This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
perlapi: Add 2 links to other parts of the pod
[perl5.git] / utf8.c
CommitLineData
a0ed51b3
LW
1/* utf8.c
2 *
1129b882 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
b94e2f88 4 * by Larry Wall and others
a0ed51b3
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550
TC
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
15 *
cdad3b53 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
a0ed51b3
LW
17 *
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
4ac71550 20 * as is the custom in the West, if you wish to be answered?'
cdad3b53 21 * --Gandalf, addressing Théoden's door wardens
4ac71550
TC
22 *
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
24 *
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
4ac71550
TC
27 *
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
29 */
30
31#include "EXTERN.h"
864dbfa3 32#define PERL_IN_UTF8_C
a0ed51b3 33#include "perl.h"
81e983c1 34#include "inline_invlist.c"
a0ed51b3 35
27da23d5
JH
36static const char unees[] =
37 "Malformed UTF-8 character (unexpected end of string)";
901b21bf 38
48ef279e 39/*
ccfc67b7 40=head1 Unicode Support
7fefc6c1 41These are various utility functions for manipulating UTF8-encoded
72d33970 42strings. For the uninitiated, this is a method of representing arbitrary
61296642 43Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
44characters in the ASCII range are unmodified, and a zero byte never appears
45within non-zero characters.
166f8a29 46
eaf7a4d2
CS
47=cut
48*/
49
50/*
7bbfa158 51=for apidoc is_invariant_string
eaf7a4d2 52
b8ed1ca9
KW
53Returns true iff the first C<len> bytes of the string C<s> are the same
54regardless of the UTF-8 encoding of the string (or UTF-EBCDIC encoding on
55EBCDIC machines). That is, if they are UTF-8 invariant. On ASCII-ish
56machines, all the ASCII characters and only the ASCII characters fit this
7bbfa158 57definition. On EBCDIC machines, the ASCII-range characters are invariant, but
b8ed1ca9
KW
58so also are the C1 controls and C<\c?> (which isn't in the ASCII range on
59EBCDIC).
eaf7a4d2 60
75200dff
KW
61If C<len> is 0, it will be calculated using C<strlen(s)>, (which means if you
62use this option, that C<s> can't have embedded C<NUL> characters and has to
63have a terminating C<NUL> byte).
9f7e3d64 64
a1433954 65See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
eaf7a4d2
CS
66
67=cut
68*/
69
70bool
7bbfa158 71Perl_is_invariant_string(const U8 *s, STRLEN len)
eaf7a4d2
CS
72{
73 const U8* const send = s + (len ? len : strlen((const char *)s));
74 const U8* x = s;
75
7bbfa158 76 PERL_ARGS_ASSERT_IS_INVARIANT_STRING;
eaf7a4d2
CS
77
78 for (; x < send; ++x) {
79 if (!UTF8_IS_INVARIANT(*x))
80 break;
81 }
82
83 return x == send;
84}
85
86/*
378516de 87=for apidoc uvoffuni_to_utf8_flags
eebe1485 88
a27992cc 89THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
90Instead, B<Almost all code should use L</uvchr_to_utf8> or
91L</uvchr_to_utf8_flags>>.
a27992cc 92
de69f3af
KW
93This function is like them, but the input is a strict Unicode
94(as opposed to native) code point. Only in very rare circumstances should code
95not be using the native code point.
949cf498 96
de69f3af 97For details, see the description for L</uvchr_to_utf8_flags>>.
949cf498 98
eebe1485
SC
99=cut
100*/
101
dfe13c55 102U8 *
378516de 103Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 104{
378516de 105 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 106
d9432125
KW
107 if (UNI_IS_INVARIANT(uv)) {
108 *d++ = (U8) LATIN1_TO_NATIVE(uv);
109 return d;
110 }
111
9415f659
KW
112#ifdef EBCDIC
113 /* Not representable in UTF-EBCDIC */
114 flags |= UNICODE_DISALLOW_FE_FF;
115#endif
116
979f77b6
KW
117 /* The first problematic code point is the first surrogate */
118 if (uv >= UNICODE_SURROGATE_FIRST
0cfa64bf 119 && ckWARN3_d(WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
979f77b6 120 {
949cf498
KW
121 if (UNICODE_IS_SURROGATE(uv)) {
122 if (flags & UNICODE_WARN_SURROGATE) {
8457b38f 123 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
949cf498
KW
124 "UTF-16 surrogate U+%04"UVXf, uv);
125 }
126 if (flags & UNICODE_DISALLOW_SURROGATE) {
127 return NULL;
128 }
129 }
130 else if (UNICODE_IS_SUPER(uv)) {
131 if (flags & UNICODE_WARN_SUPER
132 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
133 {
8457b38f 134 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
949cf498
KW
135 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
136 }
137 if (flags & UNICODE_DISALLOW_SUPER
138 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
139 {
9415f659
KW
140#ifdef EBCDIC
141 Perl_die(aTHX_ "Can't represent character for Ox%"UVXf" on this platform", uv);
661d43c4 142 NOT_REACHED; /* NOTREACHED */
9415f659 143#endif
949cf498
KW
144 return NULL;
145 }
146 }
147 else if (UNICODE_IS_NONCHAR(uv)) {
148 if (flags & UNICODE_WARN_NONCHAR) {
8457b38f 149 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
949cf498
KW
150 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
151 uv);
152 }
153 if (flags & UNICODE_DISALLOW_NONCHAR) {
154 return NULL;
155 }
156 }
507b9800 157 }
d9432125 158
2d331972 159#if defined(EBCDIC)
d9432125 160 {
5aaebcb3 161 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
162 U8 *p = d+len-1;
163 while (p > d) {
bc3632a8 164 *p-- = (U8) I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
165 uv >>= UTF_ACCUMULATION_SHIFT;
166 }
bc3632a8 167 *p = (U8) I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
168 return d+len;
169 }
170#else /* Non loop style */
a0ed51b3 171 if (uv < 0x800) {
eb160463
GS
172 *d++ = (U8)(( uv >> 6) | 0xc0);
173 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
174 return d;
175 }
176 if (uv < 0x10000) {
eb160463
GS
177 *d++ = (U8)(( uv >> 12) | 0xe0);
178 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
179 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
180 return d;
181 }
182 if (uv < 0x200000) {
eb160463
GS
183 *d++ = (U8)(( uv >> 18) | 0xf0);
184 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
185 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
186 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
187 return d;
188 }
189 if (uv < 0x4000000) {
eb160463
GS
190 *d++ = (U8)(( uv >> 24) | 0xf8);
191 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
192 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
193 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
194 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
195 return d;
196 }
197 if (uv < 0x80000000) {
eb160463
GS
198 *d++ = (U8)(( uv >> 30) | 0xfc);
199 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
200 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
201 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
202 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
203 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
204 return d;
205 }
6588300d 206#ifdef UTF8_QUAD_MAX
d7578b48 207 if (uv < UTF8_QUAD_MAX)
a0ed51b3
LW
208#endif
209 {
eb160463
GS
210 *d++ = 0xfe; /* Can't match U+FEFF! */
211 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
212 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
213 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
214 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
215 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
216 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
217 return d;
218 }
6588300d 219#ifdef UTF8_QUAD_MAX
a0ed51b3 220 {
eb160463
GS
221 *d++ = 0xff; /* Can't match U+FFFE! */
222 *d++ = 0x80; /* 6 Reserved bits */
223 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
224 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
225 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
226 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
227 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
228 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
229 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
230 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
231 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
232 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
233 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
234 return d;
235 }
236#endif
537124e4 237#endif /* Non loop style */
a0ed51b3 238}
646ca15d 239/*
07693fe6
KW
240=for apidoc uvchr_to_utf8
241
bcb1a2d4 242Adds the UTF-8 representation of the native code point C<uv> to the end
c749c9fd
KW
243of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
244C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
245the byte after the end of the new character. In other words,
07693fe6
KW
246
247 d = uvchr_to_utf8(d, uv);
248
249is the recommended wide native character-aware way of saying
250
251 *(d++) = uv;
252
de69f3af
KW
253This function accepts any UV as input. To forbid or warn on non-Unicode code
254points, or those that may be problematic, see L</uvchr_to_utf8_flags>.
255
07693fe6
KW
256=cut
257*/
258
de69f3af
KW
259/* This is also a macro */
260PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
261
07693fe6
KW
262U8 *
263Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
264{
de69f3af 265 return uvchr_to_utf8(d, uv);
07693fe6
KW
266}
267
de69f3af
KW
268/*
269=for apidoc uvchr_to_utf8_flags
270
271Adds the UTF-8 representation of the native code point C<uv> to the end
c749c9fd
KW
272of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
273C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
274the byte after the end of the new character. In other words,
de69f3af
KW
275
276 d = uvchr_to_utf8_flags(d, uv, flags);
277
278or, in most cases,
279
280 d = uvchr_to_utf8_flags(d, uv, 0);
281
282This is the Unicode-aware way of saying
283
284 *(d++) = uv;
285
286This function will convert to UTF-8 (and not warn) even code points that aren't
287legal Unicode or are problematic, unless C<flags> contains one or more of the
288following flags:
289
290If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
291the function will raise a warning, provided UTF8 warnings are enabled. If instead
292UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
293If both flags are set, the function will both warn and return NULL.
294
4c3cfd5d 295The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags
de69f3af 296affect how the function handles a Unicode non-character. And likewise, the
4c3cfd5d 297UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags affect the handling of
de69f3af
KW
298code points that are
299above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
300even less portable) can be warned and/or disallowed even if other above-Unicode
301code points are accepted, by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
302flags.
303
304And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
305above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
306DISALLOW flags.
307
308=cut
309*/
310
311/* This is also a macro */
312PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
313
07693fe6
KW
314U8 *
315Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
316{
de69f3af 317 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
318}
319
320/*
87cea99e 321=for apidoc is_utf8_string
6662521e 322
a1433954 323Returns true if the first C<len> bytes of string C<s> form a valid
9f7e3d64 324UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
75200dff
KW
325using C<strlen(s)> (which means if you use this option, that C<s> can't have
326embedded C<NUL> characters and has to have a terminating C<NUL> byte). Note
327that all characters being ASCII constitute 'a valid UTF-8 string'.
6662521e 328
7bbfa158 329See also L</is_invariant_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
768c67ee 330
6662521e
GS
331=cut
332*/
333
8e84507e 334bool
668b6d8d 335Perl_is_utf8_string(const U8 *s, STRLEN len)
6662521e 336{
35da51f7 337 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 338 const U8* x = s;
067a85ef 339
7918f24d 340 PERL_ARGS_ASSERT_IS_UTF8_STRING;
1aa99e6b 341
6662521e 342 while (x < send) {
6302f837
KW
343 STRLEN len = isUTF8_CHAR(x, send);
344 if (UNLIKELY(! len)) {
345 return FALSE;
346 }
347 x += len;
6662521e 348 }
768c67ee 349
067a85ef 350 return TRUE;
6662521e
GS
351}
352
67e989fb 353/*
814fafa7
NC
354Implemented as a macro in utf8.h
355
87cea99e 356=for apidoc is_utf8_string_loc
814fafa7 357
a1433954
KW
358Like L</is_utf8_string> but stores the location of the failure (in the
359case of "utf8ness failure") or the location C<s>+C<len> (in the case of
814fafa7
NC
360"utf8ness success") in the C<ep>.
361
a1433954 362See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
814fafa7 363
87cea99e 364=for apidoc is_utf8_string_loclen
81cd54e3 365
a1433954
KW
366Like L</is_utf8_string>() but stores the location of the failure (in the
367case of "utf8ness failure") or the location C<s>+C<len> (in the case of
768c67ee
JH
368"utf8ness success") in the C<ep>, and the number of UTF-8
369encoded characters in the C<el>.
370
a1433954 371See also L</is_utf8_string_loc>() and L</is_utf8_string>().
81cd54e3
JH
372
373=cut
374*/
375
376bool
668b6d8d 377Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
81cd54e3 378{
35da51f7 379 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 380 const U8* x = s;
3ebfea28 381 STRLEN outlen = 0;
7918f24d
NC
382
383 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
81cd54e3 384
81cd54e3 385 while (x < send) {
6302f837
KW
386 STRLEN len = isUTF8_CHAR(x, send);
387 if (UNLIKELY(! len)) {
388 goto out;
389 }
390 x += len;
391 outlen++;
81cd54e3 392 }
768c67ee
JH
393
394 out:
3ebfea28
AL
395 if (el)
396 *el = outlen;
397
768c67ee
JH
398 if (ep)
399 *ep = x;
3ebfea28 400 return (x == send);
81cd54e3
JH
401}
402
403/*
768c67ee 404
de69f3af 405=for apidoc utf8n_to_uvchr
378516de
KW
406
407THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 408Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 409
9041c2e3 410Bottom level UTF-8 decode routine.
de69f3af 411Returns the native code point value of the first character in the string C<s>,
746afd53
KW
412which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
413C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
414the length, in bytes, of that character.
949cf498
KW
415
416The value of C<flags> determines the behavior when C<s> does not point to a
417well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
418zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
419next possible position in C<s> that could begin a non-malformed character.
420Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
421
422Various ALLOW flags can be set in C<flags> to allow (and not warn on)
423individual types of malformations, such as the sequence being overlong (that
424is, when there is a shorter sequence that can express the same code point;
425overlong sequences are expressly forbidden in the UTF-8 standard due to
426potential security issues). Another malformation example is the first byte of
427a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
428flags. For allowed 0 length strings, this function returns 0; for allowed
429overlong sequences, the computed code point is returned; for all other allowed
430malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
431determinable reasonable value.
949cf498
KW
432
433The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
434flags) malformation is found. If this flag is set, the routine assumes that
435the caller will raise a warning, and this function will silently just set
d088425d
KW
436C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
437
75200dff 438Note that this API requires disambiguation between successful decoding a C<NUL>
d088425d
KW
439character, and an error return (unless the UTF8_CHECK_ONLY flag is set), as
440in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
75200dff
KW
441first byte of C<s> is 0 as well. If so, the input was a C<NUL>; if not, the
442input had an error.
949cf498
KW
443
444Certain code points are considered problematic. These are Unicode surrogates,
746afd53 445Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 446By default these are considered regular code points, but certain situations
5eafe189 447warrant special handling for them. If C<flags> contains
949cf498
KW
448UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
449malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
450UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
451maximum) can be set to disallow these categories individually.
452
453The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
454UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
455for their respective categories, but otherwise the code points are considered
456valid (not malformations). To get a category to both be treated as a
457malformation and raise a warning, specify both the WARN and DISALLOW flags.
458(But note that warnings are not raised if lexically disabled nor if
459UTF8_CHECK_ONLY is also specified.)
460
461Very large code points (above 0x7FFF_FFFF) are considered more problematic than
462the others that are above the Unicode legal maximum. There are several
eb83ed87
KW
463reasons: they requre at least 32 bits to represent them on ASCII platforms, are
464not representable at all on EBCDIC platforms, and the original UTF-8
465specification never went above this number (the current 0x10FFFF limit was
466imposed later). (The smaller ones, those that fit into 32 bits, are
467representable by a UV on ASCII platforms, but not by an IV, which means that
468the number of operations that can be performed on them is quite restricted.)
469The UTF-8 encoding on ASCII platforms for these large code points begins with a
470byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
471be treated as malformations, while allowing smaller above-Unicode code points.
472(Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
72d33970
FC
473including these, as malformations.)
474Similarly, UTF8_WARN_FE_FF acts just like
eb83ed87 475the other WARN flags, but applies just to these code points.
949cf498
KW
476
477All other code points corresponding to Unicode characters, including private
478use and those yet to be assigned, are never considered malformed and never
479warn.
67e989fb 480
37607a96
PK
481=cut
482*/
67e989fb 483
a0ed51b3 484UV
de69f3af 485Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 486{
d4c19fe8 487 const U8 * const s0 = s;
eb83ed87 488 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 489 U8 * send;
eb83ed87
KW
490 UV uv = *s;
491 STRLEN expectlen;
949cf498 492 SV* sv = NULL;
eb83ed87
KW
493 UV outlier_ret = 0; /* return value when input is in error or problematic
494 */
495 UV pack_warn = 0; /* Save result of packWARN() for later */
496 bool unexpected_non_continuation = FALSE;
497 bool overflowed = FALSE;
2f8f112e 498 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 499
eb83ed87 500 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 501
de69f3af 502 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
a0dbb045 503
eb83ed87
KW
504 /* The order of malformation tests here is important. We should consume as
505 * few bytes as possible in order to not skip any valid character. This is
506 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
507 * http://unicode.org/reports/tr36 for more discussion as to why. For
508 * example, once we've done a UTF8SKIP, we can tell the expected number of
509 * bytes, and could fail right off the bat if the input parameters indicate
510 * that there are too few available. But it could be that just that first
511 * byte is garbled, and the intended character occupies fewer bytes. If we
512 * blindly assumed that the first byte is correct, and skipped based on
513 * that number, we could skip over a valid input character. So instead, we
514 * always examine the sequence byte-by-byte.
515 *
516 * We also should not consume too few bytes, otherwise someone could inject
517 * things. For example, an input could be deliberately designed to
518 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 519 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
520 * which is actually part of of the overflowing sequence), that could look
521 * legitimate to the caller, which could discard the initial partial
522 * sequence and process the rest, inappropriately */
523
524 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 525 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
526 if (retlen) {
527 *retlen = 0;
528 }
a0dbb045 529
eb83ed87
KW
530 if (flags & UTF8_ALLOW_EMPTY) {
531 return 0;
532 }
533 if (! (flags & UTF8_CHECK_ONLY)) {
534 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
535 }
0c443dc2
JH
536 goto malformed;
537 }
538
eb83ed87
KW
539 expectlen = UTF8SKIP(s);
540
541 /* A well-formed UTF-8 character, as the vast majority of calls to this
542 * function will be for, has this expected length. For efficiency, set
543 * things up here to return it. It will be overriden only in those rare
544 * cases where a malformation is found */
545 if (retlen) {
546 *retlen = expectlen;
547 }
548
549 /* An invariant is trivially well-formed */
1d72bdf6 550 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 551 return uv;
a0ed51b3 552 }
67e989fb 553
eb83ed87 554 /* A continuation character can't start a valid sequence */
b5b9af04 555 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
556 if (flags & UTF8_ALLOW_CONTINUATION) {
557 if (retlen) {
558 *retlen = 1;
559 }
560 return UNICODE_REPLACEMENT;
561 }
ba210ebe 562
eb83ed87
KW
563 if (! (flags & UTF8_CHECK_ONLY)) {
564 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
565 }
566 curlen = 1;
ba210ebe
JH
567 goto malformed;
568 }
9041c2e3 569
dcd27b3c
KW
570 /* Here is not a continuation byte, nor an invariant. The only thing left
571 * is a start byte (possibly for an overlong) */
572
1d72bdf6 573#ifdef EBCDIC
bc3632a8 574 uv = NATIVE_UTF8_TO_I8(uv);
1d72bdf6
NIS
575#endif
576
eb83ed87
KW
577 /* Remove the leading bits that indicate the number of bytes in the
578 * character's whole UTF-8 sequence, leaving just the bits that are part of
579 * the value */
580 uv &= UTF_START_MASK(expectlen);
ba210ebe 581
eb83ed87
KW
582 /* Now, loop through the remaining bytes in the character's sequence,
583 * accumulating each into the working value as we go. Be sure to not look
584 * past the end of the input string */
0b8d30e8
KW
585 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
586
eb83ed87 587 for (s = s0 + 1; s < send; s++) {
b5b9af04 588 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
589#ifndef EBCDIC /* Can't overflow in EBCDIC */
590 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
591
592 /* The original implementors viewed this malformation as more
593 * serious than the others (though I, khw, don't understand
594 * why, since other malformations also give very very wrong
595 * results), so there is no way to turn off checking for it.
596 * Set a flag, but keep going in the loop, so that we absorb
597 * the rest of the bytes that comprise the character. */
598 overflowed = TRUE;
599 overflow_byte = *s; /* Save for warning message's use */
600 }
601#endif
8850bf83 602 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
603 }
604 else {
605 /* Here, found a non-continuation before processing all expected
606 * bytes. This byte begins a new character, so quit, even if
607 * allowing this malformation. */
608 unexpected_non_continuation = TRUE;
609 break;
610 }
611 } /* End of loop through the character's bytes */
612
613 /* Save how many bytes were actually in the character */
614 curlen = s - s0;
615
616 /* The loop above finds two types of malformations: non-continuation and/or
617 * overflow. The non-continuation malformation is really a too-short
618 * malformation, as it means that the current character ended before it was
619 * expected to (being terminated prematurely by the beginning of the next
620 * character, whereas in the too-short malformation there just are too few
621 * bytes available to hold the character. In both cases, the check below
622 * that we have found the expected number of bytes would fail if executed.)
623 * Thus the non-continuation malformation is really unnecessary, being a
624 * subset of the too-short malformation. But there may be existing
625 * applications that are expecting the non-continuation type, so we retain
626 * it, and return it in preference to the too-short malformation. (If this
627 * code were being written from scratch, the two types might be collapsed
628 * into one.) I, khw, am also giving priority to returning the
629 * non-continuation and too-short malformations over overflow when multiple
630 * ones are present. I don't know of any real reason to prefer one over
631 * the other, except that it seems to me that multiple-byte errors trumps
632 * errors from a single byte */
b5b9af04 633 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
634 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
635 if (! (flags & UTF8_CHECK_ONLY)) {
636 if (curlen == 1) {
637 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
638 }
639 else {
640 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
641 }
642 }
eb83ed87
KW
643 goto malformed;
644 }
645 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
646
647 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
648 * as what the original expectations were. */
649 do_overlong_test = FALSE;
eb83ed87
KW
650 if (retlen) {
651 *retlen = curlen;
652 }
653 }
b5b9af04 654 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
655 if (! (flags & UTF8_ALLOW_SHORT)) {
656 if (! (flags & UTF8_CHECK_ONLY)) {
657 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 658 }
eb83ed87
KW
659 goto malformed;
660 }
661 uv = UNICODE_REPLACEMENT;
2f8f112e 662 do_overlong_test = FALSE;
eb83ed87
KW
663 if (retlen) {
664 *retlen = curlen;
665 }
666 }
667
ea5ced44 668#ifndef EBCDIC /* EBCDIC can't overflow */
b5b9af04 669 if (UNLIKELY(overflowed)) {
eb83ed87 670 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 671 goto malformed;
eb83ed87
KW
672 }
673#endif
674
2f8f112e 675 if (do_overlong_test
5aaebcb3 676 && expectlen > (STRLEN) OFFUNISKIP(uv)
2f8f112e
KW
677 && ! (flags & UTF8_ALLOW_LONG))
678 {
eb83ed87
KW
679 /* The overlong malformation has lower precedence than the others.
680 * Note that if this malformation is allowed, we return the actual
681 * value, instead of the replacement character. This is because this
682 * value is actually well-defined. */
683 if (! (flags & UTF8_CHECK_ONLY)) {
5aaebcb3 684 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
685 }
686 goto malformed;
687 }
688
1a89bb6c 689 /* Here, the input is considered to be well-formed, but it still could be a
eb83ed87
KW
690 * problematic code point that is not allowed by the input parameters. */
691 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
692 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
693 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
694 {
949cf498 695 if (UNICODE_IS_SURROGATE(uv)) {
ea5ced44
KW
696
697 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
698 * generation of the sv, since no warnings are raised under CHECK */
eb83ed87 699 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
54f4afef 700 && ckWARN_d(WARN_SURROGATE))
eb83ed87 701 {
111d382d 702 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
54f4afef 703 pack_warn = packWARN(WARN_SURROGATE);
949cf498
KW
704 }
705 if (flags & UTF8_DISALLOW_SURROGATE) {
706 goto disallowed;
707 }
708 }
949cf498 709 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87 710 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
ea5ced44 711 && ckWARN_d(WARN_NON_UNICODE))
eb83ed87 712 {
111d382d 713 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
54f4afef 714 pack_warn = packWARN(WARN_NON_UNICODE);
949cf498 715 }
ea5ced44
KW
716#ifndef EBCDIC /* EBCDIC always allows FE, FF */
717
718 /* The first byte being 0xFE or 0xFF is a subset of the SUPER code
719 * points. We test for these after the regular SUPER ones, and
720 * before possibly bailing out, so that the more dire warning
721 * overrides the regular one, if applicable */
722 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
723 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
724 {
725 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY))
726 == UTF8_WARN_FE_FF
727 && ckWARN_d(WARN_UTF8))
728 {
729 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%"UVXf" is not Unicode, and not portable", uv));
730 pack_warn = packWARN(WARN_UTF8);
731 }
732 if (flags & UTF8_DISALLOW_FE_FF) {
733 goto disallowed;
734 }
735 }
736#endif
949cf498
KW
737 if (flags & UTF8_DISALLOW_SUPER) {
738 goto disallowed;
739 }
740 }
4190d317
KW
741 else if (UNICODE_IS_NONCHAR(uv)) {
742 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
54f4afef 743 && ckWARN_d(WARN_NONCHAR))
4190d317
KW
744 {
745 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
54f4afef 746 pack_warn = packWARN(WARN_NONCHAR);
4190d317
KW
747 }
748 if (flags & UTF8_DISALLOW_NONCHAR) {
749 goto disallowed;
750 }
751 }
949cf498 752
eb83ed87 753 if (sv) {
de69f3af
KW
754 outlier_ret = uv; /* Note we don't bother to convert to native,
755 as all the outlier code points are the same
756 in both ASCII and EBCDIC */
eb83ed87
KW
757 goto do_warn;
758 }
759
949cf498
KW
760 /* Here, this is not considered a malformed character, so drop through
761 * to return it */
a0ed51b3 762 }
ba210ebe 763
de69f3af 764 return UNI_TO_NATIVE(uv);
ba210ebe 765
eb83ed87
KW
766 /* There are three cases which get to beyond this point. In all 3 cases:
767 * <sv> if not null points to a string to print as a warning.
768 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
769 * set.
770 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
771 * This is done by initializing it to 0, and changing it only
772 * for case 1).
773 * The 3 cases are:
774 * 1) The input is valid but problematic, and to be warned about. The
775 * return value is the resultant code point; <*retlen> is set to
776 * <curlen>, the number of bytes that comprise the code point.
777 * <pack_warn> contains the result of packWARN() for the warning
778 * types. The entry point for this case is the label <do_warn>;
779 * 2) The input is a valid code point but disallowed by the parameters to
780 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
781 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
782 * comprise the code point. <pack_warn> contains the result of
783 * packWARN() for the warning types. The entry point for this case is
784 * the label <disallowed>.
785 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
786 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
787 * bytes that comprise the malformation. All such malformations are
788 * assumed to be warning type <utf8>. The entry point for this case
789 * is the label <malformed>.
790 */
949cf498 791
7b52d656 792 malformed:
ba210ebe 793
eb83ed87
KW
794 if (sv && ckWARN_d(WARN_UTF8)) {
795 pack_warn = packWARN(WARN_UTF8);
796 }
797
7b52d656 798 disallowed:
eb83ed87 799
fcc8fcf6 800 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 801 if (retlen)
10edeb5d 802 *retlen = ((STRLEN) -1);
ba210ebe
JH
803 return 0;
804 }
805
7b52d656 806 do_warn:
5b311467 807
eb83ed87
KW
808 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
809 if warnings are to be raised. */
f555bc63 810 const char * const string = SvPVX_const(sv);
a0dbb045 811
f555bc63
KW
812 if (PL_op)
813 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
814 else
815 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
816 }
817
eb83ed87
KW
818 if (retlen) {
819 *retlen = curlen;
820 }
ba210ebe 821
eb83ed87 822 return outlier_ret;
a0ed51b3
LW
823}
824
8e84507e 825/*
ec5f19d0
KW
826=for apidoc utf8_to_uvchr_buf
827
828Returns the native code point of the first character in the string C<s> which
829is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 830C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 831
524080c4
KW
832If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
833enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
173db420
KW
834NULL) to -1. If those warnings are off, the computed value, if well-defined
835(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
836C<*retlen> is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is
837the next possible position in C<s> that could begin a non-malformed character.
de69f3af 838See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 839returned.
ec5f19d0
KW
840
841=cut
842*/
843
844
845UV
846Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
847{
ec5f19d0
KW
848 assert(s < send);
849
850 return utf8n_to_uvchr(s, send - s, retlen,
851 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
852}
853
27d6c58a 854/* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
3986bb7c 855 * there are no malformations in the input UTF-8 string C<s>. surrogates,
57b0056d 856 * non-character code points, and non-Unicode code points are allowed. */
27d6c58a
KW
857
858UV
859Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
860{
010ab96b
KW
861 UV expectlen = UTF8SKIP(s);
862 const U8* send = s + expectlen;
9ff2f0f7 863 UV uv = *s;
3986bb7c 864
27d6c58a 865 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
81611534 866 PERL_UNUSED_CONTEXT;
27d6c58a 867
010ab96b
KW
868 if (retlen) {
869 *retlen = expectlen;
870 }
871
872 /* An invariant is trivially returned */
873 if (expectlen == 1) {
9ff2f0f7 874 return uv;
010ab96b
KW
875 }
876
9ff2f0f7
KW
877#ifdef EBCDIC
878 uv = NATIVE_UTF8_TO_I8(uv);
879#endif
880
010ab96b
KW
881 /* Remove the leading bits that indicate the number of bytes, leaving just
882 * the bits that are part of the value */
883 uv &= UTF_START_MASK(expectlen);
884
885 /* Now, loop through the remaining bytes, accumulating each into the
886 * working total as we go. (I khw tried unrolling the loop for up to 4
887 * bytes, but there was no performance improvement) */
888 for (++s; s < send; s++) {
889 uv = UTF8_ACCUMULATE(uv, *s);
890 }
891
3986bb7c 892 return UNI_TO_NATIVE(uv);
010ab96b 893
27d6c58a
KW
894}
895
ec5f19d0 896/*
ec5f19d0
KW
897=for apidoc utf8_to_uvuni_buf
898
de69f3af
KW
899Only in very rare circumstances should code need to be dealing in Unicode
900(as opposed to native) code points. In those few cases, use
901C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
902
903Returns the Unicode (not-native) code point of the first character in the
904string C<s> which
ec5f19d0
KW
905is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
906C<retlen> will be set to the length, in bytes, of that character.
907
524080c4
KW
908If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
909enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
910NULL) to -1. If those warnings are off, the computed value if well-defined (or
911the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
912is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
913next possible position in C<s> that could begin a non-malformed character.
de69f3af 914See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
915
916=cut
917*/
918
919UV
920Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
921{
922 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
923
924 assert(send > s);
925
926 /* Call the low level routine asking for checks */
de69f3af
KW
927 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
928 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
ec5f19d0
KW
929}
930
b76347f2 931/*
87cea99e 932=for apidoc utf8_length
b76347f2
JH
933
934Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
935Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
936up past C<e>, croaks.
b76347f2
JH
937
938=cut
939*/
940
941STRLEN
35a4481c 942Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
943{
944 STRLEN len = 0;
945
7918f24d
NC
946 PERL_ARGS_ASSERT_UTF8_LENGTH;
947
8850bf83
JH
948 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
949 * the bitops (especially ~) can create illegal UTF-8.
950 * In other words: in Perl UTF-8 is not just for Unicode. */
951
a3b680e6
AL
952 if (e < s)
953 goto warn_and_return;
b76347f2 954 while (s < e) {
4cbf4130 955 s += UTF8SKIP(s);
8e91ec7f
AV
956 len++;
957 }
958
959 if (e != s) {
960 len--;
961 warn_and_return:
9b387841
NC
962 if (PL_op)
963 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
964 "%s in %s", unees, OP_DESC(PL_op));
965 else
61a12c31 966 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
967 }
968
969 return len;
970}
971
b06226ff 972/*
87cea99e 973=for apidoc utf8_distance
b06226ff 974
1e54db1a 975Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
b06226ff
JH
976and C<b>.
977
978WARNING: use only if you *know* that the pointers point inside the
979same UTF-8 buffer.
980
37607a96
PK
981=cut
982*/
a0ed51b3 983
02eb7b47 984IV
35a4481c 985Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
a0ed51b3 986{
7918f24d
NC
987 PERL_ARGS_ASSERT_UTF8_DISTANCE;
988
bf1665bc 989 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
a0ed51b3
LW
990}
991
b06226ff 992/*
87cea99e 993=for apidoc utf8_hop
b06226ff 994
8850bf83
JH
995Return the UTF-8 pointer C<s> displaced by C<off> characters, either
996forward or backward.
b06226ff
JH
997
998WARNING: do not use the following unless you *know* C<off> is within
8850bf83
JH
999the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1000on the first byte of character or just after the last byte of a character.
b06226ff 1001
37607a96
PK
1002=cut
1003*/
a0ed51b3
LW
1004
1005U8 *
ddeaf645 1006Perl_utf8_hop(const U8 *s, I32 off)
a0ed51b3 1007{
7918f24d
NC
1008 PERL_ARGS_ASSERT_UTF8_HOP;
1009
8850bf83
JH
1010 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1011 * the bitops (especially ~) can create illegal UTF-8.
1012 * In other words: in Perl UTF-8 is not just for Unicode. */
1013
a0ed51b3
LW
1014 if (off >= 0) {
1015 while (off--)
1016 s += UTF8SKIP(s);
1017 }
1018 else {
1019 while (off++) {
1020 s--;
8850bf83
JH
1021 while (UTF8_IS_CONTINUATION(*s))
1022 s--;
a0ed51b3
LW
1023 }
1024 }
4373e329 1025 return (U8 *)s;
a0ed51b3
LW
1026}
1027
6940069f 1028/*
fed3ba5d
NC
1029=for apidoc bytes_cmp_utf8
1030
a1433954 1031Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1032sequence of characters (stored as UTF-8)
1033in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1034equal, -1 or -2 if the first string is less than the second string, +1 or +2
1035if the first string is greater than the second string.
1036
1037-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1038longer string. -2 or +2 is returned if
1039there was a difference between characters
fed3ba5d
NC
1040within the strings.
1041
1042=cut
1043*/
1044
1045int
1046Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1047{
1048 const U8 *const bend = b + blen;
1049 const U8 *const uend = u + ulen;
1050
1051 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1052
1053 while (b < bend && u < uend) {
1054 U8 c = *u++;
1055 if (!UTF8_IS_INVARIANT(c)) {
1056 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1057 if (u < uend) {
1058 U8 c1 = *u++;
1059 if (UTF8_IS_CONTINUATION(c1)) {
94bb8c36 1060 c = TWO_BYTE_UTF8_TO_NATIVE(c, c1);
fed3ba5d
NC
1061 } else {
1062 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1063 "Malformed UTF-8 character "
1064 "(unexpected non-continuation byte 0x%02x"
1065 ", immediately after start byte 0x%02x)"
1066 /* Dear diag.t, it's in the pod. */
1067 "%s%s", c1, c,
1068 PL_op ? " in " : "",
1069 PL_op ? OP_DESC(PL_op) : "");
1070 return -2;
1071 }
1072 } else {
1073 if (PL_op)
1074 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1075 "%s in %s", unees, OP_DESC(PL_op));
1076 else
61a12c31 1077 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1078 return -2; /* Really want to return undef :-) */
1079 }
1080 } else {
1081 return -2;
1082 }
1083 }
1084 if (*b != c) {
1085 return *b < c ? -2 : +2;
1086 }
1087 ++b;
1088 }
1089
1090 if (b == bend && u == uend)
1091 return 0;
1092
1093 return b < bend ? +1 : -1;
1094}
1095
1096/*
87cea99e 1097=for apidoc utf8_to_bytes
6940069f 1098
2bbc8d55 1099Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1100Unlike L</bytes_to_utf8>, this over-writes the original string, and
1101updates C<len> to contain the new length.
67e989fb 1102Returns zero on failure, setting C<len> to -1.
6940069f 1103
a1433954 1104If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1105
6940069f
GS
1106=cut
1107*/
1108
1109U8 *
37607a96 1110Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1111{
d4c19fe8
AL
1112 U8 * const save = s;
1113 U8 * const send = s + *len;
6940069f 1114 U8 *d;
246fae53 1115
7918f24d 1116 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 1117 PERL_UNUSED_CONTEXT;
7918f24d 1118
1e54db1a 1119 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1120 while (s < send) {
d59937ca
KW
1121 if (! UTF8_IS_INVARIANT(*s)) {
1122 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1123 *len = ((STRLEN) -1);
1124 return 0;
1125 }
1126 s++;
dcad2880 1127 }
d59937ca 1128 s++;
246fae53 1129 }
dcad2880
JH
1130
1131 d = s = save;
6940069f 1132 while (s < send) {
80e0b38f
KW
1133 U8 c = *s++;
1134 if (! UTF8_IS_INVARIANT(c)) {
1135 /* Then it is two-byte encoded */
1136 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1137 s++;
1138 }
1139 *d++ = c;
6940069f
GS
1140 }
1141 *d = '\0';
246fae53 1142 *len = d - save;
6940069f
GS
1143 return save;
1144}
1145
1146/*
87cea99e 1147=for apidoc bytes_from_utf8
f9a63242 1148
2bbc8d55 1149Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1150Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1151the newly-created string, and updates C<len> to contain the new
1152length. Returns the original string if no conversion occurs, C<len>
72d33970 1153is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55
SP
11540 if C<s> is converted or consisted entirely of characters that are invariant
1155in utf8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1156
37607a96
PK
1157=cut
1158*/
f9a63242
JH
1159
1160U8 *
e1ec3a88 1161Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1162{
f9a63242 1163 U8 *d;
e1ec3a88
AL
1164 const U8 *start = s;
1165 const U8 *send;
f9a63242
JH
1166 I32 count = 0;
1167
7918f24d 1168 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 1169 PERL_UNUSED_CONTEXT;
f9a63242 1170 if (!*is_utf8)
73d840c0 1171 return (U8 *)start;
f9a63242 1172
1e54db1a 1173 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1174 for (send = s + *len; s < send;) {
d59937ca
KW
1175 if (! UTF8_IS_INVARIANT(*s)) {
1176 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1177 return (U8 *)start;
d59937ca
KW
1178 }
1179 count++;
1180 s++;
db42d148 1181 }
d59937ca 1182 s++;
f9a63242
JH
1183 }
1184
35da51f7 1185 *is_utf8 = FALSE;
f9a63242 1186
212542aa 1187 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1188 s = start; start = d;
f9a63242
JH
1189 while (s < send) {
1190 U8 c = *s++;
1a91c45d 1191 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1192 /* Then it is two-byte encoded */
1a91c45d
KW
1193 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1194 s++;
c4d5f83a
NIS
1195 }
1196 *d++ = c;
f9a63242
JH
1197 }
1198 *d = '\0';
1199 *len = d - start;
73d840c0 1200 return (U8 *)start;
f9a63242
JH
1201}
1202
1203/*
87cea99e 1204=for apidoc bytes_to_utf8
6940069f 1205
ff97e5cf
KW
1206Converts a string C<s> of length C<len> bytes from the native encoding into
1207UTF-8.
6662521e 1208Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1209reflect the new length in bytes.
6940069f 1210
75200dff 1211A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
1212
1213If you want to convert to UTF-8 from encodings other than
1214the native (Latin1 or EBCDIC),
a1433954 1215see L</sv_recode_to_utf8>().
c9ada85f 1216
497711e7 1217=cut
6940069f
GS
1218*/
1219
c682ebef
FC
1220/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1221 likewise need duplication. */
1222
6940069f 1223U8*
35a4481c 1224Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1225{
35a4481c 1226 const U8 * const send = s + (*len);
6940069f
GS
1227 U8 *d;
1228 U8 *dst;
7918f24d
NC
1229
1230 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1231 PERL_UNUSED_CONTEXT;
6940069f 1232
212542aa 1233 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1234 dst = d;
1235
1236 while (s < send) {
55d09dc8
KW
1237 append_utf8_from_native_byte(*s, &d);
1238 s++;
6940069f
GS
1239 }
1240 *d = '\0';
6662521e 1241 *len = d-dst;
6940069f
GS
1242 return dst;
1243}
1244
a0ed51b3 1245/*
dea0fc0b 1246 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1247 *
1248 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1249 * We optimize for native, for obvious reasons. */
1250
1251U8*
dea0fc0b 1252Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1253{
dea0fc0b
JH
1254 U8* pend;
1255 U8* dstart = d;
1256
7918f24d
NC
1257 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1258
dea0fc0b 1259 if (bytelen & 1)
f5992bc4 1260 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1261
1262 pend = p + bytelen;
1263
a0ed51b3 1264 while (p < pend) {
dea0fc0b
JH
1265 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1266 p += 2;
56d37426
KW
1267 if (UNI_IS_INVARIANT(uv)) {
1268 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
1269 continue;
1270 }
56d37426
KW
1271 if (uv <= MAX_UTF8_TWO_BYTE) {
1272 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1273 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
1274 continue;
1275 }
46956fad
KW
1276#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1277#define LAST_HIGH_SURROGATE 0xDBFF
1278#define FIRST_LOW_SURROGATE 0xDC00
1279#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
1280
1281 /* This assumes that most uses will be in the first Unicode plane, not
1282 * needing surrogates */
1283 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
1284 && uv <= UNICODE_SURROGATE_LAST))
1285 {
1286 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
1287 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1288 }
1289 else {
01ea242b 1290 UV low = (p[0] << 8) + p[1];
e23c50db
KW
1291 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
1292 || UNLIKELY(low > LAST_LOW_SURROGATE))
1293 {
01ea242b 1294 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
1295 }
1296 p += 2;
46956fad
KW
1297 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1298 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 1299 }
a0ed51b3 1300 }
56d37426
KW
1301#ifdef EBCDIC
1302 d = uvoffuni_to_utf8_flags(d, uv, 0);
1303#else
a0ed51b3 1304 if (uv < 0x10000) {
eb160463
GS
1305 *d++ = (U8)(( uv >> 12) | 0xe0);
1306 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1307 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1308 continue;
1309 }
1310 else {
eb160463
GS
1311 *d++ = (U8)(( uv >> 18) | 0xf0);
1312 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1313 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1314 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1315 continue;
1316 }
56d37426 1317#endif
a0ed51b3 1318 }
dea0fc0b 1319 *newlen = d - dstart;
a0ed51b3
LW
1320 return d;
1321}
1322
1323/* Note: this one is slightly destructive of the source. */
1324
1325U8*
dea0fc0b 1326Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1327{
1328 U8* s = (U8*)p;
d4c19fe8 1329 U8* const send = s + bytelen;
7918f24d
NC
1330
1331 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1332
e0ea5e2d
NC
1333 if (bytelen & 1)
1334 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1335 (UV)bytelen);
1336
a0ed51b3 1337 while (s < send) {
d4c19fe8 1338 const U8 tmp = s[0];
a0ed51b3
LW
1339 s[0] = s[1];
1340 s[1] = tmp;
1341 s += 2;
1342 }
dea0fc0b 1343 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1344}
1345
922e8cb4
KW
1346bool
1347Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1348{
1349 U8 tmpbuf[UTF8_MAXBYTES+1];
1350 uvchr_to_utf8(tmpbuf, c);
1351 return _is_utf8_FOO(classnum, tmpbuf);
1352}
1353
f9ae8fb6
JD
1354/* Internal function so we can deprecate the external one, and call
1355 this one from other deprecated functions in this file */
1356
f2645549
KW
1357bool
1358Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 1359{
f2645549 1360 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
1361
1362 if (*p == '_')
1363 return TRUE;
f25ce844 1364 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
1365}
1366
5092f92a 1367bool
eba68aa0
KW
1368Perl__is_uni_perl_idcont(pTHX_ UV c)
1369{
1370 U8 tmpbuf[UTF8_MAXBYTES+1];
1371 uvchr_to_utf8(tmpbuf, c);
1372 return _is_utf8_perl_idcont(tmpbuf);
1373}
1374
1375bool
f91dcd13
KW
1376Perl__is_uni_perl_idstart(pTHX_ UV c)
1377{
1378 U8 tmpbuf[UTF8_MAXBYTES+1];
1379 uvchr_to_utf8(tmpbuf, c);
1380 return _is_utf8_perl_idstart(tmpbuf);
1381}
1382
3a4c58c9
KW
1383UV
1384Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1385{
1386 /* We have the latin1-range values compiled into the core, so just use
1387 * those, converting the result to utf8. The only difference between upper
1388 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1389 * either "SS" or "Ss". Which one to use is passed into the routine in
1390 * 'S_or_s' to avoid a test */
1391
1392 UV converted = toUPPER_LATIN1_MOD(c);
1393
1394 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1395
1396 assert(S_or_s == 'S' || S_or_s == 's');
1397
6f2d5cbc 1398 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 1399 characters in this range */
3a4c58c9
KW
1400 *p = (U8) converted;
1401 *lenp = 1;
1402 return converted;
1403 }
1404
1405 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1406 * which it maps to one of them, so as to only have to have one check for
1407 * it in the main case */
1408 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1409 switch (c) {
1410 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1411 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1412 break;
1413 case MICRO_SIGN:
1414 converted = GREEK_CAPITAL_LETTER_MU;
1415 break;
1416 case LATIN_SMALL_LETTER_SHARP_S:
1417 *(p)++ = 'S';
1418 *p = S_or_s;
1419 *lenp = 2;
1420 return 'S';
1421 default:
1422 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 1423 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
1424 }
1425 }
1426
1427 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1428 *p = UTF8_TWO_BYTE_LO(converted);
1429 *lenp = 2;
1430
1431 return converted;
1432}
1433
50bda2c3
KW
1434/* Call the function to convert a UTF-8 encoded character to the specified case.
1435 * Note that there may be more than one character in the result.
1436 * INP is a pointer to the first byte of the input character
1437 * OUTP will be set to the first byte of the string of changed characters. It
1438 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1439 * LENP will be set to the length in bytes of the string of changed characters
1440 *
1441 * The functions return the ordinal of the first character in the string of OUTP */
4a8240a3
KW
1442#define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "")
1443#define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "")
1444#define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "")
50bda2c3
KW
1445
1446/* This additionally has the input parameter SPECIALS, which if non-zero will
1447 * cause this to use the SPECIALS hash for folding (meaning get full case
1448 * folding); otherwise, when zero, this implies a simple case fold */
4a8240a3 1449#define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "" : NULL)
c3fd2246 1450
84afefe6
JH
1451UV
1452Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1453{
a1433954
KW
1454 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1455 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1456 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1457 * the changed version may be longer than the original character.
1458 *
1459 * The ordinal of the first character of the changed version is returned
1460 * (but note, as explained above, that there may be more.) */
1461
7918f24d
NC
1462 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1463
3a4c58c9
KW
1464 if (c < 256) {
1465 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1466 }
1467
0ebc6274 1468 uvchr_to_utf8(p, c);
3a4c58c9 1469 return CALL_UPPER_CASE(p, p, lenp);
a0ed51b3
LW
1470}
1471
84afefe6
JH
1472UV
1473Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1474{
7918f24d
NC
1475 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1476
3a4c58c9
KW
1477 if (c < 256) {
1478 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1479 }
1480
0ebc6274 1481 uvchr_to_utf8(p, c);
3a4c58c9 1482 return CALL_TITLE_CASE(p, p, lenp);
a0ed51b3
LW
1483}
1484
afc16117 1485STATIC U8
81611534 1486S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp)
afc16117
KW
1487{
1488 /* We have the latin1-range values compiled into the core, so just use
1489 * those, converting the result to utf8. Since the result is always just
a1433954 1490 * one character, we allow <p> to be NULL */
afc16117
KW
1491
1492 U8 converted = toLOWER_LATIN1(c);
1493
1494 if (p != NULL) {
6f2d5cbc 1495 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
1496 *p = converted;
1497 *lenp = 1;
1498 }
1499 else {
430c9760
KW
1500 /* Result is known to always be < 256, so can use the EIGHT_BIT
1501 * macros */
1502 *p = UTF8_EIGHT_BIT_HI(converted);
1503 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
1504 *lenp = 2;
1505 }
1506 }
1507 return converted;
1508}
1509
84afefe6
JH
1510UV
1511Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1512{
7918f24d
NC
1513 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1514
afc16117
KW
1515 if (c < 256) {
1516 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1517 }
1518
afc16117 1519 uvchr_to_utf8(p, c);
968c5e6a 1520 return CALL_LOWER_CASE(p, p, lenp);
a0ed51b3
LW
1521}
1522
84afefe6 1523UV
51910141 1524Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 1525{
51910141 1526 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 1527 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 1528 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
1529 *
1530 * Not to be used for locale folds
51910141 1531 */
f673fad4 1532
a1dde8de
KW
1533 UV converted;
1534
1535 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 1536 PERL_UNUSED_CONTEXT;
a1dde8de 1537
1ca267a5
KW
1538 assert (! (flags & FOLD_FLAGS_LOCALE));
1539
a1dde8de
KW
1540 if (c == MICRO_SIGN) {
1541 converted = GREEK_SMALL_LETTER_MU;
1542 }
51910141 1543 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1ca267a5
KW
1544
1545 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1546 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1547 * under those circumstances. */
1548 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1549 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1550 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1551 p, *lenp, U8);
1552 return LATIN_SMALL_LETTER_LONG_S;
1553 }
1554 else {
4f489194
KW
1555 *(p)++ = 's';
1556 *p = 's';
1557 *lenp = 2;
1558 return 's';
1ca267a5 1559 }
a1dde8de
KW
1560 }
1561 else { /* In this range the fold of all other characters is their lower
1562 case */
1563 converted = toLOWER_LATIN1(c);
1564 }
1565
6f2d5cbc 1566 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
1567 *p = (U8) converted;
1568 *lenp = 1;
1569 }
1570 else {
1571 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1572 *p = UTF8_TWO_BYTE_LO(converted);
1573 *lenp = 2;
1574 }
1575
1576 return converted;
1577}
1578
1579UV
31f05a37 1580Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 1581{
4b593389 1582
a0270393
KW
1583 /* Not currently externally documented, and subject to change
1584 * <flags> bits meanings:
1585 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
1586 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1587 * locale are to be used.
a0270393
KW
1588 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1589 */
4b593389 1590
36bb2ab6 1591 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1592
780fcc9f
KW
1593 if (flags & FOLD_FLAGS_LOCALE) {
1594 /* Treat a UTF-8 locale as not being in locale at all */
1595 if (IN_UTF8_CTYPE_LOCALE) {
1596 flags &= ~FOLD_FLAGS_LOCALE;
1597 }
1598 else {
1599 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 1600 goto needs_full_generality;
780fcc9f 1601 }
31f05a37
KW
1602 }
1603
a1dde8de 1604 if (c < 256) {
e7b7ac46 1605 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 1606 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
1607 }
1608
2f306ab9 1609 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
1610 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1611 uvchr_to_utf8(p, c);
1612 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1613 }
1614 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1615 the special flags. */
1616 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
1617
1618 needs_full_generality:
a0270393 1619 uvchr_to_utf8(utf8_c, c);
445bf929 1620 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
a0270393 1621 }
84afefe6
JH
1622}
1623
26483009 1624PERL_STATIC_INLINE bool
5141f98e 1625S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 1626 const char *const swashname, SV* const invlist)
bde6a22d 1627{
ea317ccb
KW
1628 /* returns a boolean giving whether or not the UTF8-encoded character that
1629 * starts at <p> is in the swash indicated by <swashname>. <swash>
1630 * contains a pointer to where the swash indicated by <swashname>
1631 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
1632 * look at <*swash> and only generate a swash if it is not null. <invlist>
1633 * is NULL or an inversion list that defines the swash. If not null, it
1634 * saves time during initialization of the swash.
ea317ccb
KW
1635 *
1636 * Note that it is assumed that the buffer length of <p> is enough to
1637 * contain all the bytes that comprise the character. Thus, <*p> should
1638 * have been checked before this call for mal-formedness enough to assure
1639 * that. */
1640
7918f24d
NC
1641 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1642
492a624f 1643 /* The API should have included a length for the UTF-8 character in <p>,
28123549 1644 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
1645 * as far as there being enough bytes available in it to accommodate the
1646 * character without reading beyond the end, and pass that number on to the
1647 * validating routine */
6302f837 1648 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
28123549
KW
1649 if (ckWARN_d(WARN_UTF8)) {
1650 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
9816f121 1651 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
28123549
KW
1652 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1653 what the malformation is */
1654 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1655 }
1656 }
1657 return FALSE;
1658 }
87367d5f
KW
1659 if (!*swash) {
1660 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
1661 *swash = _core_swash_init("utf8",
1662
1663 /* Only use the name if there is no inversion
1664 * list; otherwise will go out to disk */
1665 (invlist) ? "" : swashname,
1666
1667 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 1668 }
28123549 1669
bde6a22d
NC
1670 return swash_fetch(*swash, p, TRUE) != 0;
1671}
1672
1673bool
922e8cb4
KW
1674Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1675{
922e8cb4
KW
1676 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1677
1678 assert(classnum < _FIRST_NON_SWASH_CC);
1679
f25ce844
KW
1680 return is_utf8_common(p,
1681 &PL_utf8_swash_ptrs[classnum],
1682 swash_property_names[classnum],
1683 PL_XPosix_ptrs[classnum]);
922e8cb4
KW
1684}
1685
1686bool
f2645549 1687Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
a0ed51b3 1688{
f2645549 1689 SV* invlist = NULL;
7918f24d 1690
f2645549 1691 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
7918f24d 1692
f2645549
KW
1693 if (! PL_utf8_perl_idstart) {
1694 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1695 }
60071a22 1696 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
82686b01
JH
1697}
1698
1699bool
f2645549 1700Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 1701{
f2645549 1702 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
1703
1704 if (*p == '_')
1705 return TRUE;
f25ce844 1706 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
1707}
1708
1709bool
eba68aa0
KW
1710Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1711{
b24b43f7 1712 SV* invlist = NULL;
eba68aa0
KW
1713
1714 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1715
b24b43f7
KW
1716 if (! PL_utf8_perl_idcont) {
1717 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1718 }
60071a22 1719 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
eba68aa0
KW
1720}
1721
eba68aa0 1722bool
f2645549 1723Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 1724{
f2645549 1725 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 1726
f25ce844 1727 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
1728}
1729
1730bool
f2645549 1731Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 1732{
f2645549 1733 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 1734
f25ce844 1735 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
1736}
1737
1738bool
7dbf68d2
KW
1739Perl__is_utf8_mark(pTHX_ const U8 *p)
1740{
7dbf68d2
KW
1741 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1742
f25ce844 1743 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
1744}
1745
6b5c0936 1746/*
87cea99e 1747=for apidoc to_utf8_case
6b5c0936 1748
6fae5207 1749C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
1750the character that is being converted. This routine assumes that the character
1751at C<p> is well-formed.
6b5c0936 1752
6fae5207
KW
1753C<ustrp> is a pointer to the character buffer to put the
1754conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
1755of the result.
1756
6fae5207 1757C<swashp> is a pointer to the swash to use.
6b5c0936 1758
a1433954 1759Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
6fae5207 1760and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 1761but not always, a multicharacter mapping), is tried first.
6b5c0936 1762
4a8240a3
KW
1763C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1764any special mappings; C<""> means to use the special mappings. Values other
1765than these two are treated as the name of the hash containing the special
1766mappings, like C<"utf8::ToSpecLower">.
6b5c0936 1767
6fae5207 1768C<normal> is a string like "ToLower" which means the swash
0134edef
JH
1769%utf8::ToLower.
1770
1771=cut */
6b5c0936 1772
2104c8d9 1773UV
9a957fbc
AL
1774Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1775 SV **swashp, const char *normal, const char *special)
a0ed51b3 1776{
0134edef 1777 STRLEN len = 0;
f4cd282c 1778 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
7918f24d
NC
1779
1780 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1781
9ae3ac1a
KW
1782 /* Note that swash_fetch() doesn't output warnings for these because it
1783 * assumes we will */
8457b38f 1784 if (uv1 >= UNICODE_SURROGATE_FIRST) {
9ae3ac1a 1785 if (uv1 <= UNICODE_SURROGATE_LAST) {
8457b38f
KW
1786 if (ckWARN_d(WARN_SURROGATE)) {
1787 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1788 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1789 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1790 }
9ae3ac1a
KW
1791 }
1792 else if (UNICODE_IS_SUPER(uv1)) {
8457b38f
KW
1793 if (ckWARN_d(WARN_NON_UNICODE)) {
1794 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1795 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1796 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1797 }
9ae3ac1a
KW
1798 }
1799
1800 /* Note that non-characters are perfectly legal, so no warning should
1801 * be given */
1802 }
1803
0134edef 1804 if (!*swashp) /* load on-demand */
5ab9d2ef 1805 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 1806
a6f87d8c 1807 if (special) {
0134edef 1808 /* It might be "special" (sometimes, but not always,
2a37f04d 1809 * a multicharacter mapping) */
4a8240a3 1810 HV *hv = NULL;
b08cf34e
JH
1811 SV **svp;
1812
4a8240a3
KW
1813 /* If passed in the specials name, use that; otherwise use any
1814 * given in the swash */
1815 if (*special != '\0') {
1816 hv = get_hv(special, 0);
1817 }
1818 else {
1819 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
1820 if (svp) {
1821 hv = MUTABLE_HV(SvRV(*svp));
1822 }
1823 }
1824
176fe009
KW
1825 if (hv
1826 && (svp = hv_fetch(hv, (const char*)p, UNISKIP(uv1), FALSE))
1827 && (*svp))
1828 {
cfd0369c 1829 const char *s;
47654450 1830
cfd0369c 1831 s = SvPV_const(*svp, len);
47654450 1832 if (len == 1)
f4cd282c 1833 /* EIGHTBIT */
c80e42f3 1834 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 1835 else {
d2dcd0fb 1836 Copy(s, ustrp, len, U8);
29e98929 1837 }
983ffd37 1838 }
0134edef
JH
1839 }
1840
1841 if (!len && *swashp) {
f4cd282c 1842 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is utf8 */);
d4c19fe8 1843
0134edef
JH
1844 if (uv2) {
1845 /* It was "normal" (a single character mapping). */
f4cd282c 1846 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
1847 }
1848 }
1feea2c7 1849
cbe07460
KW
1850 if (len) {
1851 if (lenp) {
1852 *lenp = len;
1853 }
1854 return valid_utf8_to_uvchr(ustrp, 0);
1855 }
1856
1857 /* Here, there was no mapping defined, which means that the code point maps
1858 * to itself. Return the inputs */
bfdf22ec 1859 len = UTF8SKIP(p);
ca9fab46
KW
1860 if (p != ustrp) { /* Don't copy onto itself */
1861 Copy(p, ustrp, len, U8);
1862 }
0134edef 1863
2a37f04d
JH
1864 if (lenp)
1865 *lenp = len;
1866
f4cd282c 1867 return uv1;
cbe07460 1868
a0ed51b3
LW
1869}
1870
051a06d4 1871STATIC UV
357aadde 1872S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
051a06d4
KW
1873{
1874 /* This is called when changing the case of a utf8-encoded character above
31f05a37
KW
1875 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
1876 * result contains a character that crosses the 255/256 boundary, disallow
1877 * the change, and return the original code point. See L<perlfunc/lc> for
1878 * why;
051a06d4 1879 *
a1433954
KW
1880 * p points to the original string whose case was changed; assumed
1881 * by this routine to be well-formed
051a06d4
KW
1882 * result the code point of the first character in the changed-case string
1883 * ustrp points to the changed-case string (<result> represents its first char)
1884 * lenp points to the length of <ustrp> */
1885
1886 UV original; /* To store the first code point of <p> */
1887
1888 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
1889
a4f12ed7 1890 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
1891
1892 /* We know immediately if the first character in the string crosses the
1893 * boundary, so can skip */
1894 if (result > 255) {
1895
1896 /* Look at every character in the result; if any cross the
1897 * boundary, the whole thing is disallowed */
1898 U8* s = ustrp + UTF8SKIP(ustrp);
1899 U8* e = ustrp + *lenp;
1900 while (s < e) {
a4f12ed7 1901 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
1902 goto bad_crossing;
1903 }
1904 s += UTF8SKIP(s);
1905 }
1906
613abc6d
KW
1907 /* Here, no characters crossed, result is ok as-is, but we warn. */
1908 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
1909 return result;
1910 }
1911
7b52d656 1912 bad_crossing:
051a06d4
KW
1913
1914 /* Failed, have to return the original */
4b88fb76 1915 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
1916
1917 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
1918 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1919 "Can't do %s(\"\\x{%"UVXf"}\") on non-UTF-8 locale; "
1920 "resolved to \"\\x{%"UVXf"}\".",
357aadde 1921 OP_DESC(PL_op),
ab0b796c
KW
1922 original,
1923 original);
051a06d4
KW
1924 Copy(p, ustrp, *lenp, char);
1925 return original;
1926}
1927
d3e79532 1928/*
87cea99e 1929=for apidoc to_utf8_upper
d3e79532 1930
1f607577 1931Instead use L</toUPPER_utf8>.
a1433954 1932
d3e79532
JH
1933=cut */
1934
051a06d4 1935/* Not currently externally documented, and subject to change:
31f05a37
KW
1936 * <flags> is set iff iff the rules from the current underlying locale are to
1937 * be used. */
051a06d4 1938
2104c8d9 1939UV
31f05a37 1940Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 1941{
051a06d4
KW
1942 UV result;
1943
1944 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 1945
780fcc9f
KW
1946 if (flags) {
1947 /* Treat a UTF-8 locale as not being in locale at all */
1948 if (IN_UTF8_CTYPE_LOCALE) {
1949 flags = FALSE;
1950 }
1951 else {
1952 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
1953 }
31f05a37
KW
1954 }
1955
3a4c58c9 1956 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
1957 if (flags) {
1958 result = toUPPER_LC(*p);
1959 }
1960 else {
81c6c7ce 1961 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 1962 }
3a4c58c9
KW
1963 }
1964 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 1965 if (flags) {
a6d8b88b 1966 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 1967 result = toUPPER_LC(c);
051a06d4
KW
1968 }
1969 else {
94bb8c36 1970 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 1971 ustrp, lenp, 'S');
051a06d4
KW
1972 }
1973 }
1974 else { /* utf8, ord above 255 */
1975 result = CALL_UPPER_CASE(p, ustrp, lenp);
1976
1977 if (flags) {
357aadde 1978 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4
KW
1979 }
1980 return result;
1981 }
1982
1983 /* Here, used locale rules. Convert back to utf8 */
1984 if (UTF8_IS_INVARIANT(result)) {
1985 *ustrp = (U8) result;
1986 *lenp = 1;
1987 }
1988 else {
62cb07ea
KW
1989 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
1990 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 1991 *lenp = 2;
3a4c58c9 1992 }
baa60164 1993
051a06d4 1994 return result;
983ffd37 1995}
a0ed51b3 1996
d3e79532 1997/*
87cea99e 1998=for apidoc to_utf8_title
d3e79532 1999
1f607577 2000Instead use L</toTITLE_utf8>.
a1433954 2001
d3e79532
JH
2002=cut */
2003
051a06d4 2004/* Not currently externally documented, and subject to change:
31f05a37
KW
2005 * <flags> is set iff the rules from the current underlying locale are to be
2006 * used. Since titlecase is not defined in POSIX, for other than a
2007 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 2008 */
051a06d4 2009
983ffd37 2010UV
31f05a37 2011Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
983ffd37 2012{
051a06d4
KW
2013 UV result;
2014
2015 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2016
780fcc9f
KW
2017 if (flags) {
2018 /* Treat a UTF-8 locale as not being in locale at all */
2019 if (IN_UTF8_CTYPE_LOCALE) {
2020 flags = FALSE;
2021 }
2022 else {
2023 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2024 }
31f05a37
KW
2025 }
2026
3a4c58c9 2027 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2028 if (flags) {
2029 result = toUPPER_LC(*p);
2030 }
2031 else {
81c6c7ce 2032 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2033 }
3a4c58c9
KW
2034 }
2035 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2036 if (flags) {
a6d8b88b 2037 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2038 result = toUPPER_LC(c);
051a06d4
KW
2039 }
2040 else {
94bb8c36 2041 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2042 ustrp, lenp, 's');
051a06d4
KW
2043 }
2044 }
2045 else { /* utf8, ord above 255 */
2046 result = CALL_TITLE_CASE(p, ustrp, lenp);
2047
2048 if (flags) {
357aadde 2049 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4
KW
2050 }
2051 return result;
2052 }
2053
2054 /* Here, used locale rules. Convert back to utf8 */
2055 if (UTF8_IS_INVARIANT(result)) {
2056 *ustrp = (U8) result;
2057 *lenp = 1;
2058 }
2059 else {
62cb07ea
KW
2060 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2061 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2062 *lenp = 2;
3a4c58c9
KW
2063 }
2064
051a06d4 2065 return result;
a0ed51b3
LW
2066}
2067
d3e79532 2068/*
87cea99e 2069=for apidoc to_utf8_lower
d3e79532 2070
1f607577 2071Instead use L</toLOWER_utf8>.
a1433954 2072
d3e79532
JH
2073=cut */
2074
051a06d4 2075/* Not currently externally documented, and subject to change:
31f05a37
KW
2076 * <flags> is set iff iff the rules from the current underlying locale are to
2077 * be used.
2078 */
051a06d4 2079
2104c8d9 2080UV
31f05a37 2081Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2082{
051a06d4
KW
2083 UV result;
2084
051a06d4 2085 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2086
780fcc9f
KW
2087 if (flags) {
2088 /* Treat a UTF-8 locale as not being in locale at all */
2089 if (IN_UTF8_CTYPE_LOCALE) {
2090 flags = FALSE;
2091 }
2092 else {
2093 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2094 }
31f05a37
KW
2095 }
2096
968c5e6a 2097 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2098 if (flags) {
2099 result = toLOWER_LC(*p);
2100 }
2101 else {
81c6c7ce 2102 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2103 }
968c5e6a
KW
2104 }
2105 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2106 if (flags) {
a6d8b88b 2107 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2108 result = toLOWER_LC(c);
051a06d4
KW
2109 }
2110 else {
94bb8c36 2111 return to_lower_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2112 ustrp, lenp);
051a06d4 2113 }
968c5e6a 2114 }
051a06d4
KW
2115 else { /* utf8, ord above 255 */
2116 result = CALL_LOWER_CASE(p, ustrp, lenp);
2117
2118 if (flags) {
357aadde 2119 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2120 }
968c5e6a 2121
051a06d4
KW
2122 return result;
2123 }
2124
2125 /* Here, used locale rules. Convert back to utf8 */
2126 if (UTF8_IS_INVARIANT(result)) {
2127 *ustrp = (U8) result;
2128 *lenp = 1;
2129 }
2130 else {
62cb07ea
KW
2131 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2132 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2133 *lenp = 2;
2134 }
2135
051a06d4 2136 return result;
b4e400f9
JH
2137}
2138
d3e79532 2139/*
87cea99e 2140=for apidoc to_utf8_fold
d3e79532 2141
1f607577 2142Instead use L</toFOLD_utf8>.
a1433954 2143
d3e79532
JH
2144=cut */
2145
051a06d4
KW
2146/* Not currently externally documented, and subject to change,
2147 * in <flags>
31f05a37
KW
2148 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2149 * locale are to be used.
051a06d4
KW
2150 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2151 * otherwise simple folds
a0270393
KW
2152 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2153 * prohibited
445bf929 2154 */
36bb2ab6 2155
b4e400f9 2156UV
445bf929 2157Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
b4e400f9 2158{
051a06d4
KW
2159 UV result;
2160
36bb2ab6 2161 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2162
a0270393
KW
2163 /* These are mutually exclusive */
2164 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2165
50ba90ff
KW
2166 assert(p != ustrp); /* Otherwise overwrites */
2167
780fcc9f
KW
2168 if (flags & FOLD_FLAGS_LOCALE) {
2169 /* Treat a UTF-8 locale as not being in locale at all */
2170 if (IN_UTF8_CTYPE_LOCALE) {
2171 flags &= ~FOLD_FLAGS_LOCALE;
2172 }
2173 else {
2174 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2175 }
31f05a37
KW
2176 }
2177
a1dde8de 2178 if (UTF8_IS_INVARIANT(*p)) {
051a06d4 2179 if (flags & FOLD_FLAGS_LOCALE) {
d22b930b 2180 result = toFOLD_LC(*p);
051a06d4
KW
2181 }
2182 else {
81c6c7ce 2183 return _to_fold_latin1(*p, ustrp, lenp,
1ca267a5 2184 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2185 }
a1dde8de
KW
2186 }
2187 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2188 if (flags & FOLD_FLAGS_LOCALE) {
a6d8b88b 2189 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2190 result = toFOLD_LC(c);
051a06d4
KW
2191 }
2192 else {
94bb8c36 2193 return _to_fold_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
51910141 2194 ustrp, lenp,
1ca267a5 2195 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2196 }
a1dde8de 2197 }
051a06d4 2198 else { /* utf8, ord above 255 */
a0270393 2199 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2200
1ca267a5
KW
2201 if (flags & FOLD_FLAGS_LOCALE) {
2202
76f2ffcd
KW
2203# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
2204# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
2205
2206 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
2207 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
2208
538e84ed
KW
2209 /* Special case these two characters, as what normally gets
2210 * returned under locale doesn't work */
76f2ffcd
KW
2211 if (UTF8SKIP(p) == cap_sharp_s_len
2212 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 2213 {
ab0b796c
KW
2214 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2215 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2216 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
2217 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
2218 goto return_long_s;
2219 }
76f2ffcd
KW
2220 else if (UTF8SKIP(p) == long_s_t_len
2221 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 2222 {
ab0b796c
KW
2223 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2224 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2225 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
2226 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
2227 goto return_ligature_st;
2228 }
357aadde 2229 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2230 }
a0270393
KW
2231 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2232 return result;
2233 }
2234 else {
2235 /* This is called when changing the case of a utf8-encoded
9fc2026f
KW
2236 * character above the ASCII range, and the result should not
2237 * contain an ASCII character. */
a0270393
KW
2238
2239 UV original; /* To store the first code point of <p> */
2240
2241 /* Look at every character in the result; if any cross the
2242 * boundary, the whole thing is disallowed */
2243 U8* s = ustrp;
2244 U8* e = ustrp + *lenp;
2245 while (s < e) {
2246 if (isASCII(*s)) {
2247 /* Crossed, have to return the original */
2248 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 2249
9fc2026f 2250 /* But in these instances, there is an alternative we can
1ca267a5 2251 * return that is valid */
9fc2026f
KW
2252 if (original == LATIN_CAPITAL_LETTER_SHARP_S
2253 || original == LATIN_SMALL_LETTER_SHARP_S)
2254 {
1ca267a5
KW
2255 goto return_long_s;
2256 }
9fc2026f
KW
2257 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2258 goto return_ligature_st;
2259 }
a0270393
KW
2260 Copy(p, ustrp, *lenp, char);
2261 return original;
2262 }
2263 s += UTF8SKIP(s);
2264 }
051a06d4 2265
a0270393
KW
2266 /* Here, no characters crossed, result is ok as-is */
2267 return result;
2268 }
051a06d4
KW
2269 }
2270
2271 /* Here, used locale rules. Convert back to utf8 */
2272 if (UTF8_IS_INVARIANT(result)) {
2273 *ustrp = (U8) result;
2274 *lenp = 1;
2275 }
2276 else {
62cb07ea
KW
2277 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2278 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2279 *lenp = 2;
2280 }
2281
051a06d4 2282 return result;
1ca267a5
KW
2283
2284 return_long_s:
2285 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2286 * folds to a string of two of these characters. By returning this
2287 * instead, then, e.g.,
2288 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2289 * works. */
2290
2291 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2292 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2293 ustrp, *lenp, U8);
2294 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
2295
2296 return_ligature_st:
2297 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2298 * have the other one fold to it */
2299
2300 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2301 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2302 return LATIN_SMALL_LIGATURE_ST;
a0ed51b3
LW
2303}
2304
711a919c 2305/* Note:
f90a9a02 2306 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2307 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2308 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2309 */
c4a5db0c 2310
a0ed51b3 2311SV*
7fc63493 2312Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2313{
c4a5db0c
KW
2314 PERL_ARGS_ASSERT_SWASH_INIT;
2315
2316 /* Returns a copy of a swash initiated by the called function. This is the
2317 * public interface, and returning a copy prevents others from doing
2318 * mischief on the original */
2319
5d3d13d1 2320 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
2321}
2322
2323SV*
5d3d13d1 2324Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c 2325{
2c1f00b9
YO
2326
2327 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
2328 * use the following define */
2329
2330#define CORE_SWASH_INIT_RETURN(x) \
2331 PL_curpm= old_PL_curpm; \
2332 return x
2333
c4a5db0c 2334 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
2335 * by calling utf8_heavy.pl in the general case. The returned value may be
2336 * the swash's inversion list instead if the input parameters allow it.
2337 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
2338 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2339 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
2340 *
2341 * This interface should only be used by functions that won't destroy or
2342 * adversely change the swash, as doing so affects all other uses of the
2343 * swash in the program; the general public should use 'Perl_swash_init'
2344 * instead.
2345 *
2346 * pkg is the name of the package that <name> should be in.
2347 * name is the name of the swash to find. Typically it is a Unicode
2348 * property name, including user-defined ones
2349 * listsv is a string to initialize the swash with. It must be of the form
2350 * documented as the subroutine return value in
2351 * L<perlunicode/User-Defined Character Properties>
2352 * minbits is the number of bits required to represent each data element.
2353 * It is '1' for binary properties.
2354 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 2355 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
2356 * flags_p if non-NULL is the address of various input and output flag bits
2357 * to the routine, as follows: ('I' means is input to the routine;
2358 * 'O' means output from the routine. Only flags marked O are
2359 * meaningful on return.)
2360 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2361 * came from a user-defined property. (I O)
5d3d13d1
KW
2362 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2363 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
2364 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2365 * return of an inversion list instead of a swash hash if this routine
2366 * thinks that would result in faster execution of swash_fetch() later
2367 * on. (I)
9a53f6cf
KW
2368 *
2369 * Thus there are three possible inputs to find the swash: <name>,
2370 * <listsv>, and <invlist>. At least one must be specified. The result
2371 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
2372 * actions can intersect, etc. what <name> gives. To avoid going out to
2373 * disk at all, <invlist> should specify completely what the swash should
2374 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
2375 *
2376 * <invlist> is only valid for binary properties */
c4a5db0c 2377
2c1f00b9
YO
2378 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
2379
c4a5db0c 2380 SV* retval = &PL_sv_undef;
83199d38 2381 HV* swash_hv = NULL;
87367d5f
KW
2382 const int invlist_swash_boundary =
2383 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2384 ? 512 /* Based on some benchmarking, but not extensive, see commit
2385 message */
2386 : -1; /* Never return just an inversion list */
9a53f6cf
KW
2387
2388 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2389 assert(! invlist || minbits == 1);
2390
2c1f00b9
YO
2391 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
2392 that triggered the swash init and the swash init perl logic itself.
2393 See perl #122747 */
2394
9a53f6cf
KW
2395 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2396 * so */
2397 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2398 dSP;
2399 const size_t pkg_len = strlen(pkg);
2400 const size_t name_len = strlen(name);
2401 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2402 SV* errsv_save;
2403 GV *method;
2404
2405 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2406
2407 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2408 ENTER;
69794297 2409 SAVEHINTS();
2782061f 2410 save_re_context();
650f067c
JL
2411 /* We might get here via a subroutine signature which uses a utf8
2412 * parameter name, at which point PL_subname will have been set
2413 * but not yet used. */
2414 save_item(PL_subname);
69794297
KW
2415 if (PL_parser && PL_parser->error_count)
2416 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2417 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2418 if (!method) { /* demand load utf8 */
2419 ENTER;
db2c6cb3
FC
2420 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2421 GvSV(PL_errgv) = NULL;
1a419e6b 2422#ifndef NO_TAINT_SUPPORT
69794297
KW
2423 /* It is assumed that callers of this routine are not passing in
2424 * any user derived data. */
2782061f
DM
2425 /* Need to do this after save_re_context() as it will set
2426 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2427 * in Perl_magic_get). Even line to create errsv_save can turn on
2428 * PL_tainted. */
284167a5
SM
2429 SAVEBOOL(TAINT_get);
2430 TAINT_NOT;
2431#endif
69794297
KW
2432 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2433 NULL);
eed484f9 2434 {
db2c6cb3
FC
2435 /* Not ERRSV, as there is no need to vivify a scalar we are
2436 about to discard. */
2437 SV * const errsv = GvSV(PL_errgv);
2438 if (!SvTRUE(errsv)) {
2439 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2440 SvREFCNT_dec(errsv);
2441 }
eed484f9 2442 }
69794297
KW
2443 LEAVE;
2444 }
2445 SPAGAIN;
2446 PUSHMARK(SP);
2447 EXTEND(SP,5);
2448 mPUSHp(pkg, pkg_len);
2449 mPUSHp(name, name_len);
2450 PUSHs(listsv);
2451 mPUSHi(minbits);
2452 mPUSHi(none);
2453 PUTBACK;
db2c6cb3
FC
2454 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2455 GvSV(PL_errgv) = NULL;
69794297
KW
2456 /* If we already have a pointer to the method, no need to use
2457 * call_method() to repeat the lookup. */
c41800a8
KW
2458 if (method
2459 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
2460 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2461 {
2462 retval = *PL_stack_sp--;
2463 SvREFCNT_inc(retval);
2464 }
eed484f9 2465 {
db2c6cb3
FC
2466 /* Not ERRSV. See above. */
2467 SV * const errsv = GvSV(PL_errgv);
2468 if (!SvTRUE(errsv)) {
2469 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2470 SvREFCNT_dec(errsv);
2471 }
eed484f9 2472 }
ce3b816e 2473 LEAVE;
69794297
KW
2474 POPSTACK;
2475 if (IN_PERL_COMPILETIME) {
2476 CopHINTS_set(PL_curcop, PL_hints);
2477 }
2478 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2479 if (SvPOK(retval))
2480
2481 /* If caller wants to handle missing properties, let them */
5d3d13d1 2482 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2c1f00b9 2483 CORE_SWASH_INIT_RETURN(NULL);
69794297
KW
2484 }
2485 Perl_croak(aTHX_
2486 "Can't find Unicode property definition \"%"SVf"\"",
2487 SVfARG(retval));
a25b5927 2488 NOT_REACHED; /* NOTREACHED */
69794297 2489 }
9a53f6cf 2490 } /* End of calling the module to find the swash */
36eb48b4 2491
83199d38
KW
2492 /* If this operation fetched a swash, and we will need it later, get it */
2493 if (retval != &PL_sv_undef
2494 && (minbits == 1 || (flags_p
2495 && ! (*flags_p
2496 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2497 {
2498 swash_hv = MUTABLE_HV(SvRV(retval));
2499
2500 /* If we don't already know that there is a user-defined component to
2501 * this swash, and the user has indicated they wish to know if there is
2502 * one (by passing <flags_p>), find out */
2503 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2504 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2505 if (user_defined && SvUV(*user_defined)) {
2506 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2507 }
2508 }
2509 }
2510
36eb48b4
KW
2511 /* Make sure there is an inversion list for binary properties */
2512 if (minbits == 1) {
2513 SV** swash_invlistsvp = NULL;
2514 SV* swash_invlist = NULL;
9a53f6cf 2515 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
2516 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2517 an unclaimed reference count */
36eb48b4 2518
9a53f6cf 2519 /* If this operation fetched a swash, get its already existing
83199d38 2520 * inversion list, or create one for it */
36eb48b4 2521
83199d38 2522 if (swash_hv) {
5c9f4bd2 2523 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
2524 if (swash_invlistsvp) {
2525 swash_invlist = *swash_invlistsvp;
2526 invlist_in_swash_is_valid = TRUE;
2527 }
2528 else {
36eb48b4 2529 swash_invlist = _swash_to_invlist(retval);
02c85471 2530 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
2531 }
2532 }
2533
2534 /* If an inversion list was passed in, have to include it */
2535 if (invlist) {
2536
2537 /* Any fetched swash will by now have an inversion list in it;
2538 * otherwise <swash_invlist> will be NULL, indicating that we
2539 * didn't fetch a swash */
2540 if (swash_invlist) {
2541
2542 /* Add the passed-in inversion list, which invalidates the one
2543 * already stored in the swash */
2544 invlist_in_swash_is_valid = FALSE;
2545 _invlist_union(invlist, swash_invlist, &swash_invlist);
2546 }
2547 else {
2548
87367d5f
KW
2549 /* Here, there is no swash already. Set up a minimal one, if
2550 * we are going to return a swash */
2551 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 2552 swash_hv = newHV();
4aca0fe6 2553 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 2554 }
9a53f6cf
KW
2555 swash_invlist = invlist;
2556 }
9a53f6cf
KW
2557 }
2558
2559 /* Here, we have computed the union of all the passed-in data. It may
2560 * be that there was an inversion list in the swash which didn't get
538e84ed 2561 * touched; otherwise save the computed one */
87367d5f
KW
2562 if (! invlist_in_swash_is_valid
2563 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2564 {
5c9f4bd2 2565 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
2566 {
2567 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2568 }
cc34d8c5
FC
2569 /* We just stole a reference count. */
2570 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2571 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 2572 }
87367d5f 2573
dbfdbd26
KW
2574 SvREADONLY_on(swash_invlist);
2575
c41800a8 2576 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
2577 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2578 SvREFCNT_dec(retval);
02c85471
FC
2579 if (!swash_invlist_unclaimed)
2580 SvREFCNT_inc_simple_void_NN(swash_invlist);
2581 retval = newRV_noinc(swash_invlist);
87367d5f 2582 }
36eb48b4
KW
2583 }
2584
2c1f00b9
YO
2585 CORE_SWASH_INIT_RETURN(retval);
2586#undef CORE_SWASH_INIT_RETURN
a0ed51b3
LW
2587}
2588
035d37be
JH
2589
2590/* This API is wrong for special case conversions since we may need to
2591 * return several Unicode characters for a single Unicode character
2592 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2593 * the lower-level routine, and it is similarly broken for returning
38684baa
KW
2594 * multiple values. --jhi
2595 * For those, you should use to_utf8_case() instead */
b0e3252e 2596/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
2597
2598/* Note:
2599 * Returns the value of property/mapping C<swash> for the first character
2600 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3d0f8846
KW
2601 * assumed to be in well-formed utf8. If C<do_utf8> is false, the string C<ptr>
2602 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
2603 *
2604 * A "swash" is a hash which contains initially the keys/values set up by
2605 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2606 * property for all possible code points. Things are stored in a compact form
2607 * (see utf8_heavy.pl) so that calculation is required to find the actual
2608 * property value for a given code point. As code points are looked up, new
2609 * key/value pairs are added to the hash, so that the calculation doesn't have
2610 * to ever be re-done. Further, each calculation is done, not just for the
2611 * desired one, but for a whole block of code points adjacent to that one.
2612 * For binary properties on ASCII machines, the block is usually for 64 code
2613 * points, starting with a code point evenly divisible by 64. Thus if the
2614 * property value for code point 257 is requested, the code goes out and
2615 * calculates the property values for all 64 code points between 256 and 319,
2616 * and stores these as a single 64-bit long bit vector, called a "swatch",
2617 * under the key for code point 256. The key is the UTF-8 encoding for code
2618 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2619 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2620 * for code point 258 is then requested, this code realizes that it would be
2621 * stored under the key for 256, and would find that value and extract the
2622 * relevant bit, offset from 256.
2623 *
2624 * Non-binary properties are stored in as many bits as necessary to represent
2625 * their values (32 currently, though the code is more general than that), not
2626 * as single bits, but the principal is the same: the value for each key is a
2627 * vector that encompasses the property values for all code points whose UTF-8
2628 * representations are represented by the key. That is, for all code points
2629 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2630 * bytes of that.
680c470c 2631 */
a0ed51b3 2632UV
680c470c 2633Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 2634{
ef8f7699 2635 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
2636 U32 klen;
2637 U32 off;
9b56a019 2638 STRLEN slen = 0;
7d85a32c 2639 STRLEN needents;
cfd0369c 2640 const U8 *tmps = NULL;
979f2922 2641 SV *swatch;
08fb1ac5 2642 const U8 c = *ptr;
3568d838 2643
7918f24d
NC
2644 PERL_ARGS_ASSERT_SWASH_FETCH;
2645
87367d5f
KW
2646 /* If it really isn't a hash, it isn't really swash; must be an inversion
2647 * list */
2648 if (SvTYPE(hv) != SVt_PVHV) {
2649 return _invlist_contains_cp((SV*)hv,
2650 (do_utf8)
2651 ? valid_utf8_to_uvchr(ptr, NULL)
2652 : c);
2653 }
2654
08fb1ac5
KW
2655 /* We store the values in a "swatch" which is a vec() value in a swash
2656 * hash. Code points 0-255 are a single vec() stored with key length
2657 * (klen) 0. All other code points have a UTF-8 representation
2658 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2659 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2660 * length for them is the length of the encoded char - 1. ptr[klen] is the
2661 * final byte in the sequence representing the character */
2662 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2663 klen = 0;
2664 needents = 256;
2665 off = c;
3568d838 2666 }
08fb1ac5
KW
2667 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2668 klen = 0;
2669 needents = 256;
2670 off = TWO_BYTE_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
2671 }
2672 else {
08fb1ac5
KW
2673 klen = UTF8SKIP(ptr) - 1;
2674
2675 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2676 * the vec is the final byte in the sequence. (In EBCDIC this is
2677 * converted to I8 to get consecutive values.) To help you visualize
2678 * all this:
2679 * Straight 1047 After final byte
2680 * UTF-8 UTF-EBCDIC I8 transform
2681 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2682 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2683 * ...
2684 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2685 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2686 * ...
2687 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2688 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2689 * ...
2690 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2691 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2692 * ...
2693 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2694 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2695 *
2696 * (There are no discontinuities in the elided (...) entries.)
2697 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2698 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2699 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2700 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2701 * index into the vec() swatch (after subtracting 0x80, which we
2702 * actually do with an '&').
2703 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2704 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2705 * dicontinuities which go away by transforming it into I8, and we
2706 * effectively subtract 0xA0 to get the index. */
979f2922 2707 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 2708 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 2709 }
7d85a32c 2710
a0ed51b3
LW
2711 /*
2712 * This single-entry cache saves about 1/3 of the utf8 overhead in test
2713 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2714 * it's nothing to sniff at.) Pity we usually come through at least
2715 * two function calls to get here...
2716 *
2717 * NB: this code assumes that swatches are never modified, once generated!
2718 */
2719
3568d838 2720 if (hv == PL_last_swash_hv &&
a0ed51b3 2721 klen == PL_last_swash_klen &&
27da23d5 2722 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
2723 {
2724 tmps = PL_last_swash_tmps;
2725 slen = PL_last_swash_slen;
2726 }
2727 else {
2728 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 2729 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 2730
b0e3252e 2731 /* If not cached, generate it via swatch_get */
979f2922 2732 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
2733 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2734 {
2735 if (klen) {
2736 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2737 swatch = swatch_get(swash,
2738 code_point & ~((UV)needents - 1),
2739 needents);
2740 }
2741 else { /* For the first 256 code points, the swatch has a key of
2742 length 0 */
2743 swatch = swatch_get(swash, 0, needents);
2744 }
979f2922 2745
923e4eb5 2746 if (IN_PERL_COMPILETIME)
623e6609 2747 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 2748
979f2922 2749 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 2750
979f2922
ST
2751 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2752 || (slen << 3) < needents)
5637ef5b
NC
2753 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2754 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2755 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
2756 }
2757
2758 PL_last_swash_hv = hv;
16d8f38a 2759 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 2760 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
2761 /* FIXME change interpvar.h? */
2762 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
2763 PL_last_swash_slen = slen;
2764 if (klen)
2765 Copy(ptr, PL_last_swash_key, klen, U8);
2766 }
2767
9faf8d75 2768 switch ((int)((slen << 3) / needents)) {
a0ed51b3 2769 case 1:
e7aca353 2770 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
a0ed51b3 2771 case 8:
e7aca353 2772 return ((UV) tmps[off]);
a0ed51b3
LW
2773 case 16:
2774 off <<= 1;
e7aca353
JH
2775 return
2776 ((UV) tmps[off ] << 8) +
2777 ((UV) tmps[off + 1]);
a0ed51b3
LW
2778 case 32:
2779 off <<= 2;
e7aca353
JH
2780 return
2781 ((UV) tmps[off ] << 24) +
2782 ((UV) tmps[off + 1] << 16) +
2783 ((UV) tmps[off + 2] << 8) +
2784 ((UV) tmps[off + 3]);
a0ed51b3 2785 }
5637ef5b
NC
2786 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2787 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 2788 NORETURN_FUNCTION_END;
a0ed51b3 2789}
2b9d42f0 2790
319009ee
KW
2791/* Read a single line of the main body of the swash input text. These are of
2792 * the form:
2793 * 0053 0056 0073
2794 * where each number is hex. The first two numbers form the minimum and
2795 * maximum of a range, and the third is the value associated with the range.
2796 * Not all swashes should have a third number
2797 *
2798 * On input: l points to the beginning of the line to be examined; it points
2799 * to somewhere in the string of the whole input text, and is
2800 * terminated by a \n or the null string terminator.
2801 * lend points to the null terminator of that string
2802 * wants_value is non-zero if the swash expects a third number
2803 * typestr is the name of the swash's mapping, like 'ToLower'
2804 * On output: *min, *max, and *val are set to the values read from the line.
2805 * returns a pointer just beyond the line examined. If there was no
2806 * valid min number on the line, returns lend+1
2807 */
2808
2809STATIC U8*
2810S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2811 const bool wants_value, const U8* const typestr)
2812{
2813 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2814 STRLEN numlen; /* Length of the number */
02470786
KW
2815 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2816 | PERL_SCAN_DISALLOW_PREFIX
2817 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2818
2819 /* nl points to the next \n in the scan */
2820 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2821
95543e92
KW
2822 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
2823
319009ee
KW
2824 /* Get the first number on the line: the range minimum */
2825 numlen = lend - l;
2826 *min = grok_hex((char *)l, &numlen, &flags, NULL);
c88850db 2827 *max = *min; /* So can never return without setting max */
319009ee
KW
2828 if (numlen) /* If found a hex number, position past it */
2829 l += numlen;
2830 else if (nl) { /* Else, go handle next line, if any */
2831 return nl + 1; /* 1 is length of "\n" */
2832 }
2833 else { /* Else, no next line */
2834 return lend + 1; /* to LIST's end at which \n is not found */
2835 }
2836
2837 /* The max range value follows, separated by a BLANK */
2838 if (isBLANK(*l)) {
2839 ++l;
02470786
KW
2840 flags = PERL_SCAN_SILENT_ILLDIGIT
2841 | PERL_SCAN_DISALLOW_PREFIX
2842 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2843 numlen = lend - l;
2844 *max = grok_hex((char *)l, &numlen, &flags, NULL);
2845 if (numlen)
2846 l += numlen;
2847 else /* If no value here, it is a single element range */
2848 *max = *min;
2849
2850 /* Non-binary tables have a third entry: what the first element of the
24303724 2851 * range maps to. The map for those currently read here is in hex */
319009ee
KW
2852 if (wants_value) {
2853 if (isBLANK(*l)) {
2854 ++l;
f2a7d0fc
KW
2855 flags = PERL_SCAN_SILENT_ILLDIGIT
2856 | PERL_SCAN_DISALLOW_PREFIX
2857 | PERL_SCAN_SILENT_NON_PORTABLE;
2858 numlen = lend - l;
2859 *val = grok_hex((char *)l, &numlen, &flags, NULL);
2860 if (numlen)
2861 l += numlen;
2862 else
2863 *val = 0;
319009ee
KW
2864 }
2865 else {
2866 *val = 0;
2867 if (typeto) {
dcbac5bb 2868 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
2869 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
2870 typestr, l);
2871 }
2872 }
2873 }
2874 else
2875 *val = 0; /* bits == 1, then any val should be ignored */
2876 }
2877 else { /* Nothing following range min, should be single element with no
2878 mapping expected */
319009ee
KW
2879 if (wants_value) {
2880 *val = 0;
2881 if (typeto) {
dcbac5bb 2882 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
2883 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
2884 }
2885 }
2886 else
2887 *val = 0; /* bits == 1, then val should be ignored */
2888 }
2889
2890 /* Position to next line if any, or EOF */
2891 if (nl)
2892 l = nl + 1;
2893 else
2894 l = lend;
2895
2896 return l;
2897}
2898
979f2922
ST
2899/* Note:
2900 * Returns a swatch (a bit vector string) for a code point sequence
2901 * that starts from the value C<start> and comprises the number C<span>.
2902 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
2903 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
2904 */
2905STATIC SV*
b0e3252e 2906S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
2907{
2908 SV *swatch;
77f9f126 2909 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 2910 STRLEN lcur, xcur, scur;
ef8f7699 2911 HV *const hv = MUTABLE_HV(SvRV(swash));
5c9f4bd2 2912 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
36eb48b4 2913
88d45d28
KW
2914 SV** listsvp = NULL; /* The string containing the main body of the table */
2915 SV** extssvp = NULL;
2916 SV** invert_it_svp = NULL;
2917 U8* typestr = NULL;
786861f5
KW
2918 STRLEN bits;
2919 STRLEN octets; /* if bits == 1, then octets == 0 */
2920 UV none;
2921 UV end = start + span;
972dd592 2922
36eb48b4 2923 if (invlistsvp == NULL) {
786861f5
KW
2924 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
2925 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
2926 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
2927 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
2928 listsvp = hv_fetchs(hv, "LIST", FALSE);
2929 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
2930
2931 bits = SvUV(*bitssvp);
2932 none = SvUV(*nonesvp);
2933 typestr = (U8*)SvPV_nolen(*typesvp);
2934 }
36eb48b4
KW
2935 else {
2936 bits = 1;
2937 none = 0;
2938 }
786861f5 2939 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 2940
b0e3252e 2941 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 2942
979f2922 2943 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 2944 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 2945 (UV)bits);
979f2922
ST
2946 }
2947
84ea5ef6
KW
2948 /* If overflowed, use the max possible */
2949 if (end < start) {
2950 end = UV_MAX;
2951 span = end - start;
2952 }
2953
979f2922 2954 /* create and initialize $swatch */
979f2922 2955 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
2956 swatch = newSV(scur);
2957 SvPOK_on(swatch);
979f2922
ST
2958 s = (U8*)SvPVX(swatch);
2959 if (octets && none) {
0bd48802 2960 const U8* const e = s + scur;
979f2922
ST
2961 while (s < e) {
2962 if (bits == 8)
2963 *s++ = (U8)(none & 0xff);
2964 else if (bits == 16) {
2965 *s++ = (U8)((none >> 8) & 0xff);
2966 *s++ = (U8)( none & 0xff);
2967 }
2968 else if (bits == 32) {
2969 *s++ = (U8)((none >> 24) & 0xff);
2970 *s++ = (U8)((none >> 16) & 0xff);
2971 *s++ = (U8)((none >> 8) & 0xff);
2972 *s++ = (U8)( none & 0xff);
2973 }
2974 }
2975 *s = '\0';
2976 }
2977 else {
2978 (void)memzero((U8*)s, scur + 1);
2979 }
2980 SvCUR_set(swatch, scur);
2981 s = (U8*)SvPVX(swatch);
2982
36eb48b4
KW
2983 if (invlistsvp) { /* If has an inversion list set up use that */
2984 _invlist_populate_swatch(*invlistsvp, start, end, s);
2985 return swatch;
2986 }
2987
2988 /* read $swash->{LIST} */
979f2922
ST
2989 l = (U8*)SvPV(*listsvp, lcur);
2990 lend = l + lcur;
2991 while (l < lend) {
8ed25d53 2992 UV min, max, val, upper;
95543e92
KW
2993 l = swash_scan_list_line(l, lend, &min, &max, &val,
2994 cBOOL(octets), typestr);
319009ee 2995 if (l > lend) {
979f2922
ST
2996 break;
2997 }
2998
972dd592 2999 /* If looking for something beyond this range, go try the next one */
979f2922
ST
3000 if (max < start)
3001 continue;
3002
8ed25d53
KW
3003 /* <end> is generally 1 beyond where we want to set things, but at the
3004 * platform's infinity, where we can't go any higher, we want to
3005 * include the code point at <end> */
3006 upper = (max < end)
3007 ? max
3008 : (max != UV_MAX || end != UV_MAX)
3009 ? end - 1
3010 : end;
3011
979f2922 3012 if (octets) {
35da51f7 3013 UV key;
979f2922
ST
3014 if (min < start) {
3015 if (!none || val < none) {
3016 val += start - min;
3017 }
3018 min = start;
3019 }
8ed25d53 3020 for (key = min; key <= upper; key++) {
979f2922 3021 STRLEN offset;
979f2922
ST
3022 /* offset must be non-negative (start <= min <= key < end) */
3023 offset = octets * (key - start);
3024 if (bits == 8)
3025 s[offset] = (U8)(val & 0xff);
3026 else if (bits == 16) {
3027 s[offset ] = (U8)((val >> 8) & 0xff);
3028 s[offset + 1] = (U8)( val & 0xff);
3029 }
3030 else if (bits == 32) {
3031 s[offset ] = (U8)((val >> 24) & 0xff);
3032 s[offset + 1] = (U8)((val >> 16) & 0xff);
3033 s[offset + 2] = (U8)((val >> 8) & 0xff);
3034 s[offset + 3] = (U8)( val & 0xff);
3035 }
3036
3037 if (!none || val < none)
3038 ++val;
3039 }
3040 }
711a919c 3041 else { /* bits == 1, then val should be ignored */
35da51f7 3042 UV key;
979f2922
ST
3043 if (min < start)
3044 min = start;
6cb05c12 3045
8ed25d53 3046 for (key = min; key <= upper; key++) {
0bd48802 3047 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3048 s[offset >> 3] |= 1 << (offset & 7);
3049 }
3050 }
3051 } /* while */
979f2922 3052
9479a769 3053 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3054 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3055
3056 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3057 * be 0, and their inversion should also be 0, as we don't succeed any
3058 * Unicode property matches for non-Unicode code points */
3059 if (start <= PERL_UNICODE_MAX) {
3060
3061 /* The code below assumes that we never cross the
3062 * Unicode/above-Unicode boundary in a range, as otherwise we would
3063 * have to figure out where to stop flipping the bits. Since this
3064 * boundary is divisible by a large power of 2, and swatches comes
3065 * in small powers of 2, this should be a valid assumption */
3066 assert(start + span - 1 <= PERL_UNICODE_MAX);
3067
507a8485
KW
3068 send = s + scur;
3069 while (s < send) {
3070 *s = ~(*s);
3071 s++;
3072 }
0bda3001 3073 }
77f9f126
KW
3074 }
3075
d73c39c5
KW
3076 /* read $swash->{EXTRAS}
3077 * This code also copied to swash_to_invlist() below */
979f2922
ST
3078 x = (U8*)SvPV(*extssvp, xcur);
3079 xend = x + xcur;
3080 while (x < xend) {
3081 STRLEN namelen;
3082 U8 *namestr;
3083 SV** othersvp;
3084 HV* otherhv;
3085 STRLEN otherbits;
3086 SV **otherbitssvp, *other;
711a919c 3087 U8 *s, *o, *nl;
979f2922
ST
3088 STRLEN slen, olen;
3089
35da51f7 3090 const U8 opc = *x++;
979f2922
ST
3091 if (opc == '\n')
3092 continue;
3093
3094 nl = (U8*)memchr(x, '\n', xend - x);
3095
3096 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3097 if (nl) {
3098 x = nl + 1; /* 1 is length of "\n" */
3099 continue;
3100 }
3101 else {
3102 x = xend; /* to EXTRAS' end at which \n is not found */
3103 break;
3104 }
3105 }
3106
3107 namestr = x;
3108 if (nl) {
3109 namelen = nl - namestr;
3110 x = nl + 1;
3111 }
3112 else {
3113 namelen = xend - namestr;
3114 x = xend;
3115 }
3116
3117 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3118 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3119 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3120 otherbits = (STRLEN)SvUV(*otherbitssvp);
3121 if (bits < otherbits)
5637ef5b
NC
3122 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3123 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3124
3125 /* The "other" swatch must be destroyed after. */
b0e3252e 3126 other = swatch_get(*othersvp, start, span);
979f2922
ST
3127 o = (U8*)SvPV(other, olen);
3128
3129 if (!olen)
b0e3252e 3130 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3131
3132 s = (U8*)SvPV(swatch, slen);
3133 if (bits == 1 && otherbits == 1) {
3134 if (slen != olen)
5637ef5b
NC
3135 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3136 "mismatch, slen=%"UVuf", olen=%"UVuf,
3137 (UV)slen, (UV)olen);
979f2922
ST
3138
3139 switch (opc) {
3140 case '+':
3141 while (slen--)
3142 *s++ |= *o++;
3143 break;
3144 case '!':
3145 while (slen--)
3146 *s++ |= ~*o++;
3147 break;
3148 case '-':
3149 while (slen--)
3150 *s++ &= ~*o++;
3151 break;
3152 case '&':
3153 while (slen--)
3154 *s++ &= *o++;
3155 break;
3156 default:
3157 break;
3158 }
3159 }
711a919c 3160 else {
979f2922
ST
3161 STRLEN otheroctets = otherbits >> 3;
3162 STRLEN offset = 0;
35da51f7 3163 U8* const send = s + slen;
979f2922
ST
3164
3165 while (s < send) {
3166 UV otherval = 0;
3167
3168 if (otherbits == 1) {
3169 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3170 ++offset;
3171 }
3172 else {
3173 STRLEN vlen = otheroctets;
3174 otherval = *o++;
3175 while (--vlen) {
3176 otherval <<= 8;
3177 otherval |= *o++;
3178 }
3179 }
3180
711a919c 3181 if (opc == '+' && otherval)
6f207bd3 3182 NOOP; /* replace with otherval */
979f2922
ST
3183 else if (opc == '!' && !otherval)
3184 otherval = 1;
3185 else if (opc == '-' && otherval)
3186 otherval = 0;
3187 else if (opc == '&' && !otherval)
3188 otherval = 0;
3189 else {
711a919c 3190 s += octets; /* no replacement */
979f2922
ST
3191 continue;
3192 }
3193
3194 if (bits == 8)
3195 *s++ = (U8)( otherval & 0xff);
3196 else if (bits == 16) {
3197 *s++ = (U8)((otherval >> 8) & 0xff);
3198 *s++ = (U8)( otherval & 0xff);
3199 }
3200 else if (bits == 32) {
3201 *s++ = (U8)((otherval >> 24) & 0xff);
3202 *s++ = (U8)((otherval >> 16) & 0xff);
3203 *s++ = (U8)((otherval >> 8) & 0xff);
3204 *s++ = (U8)( otherval & 0xff);
3205 }
3206 }
3207 }
3208 sv_free(other); /* through with it! */
3209 } /* while */
3210 return swatch;
3211}
3212
064c021d 3213HV*
4c2e1131 3214Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3215{
3216
79a2a0e8 3217 /* Subject to change or removal. For use only in regcomp.c and regexec.c
5662e334
KW
3218 * Can't be used on a property that is subject to user override, as it
3219 * relies on the value of SPECIALS in the swash which would be set by
3220 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3221 * for overridden properties
064c021d
KW
3222 *
3223 * Returns a hash which is the inversion and closure of a swash mapping.
3224 * For example, consider the input lines:
3225 * 004B 006B
3226 * 004C 006C
3227 * 212A 006B
3228 *
3229 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3230 * 006C. The value for each key is an array. For 006C, the array would
17d5c697 3231 * have two elements, the utf8 for itself, and for 004C. For 006B, there
064c021d
KW
3232 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3233 *
538e84ed
KW
3234 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3235 * keys are only code points that are folded-to, so it isn't a full closure.
3236 *
064c021d
KW
3237 * Essentially, for any code point, it gives all the code points that map to
3238 * it, or the list of 'froms' for that point.
3239 *
5662e334
KW
3240 * Currently it ignores any additions or deletions from other swashes,
3241 * looking at just the main body of the swash, and if there are SPECIALS
3242 * in the swash, at that hash
3243 *
3244 * The specials hash can be extra code points, and most likely consists of
3245 * maps from single code points to multiple ones (each expressed as a string
3246 * of utf8 characters). This function currently returns only 1-1 mappings.
3247 * However consider this possible input in the specials hash:
3248 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3249 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3250 *
3251 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3252 * currently handle. But it also means that FB05 and FB06 are equivalent in
3253 * a 1-1 mapping which we should handle, and this relationship may not be in
3254 * the main table. Therefore this function examines all the multi-char
3255 * sequences and adds the 1-1 mappings that come out of that. */
064c021d
KW
3256
3257 U8 *l, *lend;
3258 STRLEN lcur;
3259 HV *const hv = MUTABLE_HV(SvRV(swash));
3260
923b6d4e
KW
3261 /* The string containing the main body of the table. This will have its
3262 * assertion fail if the swash has been converted to its inversion list */
064c021d
KW
3263 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3264
3265 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3266 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3267 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3268 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3269 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3270 const STRLEN bits = SvUV(*bitssvp);
3271 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3272 const UV none = SvUV(*nonesvp);
5662e334 3273 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
064c021d
KW
3274
3275 HV* ret = newHV();
3276
3277 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3278
3279 /* Must have at least 8 bits to get the mappings */
3280 if (bits != 8 && bits != 16 && bits != 32) {
3281 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3282 (UV)bits);
3283 }
3284
5662e334
KW
3285 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3286 mapping to more than one character */
3287
3288 /* Construct an inverse mapping hash for the specials */
3289 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3290 HV * specials_inverse = newHV();
3291 char *char_from; /* the lhs of the map */
3292 I32 from_len; /* its byte length */
3293 char *char_to; /* the rhs of the map */
3294 I32 to_len; /* its byte length */
3295 SV *sv_to; /* and in a sv */
3296 AV* from_list; /* list of things that map to each 'to' */
3297
3298 hv_iterinit(specials_hv);
3299
3300 /* The keys are the characters (in utf8) that map to the corresponding
3301 * utf8 string value. Iterate through the list creating the inverse
3302 * list. */
3303 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3304 SV** listp;
3305 if (! SvPOK(sv_to)) {
5637ef5b
NC
3306 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3307 "unexpectedly is not a string, flags=%lu",
3308 (unsigned long)SvFLAGS(sv_to));
5662e334 3309 }
4b88fb76 3310 /*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
3311
3312 /* Each key in the inverse list is a mapped-to value, and the key's
3313 * hash value is a list of the strings (each in utf8) that map to
3314 * it. Those strings are all one character long */
3315 if ((listp = hv_fetch(specials_inverse,
3316 SvPVX(sv_to),
3317 SvCUR(sv_to), 0)))
3318 {
3319 from_list = (AV*) *listp;
3320 }
3321 else { /* No entry yet for it: create one */
3322 from_list = newAV();
3323 if (! hv_store(specials_inverse,
3324 SvPVX(sv_to),
3325 SvCUR(sv_to),
3326 (SV*) from_list, 0))
3327 {
3328 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3329 }
3330 }
3331
3332 /* Here have the list associated with this 'to' (perhaps newly
3333 * created and empty). Just add to it. Note that we ASSUME that
3334 * the input is guaranteed to not have duplications, so we don't
3335 * check for that. Duplications just slow down execution time. */
3336 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3337 }
3338
3339 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3340 * it looking for cases like the FB05/FB06 examples above. There would
3341 * be an entry in the hash like
3342 * 'st' => [ FB05, FB06 ]
3343 * In this example we will create two lists that get stored in the
3344 * returned hash, 'ret':
3345 * FB05 => [ FB05, FB06 ]
3346 * FB06 => [ FB05, FB06 ]
3347 *
3348 * Note that there is nothing to do if the array only has one element.
3349 * (In the normal 1-1 case handled below, we don't have to worry about
3350 * two lists, as everything gets tied to the single list that is
3351 * generated for the single character 'to'. But here, we are omitting
3352 * that list, ('st' in the example), so must have multiple lists.) */
3353 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3354 &char_to, &to_len)))
3355 {
b9f2b683 3356 if (av_tindex(from_list) > 0) {
c70927a6 3357 SSize_t i;
5662e334
KW
3358
3359 /* We iterate over all combinations of i,j to place each code
3360 * point on each list */
b9f2b683 3361 for (i = 0; i <= av_tindex(from_list); i++) {
c70927a6 3362 SSize_t j;
5662e334
KW
3363 AV* i_list = newAV();
3364 SV** entryp = av_fetch(from_list, i, FALSE);
3365 if (entryp == NULL) {
3366 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3367 }
3368 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3369 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3370 }
3371 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3372 (SV*) i_list, FALSE))
3373 {
3374 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3375 }
3376
538e84ed 3377 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
b9f2b683 3378 for (j = 0; j <= av_tindex(from_list); j++) {
5662e334
KW
3379 entryp = av_fetch(from_list, j, FALSE);
3380 if (entryp == NULL) {
3381 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3382 }
3383
3384 /* When i==j this adds itself to the list */
4b88fb76
KW
3385 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3386 (U8*) SvPVX(*entryp),
3387 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3388 0)));
4637d003 3389 /*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
3390 }
3391 }
3392 }
3393 }
3394 SvREFCNT_dec(specials_inverse); /* done with it */
3395 } /* End of specials */
3396
064c021d
KW
3397 /* read $swash->{LIST} */
3398 l = (U8*)SvPV(*listsvp, lcur);
3399 lend = l + lcur;
3400
3401 /* Go through each input line */
3402 while (l < lend) {
3403 UV min, max, val;
3404 UV inverse;
95543e92
KW
3405 l = swash_scan_list_line(l, lend, &min, &max, &val,
3406 cBOOL(octets), typestr);
064c021d
KW
3407 if (l > lend) {
3408 break;
3409 }
3410
3411 /* Each element in the range is to be inverted */
3412 for (inverse = min; inverse <= max; inverse++) {
3413 AV* list;
064c021d
KW
3414 SV** listp;
3415 IV i;
3416 bool found_key = FALSE;
5662e334 3417 bool found_inverse = FALSE;
064c021d
KW
3418
3419 /* The key is the inverse mapping */
3420 char key[UTF8_MAXBYTES+1];
c80e42f3 3421 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
064c021d
KW
3422 STRLEN key_len = key_end - key;
3423
064c021d
KW
3424 /* Get the list for the map */
3425 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3426 list = (AV*) *listp;
3427 }
3428 else { /* No entry yet for it: create one */
3429 list = newAV();
3430 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3431 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3432 }
3433 }
3434
5662e334
KW
3435 /* Look through list to see if this inverse mapping already is
3436 * listed, or if there is a mapping to itself already */
b9f2b683 3437 for (i = 0; i <= av_tindex(list); i++) {
064c021d
KW
3438 SV** entryp = av_fetch(list, i, FALSE);
3439 SV* entry;
414db8a1 3440 UV uv;
064c021d
KW
3441 if (entryp == NULL) {
3442 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3443 }
3444 entry = *entryp;
414db8a1
DM
3445 uv = SvUV(entry);
3446 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, uv));*/
3447 if (uv == val) {
064c021d 3448 found_key = TRUE;
5662e334 3449 }
414db8a1 3450 if (uv == inverse) {
5662e334
KW
3451 found_inverse = TRUE;
3452 }
3453
3454 /* No need to continue searching if found everything we are
3455 * looking for */
3456 if (found_key && found_inverse) {
064c021d
KW
3457 break;
3458 }
3459 }
56ca34ca
KW
3460
3461 /* Make sure there is a mapping to itself on the list */
064c021d 3462 if (! found_key) {
d397ff6a 3463 av_push(list, newSVuv(val));
4637d003 3464 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
064c021d
KW
3465 }
3466
3467
3468 /* Simply add the value to the list */
5662e334
KW
3469 if (! found_inverse) {
3470 av_push(list, newSVuv(inverse));
4637d003 3471 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
5662e334 3472 }
064c021d 3473
b0e3252e 3474 /* swatch_get() increments the value of val for each element in the
064c021d
KW
3475 * range. That makes more compact tables possible. You can
3476 * express the capitalization, for example, of all consecutive
3477 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3478 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
bd3f2f94 3479 * and it's not documented; it appears to be used only in
b0e3252e 3480 * implementing tr//; I copied the semantics from swatch_get(), just
bd3f2f94 3481 * in case */
064c021d
KW
3482 if (!none || val < none) {
3483 ++val;
3484 }
3485 }
3486 }
3487
3488 return ret;
3489}
3490
a25abddc 3491SV*
d764b54e
KW
3492Perl__swash_to_invlist(pTHX_ SV* const swash)
3493{
3494
ed92f1b3
KW
3495 /* Subject to change or removal. For use only in one place in regcomp.c.
3496 * Ownership is given to one reference count in the returned SV* */
d764b54e
KW
3497
3498 U8 *l, *lend;
3499 char *loc;
3500 STRLEN lcur;
3501 HV *const hv = MUTABLE_HV(SvRV(swash));
3502 UV elements = 0; /* Number of elements in the inversion list */
b443038a 3503 U8 empty[] = "";
923b6d4e
KW
3504 SV** listsvp;
3505 SV** typesvp;
3506 SV** bitssvp;
3507 SV** extssvp;
3508 SV** invert_it_svp;
d764b54e 3509
923b6d4e
KW
3510 U8* typestr;
3511 STRLEN bits;
3512 STRLEN octets; /* if bits == 1, then octets == 0 */
d73c39c5
KW
3513 U8 *x, *xend;
3514 STRLEN xcur;
d764b54e 3515
a25abddc 3516 SV* invlist;
d764b54e 3517
b81740c0
KW
3518 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3519
923b6d4e
KW
3520 /* If not a hash, it must be the swash's inversion list instead */
3521 if (SvTYPE(hv) != SVt_PVHV) {
ed92f1b3 3522 return SvREFCNT_inc_simple_NN((SV*) hv);
923b6d4e
KW
3523 }
3524
3525 /* The string containing the main body of the table */
3526 listsvp = hv_fetchs(hv, "LIST", FALSE);
3527 typesvp = hv_fetchs(hv, "TYPE", FALSE);
3528 bitssvp = hv_fetchs(hv, "BITS", FALSE);
3529 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3530 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3531
3532 typestr = (U8*)SvPV_nolen(*typesvp);
3533 bits = SvUV(*bitssvp);
3534 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3535
d764b54e 3536 /* read $swash->{LIST} */
b443038a
KW
3537 if (SvPOK(*listsvp)) {
3538 l = (U8*)SvPV(*listsvp, lcur);
3539 }
3540 else {
3541 /* LIST legitimately doesn't contain a string during compilation phases
3542 * of Perl itself, before the Unicode tables are generated. In this
3543 * case, just fake things up by creating an empty list */
3544 l = empty;
3545 lcur = 0;
3546 }
d764b54e
KW
3547 loc = (char *) l;
3548 lend = l + lcur;
3549
31aa6e0b 3550 if (*l == 'V') { /* Inversion list format */
99944878 3551 const char *after_atou = (char *) lend;
31aa6e0b
KW
3552 UV element0;
3553 UV* other_elements_ptr;
3554
3555 /* The first number is a count of the rest */
3556 l++;
22ff3130
HS
3557 if (!grok_atoUV((const char *)l, &elements, &after_atou)) {
3558 Perl_croak(aTHX_ "panic: Expecting a valid count of elements at start of inversion list");
3559 }
eb092534
KW
3560 if (elements == 0) {
3561 invlist = _new_invlist(0);
3562 }
3563 else {
99944878
JH
3564 while (isSPACE(*l)) l++;
3565 l = (U8 *) after_atou;
1f9f7d4c
KW
3566
3567 /* Get the 0th element, which is needed to setup the inversion list */
99944878 3568 while (isSPACE(*l)) l++;
22ff3130
HS
3569 if (!grok_atoUV((const char *)l, &element0, &after_atou)) {
3570 Perl_croak(aTHX_ "panic: Expecting a valid 0th element for inversion list");
3571 }
99944878 3572 l = (U8 *) after_atou;
1f9f7d4c
KW
3573 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
3574 elements--;
3575
3576 /* Then just populate the rest of the input */
3577 while (elements-- > 0) {
3578 if (l > lend) {
3579 Perl_croak(aTHX_ "panic: Expecting %"UVuf" more elements than available", elements);
3580 }
99944878 3581 while (isSPACE(*l)) l++;
22ff3130
HS
3582 if (!grok_atoUV((const char *)l, other_elements_ptr++, &after_atou)) {
3583 Perl_croak(aTHX_ "panic: Expecting a valid element in inversion list");
3584 }
99944878 3585 l = (U8 *) after_atou;
1f9f7d4c 3586 }
eb092534 3587 }
31aa6e0b
KW
3588 }
3589 else {
3590
1784d2f9
KW
3591 /* Scan the input to count the number of lines to preallocate array
3592 * size based on worst possible case, which is each line in the input
3593 * creates 2 elements in the inversion list: 1) the beginning of a
3594 * range in the list; 2) the beginning of a range not in the list. */
3595 while ((loc = (strchr(loc, '\n'))) != NULL) {
3596 elements += 2;
3597 loc++;
3598 }
d764b54e 3599
1784d2f9
KW
3600 /* If the ending is somehow corrupt and isn't a new line, add another
3601 * element for the final range that isn't in the inversion list */
3602 if (! (*lend == '\n'
3603 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
3604 {
3605 elements++;
3606 }
d764b54e 3607
1784d2f9 3608 invlist = _new_invlist(elements);
d764b54e 3609
1784d2f9
KW
3610 /* Now go through the input again, adding each range to the list */
3611 while (l < lend) {
3612 UV start, end;
3613 UV val; /* Not used by this function */
d764b54e 3614
95543e92
KW
3615 l = swash_scan_list_line(l, lend, &start, &end, &val,
3616 cBOOL(octets), typestr);
d764b54e 3617
1784d2f9
KW
3618 if (l > lend) {
3619 break;
3620 }
3621
3622 invlist = _add_range_to_invlist(invlist, start, end);
3623 }
31aa6e0b 3624 }
d764b54e 3625
77f9f126
KW
3626 /* Invert if the data says it should be */
3627 if (invert_it_svp && SvUV(*invert_it_svp)) {
25151030 3628 _invlist_invert(invlist);
77f9f126
KW
3629 }
3630
b0e3252e 3631 /* This code is copied from swatch_get()
d73c39c5
KW
3632 * read $swash->{EXTRAS} */
3633 x = (U8*)SvPV(*extssvp, xcur);
3634 xend = x + xcur;
3635 while (x < xend) {
3636 STRLEN namelen;
3637 U8 *namestr;
3638 SV** othersvp;
3639 HV* otherhv;
3640 STRLEN otherbits;
3641 SV **otherbitssvp, *other;
3642 U8 *nl;
3643
3644 const U8 opc = *x++;
3645 if (opc == '\n')
3646 continue;
3647
3648 nl = (U8*)memchr(x, '\n', xend - x);
3649
3650 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3651 if (nl) {
3652 x = nl + 1; /* 1 is length of "\n" */
3653 continue;
3654 }
3655 else {
3656 x = xend; /* to EXTRAS' end at which \n is not found */
3657 break;
3658 }
3659 }
3660
3661 namestr = x;
3662 if (nl) {
3663 namelen = nl - namestr;
3664 x = nl + 1;
3665 }
3666 else {
3667 namelen = xend - namestr;
3668 x = xend;
3669 }
3670
3671 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3672 otherhv = MUTABLE_HV(SvRV(*othersvp));
3673 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3674 otherbits = (ST