) {
2607
2608 /* Add the passed-in inversion list, which invalidates the one
2609 * already stored in the swash */
2610 invlist_in_swash_is_valid = FALSE;
2611 _invlist_union(invlist, swash_invlist, &swash_invlist);
2612 }
2613 else {
2614
87367d5f
KW
2615 /* Here, there is no swash already. Set up a minimal one, if
2616 * we are going to return a swash */
2617 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 2618 swash_hv = newHV();
4aca0fe6 2619 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 2620 }
9a53f6cf
KW
2621 swash_invlist = invlist;
2622 }
9a53f6cf
KW
2623 }
2624
2625 /* Here, we have computed the union of all the passed-in data. It may
2626 * be that there was an inversion list in the swash which didn't get
538e84ed 2627 * touched; otherwise save the computed one */
87367d5f
KW
2628 if (! invlist_in_swash_is_valid
2629 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2630 {
5c9f4bd2 2631 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
2632 {
2633 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2634 }
cc34d8c5
FC
2635 /* We just stole a reference count. */
2636 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2637 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 2638 }
87367d5f 2639
dbfdbd26
KW
2640 SvREADONLY_on(swash_invlist);
2641
c41800a8 2642 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
2643 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2644 SvREFCNT_dec(retval);
02c85471
FC
2645 if (!swash_invlist_unclaimed)
2646 SvREFCNT_inc_simple_void_NN(swash_invlist);
2647 retval = newRV_noinc(swash_invlist);
87367d5f 2648 }
36eb48b4
KW
2649 }
2650
2c1f00b9
YO
2651 CORE_SWASH_INIT_RETURN(retval);
2652#undef CORE_SWASH_INIT_RETURN
a0ed51b3
LW
2653}
2654
035d37be
JH
2655
2656/* This API is wrong for special case conversions since we may need to
2657 * return several Unicode characters for a single Unicode character
2658 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2659 * the lower-level routine, and it is similarly broken for returning
38684baa
KW
2660 * multiple values. --jhi
2661 * For those, you should use to_utf8_case() instead */
b0e3252e 2662/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
2663
2664/* Note:
2665 * Returns the value of property/mapping C<swash> for the first character
2666 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
4a4088c4 2667 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
3d0f8846 2668 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
2669 *
2670 * A "swash" is a hash which contains initially the keys/values set up by
2671 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2672 * property for all possible code points. Things are stored in a compact form
2673 * (see utf8_heavy.pl) so that calculation is required to find the actual
2674 * property value for a given code point. As code points are looked up, new
2675 * key/value pairs are added to the hash, so that the calculation doesn't have
2676 * to ever be re-done. Further, each calculation is done, not just for the
2677 * desired one, but for a whole block of code points adjacent to that one.
2678 * For binary properties on ASCII machines, the block is usually for 64 code
2679 * points, starting with a code point evenly divisible by 64. Thus if the
2680 * property value for code point 257 is requested, the code goes out and
2681 * calculates the property values for all 64 code points between 256 and 319,
2682 * and stores these as a single 64-bit long bit vector, called a "swatch",
2683 * under the key for code point 256. The key is the UTF-8 encoding for code
2684 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2685 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2686 * for code point 258 is then requested, this code realizes that it would be
2687 * stored under the key for 256, and would find that value and extract the
2688 * relevant bit, offset from 256.
2689 *
2690 * Non-binary properties are stored in as many bits as necessary to represent
2691 * their values (32 currently, though the code is more general than that), not
2692 * as single bits, but the principal is the same: the value for each key is a
2693 * vector that encompasses the property values for all code points whose UTF-8
2694 * representations are represented by the key. That is, for all code points
2695 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2696 * bytes of that.
680c470c 2697 */
a0ed51b3 2698UV
680c470c 2699Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 2700{
ef8f7699 2701 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
2702 U32 klen;
2703 U32 off;
9b56a019 2704 STRLEN slen = 0;
7d85a32c 2705 STRLEN needents;
cfd0369c 2706 const U8 *tmps = NULL;
979f2922 2707 SV *swatch;
08fb1ac5 2708 const U8 c = *ptr;
3568d838 2709
7918f24d
NC
2710 PERL_ARGS_ASSERT_SWASH_FETCH;
2711
87367d5f
KW
2712 /* If it really isn't a hash, it isn't really swash; must be an inversion
2713 * list */
2714 if (SvTYPE(hv) != SVt_PVHV) {
2715 return _invlist_contains_cp((SV*)hv,
2716 (do_utf8)
2717 ? valid_utf8_to_uvchr(ptr, NULL)
2718 : c);
2719 }
2720
08fb1ac5
KW
2721 /* We store the values in a "swatch" which is a vec() value in a swash
2722 * hash. Code points 0-255 are a single vec() stored with key length
2723 * (klen) 0. All other code points have a UTF-8 representation
2724 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2725 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2726 * length for them is the length of the encoded char - 1. ptr[klen] is the
2727 * final byte in the sequence representing the character */
2728 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2729 klen = 0;
2730 needents = 256;
2731 off = c;
3568d838 2732 }
08fb1ac5
KW
2733 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2734 klen = 0;
2735 needents = 256;
a62b247b 2736 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
2737 }
2738 else {
08fb1ac5
KW
2739 klen = UTF8SKIP(ptr) - 1;
2740
2741 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2742 * the vec is the final byte in the sequence. (In EBCDIC this is
2743 * converted to I8 to get consecutive values.) To help you visualize
2744 * all this:
2745 * Straight 1047 After final byte
2746 * UTF-8 UTF-EBCDIC I8 transform
2747 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2748 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2749 * ...
2750 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2751 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2752 * ...
2753 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2754 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2755 * ...
2756 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2757 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2758 * ...
2759 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2760 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2761 *
2762 * (There are no discontinuities in the elided (...) entries.)
2763 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2764 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2765 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2766 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2767 * index into the vec() swatch (after subtracting 0x80, which we
2768 * actually do with an '&').
2769 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2770 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2771 * dicontinuities which go away by transforming it into I8, and we
2772 * effectively subtract 0xA0 to get the index. */
979f2922 2773 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 2774 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 2775 }
7d85a32c 2776
a0ed51b3 2777 /*
4a4088c4 2778 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
a0ed51b3
LW
2779 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2780 * it's nothing to sniff at.) Pity we usually come through at least
2781 * two function calls to get here...
2782 *
2783 * NB: this code assumes that swatches are never modified, once generated!
2784 */
2785
3568d838 2786 if (hv == PL_last_swash_hv &&
a0ed51b3 2787 klen == PL_last_swash_klen &&
27da23d5 2788 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
2789 {
2790 tmps = PL_last_swash_tmps;
2791 slen = PL_last_swash_slen;
2792 }
2793 else {
2794 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 2795 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 2796
b0e3252e 2797 /* If not cached, generate it via swatch_get */
979f2922 2798 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
2799 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2800 {
2801 if (klen) {
2802 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2803 swatch = swatch_get(swash,
2804 code_point & ~((UV)needents - 1),
2805 needents);
2806 }
2807 else { /* For the first 256 code points, the swatch has a key of
2808 length 0 */
2809 swatch = swatch_get(swash, 0, needents);
2810 }
979f2922 2811
923e4eb5 2812 if (IN_PERL_COMPILETIME)
623e6609 2813 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 2814
979f2922 2815 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 2816
979f2922
ST
2817 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2818 || (slen << 3) < needents)
5637ef5b
NC
2819 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2820 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2821 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
2822 }
2823
2824 PL_last_swash_hv = hv;
16d8f38a 2825 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 2826 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
2827 /* FIXME change interpvar.h? */
2828 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
2829 PL_last_swash_slen = slen;
2830 if (klen)
2831 Copy(ptr, PL_last_swash_key, klen, U8);
2832 }
2833
9faf8d75 2834 switch ((int)((slen << 3) / needents)) {
a0ed51b3 2835 case 1:
e7aca353 2836 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
a0ed51b3 2837 case 8:
e7aca353 2838 return ((UV) tmps[off]);
a0ed51b3
LW
2839 case 16:
2840 off <<= 1;
e7aca353
JH
2841 return
2842 ((UV) tmps[off ] << 8) +
2843 ((UV) tmps[off + 1]);
a0ed51b3
LW
2844 case 32:
2845 off <<= 2;
e7aca353
JH
2846 return
2847 ((UV) tmps[off ] << 24) +
2848 ((UV) tmps[off + 1] << 16) +
2849 ((UV) tmps[off + 2] << 8) +
2850 ((UV) tmps[off + 3]);
a0ed51b3 2851 }
5637ef5b
NC
2852 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2853 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 2854 NORETURN_FUNCTION_END;
a0ed51b3 2855}
2b9d42f0 2856
319009ee
KW
2857/* Read a single line of the main body of the swash input text. These are of
2858 * the form:
2859 * 0053 0056 0073
2860 * where each number is hex. The first two numbers form the minimum and
2861 * maximum of a range, and the third is the value associated with the range.
2862 * Not all swashes should have a third number
2863 *
2864 * On input: l points to the beginning of the line to be examined; it points
2865 * to somewhere in the string of the whole input text, and is
2866 * terminated by a \n or the null string terminator.
2867 * lend points to the null terminator of that string
2868 * wants_value is non-zero if the swash expects a third number
2869 * typestr is the name of the swash's mapping, like 'ToLower'
2870 * On output: *min, *max, and *val are set to the values read from the line.
2871 * returns a pointer just beyond the line examined. If there was no
2872 * valid min number on the line, returns lend+1
2873 */
2874
2875STATIC U8*
2876S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2877 const bool wants_value, const U8* const typestr)
2878{
2879 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2880 STRLEN numlen; /* Length of the number */
02470786
KW
2881 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2882 | PERL_SCAN_DISALLOW_PREFIX
2883 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2884
2885 /* nl points to the next \n in the scan */
2886 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2887
95543e92
KW
2888 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
2889
319009ee
KW
2890 /* Get the first number on the line: the range minimum */
2891 numlen = lend - l;
2892 *min = grok_hex((char *)l, &numlen, &flags, NULL);
c88850db 2893 *max = *min; /* So can never return without setting max */
319009ee
KW
2894 if (numlen) /* If found a hex number, position past it */
2895 l += numlen;
2896 else if (nl) { /* Else, go handle next line, if any */
2897 return nl + 1; /* 1 is length of "\n" */
2898 }
2899 else { /* Else, no next line */
2900 return lend + 1; /* to LIST's end at which \n is not found */
2901 }
2902
2903 /* The max range value follows, separated by a BLANK */
2904 if (isBLANK(*l)) {
2905 ++l;
02470786
KW
2906 flags = PERL_SCAN_SILENT_ILLDIGIT
2907 | PERL_SCAN_DISALLOW_PREFIX
2908 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2909 numlen = lend - l;
2910 *max = grok_hex((char *)l, &numlen, &flags, NULL);
2911 if (numlen)
2912 l += numlen;
2913 else /* If no value here, it is a single element range */
2914 *max = *min;
2915
2916 /* Non-binary tables have a third entry: what the first element of the
24303724 2917 * range maps to. The map for those currently read here is in hex */
319009ee
KW
2918 if (wants_value) {
2919 if (isBLANK(*l)) {
2920 ++l;
f2a7d0fc
KW
2921 flags = PERL_SCAN_SILENT_ILLDIGIT
2922 | PERL_SCAN_DISALLOW_PREFIX
2923 | PERL_SCAN_SILENT_NON_PORTABLE;
2924 numlen = lend - l;
2925 *val = grok_hex((char *)l, &numlen, &flags, NULL);
2926 if (numlen)
2927 l += numlen;
2928 else
2929 *val = 0;
319009ee
KW
2930 }
2931 else {
2932 *val = 0;
2933 if (typeto) {
dcbac5bb 2934 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
2935 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
2936 typestr, l);
2937 }
2938 }
2939 }
2940 else
2941 *val = 0; /* bits == 1, then any val should be ignored */
2942 }
2943 else { /* Nothing following range min, should be single element with no
2944 mapping expected */
319009ee
KW
2945 if (wants_value) {
2946 *val = 0;
2947 if (typeto) {
dcbac5bb 2948 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
2949 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
2950 }
2951 }
2952 else
2953 *val = 0; /* bits == 1, then val should be ignored */
2954 }
2955
2956 /* Position to next line if any, or EOF */
2957 if (nl)
2958 l = nl + 1;
2959 else
2960 l = lend;
2961
2962 return l;
2963}
2964
979f2922
ST
2965/* Note:
2966 * Returns a swatch (a bit vector string) for a code point sequence
2967 * that starts from the value C<start> and comprises the number C<span>.
2968 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
2969 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
2970 */
2971STATIC SV*
b0e3252e 2972S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
2973{
2974 SV *swatch;
77f9f126 2975 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 2976 STRLEN lcur, xcur, scur;
ef8f7699 2977 HV *const hv = MUTABLE_HV(SvRV(swash));
5c9f4bd2 2978 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
36eb48b4 2979
88d45d28
KW
2980 SV** listsvp = NULL; /* The string containing the main body of the table */
2981 SV** extssvp = NULL;
2982 SV** invert_it_svp = NULL;
2983 U8* typestr = NULL;
786861f5
KW
2984 STRLEN bits;
2985 STRLEN octets; /* if bits == 1, then octets == 0 */
2986 UV none;
2987 UV end = start + span;
972dd592 2988
36eb48b4 2989 if (invlistsvp == NULL) {
786861f5
KW
2990 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
2991 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
2992 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
2993 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
2994 listsvp = hv_fetchs(hv, "LIST", FALSE);
2995 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
2996
2997 bits = SvUV(*bitssvp);
2998 none = SvUV(*nonesvp);
2999 typestr = (U8*)SvPV_nolen(*typesvp);
3000 }
36eb48b4
KW
3001 else {
3002 bits = 1;
3003 none = 0;
3004 }
786861f5 3005 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 3006
b0e3252e 3007 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 3008
979f2922 3009 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 3010 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 3011 (UV)bits);
979f2922
ST
3012 }
3013
84ea5ef6
KW
3014 /* If overflowed, use the max possible */
3015 if (end < start) {
3016 end = UV_MAX;
3017 span = end - start;
3018 }
3019
979f2922 3020 /* create and initialize $swatch */
979f2922 3021 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
3022 swatch = newSV(scur);
3023 SvPOK_on(swatch);
979f2922
ST
3024 s = (U8*)SvPVX(swatch);
3025 if (octets && none) {
0bd48802 3026 const U8* const e = s + scur;
979f2922
ST
3027 while (s < e) {
3028 if (bits == 8)
3029 *s++ = (U8)(none & 0xff);
3030 else if (bits == 16) {
3031 *s++ = (U8)((none >> 8) & 0xff);
3032 *s++ = (U8)( none & 0xff);
3033 }
3034 else if (bits == 32) {
3035 *s++ = (U8)((none >> 24) & 0xff);
3036 *s++ = (U8)((none >> 16) & 0xff);
3037 *s++ = (U8)((none >> 8) & 0xff);
3038 *s++ = (U8)( none & 0xff);
3039 }
3040 }
3041 *s = '\0';
3042 }
3043 else {
3044 (void)memzero((U8*)s, scur + 1);
3045 }
3046 SvCUR_set(swatch, scur);
3047 s = (U8*)SvPVX(swatch);
3048
36eb48b4
KW
3049 if (invlistsvp) { /* If has an inversion list set up use that */
3050 _invlist_populate_swatch(*invlistsvp, start, end, s);
3051 return swatch;
3052 }
3053
3054 /* read $swash->{LIST} */
979f2922
ST
3055 l = (U8*)SvPV(*listsvp, lcur);
3056 lend = l + lcur;
3057 while (l < lend) {
8ed25d53 3058 UV min, max, val, upper;
95543e92
KW
3059 l = swash_scan_list_line(l, lend, &min, &max, &val,
3060 cBOOL(octets), typestr);
319009ee 3061 if (l > lend) {
979f2922
ST
3062 break;
3063 }
3064
972dd592 3065 /* If looking for something beyond this range, go try the next one */
979f2922
ST
3066 if (max < start)
3067 continue;
3068
8ed25d53
KW
3069 /* <end> is generally 1 beyond where we want to set things, but at the
3070 * platform's infinity, where we can't go any higher, we want to
3071 * include the code point at <end> */
3072 upper = (max < end)
3073 ? max
3074 : (max != UV_MAX || end != UV_MAX)
3075 ? end - 1
3076 : end;
3077
979f2922 3078 if (octets) {
35da51f7 3079 UV key;
979f2922
ST
3080 if (min < start) {
3081 if (!none || val < none) {
3082 val += start - min;
3083 }
3084 min = start;
3085 }
8ed25d53 3086 for (key = min; key <= upper; key++) {
979f2922 3087 STRLEN offset;
979f2922
ST
3088 /* offset must be non-negative (start <= min <= key < end) */
3089 offset = octets * (key - start);
3090 if (bits == 8)
3091 s[offset] = (U8)(val & 0xff);
3092 else if (bits == 16) {
3093 s[offset ] = (U8)((val >> 8) & 0xff);
3094 s[offset + 1] = (U8)( val & 0xff);
3095 }
3096 else if (bits == 32) {
3097 s[offset ] = (U8)((val >> 24) & 0xff);
3098 s[offset + 1] = (U8)((val >> 16) & 0xff);
3099 s[offset + 2] = (U8)((val >> 8) & 0xff);
3100 s[offset + 3] = (U8)( val & 0xff);
3101 }
3102
3103 if (!none || val < none)
3104 ++val;
3105 }
3106 }
711a919c 3107 else { /* bits == 1, then val should be ignored */
35da51f7 3108 UV key;
979f2922
ST
3109 if (min < start)
3110 min = start;
6cb05c12 3111
8ed25d53 3112 for (key = min; key <= upper; key++) {
0bd48802 3113 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3114 s[offset >> 3] |= 1 << (offset & 7);
3115 }
3116 }
3117 } /* while */
979f2922 3118
9479a769 3119 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3120 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3121
3122 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3123 * be 0, and their inversion should also be 0, as we don't succeed any
3124 * Unicode property matches for non-Unicode code points */
3125 if (start <= PERL_UNICODE_MAX) {
3126
3127 /* The code below assumes that we never cross the
3128 * Unicode/above-Unicode boundary in a range, as otherwise we would
3129 * have to figure out where to stop flipping the bits. Since this
3130 * boundary is divisible by a large power of 2, and swatches comes
3131 * in small powers of 2, this should be a valid assumption */
3132 assert(start + span - 1 <= PERL_UNICODE_MAX);
3133
507a8485
KW
3134 send = s + scur;
3135 while (s < send) {
3136 *s = ~(*s);
3137 s++;
3138 }
0bda3001 3139 }
77f9f126
KW
3140 }
3141
d73c39c5
KW
3142 /* read $swash->{EXTRAS}
3143 * This code also copied to swash_to_invlist() below */
979f2922
ST
3144 x = (U8*)SvPV(*extssvp, xcur);
3145 xend = x + xcur;
3146 while (x < xend) {
3147 STRLEN namelen;
3148 U8 *namestr;
3149 SV** othersvp;
3150 HV* otherhv;
3151 STRLEN otherbits;
3152 SV **otherbitssvp, *other;
711a919c 3153 U8 *s, *o, *nl;
979f2922
ST
3154 STRLEN slen, olen;
3155
35da51f7 3156 const U8 opc = *x++;
979f2922
ST
3157 if (opc == '\n')
3158 continue;
3159
3160 nl = (U8*)memchr(x, '\n', xend - x);
3161
3162 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3163 if (nl) {
3164 x = nl + 1; /* 1 is length of "\n" */
3165 continue;
3166 }
3167 else {
3168 x = xend; /* to EXTRAS' end at which \n is not found */
3169 break;
3170 }
3171 }
3172
3173 namestr = x;
3174 if (nl) {
3175 namelen = nl - namestr;
3176 x = nl + 1;
3177 }
3178 else {
3179 namelen = xend - namestr;
3180 x = xend;
3181 }
3182
3183 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3184 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3185 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3186 otherbits = (STRLEN)SvUV(*otherbitssvp);
3187 if (bits < otherbits)
5637ef5b
NC
3188 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3189 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3190
3191 /* The "other" swatch must be destroyed after. */
b0e3252e 3192 other = swatch_get(*othersvp, start, span);
979f2922
ST
3193 o = (U8*)SvPV(other, olen);
3194
3195 if (!olen)
b0e3252e 3196 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3197
3198 s = (U8*)SvPV(swatch, slen);
3199 if (bits == 1 && otherbits == 1) {
3200 if (slen != olen)
5637ef5b
NC
3201 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3202 "mismatch, slen=%"UVuf", olen=%"UVuf,
3203 (UV)slen, (UV)olen);
979f2922
ST
3204
3205 switch (opc) {
3206 case '+':
3207 while (slen--)
3208 *s++ |= *o++;
3209 break;
3210 case '!':
3211 while (slen--)
3212 *s++ |= ~*o++;
3213 break;
3214 case '-':
3215 while (slen--)
3216 *s++ &= ~*o++;
3217 break;
3218 case '&':
3219 while (slen--)
3220 *s++ &= *o++;
3221 break;
3222 default:
3223 break;
3224 }
3225 }
711a919c 3226 else {
979f2922
ST
3227 STRLEN otheroctets = otherbits >> 3;
3228 STRLEN offset = 0;
35da51f7 3229 U8* const send = s + slen;
979f2922
ST
3230
3231 while (s < send) {
3232 UV otherval = 0;
3233
3234 if (otherbits == 1) {
3235 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3236 ++offset;
3237 }
3238 else {
3239 STRLEN vlen = otheroctets;
3240 otherval = *o++;
3241 while (--vlen) {
3242 otherval <<= 8;
3243 otherval |= *o++;
3244 }
3245 }
3246
711a919c 3247 if (opc == '+' && otherval)
6f207bd3 3248 NOOP; /* replace with otherval */
979f2922
ST
3249 else if (opc == '!' && !otherval)
3250 otherval = 1;
3251 else if (opc == '-' && otherval)
3252 otherval = 0;
3253 else if (opc == '&' && !otherval)
3254 otherval = 0;
3255 else {
711a919c 3256 s += octets; /* no replacement */
979f2922
ST
3257 continue;
3258 }
3259
3260 if (bits == 8)
3261 *s++ = (U8)( otherval & 0xff);
3262 else if (bits == 16) {
3263 *s++ = (U8)((otherval >> 8) & 0xff);
3264 *s++ = (U8)( otherval & 0xff);
3265 }
3266 else if (bits == 32) {
3267 *s++ = (U8)((otherval >> 24) & 0xff);
3268 *s++ = (U8)((otherval >> 16) & 0xff);
3269 *s++ = (U8)((otherval >> 8) & 0xff);
3270 *s++ = (U8)( otherval & 0xff);
3271 }
3272 }
3273 }
3274 sv_free(other); /* through with it! */
3275 } /* while */
3276 return swatch;
3277}
3278
064c021d 3279HV*
4c2e1131 3280Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3281{
3282
79a2a0e8 3283 /* Subject to change or removal. For use only in regcomp.c and regexec.c
5662e334
KW
3284 * Can't be used on a property that is subject to user override, as it
3285 * relies on the value of SPECIALS in the swash which would be set by
3286 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3287 * for overridden properties
064c021d
KW
3288 *
3289 * Returns a hash which is the inversion and closure of a swash mapping.
3290 * For example, consider the input lines:
3291 * 004B 006B
3292 * 004C 006C
3293 * 212A 006B
3294 *
4a4088c4 3295 * The returned hash would have two keys, the UTF-8 for 006B and the UTF-8 for
064c021d 3296 * 006C. The value for each key is an array. For 006C, the array would
4a4088c4
KW
3297 * have two elements, the UTF-8 for itself, and for 004C. For 006B, there
3298 * would be three elements in its array, the UTF-8 for 006B, 004B and 212A.
064c021d 3299 *
538e84ed
KW
3300 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3301 * keys are only code points that are folded-to, so it isn't a full closure.
3302 *
064c021d
KW
3303 * Essentially, for any code point, it gives all the code points that map to
3304 * it, or the list of 'froms' for that point.
3305 *
5662e334
KW
3306 * Currently it ignores any additions or deletions from other swashes,
3307 * looking at just the main body of the swash, and if there are SPECIALS
3308 * in the swash, at that hash
3309 *
3310 * The specials hash can be extra code points, and most likely consists of
3311 * maps from single code points to multiple ones (each expressed as a string
4a4088c4 3312 * of UTF-8 characters). This function currently returns only 1-1 mappings.
5662e334
KW
3313 * However consider this possible input in the specials hash:
3314 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3315 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3316 *
3317 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3318 * currently handle. But it also means that FB05 and FB06 are equivalent in
3319 * a 1-1 mapping which we should handle, and this relationship may not be in
3320 * the main table. Therefore this function examines all the multi-char
74894415
KW
3321 * sequences and adds the 1-1 mappings that come out of that.
3322 *
3323 * XXX This function was originally intended to be multipurpose, but its
3324 * only use is quite likely to remain for constructing the inversion of
3325 * the CaseFolding (//i) property. If it were more general purpose for
3326 * regex patterns, it would have to do the FB05/FB06 game for simple folds,
3327 * because certain folds are prohibited under /iaa and /il. As an example,
3328 * in Unicode 3.0.1 both U+0130 and U+0131 fold to 'i', and hence are both
3329 * equivalent under /i. But under /iaa and /il, the folds to 'i' are
3330 * prohibited, so we would not figure out that they fold to each other.
3331 * Code could be written to automatically figure this out, similar to the
3332 * code that does this for multi-character folds, but this is the only case
3333 * where something like this is ever likely to happen, as all the single
3334 * char folds to The 0-255 range are now quite settled. Instead there is a
3335 * little special code that is compiled only for this Unicode version. This
3336 * is smaller and didn't require much coding time to do. But this makes
3337 * this routine strongly tied to being used just for CaseFolding. If ever
3338 * it should be generalized, this would have to be fixed */
064c021d
KW
3339
3340 U8 *l, *lend;
3341 STRLEN lcur;
3342 HV *const hv = MUTABLE_HV(SvRV(swash));
3343
923b6d4e
KW
3344 /* The string containing the main body of the table. This will have its
3345 * assertion fail if the swash has been converted to its inversion list */
064c021d
KW
3346 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3347
3348 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3349 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3350 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3351 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3352 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3353 const STRLEN bits = SvUV(*bitssvp);
3354 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3355 const UV none = SvUV(*nonesvp);
5662e334 3356 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
064c021d
KW
3357
3358 HV* ret = newHV();
3359
3360 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3361
3362 /* Must have at least 8 bits to get the mappings */
3363 if (bits != 8 && bits != 16 && bits != 32) {
3364 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3365 (UV)bits);
3366 }
3367
5662e334
KW
3368 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3369 mapping to more than one character */
3370
3371 /* Construct an inverse mapping hash for the specials */
3372 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3373 HV * specials_inverse = newHV();
3374 char *char_from; /* the lhs of the map */
3375 I32 from_len; /* its byte length */
3376 char *char_to; /* the rhs of the map */
3377 I32 to_len; /* its byte length */
3378 SV *sv_to; /* and in a sv */
3379 AV* from_list; /* list of things that map to each 'to' */
3380
3381 hv_iterinit(specials_hv);
3382
4a4088c4
KW
3383 /* The keys are the characters (in UTF-8) that map to the corresponding
3384 * UTF-8 string value. Iterate through the list creating the inverse
5662e334
KW
3385 * list. */
3386 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3387 SV** listp;
3388 if (! SvPOK(sv_to)) {
5637ef5b
NC
3389 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3390 "unexpectedly is not a string, flags=%lu",
3391 (unsigned long)SvFLAGS(sv_to));
5662e334 3392 }
4b88fb76 3393 /*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
3394
3395 /* Each key in the inverse list is a mapped-to value, and the key's
4a4088c4 3396 * hash value is a list of the strings (each in UTF-8) that map to
5662e334
KW
3397 * it. Those strings are all one character long */
3398 if ((listp = hv_fetch(specials_inverse,
3399 SvPVX(sv_to),
3400 SvCUR(sv_to), 0)))
3401 {
3402 from_list = (AV*) *listp;
3403 }
3404 else { /* No entry yet for it: create one */
3405 from_list = newAV();
3406 if (! hv_store(specials_inverse,
3407 SvPVX(sv_to),
3408 SvCUR(sv_to),
3409 (SV*) from_list, 0))
3410 {
3411 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3412 }
3413 }
3414
3415 /* Here have the list associated with this 'to' (perhaps newly
3416 * created and empty). Just add to it. Note that we ASSUME that
3417 * the input is guaranteed to not have duplications, so we don't
3418 * check for that. Duplications just slow down execution time. */
3419 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3420 }
3421
3422 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3423 * it looking for cases like the FB05/FB06 examples above. There would
3424 * be an entry in the hash like
3425 * 'st' => [ FB05, FB06 ]
3426 * In this example we will create two lists that get stored in the
3427 * returned hash, 'ret':
3428 * FB05 => [ FB05, FB06 ]
3429 * FB06 => [ FB05, FB06 ]
3430 *
3431 * Note that there is nothing to do if the array only has one element.
3432 * (In the normal 1-1 case handled below, we don't have to worry about
3433 * two lists, as everything gets tied to the single list that is
3434 * generated for the single character 'to'. But here, we are omitting
3435 * that list, ('st' in the example), so must have multiple lists.) */
3436 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3437 &char_to, &to_len)))
3438 {
b9f2b683 3439 if (av_tindex(from_list) > 0) {
c70927a6 3440 SSize_t i;
5662e334
KW
3441
3442 /* We iterate over all combinations of i,j to place each code
3443 * point on each list */
b9f2b683 3444 for (i = 0; i <= av_tindex(from_list); i++) {
c70927a6 3445 SSize_t j;
5662e334
KW
3446 AV* i_list = newAV();
3447 SV** entryp = av_fetch(from_list, i, FALSE);
3448 if (entryp == NULL) {
3449 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3450 }
3451 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3452 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3453 }
3454 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3455 (SV*) i_list, FALSE))
3456 {
3457 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3458 }
3459
538e84ed 3460 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
b9f2b683 3461 for (j = 0; j <= av_tindex(from_list); j++) {
5662e334
KW
3462 entryp = av_fetch(from_list, j, FALSE);
3463 if (entryp == NULL) {
3464 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3465 }
3466
3467 /* When i==j this adds itself to the list */
4b88fb76
KW
3468 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3469 (U8*) SvPVX(*entryp),
3470 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3471 0)));
4637d003 3472 /*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
3473 }
3474 }
3475 }
3476 }
3477 SvREFCNT_dec(specials_inverse); /* done with it */
3478 } /* End of specials */
3479
064c021d 3480 /* read $swash->{LIST} */
74894415
KW
3481
3482#if UNICODE_MAJOR_VERSION == 3 \
3483 && UNICODE_DOT_VERSION == 0 \
3484 && UNICODE_DOT_DOT_VERSION == 1
3485
3486 /* For this version only U+130 and U+131 are equivalent under qr//i. Add a
3487 * rule so that things work under /iaa and /il */
3488
3489 SV * mod_listsv = sv_mortalcopy(*listsvp);
3490 sv_catpv(mod_listsv, "130\t130\t131\n");
3491 l = (U8*)SvPV(mod_listsv, lcur);
3492
3493#else
3494
064c021d 3495 l = (U8*)SvPV(*listsvp, lcur);
74894415
KW
3496
3497#endif
3498
064c021d
KW
3499 lend = l + lcur;
3500
3501 /* Go through each input line */
3502 while (l < lend) {
3503 UV min, max, val;
3504 UV inverse;
95543e92
KW
3505 l = swash_scan_list_line(l, lend, &min, &max, &val,
3506 cBOOL(octets), typestr);
064c021d
KW
3507 if (l > lend) {
3508 break;
3509 }
3510
3511 /* Each element in the range is to be inverted */
3512 for (inverse = min; inverse <= max; inverse++) {
3513 AV* list;
064c021d
KW
3514 SV** listp;
3515 IV i;
3516 bool found_key = FALSE;
5662e334 3517 bool found_inverse = FALSE;
064c021d
KW
3518
3519 /* The key is the inverse mapping */
3520 char key[UTF8_MAXBYTES+1];
c80e42f3 3521 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
064c021d
KW
3522 STRLEN key_len = key_end - key;
3523
064c021d
KW
3524 /* Get the list for the map */
3525 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3526 list = (AV*) *listp;
3527 }
3528 else { /* No entry yet for it: create one */
3529 list = newAV();
3530 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3531 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3532 }
3533 }
3534
5662e334
KW
3535 /* Look through list to see if this inverse mapping already is
3536 * listed, or if there is a mapping to itself already */
b9f2b683 3537 for (i = 0; i <= av_tindex(list); i++) {
064c021d
KW
3538 SV** entryp = av_fetch(list, i, FALSE);
3539 SV* entry;
414db8a1 3540 UV uv;
064c021d
KW
3541 if (entryp == NULL) {
3542 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3543 }
3544 entry = *entryp;
414db8a1
DM
3545 uv = SvUV(entry);
3546 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, uv));*/
3547 if (uv == val) {
064c021d 3548 found_key = TRUE;
5662e334 3549 }
414db8a1 3550 if (uv == inverse) {
5662e334
KW
3551 found_inverse = TRUE;
3552 }
3553
3554 /* No need to continue searching if found everything we are
3555 * looking for */
3556 if (found_key && found_inverse) {
064c021d
KW
3557 break;
3558 }
3559 }
56ca34ca
KW
3560
3561 /* Make sure there is a mapping to itself on the list */
064c021d 3562 if (! found_key) {
d397ff6a 3563 av_push(list, newSVuv(val));
4637d003 3564 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
064c021d
KW
3565 }
3566
3567
3568 /* Simply add the value to the list */
5662e334
KW
3569 if (! found_inverse) {
3570 av_push(list, newSVuv(inverse));
4637d003 3571 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
5662e334 3572 }
064c021d 3573
b0e3252e 3574 /* swatch_get() increments the value of val for each element in the
064c021d
KW
3575 * range. That makes more compact tables possible. You can
3576 * express the capitalization, for example, of all consecutive
3577 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3578 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
bd3f2f94 3579 * and it's not documented; it appears to be used only in
b0e3252e 3580 * implementing tr//; I copied the semantics from swatch_get(), just
bd3f2f94 3581 * in case */
064c021d
KW
3582 if (!none || val < none) {
3583 ++val;
3584 }
3585 }
3586 }
3587
3588 return ret;
3589}
3590
a25abddc 3591SV*
d764b54e
KW
3592Perl__swash_to_invlist(pTHX_ SV* const swash)
3593{
3594
ed92f1b3
KW
3595 /* Subject to change or removal. For use only in one place in regcomp.c.
3596 * Ownership is given to one reference count in the returned SV* */
d764b54e
KW
3597
3598 U8 *l, *lend;
3599 char *loc;
3600 STRLEN lcur;
3601 HV *const hv = MUTABLE_HV(SvRV(swash));
3602 UV elements = 0; /* Number of elements in the inversion list */
b443038a 3603 U8 empty[] = "";
923b6d4e
KW
3604 SV** listsvp;
3605 SV** typesvp;
3606 SV** bitssvp;
3607 SV** extssvp;
3608 SV** invert_it_svp;
d764b54e 3609
923b6d4e
KW
3610 U8* typestr;
3611 STRLEN bits;
3612 STRLEN octets; /* if bits == 1, then octets == 0 */
d73c39c5
KW
3613 U8 *x, *xend;
3614 STRLEN xcur;
d764b54e 3615
a25abddc 3616 SV* invlist;
d764b54e 3617
b81740c0
KW
3618 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3619
923b6d4e
KW
3620 /* If not a hash, it must be the swash's inversion list instead */
3621 if (SvTYPE(hv) != SVt_PVHV) {
ed92f1b3 3622 return SvREFCNT_inc_simple_NN((SV*) hv);
923b6d4e
KW
3623 }
3624