This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
utf8.h: Combine ASCII and EBCDIC defines into one
[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)";
760c7c2f
KW
38static const char cp_above_legal_max[] =
39 "It is deprecated to use code point 0x%"UVXf"; the permissible max is 0x%"UVXf"";
40
41#define MAX_NON_DEPRECATED_CP (IV_MAX)
901b21bf 42
48ef279e 43/*
ccfc67b7 44=head1 Unicode Support
7fefc6c1 45These are various utility functions for manipulating UTF8-encoded
72d33970 46strings. For the uninitiated, this is a method of representing arbitrary
61296642 47Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
48characters in the ASCII range are unmodified, and a zero byte never appears
49within non-zero characters.
166f8a29 50
eaf7a4d2
CS
51=cut
52*/
53
54/*
7bbfa158 55=for apidoc is_invariant_string
eaf7a4d2 56
b8ed1ca9
KW
57Returns true iff the first C<len> bytes of the string C<s> are the same
58regardless of the UTF-8 encoding of the string (or UTF-EBCDIC encoding on
59EBCDIC machines). That is, if they are UTF-8 invariant. On ASCII-ish
60machines, all the ASCII characters and only the ASCII characters fit this
7bbfa158 61definition. On EBCDIC machines, the ASCII-range characters are invariant, but
b8ed1ca9
KW
62so also are the C1 controls and C<\c?> (which isn't in the ASCII range on
63EBCDIC).
eaf7a4d2 64
75200dff
KW
65If C<len> is 0, it will be calculated using C<strlen(s)>, (which means if you
66use this option, that C<s> can't have embedded C<NUL> characters and has to
67have a terminating C<NUL> byte).
9f7e3d64 68
a1433954 69See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
eaf7a4d2
CS
70
71=cut
72*/
73
74bool
7bbfa158 75Perl_is_invariant_string(const U8 *s, STRLEN len)
eaf7a4d2
CS
76{
77 const U8* const send = s + (len ? len : strlen((const char *)s));
78 const U8* x = s;
79
7bbfa158 80 PERL_ARGS_ASSERT_IS_INVARIANT_STRING;
eaf7a4d2
CS
81
82 for (; x < send; ++x) {
83 if (!UTF8_IS_INVARIANT(*x))
84 break;
85 }
86
87 return x == send;
88}
89
90/*
378516de 91=for apidoc uvoffuni_to_utf8_flags
eebe1485 92
a27992cc 93THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
94Instead, B<Almost all code should use L</uvchr_to_utf8> or
95L</uvchr_to_utf8_flags>>.
a27992cc 96
de69f3af
KW
97This function is like them, but the input is a strict Unicode
98(as opposed to native) code point. Only in very rare circumstances should code
99not be using the native code point.
949cf498 100
efa9cd84 101For details, see the description for L</uvchr_to_utf8_flags>.
949cf498 102
eebe1485
SC
103=cut
104*/
105
dfe13c55 106U8 *
378516de 107Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 108{
378516de 109 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 110
2d1545e5 111 if (OFFUNI_IS_INVARIANT(uv)) {
d9432125
KW
112 *d++ = (U8) LATIN1_TO_NATIVE(uv);
113 return d;
114 }
115
979f77b6 116 /* The first problematic code point is the first surrogate */
760c7c2f 117 if (uv >= UNICODE_SURROGATE_FIRST) {
949cf498
KW
118 if (UNICODE_IS_SURROGATE(uv)) {
119 if (flags & UNICODE_WARN_SURROGATE) {
8457b38f 120 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
949cf498
KW
121 "UTF-16 surrogate U+%04"UVXf, uv);
122 }
123 if (flags & UNICODE_DISALLOW_SURROGATE) {
124 return NULL;
125 }
126 }
127 else if (UNICODE_IS_SUPER(uv)) {
760c7c2f
KW
128 if ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
129 && ckWARN_d(WARN_DEPRECATED))
130 {
131 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
132 cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP);
133 }
0bcdd8f6 134 if ( (flags & UNICODE_WARN_SUPER)
d35f2ca5 135 || (UNICODE_IS_ABOVE_31_BIT(uv) && (flags & UNICODE_WARN_ABOVE_31_BIT)))
0bcdd8f6
KW
136 {
137 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
138
139 /* Choose the more dire applicable warning */
d35f2ca5 140 (UNICODE_IS_ABOVE_31_BIT(uv))
0bcdd8f6
KW
141 ? "Code point 0x%"UVXf" is not Unicode, and not portable"
142 : "Code point 0x%"UVXf" is not Unicode, may not be portable",
143 uv);
949cf498
KW
144 }
145 if (flags & UNICODE_DISALLOW_SUPER
d35f2ca5 146 || (UNICODE_IS_ABOVE_31_BIT(uv) && (flags & UNICODE_DISALLOW_ABOVE_31_BIT)))
949cf498
KW
147 {
148 return NULL;
149 }
150 }
151 else if (UNICODE_IS_NONCHAR(uv)) {
152 if (flags & UNICODE_WARN_NONCHAR) {
8457b38f 153 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
ba707cdc 154 "Unicode non-character U+%04"UVXf" is not recommended for open interchange",
949cf498
KW
155 uv);
156 }
157 if (flags & UNICODE_DISALLOW_NONCHAR) {
158 return NULL;
159 }
160 }
507b9800 161 }
d9432125 162
2d331972 163#if defined(EBCDIC)
d9432125 164 {
5aaebcb3 165 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
166 U8 *p = d+len-1;
167 while (p > d) {
bc3632a8 168 *p-- = (U8) I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
169 uv >>= UTF_ACCUMULATION_SHIFT;
170 }
bc3632a8 171 *p = (U8) I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
172 return d+len;
173 }
174#else /* Non loop style */
a0ed51b3 175 if (uv < 0x800) {
eb160463
GS
176 *d++ = (U8)(( uv >> 6) | 0xc0);
177 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
178 return d;
179 }
180 if (uv < 0x10000) {
eb160463
GS
181 *d++ = (U8)(( uv >> 12) | 0xe0);
182 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
183 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
184 return d;
185 }
186 if (uv < 0x200000) {
eb160463
GS
187 *d++ = (U8)(( uv >> 18) | 0xf0);
188 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
189 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
190 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
191 return d;
192 }
193 if (uv < 0x4000000) {
eb160463
GS
194 *d++ = (U8)(( uv >> 24) | 0xf8);
195 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
196 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
197 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
198 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
199 return d;
200 }
201 if (uv < 0x80000000) {
eb160463
GS
202 *d++ = (U8)(( uv >> 30) | 0xfc);
203 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
204 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
205 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
206 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
207 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
208 return d;
209 }
6588300d 210#ifdef UTF8_QUAD_MAX
d7578b48 211 if (uv < UTF8_QUAD_MAX)
a0ed51b3
LW
212#endif
213 {
eb160463
GS
214 *d++ = 0xfe; /* Can't match U+FEFF! */
215 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
217 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
218 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
219 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
220 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
221 return d;
222 }
6588300d 223#ifdef UTF8_QUAD_MAX
a0ed51b3 224 {
eb160463
GS
225 *d++ = 0xff; /* Can't match U+FFFE! */
226 *d++ = 0x80; /* 6 Reserved bits */
227 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
228 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
229 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
230 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
231 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
232 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
233 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
234 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
235 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
236 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
237 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
238 return d;
239 }
240#endif
537124e4 241#endif /* Non loop style */
a0ed51b3 242}
646ca15d 243/*
07693fe6
KW
244=for apidoc uvchr_to_utf8
245
bcb1a2d4 246Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 247of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
248C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
249the byte after the end of the new character. In other words,
07693fe6
KW
250
251 d = uvchr_to_utf8(d, uv);
252
253is the recommended wide native character-aware way of saying
254
255 *(d++) = uv;
256
760c7c2f
KW
257This function accepts any UV as input, but very high code points (above
258C<IV_MAX> on the platform) will raise a deprecation warning. This is
259typically 0x7FFF_FFFF in a 32-bit word.
260
261It is possible to forbid or warn on non-Unicode code points, or those that may
262be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 263
07693fe6
KW
264=cut
265*/
266
de69f3af
KW
267/* This is also a macro */
268PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
269
07693fe6
KW
270U8 *
271Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
272{
de69f3af 273 return uvchr_to_utf8(d, uv);
07693fe6
KW
274}
275
de69f3af
KW
276/*
277=for apidoc uvchr_to_utf8_flags
278
279Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 280of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
281C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
282the byte after the end of the new character. In other words,
de69f3af
KW
283
284 d = uvchr_to_utf8_flags(d, uv, flags);
285
286or, in most cases,
287
288 d = uvchr_to_utf8_flags(d, uv, 0);
289
290This is the Unicode-aware way of saying
291
292 *(d++) = uv;
293
760c7c2f
KW
294If C<flags> is 0, this function accepts any UV as input, but very high code
295points (above C<IV_MAX> for the platform) will raise a deprecation warning.
296This is typically 0x7FFF_FFFF in a 32-bit word.
297
298Specifying C<flags> can further restrict what is allowed and not warned on, as
299follows:
de69f3af 300
796b6530 301If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
302the function will raise a warning, provided UTF8 warnings are enabled. If
303instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
304NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 305
760c7c2f
KW
306Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
307affect how the function handles a Unicode non-character.
93e6dbd6 308
760c7c2f
KW
309And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
310affect the handling of code points that are above the Unicode maximum of
3110x10FFFF. Languages other than Perl may not be able to accept files that
312contain these.
93e6dbd6
KW
313
314The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
315the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
316three DISALLOW flags.
317
ab8e6d41
KW
318Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
319so using them is more problematic than other above-Unicode code points. Perl
320invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
321likely that non-Perl languages will not be able to read files that contain
322these that written by the perl interpreter; nor would Perl understand files
323written by something that uses a different extension. For these reasons, there
324is a separate set of flags that can warn and/or disallow these extremely high
325code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
326C<UNICODE_WARN_ABOVE_31_BIT> and C<UNICODE_DISALLOW_ABOVE_31_BIT> flags. These
327are entirely independent from the deprecation warning for code points above
328C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
329code point that needs more than 31 bits to represent. When that happens,
330effectively the C<UNICODE_DISALLOW_ABOVE_31_BIT> flag will always be set on
33132-bit machines. (Of course C<UNICODE_DISALLOW_SUPER> will treat all
ab8e6d41
KW
332above-Unicode code points, including these, as malformations; and
333C<UNICODE_WARN_SUPER> warns on these.)
334
335On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
336extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
337than on ASCII. Prior to that, code points 2**31 and higher were simply
338unrepresentable, and a different, incompatible method was used to represent
339code points between 2**30 and 2**31 - 1. The flags C<UNICODE_WARN_ABOVE_31_BIT>
340and C<UNICODE_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
341platforms, warning and disallowing 2**31 and higher.
de69f3af 342
de69f3af
KW
343=cut
344*/
345
346/* This is also a macro */
347PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
348
07693fe6
KW
349U8 *
350Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
351{
de69f3af 352 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
353}
354
355/*
87cea99e 356=for apidoc is_utf8_string
6662521e 357
a1433954 358Returns true if the first C<len> bytes of string C<s> form a valid
9f7e3d64 359UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
75200dff
KW
360using C<strlen(s)> (which means if you use this option, that C<s> can't have
361embedded C<NUL> characters and has to have a terminating C<NUL> byte). Note
362that all characters being ASCII constitute 'a valid UTF-8 string'.
6662521e 363
7bbfa158 364See also L</is_invariant_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
768c67ee 365
6662521e
GS
366=cut
367*/
368
8e84507e 369bool
668b6d8d 370Perl_is_utf8_string(const U8 *s, STRLEN len)
6662521e 371{
35da51f7 372 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 373 const U8* x = s;
067a85ef 374
7918f24d 375 PERL_ARGS_ASSERT_IS_UTF8_STRING;
1aa99e6b 376
6662521e 377 while (x < send) {
6302f837
KW
378 STRLEN len = isUTF8_CHAR(x, send);
379 if (UNLIKELY(! len)) {
380 return FALSE;
381 }
382 x += len;
6662521e 383 }
768c67ee 384
067a85ef 385 return TRUE;
6662521e
GS
386}
387
67e989fb 388/*
814fafa7
NC
389Implemented as a macro in utf8.h
390
87cea99e 391=for apidoc is_utf8_string_loc
814fafa7 392
a1433954
KW
393Like L</is_utf8_string> but stores the location of the failure (in the
394case of "utf8ness failure") or the location C<s>+C<len> (in the case of
814fafa7
NC
395"utf8ness success") in the C<ep>.
396
a1433954 397See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
814fafa7 398
87cea99e 399=for apidoc is_utf8_string_loclen
81cd54e3 400
a1433954
KW
401Like L</is_utf8_string>() but stores the location of the failure (in the
402case of "utf8ness failure") or the location C<s>+C<len> (in the case of
768c67ee
JH
403"utf8ness success") in the C<ep>, and the number of UTF-8
404encoded characters in the C<el>.
405
a1433954 406See also L</is_utf8_string_loc>() and L</is_utf8_string>().
81cd54e3
JH
407
408=cut
409*/
410
411bool
668b6d8d 412Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
81cd54e3 413{
35da51f7 414 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 415 const U8* x = s;
3ebfea28 416 STRLEN outlen = 0;
7918f24d
NC
417
418 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
81cd54e3 419
81cd54e3 420 while (x < send) {
6302f837
KW
421 STRLEN len = isUTF8_CHAR(x, send);
422 if (UNLIKELY(! len)) {
423 goto out;
424 }
425 x += len;
426 outlen++;
81cd54e3 427 }
768c67ee
JH
428
429 out:
3ebfea28
AL
430 if (el)
431 *el = outlen;
432
768c67ee
JH
433 if (ep)
434 *ep = x;
3ebfea28 435 return (x == send);
81cd54e3
JH
436}
437
438/*
768c67ee 439
de69f3af 440=for apidoc utf8n_to_uvchr
378516de
KW
441
442THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 443Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 444
9041c2e3 445Bottom level UTF-8 decode routine.
de69f3af 446Returns the native code point value of the first character in the string C<s>,
746afd53
KW
447which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
448C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
449the length, in bytes, of that character.
949cf498
KW
450
451The value of C<flags> determines the behavior when C<s> does not point to a
452well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
453zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
454next possible position in C<s> that could begin a non-malformed character.
455Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
456
457Various ALLOW flags can be set in C<flags> to allow (and not warn on)
458individual types of malformations, such as the sequence being overlong (that
459is, when there is a shorter sequence that can express the same code point;
460overlong sequences are expressly forbidden in the UTF-8 standard due to
461potential security issues). Another malformation example is the first byte of
462a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
463flags. For allowed 0 length strings, this function returns 0; for allowed
464overlong sequences, the computed code point is returned; for all other allowed
465malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
466determinable reasonable value.
949cf498 467
796b6530 468The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
469flags) malformation is found. If this flag is set, the routine assumes that
470the caller will raise a warning, and this function will silently just set
d088425d
KW
471C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
472
75200dff 473Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 474character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
d088425d 475in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
75200dff
KW
476first byte of C<s> is 0 as well. If so, the input was a C<NUL>; if not, the
477input had an error.
949cf498
KW
478
479Certain code points are considered problematic. These are Unicode surrogates,
746afd53 480Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 481By default these are considered regular code points, but certain situations
5eafe189 482warrant special handling for them. If C<flags> contains
796b6530
KW
483C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all three classes are treated as
484malformations and handled as such. The flags C<UTF8_DISALLOW_SURROGATE>,
485C<UTF8_DISALLOW_NONCHAR>, and C<UTF8_DISALLOW_SUPER> (meaning above the legal
486Unicode maximum) can be set to disallow these categories individually.
487
488The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
489C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
490raised for their respective categories, but otherwise the code points are
491considered valid (not malformations). To get a category to both be treated as
492a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 493(But note that warnings are not raised if lexically disabled nor if
796b6530 494C<UTF8_CHECK_ONLY> is also specified.)
949cf498 495
760c7c2f
KW
496It is now deprecated to have very high code points (above C<IV_MAX> on the
497platforms) and this function will raise a deprecation warning for these (unless
498such warnings are turned off). This value, is typically 0x7FFF_FFFF (2**31 -1)
499in a 32-bit word.
ab8e6d41
KW
500
501Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
502so using them is more problematic than other above-Unicode code points. Perl
503invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
504likely that non-Perl languages will not be able to read files that contain
505these that written by the perl interpreter; nor would Perl understand files
506written by something that uses a different extension. For these reasons, there
507is a separate set of flags that can warn and/or disallow these extremely high
508code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
509C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These
510are entirely independent from the deprecation warning for code points above
511C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
512code point that needs more than 31 bits to represent. When that happens,
513effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on
51432-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all
ab8e6d41
KW
515above-Unicode code points, including these, as malformations; and
516C<UTF8_WARN_SUPER> warns on these.)
517
518On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
519extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
520than on ASCII. Prior to that, code points 2**31 and higher were simply
521unrepresentable, and a different, incompatible method was used to represent
522code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT>
523and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
524platforms, warning and disallowing 2**31 and higher.
949cf498
KW
525
526All other code points corresponding to Unicode characters, including private
527use and those yet to be assigned, are never considered malformed and never
528warn.
67e989fb 529
37607a96
PK
530=cut
531*/
67e989fb 532
a0ed51b3 533UV
de69f3af 534Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 535{
d4c19fe8 536 const U8 * const s0 = s;
eb83ed87 537 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 538 U8 * send;
eb83ed87
KW
539 UV uv = *s;
540 STRLEN expectlen;
949cf498 541 SV* sv = NULL;
eb83ed87
KW
542 UV outlier_ret = 0; /* return value when input is in error or problematic
543 */
544 UV pack_warn = 0; /* Save result of packWARN() for later */
545 bool unexpected_non_continuation = FALSE;
546 bool overflowed = FALSE;
2f8f112e 547 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 548
eb83ed87 549 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 550
de69f3af 551 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
a0dbb045 552
eb83ed87
KW
553 /* The order of malformation tests here is important. We should consume as
554 * few bytes as possible in order to not skip any valid character. This is
555 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
556 * http://unicode.org/reports/tr36 for more discussion as to why. For
557 * example, once we've done a UTF8SKIP, we can tell the expected number of
558 * bytes, and could fail right off the bat if the input parameters indicate
559 * that there are too few available. But it could be that just that first
560 * byte is garbled, and the intended character occupies fewer bytes. If we
561 * blindly assumed that the first byte is correct, and skipped based on
562 * that number, we could skip over a valid input character. So instead, we
563 * always examine the sequence byte-by-byte.
564 *
565 * We also should not consume too few bytes, otherwise someone could inject
566 * things. For example, an input could be deliberately designed to
567 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 568 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
569 * which is actually part of of the overflowing sequence), that could look
570 * legitimate to the caller, which could discard the initial partial
571 * sequence and process the rest, inappropriately */
572
573 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 574 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
575 if (retlen) {
576 *retlen = 0;
577 }
a0dbb045 578
eb83ed87
KW
579 if (flags & UTF8_ALLOW_EMPTY) {
580 return 0;
581 }
582 if (! (flags & UTF8_CHECK_ONLY)) {
583 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
584 }
0c443dc2
JH
585 goto malformed;
586 }
587
eb83ed87
KW
588 expectlen = UTF8SKIP(s);
589
590 /* A well-formed UTF-8 character, as the vast majority of calls to this
591 * function will be for, has this expected length. For efficiency, set
592 * things up here to return it. It will be overriden only in those rare
593 * cases where a malformation is found */
594 if (retlen) {
595 *retlen = expectlen;
596 }
597
598 /* An invariant is trivially well-formed */
1d72bdf6 599 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 600 return uv;
a0ed51b3 601 }
67e989fb 602
eb83ed87 603 /* A continuation character can't start a valid sequence */
b5b9af04 604 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
605 if (flags & UTF8_ALLOW_CONTINUATION) {
606 if (retlen) {
607 *retlen = 1;
608 }
609 return UNICODE_REPLACEMENT;
610 }
ba210ebe 611
eb83ed87
KW
612 if (! (flags & UTF8_CHECK_ONLY)) {
613 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
614 }
615 curlen = 1;
ba210ebe
JH
616 goto malformed;
617 }
9041c2e3 618
dcd27b3c
KW
619 /* Here is not a continuation byte, nor an invariant. The only thing left
620 * is a start byte (possibly for an overlong) */
621
1d72bdf6 622#ifdef EBCDIC
bc3632a8 623 uv = NATIVE_UTF8_TO_I8(uv);
1d72bdf6
NIS
624#endif
625
eb83ed87
KW
626 /* Remove the leading bits that indicate the number of bytes in the
627 * character's whole UTF-8 sequence, leaving just the bits that are part of
628 * the value */
629 uv &= UTF_START_MASK(expectlen);
ba210ebe 630
eb83ed87
KW
631 /* Now, loop through the remaining bytes in the character's sequence,
632 * accumulating each into the working value as we go. Be sure to not look
633 * past the end of the input string */
0b8d30e8
KW
634 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
635
eb83ed87 636 for (s = s0 + 1; s < send; s++) {
b5b9af04 637 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
638 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
639
640 /* The original implementors viewed this malformation as more
641 * serious than the others (though I, khw, don't understand
642 * why, since other malformations also give very very wrong
643 * results), so there is no way to turn off checking for it.
644 * Set a flag, but keep going in the loop, so that we absorb
645 * the rest of the bytes that comprise the character. */
646 overflowed = TRUE;
647 overflow_byte = *s; /* Save for warning message's use */
648 }
8850bf83 649 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
650 }
651 else {
652 /* Here, found a non-continuation before processing all expected
653 * bytes. This byte begins a new character, so quit, even if
654 * allowing this malformation. */
655 unexpected_non_continuation = TRUE;
656 break;
657 }
658 } /* End of loop through the character's bytes */
659
660 /* Save how many bytes were actually in the character */
661 curlen = s - s0;
662
663 /* The loop above finds two types of malformations: non-continuation and/or
664 * overflow. The non-continuation malformation is really a too-short
665 * malformation, as it means that the current character ended before it was
666 * expected to (being terminated prematurely by the beginning of the next
667 * character, whereas in the too-short malformation there just are too few
668 * bytes available to hold the character. In both cases, the check below
669 * that we have found the expected number of bytes would fail if executed.)
670 * Thus the non-continuation malformation is really unnecessary, being a
671 * subset of the too-short malformation. But there may be existing
672 * applications that are expecting the non-continuation type, so we retain
673 * it, and return it in preference to the too-short malformation. (If this
674 * code were being written from scratch, the two types might be collapsed
675 * into one.) I, khw, am also giving priority to returning the
676 * non-continuation and too-short malformations over overflow when multiple
677 * ones are present. I don't know of any real reason to prefer one over
678 * the other, except that it seems to me that multiple-byte errors trumps
679 * errors from a single byte */
b5b9af04 680 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
681 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
682 if (! (flags & UTF8_CHECK_ONLY)) {
683 if (curlen == 1) {
684 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
685 }
686 else {
687 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
688 }
689 }
eb83ed87
KW
690 goto malformed;
691 }
692 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
693
694 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
695 * as what the original expectations were. */
696 do_overlong_test = FALSE;
eb83ed87
KW
697 if (retlen) {
698 *retlen = curlen;
699 }
700 }
b5b9af04 701 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
702 if (! (flags & UTF8_ALLOW_SHORT)) {
703 if (! (flags & UTF8_CHECK_ONLY)) {
704 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 705 }
eb83ed87
KW
706 goto malformed;
707 }
708 uv = UNICODE_REPLACEMENT;
2f8f112e 709 do_overlong_test = FALSE;
eb83ed87
KW
710 if (retlen) {
711 *retlen = curlen;
712 }
713 }
714
b5b9af04 715 if (UNLIKELY(overflowed)) {
eb83ed87 716 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 717 goto malformed;
eb83ed87 718 }
eb83ed87 719
2f8f112e 720 if (do_overlong_test
5aaebcb3 721 && expectlen > (STRLEN) OFFUNISKIP(uv)
2f8f112e
KW
722 && ! (flags & UTF8_ALLOW_LONG))
723 {
eb83ed87
KW
724 /* The overlong malformation has lower precedence than the others.
725 * Note that if this malformation is allowed, we return the actual
726 * value, instead of the replacement character. This is because this
727 * value is actually well-defined. */
728 if (! (flags & UTF8_CHECK_ONLY)) {
5aaebcb3 729 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
730 }
731 goto malformed;
732 }
733
1a89bb6c 734 /* Here, the input is considered to be well-formed, but it still could be a
eb83ed87
KW
735 * problematic code point that is not allowed by the input parameters. */
736 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
760c7c2f
KW
737 && ((flags & ( UTF8_DISALLOW_NONCHAR
738 |UTF8_DISALLOW_SURROGATE
739 |UTF8_DISALLOW_SUPER
740 |UTF8_DISALLOW_ABOVE_31_BIT
741 |UTF8_WARN_NONCHAR
742 |UTF8_WARN_SURROGATE
743 |UTF8_WARN_SUPER
744 |UTF8_WARN_ABOVE_31_BIT))
745 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
746 && ckWARN_d(WARN_DEPRECATED))))
eb83ed87 747 {
949cf498 748 if (UNICODE_IS_SURROGATE(uv)) {
ea5ced44
KW
749
750 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
751 * generation of the sv, since no warnings are raised under CHECK */
eb83ed87 752 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
54f4afef 753 && ckWARN_d(WARN_SURROGATE))
eb83ed87 754 {
111d382d 755 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
54f4afef 756 pack_warn = packWARN(WARN_SURROGATE);
949cf498
KW
757 }
758 if (flags & UTF8_DISALLOW_SURROGATE) {
759 goto disallowed;
760 }
761 }
949cf498 762 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87 763 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
ea5ced44 764 && ckWARN_d(WARN_NON_UNICODE))
eb83ed87 765 {
0bcdd8f6
KW
766 sv = sv_2mortal(Perl_newSVpvf(aTHX_
767 "Code point 0x%04"UVXf" is not Unicode, may not be portable",
768 uv));
54f4afef 769 pack_warn = packWARN(WARN_NON_UNICODE);
949cf498 770 }
c0236afe
KW
771
772 /* The maximum code point ever specified by a standard was
773 * 2**31 - 1. Anything larger than that is a Perl extension that
774 * very well may not be understood by other applications (including
775 * earlier perl versions on EBCDIC platforms). On ASCII platforms,
776 * these code points are indicated by the first UTF-8 byte being
d35f2ca5
KW
777 * 0xFE or 0xFF. We test for these after the regular SUPER ones,
778 * and before possibly bailing out, so that the slightly more dire
c0236afe
KW
779 * warning will override the regular one. */
780 if (
781#ifndef EBCDIC
782 (*s0 & 0xFE) == 0xFE /* matches both FE, FF */
783#else
784 /* The I8 for 2**31 (U+80000000) is
785 * \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
786 * and it turns out that on all EBCDIC pages recognized that
787 * the UTF-EBCDIC for that code point is
788 * \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
789 * For the next lower code point, the 1047 UTF-EBCDIC is
790 * \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
791 * The other code pages differ only in the bytes following
792 * \x42. Thus the following works (the minimum continuation
793 * byte is \x41). */
794 *s0 == 0xFE && send - s0 > 7 && ( s0[1] > 0x41
795 || s0[2] > 0x41
796 || s0[3] > 0x41
797 || s0[4] > 0x41
798 || s0[5] > 0x41
799 || s0[6] > 0x41
800 || s0[7] > 0x42)
801#endif
d35f2ca5
KW
802 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER
803 |UTF8_DISALLOW_ABOVE_31_BIT)))
ea5ced44 804 {
0bcdd8f6 805 if ( ! (flags & UTF8_CHECK_ONLY)
d35f2ca5 806 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
0bcdd8f6 807 && ckWARN_d(WARN_UTF8))
ea5ced44 808 {
0bcdd8f6
KW
809 sv = sv_2mortal(Perl_newSVpvf(aTHX_
810 "Code point 0x%"UVXf" is not Unicode, and not portable",
811 uv));
ea5ced44
KW
812 pack_warn = packWARN(WARN_UTF8);
813 }
d35f2ca5 814 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
ea5ced44
KW
815 goto disallowed;
816 }
817 }
c0236afe 818
949cf498
KW
819 if (flags & UTF8_DISALLOW_SUPER) {
820 goto disallowed;
821 }
760c7c2f
KW
822
823 /* The deprecated warning overrides any non-deprecated one */
824 if (UNLIKELY(uv > MAX_NON_DEPRECATED_CP) && ckWARN_d(WARN_DEPRECATED))
825 {
826 sv = sv_2mortal(Perl_newSVpvf(aTHX_ cp_above_legal_max,
827 uv, MAX_NON_DEPRECATED_CP));
828 pack_warn = packWARN(WARN_DEPRECATED);
829 }
949cf498 830 }
4190d317
KW
831 else if (UNICODE_IS_NONCHAR(uv)) {
832 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
54f4afef 833 && ckWARN_d(WARN_NONCHAR))
4190d317 834 {
ba707cdc 835 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is not recommended for open interchange", uv));
54f4afef 836 pack_warn = packWARN(WARN_NONCHAR);
4190d317
KW
837 }
838 if (flags & UTF8_DISALLOW_NONCHAR) {
839 goto disallowed;
840 }
841 }
949cf498 842
eb83ed87 843 if (sv) {
de69f3af
KW
844 outlier_ret = uv; /* Note we don't bother to convert to native,
845 as all the outlier code points are the same
846 in both ASCII and EBCDIC */
eb83ed87
KW
847 goto do_warn;
848 }
849
949cf498
KW
850 /* Here, this is not considered a malformed character, so drop through
851 * to return it */
a0ed51b3 852 }
ba210ebe 853
de69f3af 854 return UNI_TO_NATIVE(uv);
ba210ebe 855
eb83ed87
KW
856 /* There are three cases which get to beyond this point. In all 3 cases:
857 * <sv> if not null points to a string to print as a warning.
858 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
859 * set.
860 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
861 * This is done by initializing it to 0, and changing it only
862 * for case 1).
863 * The 3 cases are:
864 * 1) The input is valid but problematic, and to be warned about. The
865 * return value is the resultant code point; <*retlen> is set to
866 * <curlen>, the number of bytes that comprise the code point.
867 * <pack_warn> contains the result of packWARN() for the warning
868 * types. The entry point for this case is the label <do_warn>;
869 * 2) The input is a valid code point but disallowed by the parameters to
870 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
871 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
872 * comprise the code point. <pack_warn> contains the result of
873 * packWARN() for the warning types. The entry point for this case is
874 * the label <disallowed>.
875 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
876 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
877 * bytes that comprise the malformation. All such malformations are
878 * assumed to be warning type <utf8>. The entry point for this case
879 * is the label <malformed>.
880 */
949cf498 881
7b52d656 882 malformed:
ba210ebe 883
eb83ed87
KW
884 if (sv && ckWARN_d(WARN_UTF8)) {
885 pack_warn = packWARN(WARN_UTF8);
886 }
887
7b52d656 888 disallowed:
eb83ed87 889
fcc8fcf6 890 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 891 if (retlen)
10edeb5d 892 *retlen = ((STRLEN) -1);
ba210ebe
JH
893 return 0;
894 }
895
7b52d656 896 do_warn:
5b311467 897
eb83ed87
KW
898 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
899 if warnings are to be raised. */
f555bc63 900 const char * const string = SvPVX_const(sv);
a0dbb045 901
f555bc63
KW
902 if (PL_op)
903 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
904 else
905 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
906 }
907
eb83ed87
KW
908 if (retlen) {
909 *retlen = curlen;
910 }
ba210ebe 911
eb83ed87 912 return outlier_ret;
a0ed51b3
LW
913}
914
8e84507e 915/*
ec5f19d0
KW
916=for apidoc utf8_to_uvchr_buf
917
918Returns the native code point of the first character in the string C<s> which
919is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 920C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 921
524080c4
KW
922If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
923enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 924C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 925(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 926C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 927the next possible position in C<s> that could begin a non-malformed character.
de69f3af 928See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 929returned.
ec5f19d0 930
760c7c2f
KW
931Code points above the platform's C<IV_MAX> will raise a deprecation warning,
932unless those are turned off.
933
ec5f19d0
KW
934=cut
935*/
936
937
938UV
939Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
940{
ec5f19d0
KW
941 assert(s < send);
942
943 return utf8n_to_uvchr(s, send - s, retlen,
944 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
945}
946
27d6c58a 947/* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
3986bb7c 948 * there are no malformations in the input UTF-8 string C<s>. surrogates,
57b0056d 949 * non-character code points, and non-Unicode code points are allowed. */
27d6c58a
KW
950
951UV
952Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
953{
010ab96b
KW
954 UV expectlen = UTF8SKIP(s);
955 const U8* send = s + expectlen;
9ff2f0f7 956 UV uv = *s;
3986bb7c 957
27d6c58a 958 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
81611534 959 PERL_UNUSED_CONTEXT;
27d6c58a 960
010ab96b
KW
961 if (retlen) {
962 *retlen = expectlen;
963 }
964
965 /* An invariant is trivially returned */
966 if (expectlen == 1) {
9ff2f0f7 967 return uv;
010ab96b
KW
968 }
969
9ff2f0f7
KW
970#ifdef EBCDIC
971 uv = NATIVE_UTF8_TO_I8(uv);
972#endif
973
010ab96b
KW
974 /* Remove the leading bits that indicate the number of bytes, leaving just
975 * the bits that are part of the value */
976 uv &= UTF_START_MASK(expectlen);
977
978 /* Now, loop through the remaining bytes, accumulating each into the
979 * working total as we go. (I khw tried unrolling the loop for up to 4
980 * bytes, but there was no performance improvement) */
981 for (++s; s < send; s++) {
982 uv = UTF8_ACCUMULATE(uv, *s);
983 }
984
3986bb7c 985 return UNI_TO_NATIVE(uv);
010ab96b 986
27d6c58a
KW
987}
988
ec5f19d0 989/*
ec5f19d0
KW
990=for apidoc utf8_to_uvuni_buf
991
de69f3af
KW
992Only in very rare circumstances should code need to be dealing in Unicode
993(as opposed to native) code points. In those few cases, use
994C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
995
996Returns the Unicode (not-native) code point of the first character in the
997string C<s> which
ec5f19d0
KW
998is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
999C<retlen> will be set to the length, in bytes, of that character.
1000
524080c4
KW
1001If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1002enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1003NULL) to -1. If those warnings are off, the computed value if well-defined (or
1004the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1005is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1006next possible position in C<s> that could begin a non-malformed character.
de69f3af 1007See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0 1008
760c7c2f
KW
1009Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1010unless those are turned off.
1011
ec5f19d0
KW
1012=cut
1013*/
1014
1015UV
1016Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1017{
1018 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1019
1020 assert(send > s);
1021
1022 /* Call the low level routine asking for checks */
de69f3af
KW
1023 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
1024 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
ec5f19d0
KW
1025}
1026
b76347f2 1027/*
87cea99e 1028=for apidoc utf8_length
b76347f2
JH
1029
1030Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1031Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1032up past C<e>, croaks.
b76347f2
JH
1033
1034=cut
1035*/
1036
1037STRLEN
35a4481c 1038Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
1039{
1040 STRLEN len = 0;
1041
7918f24d
NC
1042 PERL_ARGS_ASSERT_UTF8_LENGTH;
1043
8850bf83
JH
1044 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1045 * the bitops (especially ~) can create illegal UTF-8.
1046 * In other words: in Perl UTF-8 is not just for Unicode. */
1047
a3b680e6
AL
1048 if (e < s)
1049 goto warn_and_return;
b76347f2 1050 while (s < e) {
4cbf4130 1051 s += UTF8SKIP(s);
8e91ec7f
AV
1052 len++;
1053 }
1054
1055 if (e != s) {
1056 len--;
1057 warn_and_return:
9b387841
NC
1058 if (PL_op)
1059 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1060 "%s in %s", unees, OP_DESC(PL_op));
1061 else
61a12c31 1062 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1063 }
1064
1065 return len;
1066}
1067
b06226ff 1068/*
87cea99e 1069=for apidoc utf8_distance
b06226ff 1070
1e54db1a 1071Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
b06226ff
JH
1072and C<b>.
1073
1074WARNING: use only if you *know* that the pointers point inside the
1075same UTF-8 buffer.
1076
37607a96
PK
1077=cut
1078*/
a0ed51b3 1079
02eb7b47 1080IV
35a4481c 1081Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
a0ed51b3 1082{
7918f24d
NC
1083 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1084
bf1665bc 1085 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
a0ed51b3
LW
1086}
1087
b06226ff 1088/*
87cea99e 1089=for apidoc utf8_hop
b06226ff 1090
8850bf83
JH
1091Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1092forward or backward.
b06226ff
JH
1093
1094WARNING: do not use the following unless you *know* C<off> is within
8850bf83
JH
1095the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1096on the first byte of character or just after the last byte of a character.
b06226ff 1097
37607a96
PK
1098=cut
1099*/
a0ed51b3
LW
1100
1101U8 *
ddeaf645 1102Perl_utf8_hop(const U8 *s, I32 off)
a0ed51b3 1103{
7918f24d
NC
1104 PERL_ARGS_ASSERT_UTF8_HOP;
1105
8850bf83
JH
1106 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1107 * the bitops (especially ~) can create illegal UTF-8.
1108 * In other words: in Perl UTF-8 is not just for Unicode. */
1109
a0ed51b3
LW
1110 if (off >= 0) {
1111 while (off--)
1112 s += UTF8SKIP(s);
1113 }
1114 else {
1115 while (off++) {
1116 s--;
8850bf83
JH
1117 while (UTF8_IS_CONTINUATION(*s))
1118 s--;
a0ed51b3
LW
1119 }
1120 }
4373e329 1121 return (U8 *)s;
a0ed51b3
LW
1122}
1123
6940069f 1124/*
fed3ba5d
NC
1125=for apidoc bytes_cmp_utf8
1126
a1433954 1127Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1128sequence of characters (stored as UTF-8)
1129in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1130equal, -1 or -2 if the first string is less than the second string, +1 or +2
1131if the first string is greater than the second string.
1132
1133-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1134longer string. -2 or +2 is returned if
1135there was a difference between characters
fed3ba5d
NC
1136within the strings.
1137
1138=cut
1139*/
1140
1141int
1142Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1143{
1144 const U8 *const bend = b + blen;
1145 const U8 *const uend = u + ulen;
1146
1147 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1148
1149 while (b < bend && u < uend) {
1150 U8 c = *u++;
1151 if (!UTF8_IS_INVARIANT(c)) {
1152 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1153 if (u < uend) {
1154 U8 c1 = *u++;
1155 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 1156 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d
NC
1157 } else {
1158 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1159 "Malformed UTF-8 character "
1160 "(unexpected non-continuation byte 0x%02x"
1161 ", immediately after start byte 0x%02x)"
1162 /* Dear diag.t, it's in the pod. */
1163 "%s%s", c1, c,
1164 PL_op ? " in " : "",
1165 PL_op ? OP_DESC(PL_op) : "");
1166 return -2;
1167 }
1168 } else {
1169 if (PL_op)
1170 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1171 "%s in %s", unees, OP_DESC(PL_op));
1172 else
61a12c31 1173 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1174 return -2; /* Really want to return undef :-) */
1175 }
1176 } else {
1177 return -2;
1178 }
1179 }
1180 if (*b != c) {
1181 return *b < c ? -2 : +2;
1182 }
1183 ++b;
1184 }
1185
1186 if (b == bend && u == uend)
1187 return 0;
1188
1189 return b < bend ? +1 : -1;
1190}
1191
1192/*
87cea99e 1193=for apidoc utf8_to_bytes
6940069f 1194
2bbc8d55 1195Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1196Unlike L</bytes_to_utf8>, this over-writes the original string, and
1197updates C<len> to contain the new length.
67e989fb 1198Returns zero on failure, setting C<len> to -1.
6940069f 1199
a1433954 1200If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1201
6940069f
GS
1202=cut
1203*/
1204
1205U8 *
37607a96 1206Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1207{
d4c19fe8
AL
1208 U8 * const save = s;
1209 U8 * const send = s + *len;
6940069f 1210 U8 *d;
246fae53 1211
7918f24d 1212 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 1213 PERL_UNUSED_CONTEXT;
7918f24d 1214
1e54db1a 1215 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1216 while (s < send) {
d59937ca
KW
1217 if (! UTF8_IS_INVARIANT(*s)) {
1218 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1219 *len = ((STRLEN) -1);
1220 return 0;
1221 }
1222 s++;
dcad2880 1223 }
d59937ca 1224 s++;
246fae53 1225 }
dcad2880
JH
1226
1227 d = s = save;
6940069f 1228 while (s < send) {
80e0b38f
KW
1229 U8 c = *s++;
1230 if (! UTF8_IS_INVARIANT(c)) {
1231 /* Then it is two-byte encoded */
a62b247b 1232 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
80e0b38f
KW
1233 s++;
1234 }
1235 *d++ = c;
6940069f
GS
1236 }
1237 *d = '\0';
246fae53 1238 *len = d - save;
6940069f
GS
1239 return save;
1240}
1241
1242/*
87cea99e 1243=for apidoc bytes_from_utf8
f9a63242 1244
2bbc8d55 1245Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1246Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1247the newly-created string, and updates C<len> to contain the new
1248length. Returns the original string if no conversion occurs, C<len>
72d33970 1249is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55 12500 if C<s> is converted or consisted entirely of characters that are invariant
4a4088c4 1251in UTF-8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1252
37607a96
PK
1253=cut
1254*/
f9a63242
JH
1255
1256U8 *
e1ec3a88 1257Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1258{
f9a63242 1259 U8 *d;
e1ec3a88
AL
1260 const U8 *start = s;
1261 const U8 *send;
f9a63242
JH
1262 I32 count = 0;
1263
7918f24d 1264 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 1265 PERL_UNUSED_CONTEXT;
f9a63242 1266 if (!*is_utf8)
73d840c0 1267 return (U8 *)start;
f9a63242 1268
1e54db1a 1269 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1270 for (send = s + *len; s < send;) {
d59937ca
KW
1271 if (! UTF8_IS_INVARIANT(*s)) {
1272 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1273 return (U8 *)start;
d59937ca
KW
1274 }
1275 count++;
1276 s++;
db42d148 1277 }
d59937ca 1278 s++;
f9a63242
JH
1279 }
1280
35da51f7 1281 *is_utf8 = FALSE;
f9a63242 1282
212542aa 1283 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1284 s = start; start = d;
f9a63242
JH
1285 while (s < send) {
1286 U8 c = *s++;
1a91c45d 1287 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1288 /* Then it is two-byte encoded */
a62b247b 1289 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1a91c45d 1290 s++;
c4d5f83a
NIS
1291 }
1292 *d++ = c;
f9a63242
JH
1293 }
1294 *d = '\0';
1295 *len = d - start;
73d840c0 1296 return (U8 *)start;
f9a63242
JH
1297}
1298
1299/*
87cea99e 1300=for apidoc bytes_to_utf8
6940069f 1301
ff97e5cf
KW
1302Converts a string C<s> of length C<len> bytes from the native encoding into
1303UTF-8.
6662521e 1304Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1305reflect the new length in bytes.
6940069f 1306
75200dff 1307A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
1308
1309If you want to convert to UTF-8 from encodings other than
1310the native (Latin1 or EBCDIC),
a1433954 1311see L</sv_recode_to_utf8>().
c9ada85f 1312
497711e7 1313=cut
6940069f
GS
1314*/
1315
c682ebef
FC
1316/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1317 likewise need duplication. */
1318
6940069f 1319U8*
35a4481c 1320Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1321{
35a4481c 1322 const U8 * const send = s + (*len);
6940069f
GS
1323 U8 *d;
1324 U8 *dst;
7918f24d
NC
1325
1326 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1327 PERL_UNUSED_CONTEXT;
6940069f 1328
212542aa 1329 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1330 dst = d;
1331
1332 while (s < send) {
55d09dc8
KW
1333 append_utf8_from_native_byte(*s, &d);
1334 s++;
6940069f
GS
1335 }
1336 *d = '\0';
6662521e 1337 *len = d-dst;
6940069f
GS
1338 return dst;
1339}
1340
a0ed51b3 1341/*
dea0fc0b 1342 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1343 *
1344 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1345 * We optimize for native, for obvious reasons. */
1346
1347U8*
dea0fc0b 1348Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1349{
dea0fc0b
JH
1350 U8* pend;
1351 U8* dstart = d;
1352
7918f24d
NC
1353 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1354
dea0fc0b 1355 if (bytelen & 1)
f5992bc4 1356 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1357
1358 pend = p + bytelen;
1359
a0ed51b3 1360 while (p < pend) {
dea0fc0b
JH
1361 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1362 p += 2;
2d1545e5 1363 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 1364 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
1365 continue;
1366 }
56d37426
KW
1367 if (uv <= MAX_UTF8_TWO_BYTE) {
1368 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1369 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
1370 continue;
1371 }
46956fad
KW
1372#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1373#define LAST_HIGH_SURROGATE 0xDBFF
1374#define FIRST_LOW_SURROGATE 0xDC00
1375#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
1376
1377 /* This assumes that most uses will be in the first Unicode plane, not
1378 * needing surrogates */
1379 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
1380 && uv <= UNICODE_SURROGATE_LAST))
1381 {
1382 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
1383 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1384 }
1385 else {
01ea242b 1386 UV low = (p[0] << 8) + p[1];
e23c50db
KW
1387 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
1388 || UNLIKELY(low > LAST_LOW_SURROGATE))
1389 {
01ea242b 1390 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
1391 }
1392 p += 2;
46956fad
KW
1393 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1394 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 1395 }
a0ed51b3 1396 }
56d37426
KW
1397#ifdef EBCDIC
1398 d = uvoffuni_to_utf8_flags(d, uv, 0);
1399#else
a0ed51b3 1400 if (uv < 0x10000) {
eb160463
GS
1401 *d++ = (U8)(( uv >> 12) | 0xe0);
1402 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1403 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1404 continue;
1405 }
1406 else {
eb160463
GS
1407 *d++ = (U8)(( uv >> 18) | 0xf0);
1408 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1409 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1410 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1411 continue;
1412 }
56d37426 1413#endif
a0ed51b3 1414 }
dea0fc0b 1415 *newlen = d - dstart;
a0ed51b3
LW
1416 return d;
1417}
1418
1419/* Note: this one is slightly destructive of the source. */
1420
1421U8*
dea0fc0b 1422Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1423{
1424 U8* s = (U8*)p;
d4c19fe8 1425 U8* const send = s + bytelen;
7918f24d
NC
1426
1427 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1428
e0ea5e2d
NC
1429 if (bytelen & 1)
1430 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1431 (UV)bytelen);
1432
a0ed51b3 1433 while (s < send) {
d4c19fe8 1434 const U8 tmp = s[0];
a0ed51b3
LW
1435 s[0] = s[1];
1436 s[1] = tmp;
1437 s += 2;
1438 }
dea0fc0b 1439 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1440}
1441
922e8cb4
KW
1442bool
1443Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1444{
1445 U8 tmpbuf[UTF8_MAXBYTES+1];
1446 uvchr_to_utf8(tmpbuf, c);
1447 return _is_utf8_FOO(classnum, tmpbuf);
1448}
1449
f9ae8fb6
JD
1450/* Internal function so we can deprecate the external one, and call
1451 this one from other deprecated functions in this file */
1452
f2645549
KW
1453bool
1454Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 1455{
f2645549 1456 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
1457
1458 if (*p == '_')
1459 return TRUE;
f25ce844 1460 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
1461}
1462
5092f92a 1463bool
eba68aa0
KW
1464Perl__is_uni_perl_idcont(pTHX_ UV c)
1465{
1466 U8 tmpbuf[UTF8_MAXBYTES+1];
1467 uvchr_to_utf8(tmpbuf, c);
1468 return _is_utf8_perl_idcont(tmpbuf);
1469}
1470
1471bool
f91dcd13
KW
1472Perl__is_uni_perl_idstart(pTHX_ UV c)
1473{
1474 U8 tmpbuf[UTF8_MAXBYTES+1];
1475 uvchr_to_utf8(tmpbuf, c);
1476 return _is_utf8_perl_idstart(tmpbuf);
1477}
1478
3a4c58c9
KW
1479UV
1480Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1481{
1482 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 1483 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
1484 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1485 * either "SS" or "Ss". Which one to use is passed into the routine in
1486 * 'S_or_s' to avoid a test */
1487
1488 UV converted = toUPPER_LATIN1_MOD(c);
1489
1490 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1491
1492 assert(S_or_s == 'S' || S_or_s == 's');
1493
6f2d5cbc 1494 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 1495 characters in this range */
3a4c58c9
KW
1496 *p = (U8) converted;
1497 *lenp = 1;
1498 return converted;
1499 }
1500
1501 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1502 * which it maps to one of them, so as to only have to have one check for
1503 * it in the main case */
1504 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1505 switch (c) {
1506 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1507 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1508 break;
1509 case MICRO_SIGN:
1510 converted = GREEK_CAPITAL_LETTER_MU;
1511 break;
79e064b9
KW
1512#if UNICODE_MAJOR_VERSION > 2 \
1513 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
1514 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
1515 case LATIN_SMALL_LETTER_SHARP_S:
1516 *(p)++ = 'S';
1517 *p = S_or_s;
1518 *lenp = 2;
1519 return 'S';
79e064b9 1520#endif
3a4c58c9
KW
1521 default:
1522 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 1523 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
1524 }
1525 }
1526
1527 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1528 *p = UTF8_TWO_BYTE_LO(converted);
1529 *lenp = 2;
1530
1531 return converted;
1532}
1533
50bda2c3
KW
1534/* Call the function to convert a UTF-8 encoded character to the specified case.
1535 * Note that there may be more than one character in the result.
1536 * INP is a pointer to the first byte of the input character
1537 * OUTP will be set to the first byte of the string of changed characters. It
1538 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1539 * LENP will be set to the length in bytes of the string of changed characters
1540 *
1541 * The functions return the ordinal of the first character in the string of OUTP */
4a8240a3
KW
1542#define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "")
1543#define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "")
1544#define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "")
50bda2c3
KW
1545
1546/* This additionally has the input parameter SPECIALS, which if non-zero will
1547 * cause this to use the SPECIALS hash for folding (meaning get full case
1548 * folding); otherwise, when zero, this implies a simple case fold */
4a8240a3 1549#define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "" : NULL)
c3fd2246 1550
84afefe6
JH
1551UV
1552Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1553{
a1433954
KW
1554 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1555 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1556 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1557 * the changed version may be longer than the original character.
1558 *
1559 * The ordinal of the first character of the changed version is returned
1560 * (but note, as explained above, that there may be more.) */
1561
7918f24d
NC
1562 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1563
3a4c58c9
KW
1564 if (c < 256) {
1565 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1566 }
1567
0ebc6274 1568 uvchr_to_utf8(p, c);
3a4c58c9 1569 return CALL_UPPER_CASE(p, p, lenp);
a0ed51b3
LW
1570}
1571
84afefe6
JH
1572UV
1573Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1574{
7918f24d
NC
1575 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1576
3a4c58c9
KW
1577 if (c < 256) {
1578 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1579 }
1580
0ebc6274 1581 uvchr_to_utf8(p, c);
3a4c58c9 1582 return CALL_TITLE_CASE(p, p, lenp);
a0ed51b3
LW
1583}
1584
afc16117 1585STATIC U8
81611534 1586S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp)
afc16117
KW
1587{
1588 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 1589 * those, converting the result to UTF-8. Since the result is always just
a1433954 1590 * one character, we allow <p> to be NULL */
afc16117
KW
1591
1592 U8 converted = toLOWER_LATIN1(c);
1593
1594 if (p != NULL) {
6f2d5cbc 1595 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
1596 *p = converted;
1597 *lenp = 1;
1598 }
1599 else {
430c9760
KW
1600 /* Result is known to always be < 256, so can use the EIGHT_BIT
1601 * macros */
1602 *p = UTF8_EIGHT_BIT_HI(converted);
1603 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
1604 *lenp = 2;
1605 }
1606 }
1607 return converted;
1608}
1609
84afefe6
JH
1610UV
1611Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1612{
7918f24d
NC
1613 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1614
afc16117
KW
1615 if (c < 256) {
1616 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1617 }
1618
afc16117 1619 uvchr_to_utf8(p, c);
968c5e6a 1620 return CALL_LOWER_CASE(p, p, lenp);
a0ed51b3
LW
1621}
1622
84afefe6 1623UV
51910141 1624Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 1625{
51910141 1626 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 1627 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 1628 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
1629 *
1630 * Not to be used for locale folds
51910141 1631 */
f673fad4 1632
a1dde8de
KW
1633 UV converted;
1634
1635 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 1636 PERL_UNUSED_CONTEXT;
a1dde8de 1637
1ca267a5
KW
1638 assert (! (flags & FOLD_FLAGS_LOCALE));
1639
a1dde8de
KW
1640 if (c == MICRO_SIGN) {
1641 converted = GREEK_SMALL_LETTER_MU;
1642 }
9b63e895
KW
1643#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
1644 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
1645 || UNICODE_DOT_DOT_VERSION > 0)
51910141 1646 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1ca267a5
KW
1647
1648 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1649 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1650 * under those circumstances. */
1651 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1652 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1653 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1654 p, *lenp, U8);
1655 return LATIN_SMALL_LETTER_LONG_S;
1656 }
1657 else {
4f489194
KW
1658 *(p)++ = 's';
1659 *p = 's';
1660 *lenp = 2;
1661 return 's';
1ca267a5 1662 }
a1dde8de 1663 }
9b63e895 1664#endif
a1dde8de
KW
1665 else { /* In this range the fold of all other characters is their lower
1666 case */
1667 converted = toLOWER_LATIN1(c);
1668 }
1669
6f2d5cbc 1670 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
1671 *p = (U8) converted;
1672 *lenp = 1;
1673 }
1674 else {
1675 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1676 *p = UTF8_TWO_BYTE_LO(converted);
1677 *lenp = 2;
1678 }
1679
1680 return converted;
1681}
1682
1683UV
31f05a37 1684Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 1685{
4b593389 1686
a0270393
KW
1687 /* Not currently externally documented, and subject to change
1688 * <flags> bits meanings:
1689 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
1690 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1691 * locale are to be used.
a0270393
KW
1692 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1693 */
4b593389 1694
36bb2ab6 1695 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1696
780fcc9f
KW
1697 if (flags & FOLD_FLAGS_LOCALE) {
1698 /* Treat a UTF-8 locale as not being in locale at all */
1699 if (IN_UTF8_CTYPE_LOCALE) {
1700 flags &= ~FOLD_FLAGS_LOCALE;
1701 }
1702 else {
1703 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 1704 goto needs_full_generality;
780fcc9f 1705 }
31f05a37
KW
1706 }
1707
a1dde8de 1708 if (c < 256) {
e7b7ac46 1709 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 1710 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
1711 }
1712
2f306ab9 1713 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
1714 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1715 uvchr_to_utf8(p, c);
1716 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1717 }
1718 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1719 the special flags. */
1720 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
1721
1722 needs_full_generality:
a0270393 1723 uvchr_to_utf8(utf8_c, c);
445bf929 1724 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
a0270393 1725 }
84afefe6
JH
1726}
1727
26483009 1728PERL_STATIC_INLINE bool
5141f98e 1729S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 1730 const char *const swashname, SV* const invlist)
bde6a22d 1731{
ea317ccb
KW
1732 /* returns a boolean giving whether or not the UTF8-encoded character that
1733 * starts at <p> is in the swash indicated by <swashname>. <swash>
1734 * contains a pointer to where the swash indicated by <swashname>
1735 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
1736 * look at <*swash> and only generate a swash if it is not null. <invlist>
1737 * is NULL or an inversion list that defines the swash. If not null, it
1738 * saves time during initialization of the swash.
ea317ccb
KW
1739 *
1740 * Note that it is assumed that the buffer length of <p> is enough to
1741 * contain all the bytes that comprise the character. Thus, <*p> should
1742 * have been checked before this call for mal-formedness enough to assure
1743 * that. */
1744
7918f24d
NC
1745 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1746
492a624f 1747 /* The API should have included a length for the UTF-8 character in <p>,
28123549 1748 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
1749 * as far as there being enough bytes available in it to accommodate the
1750 * character without reading beyond the end, and pass that number on to the
1751 * validating routine */
6302f837 1752 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
28123549
KW
1753 if (ckWARN_d(WARN_UTF8)) {
1754 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
9816f121 1755 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
28123549
KW
1756 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1757 what the malformation is */
1758 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1759 }
1760 }
1761 return FALSE;
1762 }
87367d5f
KW
1763 if (!*swash) {
1764 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
1765 *swash = _core_swash_init("utf8",
1766
1767 /* Only use the name if there is no inversion
1768 * list; otherwise will go out to disk */
1769 (invlist) ? "" : swashname,
1770
1771 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 1772 }
28123549 1773
bde6a22d
NC
1774 return swash_fetch(*swash, p, TRUE) != 0;
1775}
1776
1777bool
922e8cb4
KW
1778Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1779{
922e8cb4
KW
1780 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1781
1782 assert(classnum < _FIRST_NON_SWASH_CC);
1783
f25ce844
KW
1784 return is_utf8_common(p,
1785 &PL_utf8_swash_ptrs[classnum],
1786 swash_property_names[classnum],
1787 PL_XPosix_ptrs[classnum]);
922e8cb4
KW
1788}
1789
1790bool
f2645549 1791Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
a0ed51b3 1792{
f2645549 1793 SV* invlist = NULL;
7918f24d 1794
f2645549 1795 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
7918f24d 1796
f2645549
KW
1797 if (! PL_utf8_perl_idstart) {
1798 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1799 }
60071a22 1800 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
82686b01
JH
1801}
1802
1803bool
f2645549 1804Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 1805{
f2645549 1806 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
1807
1808 if (*p == '_')
1809 return TRUE;
f25ce844 1810 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
1811}
1812
1813bool
eba68aa0
KW
1814Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1815{
b24b43f7 1816 SV* invlist = NULL;
eba68aa0
KW
1817
1818 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1819
b24b43f7
KW
1820 if (! PL_utf8_perl_idcont) {
1821 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1822 }
60071a22 1823 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
eba68aa0
KW
1824}
1825
eba68aa0 1826bool
f2645549 1827Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 1828{
f2645549 1829 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 1830
f25ce844 1831 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
1832}
1833
1834bool
f2645549 1835Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 1836{
f2645549 1837 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 1838
f25ce844 1839 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
1840}
1841
1842bool
7dbf68d2
KW
1843Perl__is_utf8_mark(pTHX_ const U8 *p)
1844{
7dbf68d2
KW
1845 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1846
f25ce844 1847 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
1848}
1849
6b5c0936 1850/*
87cea99e 1851=for apidoc to_utf8_case
6b5c0936 1852
6fae5207 1853C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
1854the character that is being converted. This routine assumes that the character
1855at C<p> is well-formed.
6b5c0936 1856
6fae5207
KW
1857C<ustrp> is a pointer to the character buffer to put the
1858conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
1859of the result.
1860
6fae5207 1861C<swashp> is a pointer to the swash to use.
6b5c0936 1862
a1433954 1863Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
796b6530 1864and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 1865but not always, a multicharacter mapping), is tried first.
6b5c0936 1866
4a8240a3
KW
1867C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1868any special mappings; C<""> means to use the special mappings. Values other
1869than these two are treated as the name of the hash containing the special
1870mappings, like C<"utf8::ToSpecLower">.
6b5c0936 1871
796b6530
KW
1872C<normal> is a string like C<"ToLower"> which means the swash
1873C<%utf8::ToLower>.
0134edef 1874
760c7c2f
KW
1875Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1876unless those are turned off.
1877
0134edef 1878=cut */
6b5c0936 1879
2104c8d9 1880UV
9a957fbc
AL
1881Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1882 SV **swashp, const char *normal, const char *special)
a0ed51b3 1883{
0134edef 1884 STRLEN len = 0;
f4cd282c 1885 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
7918f24d
NC
1886
1887 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1888
9ae3ac1a
KW
1889 /* Note that swash_fetch() doesn't output warnings for these because it
1890 * assumes we will */
8457b38f 1891 if (uv1 >= UNICODE_SURROGATE_FIRST) {
9ae3ac1a 1892 if (uv1 <= UNICODE_SURROGATE_LAST) {
8457b38f
KW
1893 if (ckWARN_d(WARN_SURROGATE)) {
1894 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1895 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1896 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1897 }
9ae3ac1a
KW
1898 }
1899 else if (UNICODE_IS_SUPER(uv1)) {
760c7c2f
KW
1900 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
1901 && ckWARN_d(WARN_DEPRECATED))
1902 {
1903 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
1904 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
1905 }
8457b38f
KW
1906 if (ckWARN_d(WARN_NON_UNICODE)) {
1907 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1908 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1909 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1910 }
9ae3ac1a
KW
1911 }
1912
1913 /* Note that non-characters are perfectly legal, so no warning should
1914 * be given */
1915 }
1916
0134edef 1917 if (!*swashp) /* load on-demand */
5ab9d2ef 1918 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 1919
a6f87d8c 1920 if (special) {
0134edef 1921 /* It might be "special" (sometimes, but not always,
2a37f04d 1922 * a multicharacter mapping) */
4a8240a3 1923 HV *hv = NULL;
b08cf34e
JH
1924 SV **svp;
1925
4a8240a3
KW
1926 /* If passed in the specials name, use that; otherwise use any
1927 * given in the swash */
1928 if (*special != '\0') {
1929 hv = get_hv(special, 0);
1930 }
1931 else {
1932 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
1933 if (svp) {
1934 hv = MUTABLE_HV(SvRV(*svp));
1935 }
1936 }
1937
176fe009 1938 if (hv
5f560d8a 1939 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
1940 && (*svp))
1941 {
cfd0369c 1942 const char *s;
47654450 1943
cfd0369c 1944 s = SvPV_const(*svp, len);
47654450 1945 if (len == 1)
f4cd282c 1946 /* EIGHTBIT */
c80e42f3 1947 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 1948 else {
d2dcd0fb 1949 Copy(s, ustrp, len, U8);
29e98929 1950 }
983ffd37 1951 }
0134edef
JH
1952 }
1953
1954 if (!len && *swashp) {
4a4088c4 1955 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 1956
0134edef
JH
1957 if (uv2) {
1958 /* It was "normal" (a single character mapping). */
f4cd282c 1959 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
1960 }
1961 }
1feea2c7 1962
cbe07460
KW
1963 if (len) {
1964 if (lenp) {
1965 *lenp = len;
1966 }
1967 return valid_utf8_to_uvchr(ustrp, 0);
1968 }
1969
1970 /* Here, there was no mapping defined, which means that the code point maps
1971 * to itself. Return the inputs */
bfdf22ec 1972 len = UTF8SKIP(p);
ca9fab46
KW
1973 if (p != ustrp) { /* Don't copy onto itself */
1974 Copy(p, ustrp, len, U8);
1975 }
0134edef 1976
2a37f04d
JH
1977 if (lenp)
1978 *lenp = len;
1979
f4cd282c 1980 return uv1;
cbe07460 1981
a0ed51b3
LW
1982}
1983
051a06d4 1984STATIC UV
357aadde 1985S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
051a06d4 1986{
4a4088c4 1987 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
1988 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
1989 * result contains a character that crosses the 255/256 boundary, disallow
1990 * the change, and return the original code point. See L<perlfunc/lc> for
1991 * why;
051a06d4 1992 *
a1433954
KW
1993 * p points to the original string whose case was changed; assumed
1994 * by this routine to be well-formed
051a06d4
KW
1995 * result the code point of the first character in the changed-case string
1996 * ustrp points to the changed-case string (<result> represents its first char)
1997 * lenp points to the length of <ustrp> */
1998
1999 UV original; /* To store the first code point of <p> */
2000
2001 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2002
a4f12ed7 2003 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
2004
2005 /* We know immediately if the first character in the string crosses the
2006 * boundary, so can skip */
2007 if (result > 255) {
2008
2009 /* Look at every character in the result; if any cross the
2010 * boundary, the whole thing is disallowed */
2011 U8* s = ustrp + UTF8SKIP(ustrp);
2012 U8* e = ustrp + *lenp;
2013 while (s < e) {
a4f12ed7 2014 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
2015 goto bad_crossing;
2016 }
2017 s += UTF8SKIP(s);
2018 }
2019
613abc6d
KW
2020 /* Here, no characters crossed, result is ok as-is, but we warn. */
2021 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
2022 return result;
2023 }
2024
7b52d656 2025 bad_crossing:
051a06d4
KW
2026
2027 /* Failed, have to return the original */
4b88fb76 2028 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
2029
2030 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2031 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2032 "Can't do %s(\"\\x{%"UVXf"}\") on non-UTF-8 locale; "
2033 "resolved to \"\\x{%"UVXf"}\".",
357aadde 2034 OP_DESC(PL_op),
ab0b796c
KW
2035 original,
2036 original);
051a06d4
KW
2037 Copy(p, ustrp, *lenp, char);
2038 return original;
2039}
2040
d3e79532 2041/*
87cea99e 2042=for apidoc to_utf8_upper
d3e79532 2043
1f607577 2044Instead use L</toUPPER_utf8>.
a1433954 2045
d3e79532
JH
2046=cut */
2047
051a06d4 2048/* Not currently externally documented, and subject to change:
31f05a37
KW
2049 * <flags> is set iff iff the rules from the current underlying locale are to
2050 * be used. */
051a06d4 2051
2104c8d9 2052UV
31f05a37 2053Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2054{
051a06d4
KW
2055 UV result;
2056
2057 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 2058
780fcc9f
KW
2059 if (flags) {
2060 /* Treat a UTF-8 locale as not being in locale at all */
2061 if (IN_UTF8_CTYPE_LOCALE) {
2062 flags = FALSE;
2063 }
2064 else {
2065 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2066 }
31f05a37
KW
2067 }
2068
3a4c58c9 2069 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2070 if (flags) {
2071 result = toUPPER_LC(*p);
2072 }
2073 else {
81c6c7ce 2074 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 2075 }
3a4c58c9
KW
2076 }
2077 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2078 if (flags) {
a62b247b 2079 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2080 result = toUPPER_LC(c);
051a06d4
KW
2081 }
2082 else {
a62b247b 2083 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2084 ustrp, lenp, 'S');
051a06d4
KW
2085 }
2086 }
4a4088c4 2087 else { /* UTF-8, ord above 255 */
051a06d4
KW
2088 result = CALL_UPPER_CASE(p, ustrp, lenp);
2089
2090 if (flags) {
357aadde 2091 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4
KW
2092 }
2093 return result;
2094 }
2095
4a4088c4 2096 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2097 if (UTF8_IS_INVARIANT(result)) {
2098 *ustrp = (U8) result;
2099 *lenp = 1;
2100 }
2101 else {
62cb07ea
KW
2102 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2103 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2104 *lenp = 2;
3a4c58c9 2105 }
baa60164 2106
051a06d4 2107 return result;
983ffd37 2108}
a0ed51b3 2109
d3e79532 2110/*
87cea99e 2111=for apidoc to_utf8_title
d3e79532 2112
1f607577 2113Instead use L</toTITLE_utf8>.
a1433954 2114
d3e79532
JH
2115=cut */
2116
051a06d4 2117/* Not currently externally documented, and subject to change:
31f05a37
KW
2118 * <flags> is set iff the rules from the current underlying locale are to be
2119 * used. Since titlecase is not defined in POSIX, for other than a
2120 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 2121 */
051a06d4 2122
983ffd37 2123UV
31f05a37 2124Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
983ffd37 2125{
051a06d4
KW
2126 UV result;
2127
2128 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2129
780fcc9f
KW
2130 if (flags) {
2131 /* Treat a UTF-8 locale as not being in locale at all */
2132 if (IN_UTF8_CTYPE_LOCALE) {
2133 flags = FALSE;
2134 }
2135 else {
2136 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2137 }
31f05a37
KW
2138 }
2139
3a4c58c9 2140 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2141 if (flags) {
2142 result = toUPPER_LC(*p);
2143 }
2144 else {
81c6c7ce 2145 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2146 }
3a4c58c9
KW
2147 }
2148 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2149 if (flags) {
a62b247b 2150 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2151 result = toUPPER_LC(c);
051a06d4
KW
2152 }
2153 else {
a62b247b 2154 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2155 ustrp, lenp, 's');
051a06d4
KW
2156 }
2157 }
4a4088c4 2158 else { /* UTF-8, ord above 255 */
051a06d4
KW
2159 result = CALL_TITLE_CASE(p, ustrp, lenp);
2160
2161 if (flags) {
357aadde 2162 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4
KW
2163 }
2164 return result;
2165 }
2166
4a4088c4 2167 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2168 if (UTF8_IS_INVARIANT(result)) {
2169 *ustrp = (U8) result;
2170 *lenp = 1;
2171 }
2172 else {
62cb07ea
KW
2173 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2174 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2175 *lenp = 2;
3a4c58c9
KW
2176 }
2177
051a06d4 2178 return result;
a0ed51b3
LW
2179}
2180
d3e79532 2181/*
87cea99e 2182=for apidoc to_utf8_lower
d3e79532 2183
1f607577 2184Instead use L</toLOWER_utf8>.
a1433954 2185
d3e79532
JH
2186=cut */
2187
051a06d4 2188/* Not currently externally documented, and subject to change:
31f05a37
KW
2189 * <flags> is set iff iff the rules from the current underlying locale are to
2190 * be used.
2191 */
051a06d4 2192
2104c8d9 2193UV
31f05a37 2194Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2195{
051a06d4
KW
2196 UV result;
2197
051a06d4 2198 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2199
780fcc9f
KW
2200 if (flags) {
2201 /* Treat a UTF-8 locale as not being in locale at all */
2202 if (IN_UTF8_CTYPE_LOCALE) {
2203 flags = FALSE;
2204 }
2205 else {
2206 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2207 }
31f05a37
KW
2208 }
2209
968c5e6a 2210 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2211 if (flags) {
2212 result = toLOWER_LC(*p);
2213 }
2214 else {
81c6c7ce 2215 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2216 }
968c5e6a
KW
2217 }
2218 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2219 if (flags) {
a62b247b 2220 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2221 result = toLOWER_LC(c);
051a06d4
KW
2222 }
2223 else {
a62b247b 2224 return to_lower_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2225 ustrp, lenp);
051a06d4 2226 }
968c5e6a 2227 }
4a4088c4 2228 else { /* UTF-8, ord above 255 */
051a06d4
KW
2229 result = CALL_LOWER_CASE(p, ustrp, lenp);
2230
2231 if (flags) {
357aadde 2232 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2233 }
968c5e6a 2234
051a06d4
KW
2235 return result;
2236 }
2237
4a4088c4 2238 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2239 if (UTF8_IS_INVARIANT(result)) {
2240 *ustrp = (U8) result;
2241 *lenp = 1;
2242 }
2243 else {
62cb07ea
KW
2244 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2245 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2246 *lenp = 2;
2247 }
2248
051a06d4 2249 return result;
b4e400f9
JH
2250}
2251
d3e79532 2252/*
87cea99e 2253=for apidoc to_utf8_fold
d3e79532 2254
1f607577 2255Instead use L</toFOLD_utf8>.
a1433954 2256
d3e79532
JH
2257=cut */
2258
051a06d4
KW
2259/* Not currently externally documented, and subject to change,
2260 * in <flags>
31f05a37
KW
2261 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2262 * locale are to be used.
051a06d4
KW
2263 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2264 * otherwise simple folds
a0270393
KW
2265 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2266 * prohibited
445bf929 2267 */
36bb2ab6 2268
b4e400f9 2269UV
445bf929 2270Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
b4e400f9 2271{
051a06d4
KW
2272 UV result;
2273
36bb2ab6 2274 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2275
a0270393
KW
2276 /* These are mutually exclusive */
2277 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2278
50ba90ff
KW
2279 assert(p != ustrp); /* Otherwise overwrites */
2280
780fcc9f
KW
2281 if (flags & FOLD_FLAGS_LOCALE) {
2282 /* Treat a UTF-8 locale as not being in locale at all */
2283 if (IN_UTF8_CTYPE_LOCALE) {
2284 flags &= ~FOLD_FLAGS_LOCALE;
2285 }
2286 else {
2287 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2288 }
31f05a37
KW
2289 }
2290
a1dde8de 2291 if (UTF8_IS_INVARIANT(*p)) {
051a06d4 2292 if (flags & FOLD_FLAGS_LOCALE) {
d22b930b 2293 result = toFOLD_LC(*p);
051a06d4
KW
2294 }
2295 else {
81c6c7ce 2296 return _to_fold_latin1(*p, ustrp, lenp,
1ca267a5 2297 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2298 }
a1dde8de
KW
2299 }
2300 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2301 if (flags & FOLD_FLAGS_LOCALE) {
a62b247b 2302 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2303 result = toFOLD_LC(c);
051a06d4
KW
2304 }
2305 else {
a62b247b 2306 return _to_fold_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
51910141 2307 ustrp, lenp,
1ca267a5 2308 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2309 }
a1dde8de 2310 }
4a4088c4 2311 else { /* UTF-8, ord above 255 */
a0270393 2312 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2313
1ca267a5
KW
2314 if (flags & FOLD_FLAGS_LOCALE) {
2315
76f2ffcd 2316# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
2317 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
2318
2319# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
2320# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd
KW
2321
2322 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
76f2ffcd 2323
538e84ed
KW
2324 /* Special case these two characters, as what normally gets
2325 * returned under locale doesn't work */
76f2ffcd
KW
2326 if (UTF8SKIP(p) == cap_sharp_s_len
2327 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 2328 {
ab0b796c
KW
2329 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2330 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2331 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
2332 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
2333 goto return_long_s;
2334 }
0766489e
KW
2335 else
2336#endif
2337 if (UTF8SKIP(p) == long_s_t_len
76f2ffcd 2338 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 2339 {
ab0b796c
KW
2340 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2341 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2342 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
2343 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
2344 goto return_ligature_st;
2345 }
74894415
KW
2346
2347#if UNICODE_MAJOR_VERSION == 3 \
2348 && UNICODE_DOT_VERSION == 0 \
2349 && UNICODE_DOT_DOT_VERSION == 1
2350# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
2351
2352 /* And special case this on this Unicode version only, for the same
2353 * reaons the other two are special cased. They would cross the
2354 * 255/256 boundary which is forbidden under /l, and so the code
2355 * wouldn't catch that they are equivalent (which they are only in
2356 * this release) */
2357 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
2358 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
2359 {
2360 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2361 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2362 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
2363 "resolved to \"\\x{0131}\".");
2364 goto return_dotless_i;
2365 }
2366#endif
2367
357aadde 2368 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2369 }
a0270393
KW
2370 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2371 return result;
2372 }
2373 else {
4a4088c4 2374 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
2375 * character above the ASCII range, and the result should not
2376 * contain an ASCII character. */
a0270393
KW
2377
2378 UV original; /* To store the first code point of <p> */
2379
2380 /* Look at every character in the result; if any cross the
2381 * boundary, the whole thing is disallowed */
2382 U8* s = ustrp;
2383 U8* e = ustrp + *lenp;
2384 while (s < e) {
2385 if (isASCII(*s)) {
2386 /* Crossed, have to return the original */
2387 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 2388
9fc2026f 2389 /* But in these instances, there is an alternative we can
1ca267a5 2390 * return that is valid */
0766489e
KW
2391 if (original == LATIN_SMALL_LETTER_SHARP_S
2392#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
2393 || original == LATIN_CAPITAL_LETTER_SHARP_S
2394#endif
2395 ) {
1ca267a5
KW
2396 goto return_long_s;
2397 }
9fc2026f
KW
2398 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2399 goto return_ligature_st;
2400 }
74894415
KW
2401#if UNICODE_MAJOR_VERSION == 3 \
2402 && UNICODE_DOT_VERSION == 0 \
2403 && UNICODE_DOT_DOT_VERSION == 1
2404
2405 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
2406 goto return_dotless_i;
2407 }
2408#endif
a0270393
KW
2409 Copy(p, ustrp, *lenp, char);
2410 return original;
2411 }
2412 s += UTF8SKIP(s);
2413 }
051a06d4 2414
a0270393
KW
2415 /* Here, no characters crossed, result is ok as-is */
2416 return result;
2417 }
051a06d4
KW
2418 }
2419
4a4088c4 2420 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2421 if (UTF8_IS_INVARIANT(result)) {
2422 *ustrp = (U8) result;
2423 *lenp = 1;
2424 }
2425 else {
62cb07ea
KW
2426 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2427 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2428 *lenp = 2;
2429 }
2430
051a06d4 2431 return result;
1ca267a5
KW
2432
2433 return_long_s:
2434 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2435 * folds to a string of two of these characters. By returning this
2436 * instead, then, e.g.,
2437 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2438 * works. */
2439
2440 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2441 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2442 ustrp, *lenp, U8);
2443 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
2444
2445 return_ligature_st:
2446 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2447 * have the other one fold to it */
2448
2449 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2450 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2451 return LATIN_SMALL_LIGATURE_ST;
74894415
KW
2452
2453#if UNICODE_MAJOR_VERSION == 3 \
2454 && UNICODE_DOT_VERSION == 0 \
2455 && UNICODE_DOT_DOT_VERSION == 1
2456
2457 return_dotless_i:
2458 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
2459 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
2460 return LATIN_SMALL_LETTER_DOTLESS_I;
2461
2462#endif
2463
a0ed51b3
LW
2464}
2465
711a919c 2466/* Note:
f90a9a02 2467 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2468 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2469 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2470 */
c4a5db0c 2471
a0ed51b3 2472SV*
7fc63493 2473Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2474{
c4a5db0c
KW
2475 PERL_ARGS_ASSERT_SWASH_INIT;
2476
2477 /* Returns a copy of a swash initiated by the called function. This is the
2478 * public interface, and returning a copy prevents others from doing
2479 * mischief on the original */
2480
5d3d13d1 2481 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
2482}
2483
2484SV*
5d3d13d1 2485Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c 2486{
2c1f00b9
YO
2487
2488 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
2489 * use the following define */
2490
2491#define CORE_SWASH_INIT_RETURN(x) \
2492 PL_curpm= old_PL_curpm; \
2493 return x
2494
c4a5db0c 2495 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
2496 * by calling utf8_heavy.pl in the general case. The returned value may be
2497 * the swash's inversion list instead if the input parameters allow it.
2498 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
2499 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2500 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
2501 *
2502 * This interface should only be used by functions that won't destroy or
2503 * adversely change the swash, as doing so affects all other uses of the
2504 * swash in the program; the general public should use 'Perl_swash_init'
2505 * instead.
2506 *
2507 * pkg is the name of the package that <name> should be in.
2508 * name is the name of the swash to find. Typically it is a Unicode
2509 * property name, including user-defined ones
2510 * listsv is a string to initialize the swash with. It must be of the form
2511 * documented as the subroutine return value in
2512 * L<perlunicode/User-Defined Character Properties>
2513 * minbits is the number of bits required to represent each data element.
2514 * It is '1' for binary properties.
2515 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 2516 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
2517 * flags_p if non-NULL is the address of various input and output flag bits
2518 * to the routine, as follows: ('I' means is input to the routine;
2519 * 'O' means output from the routine. Only flags marked O are
2520 * meaningful on return.)
2521 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2522 * came from a user-defined property. (I O)
5d3d13d1
KW
2523 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2524 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
2525 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2526 * return of an inversion list instead of a swash hash if this routine
2527 * thinks that would result in faster execution of swash_fetch() later
2528 * on. (I)
9a53f6cf
KW
2529 *
2530 * Thus there are three possible inputs to find the swash: <name>,
2531 * <listsv>, and <invlist>. At least one must be specified. The result
2532 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
2533 * actions can intersect, etc. what <name> gives. To avoid going out to
2534 * disk at all, <invlist> should specify completely what the swash should
2535 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
2536 *
2537 * <invlist> is only valid for binary properties */
c4a5db0c 2538
2c1f00b9
YO
2539 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
2540
c4a5db0c 2541 SV* retval = &PL_sv_undef;
83199d38 2542 HV* swash_hv = NULL;
87367d5f
KW
2543 const int invlist_swash_boundary =
2544 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2545 ? 512 /* Based on some benchmarking, but not extensive, see commit
2546 message */
2547 : -1; /* Never return just an inversion list */
9a53f6cf
KW
2548
2549 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2550 assert(! invlist || minbits == 1);
2551
2c1f00b9
YO
2552 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
2553 that triggered the swash init and the swash init perl logic itself.
2554 See perl #122747 */
2555
9a53f6cf
KW
2556 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2557 * so */
2558 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2559 dSP;
2560 const size_t pkg_len = strlen(pkg);
2561 const size_t name_len = strlen(name);
2562 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2563 SV* errsv_save;
2564 GV *method;
2565
2566 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2567
2568 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2569 ENTER;
69794297 2570 SAVEHINTS();
2782061f 2571 save_re_context();
650f067c
JL
2572 /* We might get here via a subroutine signature which uses a utf8
2573 * parameter name, at which point PL_subname will have been set
2574 * but not yet used. */
2575 save_item(PL_subname);
69794297
KW
2576 if (PL_parser && PL_parser->error_count)
2577 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2578 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
4a4088c4 2579 if (!method) { /* demand load UTF-8 */
69794297 2580 ENTER;
db2c6cb3
FC
2581 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2582 GvSV(PL_errgv) = NULL;
1a419e6b 2583#ifndef NO_TAINT_SUPPORT
69794297
KW
2584 /* It is assumed that callers of this routine are not passing in
2585 * any user derived data. */
2782061f
DM
2586 /* Need to do this after save_re_context() as it will set
2587 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2588 * in Perl_magic_get). Even line to create errsv_save can turn on
2589 * PL_tainted. */
284167a5
SM
2590 SAVEBOOL(TAINT_get);
2591 TAINT_NOT;
2592#endif
69794297
KW
2593 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2594 NULL);
eed484f9 2595 {
db2c6cb3
FC
2596 /* Not ERRSV, as there is no need to vivify a scalar we are
2597 about to discard. */
2598 SV * const errsv = GvSV(PL_errgv);
2599 if (!SvTRUE(errsv)) {
2600 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2601 SvREFCNT_dec(errsv);
2602 }
eed484f9 2603 }
69794297
KW
2604 LEAVE;
2605 }
2606 SPAGAIN;
2607 PUSHMARK(SP);
2608 EXTEND(SP,5);
2609 mPUSHp(pkg, pkg_len);
2610 mPUSHp(name, name_len);
2611 PUSHs(listsv);
2612 mPUSHi(minbits);
2613 mPUSHi(none);
2614 PUTBACK;
db2c6cb3
FC
2615 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2616 GvSV(PL_errgv) = NULL;
69794297
KW
2617 /* If we already have a pointer to the method, no need to use
2618 * call_method() to repeat the lookup. */
c41800a8
KW
2619 if (method
2620 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
2621 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2622 {
2623 retval = *PL_stack_sp--;
2624 SvREFCNT_inc(retval);
2625 }
eed484f9 2626 {
db2c6cb3
FC
2627 /* Not ERRSV. See above. */
2628 SV * const errsv = GvSV(PL_errgv);
2629 if (!SvTRUE(errsv)) {
2630 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2631 SvREFCNT_dec(errsv);
2632 }
eed484f9 2633 }
ce3b816e 2634 LEAVE;
69794297
KW
2635 POPSTACK;
2636 if (IN_PERL_COMPILETIME) {
2637 CopHINTS_set(PL_curcop, PL_hints);
2638 }
2639 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2640 if (SvPOK(retval))
2641
2642 /* If caller wants to handle missing properties, let them */
5d3d13d1 2643 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2c1f00b9 2644 CORE_SWASH_INIT_RETURN(NULL);
69794297
KW
2645 }
2646 Perl_croak(aTHX_
2647 "Can't find Unicode property definition \"%"SVf"\"",
2648 SVfARG(retval));
a25b5927 2649 NOT_REACHED; /* NOTREACHED */
69794297 2650 }
9a53f6cf 2651 } /* End of calling the module to find the swash */
36eb48b4 2652
83199d38
KW
2653 /* If this operation fetched a swash, and we will need it later, get it */
2654 if (retval != &PL_sv_undef
2655 && (minbits == 1 || (flags_p
2656 && ! (*flags_p
2657 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2658 {
2659 swash_hv = MUTABLE_HV(SvRV(retval));
2660
2661 /* If we don't already know that there is a user-defined component to
2662 * this swash, and the user has indicated they wish to know if there is
2663 * one (by passing <flags_p>), find out */
2664 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2665 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2666 if (user_defined && SvUV(*user_defined)) {
2667 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2668 }
2669 }
2670 }
2671
36eb48b4
KW
2672 /* Make sure there is an inversion list for binary properties */
2673 if (minbits == 1) {
2674 SV** swash_invlistsvp = NULL;
2675 SV* swash_invlist = NULL;
9a53f6cf 2676 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
2677 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2678 an unclaimed reference count */
36eb48b4 2679
9a53f6cf 2680 /* If this operation fetched a swash, get its already existing
83199d38 2681 * inversion list, or create one for it */
36eb48b4 2682
83199d38 2683 if (swash_hv) {
5c9f4bd2 2684 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
2685 if (swash_invlistsvp) {
2686 swash_invlist = *swash_invlistsvp;
2687 invlist_in_swash_is_valid = TRUE;
2688 }
2689 else {
36eb48b4 2690 swash_invlist = _swash_to_invlist(retval);
02c85471 2691 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
2692 }
2693 }
2694
2695 /* If an inversion list was passed in, have to include it */
2696 if (invlist) {
2697
2698 /* Any fetched swash will by now have an inversion list in it;
2699 * otherwise <swash_invlist> will be NULL, indicating that we
2700 * didn't fetch a swash */
2701 if (swash_invlist) {
2702
2703 /* Add the passed-in inversion list, which invalidates the one
2704 * already stored in the swash */
2705 invlist_in_swash_is_valid = FALSE;
2706 _invlist_union(invlist, swash_invlist, &swash_invlist);
2707 }
2708 else {
2709
87367d5f
KW
2710 /* Here, there is no swash already. Set up a minimal one, if
2711 * we are going to return a swash */
2712 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 2713 swash_hv = newHV();
4aca0fe6 2714 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 2715 }
9a53f6cf
KW
2716 swash_invlist = invlist;
2717 }
9a53f6cf
KW
2718 }
2719
2720 /* Here, we have computed the union of all the passed-in data. It may
2721 * be that there was an inversion list in the swash which didn't get
538e84ed 2722 * touched; otherwise save the computed one */
87367d5f
KW
2723 if (! invlist_in_swash_is_valid
2724 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2725 {
5c9f4bd2 2726 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
2727 {
2728 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2729 }
cc34d8c5
FC
2730 /* We just stole a reference count. */
2731 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2732 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 2733 }
87367d5f 2734
dbfdbd26
KW
2735 SvREADONLY_on(swash_invlist);
2736
c41800a8 2737 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
2738 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2739 SvREFCNT_dec(retval);
02c85471
FC
2740 if (!swash_invlist_unclaimed)
2741 SvREFCNT_inc_simple_void_NN(swash_invlist);
2742 retval = newRV_noinc(swash_invlist);
87367d5f 2743 }
36eb48b4
KW
2744 }
2745
2c1f00b9
YO
2746 CORE_SWASH_INIT_RETURN(retval);
2747#undef CORE_SWASH_INIT_RETURN
a0ed51b3
LW
2748}
2749
035d37be
JH
2750
2751/* This API is wrong for special case conversions since we may need to
2752 * return several Unicode characters for a single Unicode character
2753 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2754 * the lower-level routine, and it is similarly broken for returning
38684baa
KW
2755 * multiple values. --jhi
2756 * For those, you should use to_utf8_case() instead */
b0e3252e 2757/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
2758
2759/* Note:
2760 * Returns the value of property/mapping C<swash> for the first character
2761 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
4a4088c4 2762 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
3d0f8846 2763 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
2764 *
2765 * A "swash" is a hash which contains initially the keys/values set up by
2766 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2767 * property for all possible code points. Things are stored in a compact form
2768 * (see utf8_heavy.pl) so that calculation is required to find the actual
2769 * property value for a given code point. As code points are looked up, new
2770 * key/value pairs are added to the hash, so that the calculation doesn't have
2771 * to ever be re-done. Further, each calculation is done, not just for the
2772 * desired one, but for a whole block of code points adjacent to that one.
2773 * For binary properties on ASCII machines, the block is usually for 64 code
2774 * points, starting with a code point evenly divisible by 64. Thus if the
2775 * property value for code point 257 is requested, the code goes out and
2776 * calculates the property values for all 64 code points between 256 and 319,
2777 * and stores these as a single 64-bit long bit vector, called a "swatch",
2778 * under the key for code point 256. The key is the UTF-8 encoding for code
2779 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2780 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2781 * for code point 258 is then requested, this code realizes that it would be
2782 * stored under the key for 256, and would find that value and extract the
2783 * relevant bit, offset from 256.
2784 *
2785 * Non-binary properties are stored in as many bits as necessary to represent
2786 * their values (32 currently, though the code is more general than that), not
2787 * as single bits, but the principal is the same: the value for each key is a
2788 * vector that encompasses the property values for all code points whose UTF-8
2789 * representations are represented by the key. That is, for all code points
2790 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2791 * bytes of that.
680c470c 2792 */
a0ed51b3 2793UV
680c470c 2794Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 2795{
ef8f7699 2796 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
2797 U32 klen;
2798 U32 off;
9b56a019 2799 STRLEN slen = 0;
7d85a32c 2800 STRLEN needents;
cfd0369c 2801 const U8 *tmps = NULL;
979f2922 2802 SV *swatch;
08fb1ac5 2803 const U8 c = *ptr;
3568d838 2804
7918f24d
NC
2805 PERL_ARGS_ASSERT_SWASH_FETCH;
2806
87367d5f
KW
2807 /* If it really isn't a hash, it isn't really swash; must be an inversion
2808 * list */
2809 if (SvTYPE(hv) != SVt_PVHV) {
2810 return _invlist_contains_cp((SV*)hv,
2811 (do_utf8)
2812 ? valid_utf8_to_uvchr(ptr, NULL)
2813 : c);
2814 }
2815
08fb1ac5
KW
2816 /* We store the values in a "swatch" which is a vec() value in a swash
2817 * hash. Code points 0-255 are a single vec() stored with key length
2818 * (klen) 0. All other code points have a UTF-8 representation
2819 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2820 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2821 * length for them is the length of the encoded char - 1. ptr[klen] is the
2822 * final byte in the sequence representing the character */
2823 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2824 klen = 0;
2825 needents = 256;
2826 off = c;
3568d838 2827 }
08fb1ac5
KW
2828 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2829 klen = 0;
2830 needents = 256;
a62b247b 2831 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
2832 }
2833 else {
08fb1ac5
KW
2834 klen = UTF8SKIP(ptr) - 1;
2835
2836 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2837 * the vec is the final byte in the sequence. (In EBCDIC this is
2838 * converted to I8 to get consecutive values.) To help you visualize
2839 * all this:
2840 * Straight 1047 After final byte
2841 * UTF-8 UTF-EBCDIC I8 transform
2842 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2843 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2844 * ...
2845 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2846 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2847 * ...
2848 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2849 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2850 * ...
2851 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2852 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2853 * ...
2854 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2855 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2856 *
2857 * (There are no discontinuities in the elided (...) entries.)
2858 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2859 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2860 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2861 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2862 * index into the vec() swatch (after subtracting 0x80, which we
2863 * actually do with an '&').
2864 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2865 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2866 * dicontinuities which go away by transforming it into I8, and we
2867 * effectively subtract 0xA0 to get the index. */
979f2922 2868 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 2869 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 2870 }
7d85a32c 2871
a0ed51b3 2872 /*
4a4088c4 2873 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
a0ed51b3
LW
2874 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2875 * it's nothing to sniff at.) Pity we usually come through at least
2876 * two function calls to get here...
2877 *
2878 * NB: this code assumes that swatches are never modified, once generated!
2879 */
2880
3568d838 2881 if (hv == PL_last_swash_hv &&
a0ed51b3 2882 klen == PL_last_swash_klen &&
27da23d5 2883 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
2884 {
2885 tmps = PL_last_swash_tmps;
2886 slen = PL_last_swash_slen;
2887 }
2888 else {
2889 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 2890 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 2891
b0e3252e 2892 /* If not cached, generate it via swatch_get */
979f2922 2893 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
2894 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2895 {
2896 if (klen) {
2897 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2898 swatch = swatch_get(swash,
2899 code_point & ~((UV)needents - 1),
2900 needents);
2901 }
2902 else { /* For the first 256 code points, the swatch has a key of
2903 length 0 */
2904 swatch = swatch_get(swash, 0, needents);
2905 }
979f2922 2906
923e4eb5 2907 if (IN_PERL_COMPILETIME)
623e6609 2908 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 2909
979f2922 2910 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 2911
979f2922
ST
2912 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2913 || (slen << 3) < needents)
5637ef5b
NC
2914 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2915 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2916 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
2917 }
2918
2919 PL_last_swash_hv = hv;
16d8f38a 2920 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 2921 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
2922 /* FIXME change interpvar.h? */
2923 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
2924 PL_last_swash_slen = slen;
2925 if (klen)
2926 Copy(ptr, PL_last_swash_key, klen, U8);
2927 }
2928
9faf8d75 2929 switch ((int)((slen << 3) / needents)) {
a0ed51b3 2930 case 1:
e7aca353 2931 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
a0ed51b3 2932 case 8:
e7aca353 2933 return ((UV) tmps[off]);
a0ed51b3
LW
2934 case 16:
2935 off <<= 1;
e7aca353
JH
2936 return
2937 ((UV) tmps[off ] << 8) +
2938 ((UV) tmps[off + 1]);
a0ed51b3
LW
2939 case 32:
2940 off <<= 2;
e7aca353
JH
2941 return
2942 ((UV) tmps[off ] << 24) +
2943 ((UV) tmps[off + 1] << 16) +
2944 ((UV) tmps[off + 2] << 8) +
2945 ((UV) tmps[off + 3]);
a0ed51b3 2946 }
5637ef5b
NC
2947 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2948 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 2949 NORETURN_FUNCTION_END;
a0ed51b3 2950}
2b9d42f0 2951
319009ee
KW
2952/* Read a single line of the main body of the swash input text. These are of
2953 * the form:
2954 * 0053 0056 0073
2955 * where each number is hex. The first two numbers form the minimum and
2956 * maximum of a range, and the third is the value associated with the range.
2957 * Not all swashes should have a third number
2958 *
2959 * On input: l points to the beginning of the line to be examined; it points
2960 * to somewhere in the string of the whole input text, and is
2961 * terminated by a \n or the null string terminator.
2962 * lend points to the null terminator of that string
2963 * wants_value is non-zero if the swash expects a third number
2964 * typestr is the name of the swash's mapping, like 'ToLower'
2965 * On output: *min, *max, and *val are set to the values read from the line.
2966 * returns a pointer just beyond the line examined. If there was no
2967 * valid min number on the line, returns lend+1
2968 */
2969
2970STATIC U8*
2971S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2972 const bool wants_value, const U8* const typestr)
2973{
2974 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2975 STRLEN numlen; /* Length of the number */
02470786
KW
2976 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2977 | PERL_SCAN_DISALLOW_PREFIX
2978 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2979
2980 /* nl points to the next \n in the scan */
2981 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2982
95543e92
KW
2983 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
2984
319009ee
KW
2985 /* Get the first number on the line: the range minimum */
2986 numlen = lend - l;
2987 *min = grok_hex((char *)l, &numlen, &flags, NULL);
c88850db 2988 *max = *min; /* So can never return without setting max */
319009ee
KW
2989 if (numlen) /* If found a hex number, position past it */
2990 l += numlen;
2991 else if (nl) { /* Else, go handle next line, if any */
2992 return nl + 1; /* 1 is length of "\n" */
2993 }
2994 else { /* Else, no next line */
2995 return lend + 1; /* to LIST's end at which \n is not found */
2996 }
2997
2998 /* The max range value follows, separated by a BLANK */
2999 if (isBLANK(*l)) {
3000 ++l;
02470786
KW
3001 flags = PERL_SCAN_SILENT_ILLDIGIT
3002 | PERL_SCAN_DISALLOW_PREFIX
3003 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3004 numlen = lend - l;
3005 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3006 if (numlen)
3007 l += numlen;
3008 else /* If no value here, it is a single element range */
3009 *max = *min;
3010
3011 /* Non-binary tables have a third entry: what the first element of the
24303724 3012 * range maps to. The map for those currently read here is in hex */
319009ee
KW
3013 if (wants_value) {
3014 if (isBLANK(*l)) {
3015 ++l;
f2a7d0fc
KW
3016 flags = PERL_SCAN_SILENT_ILLDIGIT
3017 | PERL_SCAN_DISALLOW_PREFIX
3018 | PERL_SCAN_SILENT_NON_PORTABLE;
3019 numlen = lend - l;
3020 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3021 if (numlen)
3022 l += numlen;
3023 else
3024 *val = 0;
319009ee
KW
3025 }
3026 else {
3027 *val = 0;
3028 if (typeto) {
dcbac5bb 3029 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3030 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3031 typestr, l);
3032 }
3033 }
3034 }
3035 else
3036 *val = 0; /* bits == 1, then any val should be ignored */
3037 }
3038 else { /* Nothing following range min, should be single element with no
3039 mapping expected */
319009ee
KW
3040 if (wants_value) {
3041 *val = 0;
3042 if (typeto) {
dcbac5bb 3043 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3044 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3045 }
3046 }
3047 else
3048 *val = 0; /* bits == 1, then val should be ignored */
3049 }
3050
3051 /* Position to next line if any, or EOF */
3052 if (nl)
3053 l = nl + 1;
3054 else
3055 l = lend;
3056
3057 return l;
3058}
3059
979f2922
ST
3060/* Note:
3061 * Returns a swatch (a bit vector string) for a code point sequence
3062 * that starts from the value C<start> and comprises the number C<span>.
3063 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3064 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3065 */
3066STATIC SV*
b0e3252e 3067S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
3068{
3069 SV *swatch;
77f9f126 3070 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 3071 STRLEN lcur, xcur, scur;
ef8f7699 3072 HV *const hv = MUTABLE_HV(SvRV(swash));
5c9f4bd2 3073 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
36eb48b4 3074
88d45d28
KW
3075 SV** listsvp = NULL; /* The string containing the main body of the table */
3076 SV** extssvp = NULL;
3077 SV** invert_it_svp = NULL;
3078 U8* typestr = NULL;
786861f5
KW
3079 STRLEN bits;
3080 STRLEN octets; /* if bits == 1, then octets == 0 */
3081 UV none;
3082 UV end = start + span;
972dd592 3083
36eb48b4 3084 if (invlistsvp == NULL) {
786861f5
KW
3085 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3086 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3087 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3088 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3089 listsvp = hv_fetchs(hv, "LIST", FALSE);
3090 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3091
3092 bits = SvUV(*bitssvp);
3093 none = SvUV(*nonesvp);
3094 typestr = (U8*)SvPV_nolen(*typesvp);
3095 }
36eb48b4
KW
3096 else {
3097 bits = 1;
3098 none = 0;
3099 }
786861f5 3100 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 3101
b0e3252e 3102 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 3103
979f2922 3104 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 3105 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 3106 (UV)bits);
979f2922
ST
3107 }
3108
84ea5ef6
KW
3109 /* If overflowed, use the max possible */
3110 if (end < start) {
3111 end = UV_MAX;
3112 span = end - start;
3113 }
3114
979f2922 3115 /* create and initialize $swatch */
979f2922 3116 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
3117 swatch = newSV(scur);
3118 SvPOK_on(swatch);
979f2922
ST
3119 s = (U8*)SvPVX(swatch);
3120 if (octets && none) {
0bd48802 3121 const U8* const e = s + scur;
979f2922
ST
3122 while (s < e) {
3123 if (bits == 8)
3124 *s++ = (U8)(none & 0xff);
3125 else if (bits == 16) {
3126 *s++ = (U8)((none >> 8) & 0xff);
3127 *s++ = (U8)( none & 0xff);
3128 }
3129 else if (bits == 32) {
3130 *s++ = (U8)((none >> 24) & 0xff);
3131 *s++ = (U8)((none >> 16) & 0xff);
3132 *s++ = (U8)((none >> 8) & 0xff);
3133 *s++ = (U8)( none & 0xff);
3134 }
3135 }
3136 *s = '\0';
3137 }
3138 else {
3139 (void)memzero((U8*)s, scur + 1);
3140 }
3141 SvCUR_set(swatch, scur);
3142 s = (U8*)SvPVX(swatch);
3143
36eb48b4
KW
3144 if (invlistsvp) { /* If has an inversion list set up use that */
3145 _invlist_populate_swatch(*invlistsvp, start, end, s);
3146 return swatch;
3147 }
3148
3149 /* read $swash->{LIST} */
979f2922
ST
3150 l = (U8*)SvPV(*listsvp, lcur);
3151 lend = l + lcur;
3152 while (l < lend) {
8ed25d53 3153 UV min, max, val, upper;
95543e92
KW
3154 l = swash_scan_list_line(l, lend, &min, &max, &val,
3155 cBOOL(octets), typestr);
319009ee 3156 if (l > lend) {
979f2922
ST
3157 break;
3158 }
3159
972dd592 3160 /* If looking for something beyond this range, go try the next one */
979f2922
ST
3161 if (max < start)
3162 continue;
3163
8ed25d53
KW
3164 /* <end> is generally 1 beyond where we want to set things, but at the
3165 * platform's infinity, where we can't go any higher, we want to
3166 * include the code point at <end> */
3167 upper = (max < end)
3168 ? max
3169 : (max != UV_MAX || end != UV_MAX)
3170 ? end - 1
3171 : end;
3172
979f2922 3173 if (octets) {
35da51f7 3174 UV key;
979f2922
ST
3175 if (min < start) {
3176 if (!none || val < none) {
3177 val += start - min;
3178 }
3179 min = start;
3180 }
8ed25d53 3181 for (key = min; key <= upper; key++) {
979f2922 3182 STRLEN offset;
979f2922
ST
3183 /* offset must be non-negative (start <= min <= key < end) */
3184 offset = octets * (key - start);
3185 if (bits == 8)
3186 s[offset] = (U8)(val & 0xff);
3187 else if (bits == 16) {
3188 s[offset ] = (U8)((val >> 8) & 0xff);
3189 s[offset + 1] = (U8)( val & 0xff);
3190 }
3191 else if (bits == 32) {
3192 s[offset ] = (U8)((val >> 24) & 0xff);
3193 s[offset + 1] = (U8)((val >> 16) & 0xff);
3194 s[offset + 2] = (U8)((val >> 8) & 0xff);
3195 s[offset + 3] = (U8)( val & 0xff);
3196 }
3197
3198 if (!none || val < none)
3199 ++val;
3200 }
3201 }
711a919c 3202 else { /* bits == 1, then val should be ignored */
35da51f7 3203 UV key;
979f2922
ST
3204 if (min < start)
3205 min = start;
6cb05c12 3206
8ed25d53 3207 for (key = min; key <= upper; key++) {
0bd48802 3208 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3209 s[offset >> 3] |= 1 << (offset & 7);
3210 }
3211 }
3212 } /* while */
979f2922 3213
9479a769 3214 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3215 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3216
3217 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3218 * be 0, and their inversion should also be 0, as we don't succeed any
3219 * Unicode property matches for non-Unicode code points */
3220 if (start <= PERL_UNICODE_MAX) {
3221
3222 /* The code below assumes that we never cross the
3223 * Unicode/above-Unicode boundary in a range, as otherwise we would
3224 * have to figure out where to stop flipping the bits. Since this
3225 * boundary is divisible by a large power of 2, and swatches comes
3226 * in small powers of 2, this should be a valid assumption */
3227 assert(start + span - 1 <= PERL_UNICODE_MAX);
3228
507a8485
KW
3229 send = s + scur;
3230 while (s < send) {
3231 *s = ~(*s);
3232 s++;
3233 }
0bda3001 3234 }
77f9f126
KW
3235 }
3236
d73c39c5
KW
3237 /* read $swash->{EXTRAS}
3238 * This code also copied to swash_to_invlist() below */
979f2922
ST
3239 x = (U8*)SvPV(*extssvp, xcur);
3240 xend = x + xcur;
3241 while (x < xend) {
3242 STRLEN namelen;
3243 U8 *namestr;
3244 SV** othersvp;
3245 HV* otherhv;
3246 STRLEN otherbits;
3247 SV **otherbitssvp, *other;
711a919c 3248 U8 *s, *o, *nl;
979f2922
ST
3249 STRLEN slen, olen;
3250
35da51f7 3251 const U8 opc = *x++;
979f2922
ST
3252 if (opc == '\n')
3253 continue;
3254
3255 nl = (U8*)memchr(x, '\n', xend - x);
3256
3257 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3258 if (nl) {
3259 x = nl + 1; /* 1 is length of "\n" */
3260 continue;
3261 }
3262 else {
3263 x = xend; /* to EXTRAS' end at which \n is not found */
3264 break;
3265 }
3266 }
3267
3268 namestr = x;
3269 if (nl) {
3270 namelen = nl - namestr;
3271 x = nl + 1;
3272 }
3273 else {
3274 namelen = xend - namestr;
3275 x = xend;
3276 }
3277
3278 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3279 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3280 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3281 otherbits = (STRLEN)SvUV(*otherbitssvp);
3282 if (bits < otherbits)
5637ef5b
NC
3283 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3284 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3285
3286 /* The "other" swatch must be destroyed after. */
b0e3252e 3287 other = swatch_get(*othersvp, start, span);
979f2922
ST
3288 o = (U8*)SvPV(other, olen);
3289
3290 if (!olen)
b0e3252e 3291 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3292
3293 s = (U8*)SvPV(swatch, slen);
3294 if (bits == 1 && otherbits == 1) {
3295 if (slen != olen)
5637ef5b
NC
3296 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3297 "mismatch, slen=%"UVuf", olen=%"UVuf,
3298 (UV)slen, (UV)olen);
979f2922
ST
3299
3300 switch (opc) {
3301 case '+':
3302 while (slen--)
3303 *s++ |= *o++;
3304 break;
3305 case '!':
3306 while (slen--)
3307 *s++ |= ~*o++;
3308 break;
3309 case '-':
3310 while (slen--)
3311 *s++ &= ~*o++;
3312 break;
3313 case '&':
3314 while (slen--)
3315 *s++ &= *o++;
3316 break;
3317 default:
3318 break;
3319 }
3320 }
711a919c 3321 else {
979f2922
ST
3322 STRLEN otheroctets = otherbits >> 3;
3323 STRLEN offset = 0;
35da51f7 3324 U8* const send = s + slen;
979f2922
ST
3325
3326 while (s < send) {
3327 UV otherval = 0;
3328
3329 if (otherbits == 1) {
3330 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3331 ++offset;
3332 }
3333 else {
3334 STRLEN vlen = otheroctets;
3335 otherval = *o++;
3336 while (--vlen) {
3337 otherval <<= 8;
3338 otherval |= *o++;
3339 }
3340 }
3341
711a919c 3342 if (opc == '+' && otherval)
6f207bd3 3343 NOOP; /* replace with otherval */
979f2922
ST
3344 else if (opc == '!' && !otherval)
3345 otherval = 1;
3346 else if (opc == '-' && otherval)
3347 otherval = 0;
3348 else if (opc == '&' && !otherval)
3349 otherval = 0;
3350 else {
711a919c 3351 s += octets; /* no replacement */
979f2922
ST
3352 continue;
3353 }
3354
3355 if (bits == 8)
3356 *s++ = (U8)( otherval & 0xff);
3357 else if (bits == 16) {
3358 *s++ = (U8)((otherval >> 8) & 0xff);
3359 *s++ = (U8)( otherval & 0xff);
3360 }
3361 else if (bits == 32) {
3362 *s++ = (U8)((otherval >> 24) & 0xff);
3363 *s++ = (U8)((otherval >> 16) & 0xff);
3364 *s++ = (U8)((otherval >> 8) & 0xff);
3365 *s++ = (U8)( otherval & 0xff);
3366 }
3367 }
3368 }
3369 sv_free(other); /* through with it! */
3370 } /* while */
3371 return swatch;
3372}
3373
064c021d 3374HV*
4c2e1131 3375Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3376{
3377
79a2a0e8 3378 /* Subject to change or removal. For use only in regcomp.c and regexec.c
5662e334
KW
3379 * Can't be used on a property that is subject to user override, as it
3380 * relies on the value of SPECIALS in the swash which would be set by
3381 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3382 * for overridden properties
064c021d
KW
3383 *
3384 * Returns a hash which is the inversion and closure of a swash mapping.
3385 * For example, consider the input lines:
3386 * 004B 006B
3387 * 004C 006C
3388 * 212A 006B
3389 *
4a4088c4 3390 * The returned hash would have two keys, the UTF-8 for 006B and the UTF-8 for
064c021d 3391 * 006C. The value for each key is an array. For 006C, the array would
4a4088c4
KW
3392 * have two elements, the UTF-8 for itself, and for 004C. For 006B, there
3393 * would be three elements in its array, the UTF-8 for 006B, 004B and 212A.
064c021d 3394 *
538e84ed
KW
3395 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3396 * keys are only code points that are folded-to, so it isn't a full closure.
3397 *
064c021d
KW
3398 * Essentially, for any code point, it gives all the code points that map to
3399 * it, or the list of 'froms' for that point.
3400 *
5662e334
KW
3401 * Currently it ignores any additions or deletions from other swashes,
3402 * looking at just the main body of the swash, and if there are SPECIALS
3403 * in the swash, at that hash
3404 *
3405 * The specials hash can be extra code points, and most likely consists of
3406 * maps from single code points to multiple ones (each expressed as a string
4a4088c4 3407 * of UTF-8 characters). This function currently returns only 1-1 mappings.
5662e334
KW
3408 * However consider this possible input in the specials hash:
3409 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3410 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3411 *
3412 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3413 * currently handle. But it also means that FB05 and FB06 are equivalent in
3414 * a 1-1 mapping which we should handle, and this relationship may not be in
3415 * the main table. Therefore this function examines all the multi-char
74894415
KW
3416 * sequences and adds the 1-1 mappings that come out of that.
3417 *
3418 * XXX This function was originally intended to be multipurpose, but its
3419 * only use is quite likely to remain for constructing the inversion of
3420 * the CaseFolding (//i) property. If it were more general purpose for
3421 * regex patterns, it would have to do the FB05/FB06 game for simple folds,
3422 * because certain folds are prohibited under /iaa and /il. As an example,
3423 * in Unicode 3.0.1 both U+0130 and U+0131 fold to 'i', and hence are both
3424 * equivalent under /i. But under /iaa and /il, the folds to 'i' are
3425 * prohibited, so we would not figure out that they fold to each other.
3426 * Code could be written to automatically figure this out, similar to the
3427 * code that does this for multi-character folds, but this is the only case
3428 * where something like this is ever likely to happen, as all the single
7ee537e6 3429 * char folds to the 0-255 range are now quite settled. Instead there is a
74894415
KW
3430 * little special code that is compiled only for this Unicode version. This
3431 * is smaller and didn't require much coding time to do. But this makes
3432 * this routine strongly tied to being used just for CaseFolding. If ever
3433 * it should be generalized, this would have to be fixed */
064c021d
KW
3434
3435 U8 *l, *lend;
3436 STRLEN lcur;
3437 HV *const hv = MUTABLE_HV(SvRV(swash));
3438
923b6d4e
KW
3439 /* The string containing the main body of the table. This will have its
3440 * assertion fail if the swash has been converted to its inversion list */
064c021d
KW
3441 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3442
3443 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3444 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3445 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3446 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3447 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3448 const STRLEN bits = SvUV(*bitssvp);
3449 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3450 const UV none = SvUV(*nonesvp);
5662e334 3451 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
064c021d
KW
3452
3453 HV* ret = newHV();
3454
3455 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3456
3457 /* Must have at least 8 bits to get the mappings */
3458 if (bits != 8 && bits != 16 && bits != 32) {
3459 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3460 (UV)bits);
3461 }
3462
5662e334
KW
3463 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3464 mapping to more than one character */
3465
3466 /* Construct an inverse mapping hash for the specials */
3467 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3468 HV * specials_inverse = newHV();
3469 char *char_from; /* the lhs of the map */
3470 I32 from_len; /* its byte length */
3471 char *char_to; /* the rhs of the map */
3472 I32 to_len; /* its byte length */
3473 SV *sv_to; /* and in a sv */
3474 AV* from_list; /* list of things that map to each 'to' */
3475
3476 hv_iterinit(specials_hv);
3477
4a4088c4
KW
3478 /* The keys are the characters (in UTF-8) that map to the corresponding
3479 * UTF-8 string value. Iterate through the list creating the inverse
5662e334
KW
3480 * list. */
3481 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3482 SV** listp;
3483 if (! SvPOK(sv_to)) {
5637ef5b
NC
3484 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3485 "unexpectedly is not a string, flags=%lu",
3486 (unsigned long)SvFLAGS(sv_to));
5662e334 3487 }
4b88fb76 3488 /*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
3489
3490 /* Each key in the inverse list is a mapped-to value, and the key's
4a4088c4 3491 * hash value is a list of the strings (each in UTF-8) that map to
5662e334
KW
3492 * it. Those strings are all one character long */
3493 if ((listp = hv_fetch(specials_inverse,
3494 SvPVX(sv_to),
3495 SvCUR(sv_to), 0)))
3496 {
3497 from_list = (AV*) *listp;
3498 }
3499 else { /* No entry yet for it: create one */
3500 from_list = newAV();
3501 if (! hv_store(specials_inverse,
3502 SvPVX(sv_to),
3503 SvCUR(sv_to),
3504 (SV*) from_list, 0))
3505 {
3506 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3507 }
3508 }
3509
3510 /* Here have the list associated with this 'to' (perhaps newly
3511 * created and empty). Just add to it. Note that we ASSUME that
3512 * the input is guaranteed to not have duplications, so we don't
3513 * check for that. Duplications just slow down execution time. */
3514 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3515 }
3516
3517 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3518 * it looking for cases like the FB05/FB06 examples above. There would
3519 * be an entry in the hash like
3520 * 'st' => [ FB05, FB06 ]
3521 * In this example we will create two lists that get stored in the
3522 * returned hash, 'ret':
3523 * FB05 => [ FB05, FB06 ]
3524 * FB06 => [ FB05, FB06 ]
3525 *
3526 * Note that there is nothing to do if the array only has one element.
3527 * (In the normal 1-1 case handled below, we don't have to worry about
3528 * two lists, as everything gets tied to the single list that is
3529 * generated for the single character 'to'. But here, we are omitting
3530 * that list, ('st' in the example), so must have multiple lists.) */
3531 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3532 &char_to, &to_len)))
3533 {
b9f2b683 3534 if (av_tindex(from_list) > 0) {
c70927a6 3535 SSize_t i;
5662e334
KW
3536
3537 /* We iterate over all combinations of i,j to place each code
3538 * point on each list */
b9f2b683 3539 for (i = 0; i <= av_tindex(from_list); i++) {
c70927a6 3540 SSize_t j;
5662e334
KW
3541 AV* i_list = newAV();
3542 SV** entryp = av_fetch(from_list, i, FALSE);
3543 if (entryp == NULL) {
3544 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3545 }
3546 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3547 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3548 }
3549 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3550 (SV*) i_list, FALSE))
3551 {
3552 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3553 }
3554
538e84ed 3555 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
b9f2b683 3556 for (j = 0; j <= av_tindex(from_list); j++) {
5662e334
KW
3557 entryp = av_fetch(from_list, j, FALSE);
3558 if (entryp == NULL) {
3559 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3560 }
3561
3562 /* When i==j this adds itself to the list */
4b88fb76
KW
3563 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3564 (U8*) SvPVX(*entryp),
3565 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3566 0)));
4637d003 3567 /*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
3568 }
3569 }
3570 }
3571 }
3572 SvREFCNT_dec(specials_inverse); /* done with it */
3573 } /* End of specials */
3574
064c021d 3575 /* read $swash->{LIST} */
74894415
KW
3576
3577#if UNICODE_MAJOR_VERSION == 3 \
3578 && UNICODE_DOT_VERSION == 0 \
3579 && UNICODE_DOT_DOT_VERSION == 1
3580
3581 /* For this version only U+130 and U+131 are equivalent under qr//i. Add a
3582 * rule so that things work under /iaa and /il */
3583
3584 SV * mod_listsv = sv_mortalcopy(*listsvp);
3585 sv_catpv(mod_listsv, "130\t130\t131\n");
3586 l = (U8*)SvPV(mod_listsv, lcur);
3587
3588#else
3589
064c021d 3590 l = (U8*)SvPV(*listsvp, lcur);
74894415
KW
3591
3592#endif
3593
064c021d
KW
3594 lend = l + lcur;
3595
3596 /* Go through each input line */
3597 while (l < lend) {
3598 UV min, max, val;
3599 UV inverse;
95543e92
KW
3600 l = swash_scan_list_line(l, lend, &min, &max, &val,
3601 cBOOL(octets), typestr);
064c021d
KW
3602 if (l > lend) {
3603 break;
3604 }
3605
3606 /* Each element in the range is to be inverted */
3607 for (inverse = min; inverse <= max; inverse++) {
3608 AV* list;
064c021d
KW
3609 SV** listp;
3610 IV i;
3611 bool found_key = FALSE;
5662e334 3612 bool found_inverse = FALSE;
064c021d
KW
3613
3614 /* The key is the inverse mapping */
3615 char key[UTF8_MAXBYTES+1];
c80e42f3 3616 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
064c021d
KW
3617 STRLEN key_len = key_end - key;
3618
064c021d
KW
3619 /* Get the list for the map */
3620 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3621 list = (AV*) *listp;
3622 }
3623 else { /* No entry yet for it: create one */
3624 list = newAV();
3625 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3626 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3627 }
3628 }
3629
5662e334
KW