This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Move a VMS-specific include file in sv.c.
[perl5.git] / utf8.c
CommitLineData
a0ed51b3
LW
1/* utf8.c
2 *
1129b882 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
b94e2f88 4 * by Larry Wall and others
a0ed51b3
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550
TC
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
15 *
cdad3b53 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
a0ed51b3
LW
17 *
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
4ac71550 20 * as is the custom in the West, if you wish to be answered?'
cdad3b53 21 * --Gandalf, addressing Théoden's door wardens
4ac71550
TC
22 *
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
24 *
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
4ac71550
TC
27 *
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
29 */
30
31#include "EXTERN.h"
864dbfa3 32#define PERL_IN_UTF8_C
a0ed51b3 33#include "perl.h"
81e983c1 34#include "inline_invlist.c"
b24b43f7 35#include "charclass_invlists.h"
a0ed51b3 36
27da23d5
JH
37static const char unees[] =
38 "Malformed UTF-8 character (unexpected end of string)";
901b21bf 39
48ef279e 40/*
ccfc67b7 41=head1 Unicode Support
a0ed51b3 42
166f8a29 43This file contains various utility functions for manipulating UTF8-encoded
72d33970 44strings. For the uninitiated, this is a method of representing arbitrary
61296642 45Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
46characters in the ASCII range are unmodified, and a zero byte never appears
47within non-zero characters.
166f8a29 48
eaf7a4d2
CS
49=cut
50*/
51
52/*
53=for apidoc is_ascii_string
54
a1433954 55Returns true if the first C<len> bytes of the string C<s> are the same whether
970ea3cb
KW
56or not the string is encoded in UTF-8 (or UTF-EBCDIC on EBCDIC machines). That
57is, if they are invariant. On ASCII-ish machines, only ASCII characters
58fit this definition, hence the function's name.
eaf7a4d2 59
75200dff
KW
60If C<len> is 0, it will be calculated using C<strlen(s)>, (which means if you
61use this option, that C<s> can't have embedded C<NUL> characters and has to
62have a terminating C<NUL> byte).
9f7e3d64 63
a1433954 64See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
eaf7a4d2
CS
65
66=cut
67*/
68
69bool
668b6d8d 70Perl_is_ascii_string(const U8 *s, STRLEN len)
eaf7a4d2
CS
71{
72 const U8* const send = s + (len ? len : strlen((const char *)s));
73 const U8* x = s;
74
75 PERL_ARGS_ASSERT_IS_ASCII_STRING;
eaf7a4d2
CS
76
77 for (; x < send; ++x) {
78 if (!UTF8_IS_INVARIANT(*x))
79 break;
80 }
81
82 return x == send;
83}
84
85/*
378516de 86=for apidoc uvoffuni_to_utf8_flags
eebe1485 87
a27992cc 88THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
89Instead, B<Almost all code should use L</uvchr_to_utf8> or
90L</uvchr_to_utf8_flags>>.
a27992cc 91
de69f3af
KW
92This function is like them, but the input is a strict Unicode
93(as opposed to native) code point. Only in very rare circumstances should code
94not be using the native code point.
949cf498 95
de69f3af 96For details, see the description for L</uvchr_to_utf8_flags>>.
949cf498 97
eebe1485
SC
98=cut
99*/
100
dfe13c55 101U8 *
378516de 102Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 103{
378516de 104 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 105
d9432125
KW
106 if (UNI_IS_INVARIANT(uv)) {
107 *d++ = (U8) LATIN1_TO_NATIVE(uv);
108 return d;
109 }
110
979f77b6
KW
111 /* The first problematic code point is the first surrogate */
112 if (uv >= UNICODE_SURROGATE_FIRST
0cfa64bf 113 && ckWARN3_d(WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
979f77b6 114 {
949cf498
KW
115 if (UNICODE_IS_SURROGATE(uv)) {
116 if (flags & UNICODE_WARN_SURROGATE) {
8457b38f 117 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
949cf498
KW
118 "UTF-16 surrogate U+%04"UVXf, uv);
119 }
120 if (flags & UNICODE_DISALLOW_SURROGATE) {
121 return NULL;
122 }
123 }
124 else if (UNICODE_IS_SUPER(uv)) {
125 if (flags & UNICODE_WARN_SUPER
126 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
127 {
8457b38f 128 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
949cf498
KW
129 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
130 }
131 if (flags & UNICODE_DISALLOW_SUPER
132 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
133 {
134 return NULL;
135 }
136 }
137 else if (UNICODE_IS_NONCHAR(uv)) {
138 if (flags & UNICODE_WARN_NONCHAR) {
8457b38f 139 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
949cf498
KW
140 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
141 uv);
142 }
143 if (flags & UNICODE_DISALLOW_NONCHAR) {
144 return NULL;
145 }
146 }
507b9800 147 }
d9432125 148
2d331972 149#if defined(EBCDIC)
d9432125 150 {
5aaebcb3 151 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
152 U8 *p = d+len-1;
153 while (p > d) {
bc3632a8 154 *p-- = (U8) I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
155 uv >>= UTF_ACCUMULATION_SHIFT;
156 }
bc3632a8 157 *p = (U8) I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
158 return d+len;
159 }
160#else /* Non loop style */
a0ed51b3 161 if (uv < 0x800) {
eb160463
GS
162 *d++ = (U8)(( uv >> 6) | 0xc0);
163 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
164 return d;
165 }
166 if (uv < 0x10000) {
eb160463
GS
167 *d++ = (U8)(( uv >> 12) | 0xe0);
168 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
169 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
170 return d;
171 }
172 if (uv < 0x200000) {
eb160463
GS
173 *d++ = (U8)(( uv >> 18) | 0xf0);
174 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
175 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
176 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
177 return d;
178 }
179 if (uv < 0x4000000) {
eb160463
GS
180 *d++ = (U8)(( uv >> 24) | 0xf8);
181 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
182 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
183 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
184 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
185 return d;
186 }
187 if (uv < 0x80000000) {
eb160463
GS
188 *d++ = (U8)(( uv >> 30) | 0xfc);
189 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
190 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
191 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
192 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
193 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
194 return d;
195 }
6588300d 196#ifdef UTF8_QUAD_MAX
d7578b48 197 if (uv < UTF8_QUAD_MAX)
a0ed51b3
LW
198#endif
199 {
eb160463
GS
200 *d++ = 0xfe; /* Can't match U+FEFF! */
201 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
202 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
203 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
204 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
205 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
206 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
207 return d;
208 }
6588300d 209#ifdef UTF8_QUAD_MAX
a0ed51b3 210 {
eb160463
GS
211 *d++ = 0xff; /* Can't match U+FFFE! */
212 *d++ = 0x80; /* 6 Reserved bits */
213 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
214 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
215 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
217 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
218 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
219 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
220 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
221 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
222 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
223 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
224 return d;
225 }
226#endif
537124e4 227#endif /* Non loop style */
a0ed51b3 228}
646ca15d 229/*
07693fe6
KW
230=for apidoc uvchr_to_utf8
231
bcb1a2d4 232Adds the UTF-8 representation of the native code point C<uv> to the end
c749c9fd
KW
233of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
234C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
235the byte after the end of the new character. In other words,
07693fe6
KW
236
237 d = uvchr_to_utf8(d, uv);
238
239is the recommended wide native character-aware way of saying
240
241 *(d++) = uv;
242
de69f3af
KW
243This function accepts any UV as input. To forbid or warn on non-Unicode code
244points, or those that may be problematic, see L</uvchr_to_utf8_flags>.
245
07693fe6
KW
246=cut
247*/
248
de69f3af
KW
249/* This is also a macro */
250PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
251
07693fe6
KW
252U8 *
253Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
254{
de69f3af 255 return uvchr_to_utf8(d, uv);
07693fe6
KW
256}
257
de69f3af
KW
258/*
259=for apidoc uvchr_to_utf8_flags
260
261Adds the UTF-8 representation of the native code point C<uv> to the end
c749c9fd
KW
262of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
263C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
264the byte after the end of the new character. In other words,
de69f3af
KW
265
266 d = uvchr_to_utf8_flags(d, uv, flags);
267
268or, in most cases,
269
270 d = uvchr_to_utf8_flags(d, uv, 0);
271
272This is the Unicode-aware way of saying
273
274 *(d++) = uv;
275
276This function will convert to UTF-8 (and not warn) even code points that aren't
277legal Unicode or are problematic, unless C<flags> contains one or more of the
278following flags:
279
280If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
281the function will raise a warning, provided UTF8 warnings are enabled. If instead
282UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
283If both flags are set, the function will both warn and return NULL.
284
4c3cfd5d 285The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags
de69f3af 286affect how the function handles a Unicode non-character. And likewise, the
4c3cfd5d 287UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags affect the handling of
de69f3af
KW
288code points that are
289above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
290even less portable) can be warned and/or disallowed even if other above-Unicode
291code points are accepted, by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
292flags.
293
294And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
295above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
296DISALLOW flags.
297
298=cut
299*/
300
301/* This is also a macro */
302PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
303
07693fe6
KW
304U8 *
305Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
306{
de69f3af 307 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
308}
309
310/*
492a624f
KW
311=for apidoc is_utf8_char_buf
312
3cedd9d9 313This is identical to the macro L</isUTF8_CHAR>.
492a624f
KW
314
315=cut */
316
317STRLEN
318Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
319{
320
492a624f
KW
321 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
322
6302f837 323 return isUTF8_CHAR(buf, buf_end);
492a624f
KW
324}
325
326/*
87cea99e 327=for apidoc is_utf8_string
6662521e 328
a1433954 329Returns true if the first C<len> bytes of string C<s> form a valid
9f7e3d64 330UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
75200dff
KW
331using C<strlen(s)> (which means if you use this option, that C<s> can't have
332embedded C<NUL> characters and has to have a terminating C<NUL> byte). Note
333that all characters being ASCII constitute 'a valid UTF-8 string'.
6662521e 334
a1433954 335See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
768c67ee 336
6662521e
GS
337=cut
338*/
339
8e84507e 340bool
668b6d8d 341Perl_is_utf8_string(const U8 *s, STRLEN len)
6662521e 342{
35da51f7 343 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 344 const U8* x = s;
067a85ef 345
7918f24d 346 PERL_ARGS_ASSERT_IS_UTF8_STRING;
1aa99e6b 347
6662521e 348 while (x < send) {
6302f837
KW
349 STRLEN len = isUTF8_CHAR(x, send);
350 if (UNLIKELY(! len)) {
351 return FALSE;
352 }
353 x += len;
6662521e 354 }
768c67ee 355
067a85ef 356 return TRUE;
6662521e
GS
357}
358
67e989fb 359/*
814fafa7
NC
360Implemented as a macro in utf8.h
361
87cea99e 362=for apidoc is_utf8_string_loc
814fafa7 363
a1433954
KW
364Like L</is_utf8_string> but stores the location of the failure (in the
365case of "utf8ness failure") or the location C<s>+C<len> (in the case of
814fafa7
NC
366"utf8ness success") in the C<ep>.
367
a1433954 368See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
814fafa7 369
87cea99e 370=for apidoc is_utf8_string_loclen
81cd54e3 371
a1433954
KW
372Like L</is_utf8_string>() but stores the location of the failure (in the
373case of "utf8ness failure") or the location C<s>+C<len> (in the case of
768c67ee
JH
374"utf8ness success") in the C<ep>, and the number of UTF-8
375encoded characters in the C<el>.
376
a1433954 377See also L</is_utf8_string_loc>() and L</is_utf8_string>().
81cd54e3
JH
378
379=cut
380*/
381
382bool
668b6d8d 383Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
81cd54e3 384{
35da51f7 385 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 386 const U8* x = s;
3ebfea28 387 STRLEN outlen = 0;
7918f24d
NC
388
389 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
81cd54e3 390
81cd54e3 391 while (x < send) {
6302f837
KW
392 STRLEN len = isUTF8_CHAR(x, send);
393 if (UNLIKELY(! len)) {
394 goto out;
395 }
396 x += len;
397 outlen++;
81cd54e3 398 }
768c67ee
JH
399
400 out:
3ebfea28
AL
401 if (el)
402 *el = outlen;
403
768c67ee
JH
404 if (ep)
405 *ep = x;
3ebfea28 406 return (x == send);
81cd54e3
JH
407}
408
409/*
768c67ee 410
de69f3af 411=for apidoc utf8n_to_uvchr
378516de
KW
412
413THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 414Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 415
9041c2e3 416Bottom level UTF-8 decode routine.
de69f3af 417Returns the native code point value of the first character in the string C<s>,
746afd53
KW
418which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
419C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
420the length, in bytes, of that character.
949cf498
KW
421
422The value of C<flags> determines the behavior when C<s> does not point to a
423well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
424zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
425next possible position in C<s> that could begin a non-malformed character.
426Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
427
428Various ALLOW flags can be set in C<flags> to allow (and not warn on)
429individual types of malformations, such as the sequence being overlong (that
430is, when there is a shorter sequence that can express the same code point;
431overlong sequences are expressly forbidden in the UTF-8 standard due to
432potential security issues). Another malformation example is the first byte of
433a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
434flags. For allowed 0 length strings, this function returns 0; for allowed
435overlong sequences, the computed code point is returned; for all other allowed
436malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
437determinable reasonable value.
949cf498
KW
438
439The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
440flags) malformation is found. If this flag is set, the routine assumes that
441the caller will raise a warning, and this function will silently just set
d088425d
KW
442C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
443
75200dff 444Note that this API requires disambiguation between successful decoding a C<NUL>
d088425d
KW
445character, and an error return (unless the UTF8_CHECK_ONLY flag is set), as
446in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
75200dff
KW
447first byte of C<s> is 0 as well. If so, the input was a C<NUL>; if not, the
448input had an error.
949cf498
KW
449
450Certain code points are considered problematic. These are Unicode surrogates,
746afd53 451Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 452By default these are considered regular code points, but certain situations
5eafe189 453warrant special handling for them. If C<flags> contains
949cf498
KW
454UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
455malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
456UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
457maximum) can be set to disallow these categories individually.
458
459The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
460UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
461for their respective categories, but otherwise the code points are considered
462valid (not malformations). To get a category to both be treated as a
463malformation and raise a warning, specify both the WARN and DISALLOW flags.
464(But note that warnings are not raised if lexically disabled nor if
465UTF8_CHECK_ONLY is also specified.)
466
467Very large code points (above 0x7FFF_FFFF) are considered more problematic than
468the others that are above the Unicode legal maximum. There are several
eb83ed87
KW
469reasons: they requre at least 32 bits to represent them on ASCII platforms, are
470not representable at all on EBCDIC platforms, and the original UTF-8
471specification never went above this number (the current 0x10FFFF limit was
472imposed later). (The smaller ones, those that fit into 32 bits, are
473representable by a UV on ASCII platforms, but not by an IV, which means that
474the number of operations that can be performed on them is quite restricted.)
475The UTF-8 encoding on ASCII platforms for these large code points begins with a
476byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
477be treated as malformations, while allowing smaller above-Unicode code points.
478(Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
72d33970
FC
479including these, as malformations.)
480Similarly, UTF8_WARN_FE_FF acts just like
eb83ed87 481the other WARN flags, but applies just to these code points.
949cf498
KW
482
483All other code points corresponding to Unicode characters, including private
484use and those yet to be assigned, are never considered malformed and never
485warn.
67e989fb 486
37607a96
PK
487=cut
488*/
67e989fb 489
a0ed51b3 490UV
de69f3af 491Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 492{
97aff369 493 dVAR;
d4c19fe8 494 const U8 * const s0 = s;
eb83ed87 495 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 496 U8 * send;
eb83ed87
KW
497 UV uv = *s;
498 STRLEN expectlen;
949cf498 499 SV* sv = NULL;
eb83ed87
KW
500 UV outlier_ret = 0; /* return value when input is in error or problematic
501 */
502 UV pack_warn = 0; /* Save result of packWARN() for later */
503 bool unexpected_non_continuation = FALSE;
504 bool overflowed = FALSE;
2f8f112e 505 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 506
eb83ed87 507 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 508
de69f3af 509 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
a0dbb045 510
eb83ed87
KW
511 /* The order of malformation tests here is important. We should consume as
512 * few bytes as possible in order to not skip any valid character. This is
513 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
514 * http://unicode.org/reports/tr36 for more discussion as to why. For
515 * example, once we've done a UTF8SKIP, we can tell the expected number of
516 * bytes, and could fail right off the bat if the input parameters indicate
517 * that there are too few available. But it could be that just that first
518 * byte is garbled, and the intended character occupies fewer bytes. If we
519 * blindly assumed that the first byte is correct, and skipped based on
520 * that number, we could skip over a valid input character. So instead, we
521 * always examine the sequence byte-by-byte.
522 *
523 * We also should not consume too few bytes, otherwise someone could inject
524 * things. For example, an input could be deliberately designed to
525 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 526 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
527 * which is actually part of of the overflowing sequence), that could look
528 * legitimate to the caller, which could discard the initial partial
529 * sequence and process the rest, inappropriately */
530
531 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 532 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
533 if (retlen) {
534 *retlen = 0;
535 }
a0dbb045 536
eb83ed87
KW
537 if (flags & UTF8_ALLOW_EMPTY) {
538 return 0;
539 }
540 if (! (flags & UTF8_CHECK_ONLY)) {
541 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
542 }
0c443dc2
JH
543 goto malformed;
544 }
545
eb83ed87
KW
546 expectlen = UTF8SKIP(s);
547
548 /* A well-formed UTF-8 character, as the vast majority of calls to this
549 * function will be for, has this expected length. For efficiency, set
550 * things up here to return it. It will be overriden only in those rare
551 * cases where a malformation is found */
552 if (retlen) {
553 *retlen = expectlen;
554 }
555
556 /* An invariant is trivially well-formed */
1d72bdf6 557 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 558 return uv;
a0ed51b3 559 }
67e989fb 560
eb83ed87 561 /* A continuation character can't start a valid sequence */
b5b9af04 562 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
563 if (flags & UTF8_ALLOW_CONTINUATION) {
564 if (retlen) {
565 *retlen = 1;
566 }
567 return UNICODE_REPLACEMENT;
568 }
ba210ebe 569
eb83ed87
KW
570 if (! (flags & UTF8_CHECK_ONLY)) {
571 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
572 }
573 curlen = 1;
ba210ebe
JH
574 goto malformed;
575 }
9041c2e3 576
dcd27b3c
KW
577 /* Here is not a continuation byte, nor an invariant. The only thing left
578 * is a start byte (possibly for an overlong) */
579
1d72bdf6 580#ifdef EBCDIC
bc3632a8 581 uv = NATIVE_UTF8_TO_I8(uv);
1d72bdf6
NIS
582#endif
583
eb83ed87
KW
584 /* Remove the leading bits that indicate the number of bytes in the
585 * character's whole UTF-8 sequence, leaving just the bits that are part of
586 * the value */
587 uv &= UTF_START_MASK(expectlen);
ba210ebe 588
eb83ed87
KW
589 /* Now, loop through the remaining bytes in the character's sequence,
590 * accumulating each into the working value as we go. Be sure to not look
591 * past the end of the input string */
0b8d30e8
KW
592 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
593
eb83ed87 594 for (s = s0 + 1; s < send; s++) {
b5b9af04 595 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
596#ifndef EBCDIC /* Can't overflow in EBCDIC */
597 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
598
599 /* The original implementors viewed this malformation as more
600 * serious than the others (though I, khw, don't understand
601 * why, since other malformations also give very very wrong
602 * results), so there is no way to turn off checking for it.
603 * Set a flag, but keep going in the loop, so that we absorb
604 * the rest of the bytes that comprise the character. */
605 overflowed = TRUE;
606 overflow_byte = *s; /* Save for warning message's use */
607 }
608#endif
8850bf83 609 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
610 }
611 else {
612 /* Here, found a non-continuation before processing all expected
613 * bytes. This byte begins a new character, so quit, even if
614 * allowing this malformation. */
615 unexpected_non_continuation = TRUE;
616 break;
617 }
618 } /* End of loop through the character's bytes */
619
620 /* Save how many bytes were actually in the character */
621 curlen = s - s0;
622
623 /* The loop above finds two types of malformations: non-continuation and/or
624 * overflow. The non-continuation malformation is really a too-short
625 * malformation, as it means that the current character ended before it was
626 * expected to (being terminated prematurely by the beginning of the next
627 * character, whereas in the too-short malformation there just are too few
628 * bytes available to hold the character. In both cases, the check below
629 * that we have found the expected number of bytes would fail if executed.)
630 * Thus the non-continuation malformation is really unnecessary, being a
631 * subset of the too-short malformation. But there may be existing
632 * applications that are expecting the non-continuation type, so we retain
633 * it, and return it in preference to the too-short malformation. (If this
634 * code were being written from scratch, the two types might be collapsed
635 * into one.) I, khw, am also giving priority to returning the
636 * non-continuation and too-short malformations over overflow when multiple
637 * ones are present. I don't know of any real reason to prefer one over
638 * the other, except that it seems to me that multiple-byte errors trumps
639 * errors from a single byte */
b5b9af04 640 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
641 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
642 if (! (flags & UTF8_CHECK_ONLY)) {
643 if (curlen == 1) {
644 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
645 }
646 else {
647 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
648 }
649 }
eb83ed87
KW
650 goto malformed;
651 }
652 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
653
654 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
655 * as what the original expectations were. */
656 do_overlong_test = FALSE;
eb83ed87
KW
657 if (retlen) {
658 *retlen = curlen;
659 }
660 }
b5b9af04 661 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
662 if (! (flags & UTF8_ALLOW_SHORT)) {
663 if (! (flags & UTF8_CHECK_ONLY)) {
664 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 665 }
eb83ed87
KW
666 goto malformed;
667 }
668 uv = UNICODE_REPLACEMENT;
2f8f112e 669 do_overlong_test = FALSE;
eb83ed87
KW
670 if (retlen) {
671 *retlen = curlen;
672 }
673 }
674
ea5ced44 675#ifndef EBCDIC /* EBCDIC can't overflow */
b5b9af04 676 if (UNLIKELY(overflowed)) {
eb83ed87 677 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 678 goto malformed;
eb83ed87
KW
679 }
680#endif
681
2f8f112e 682 if (do_overlong_test
5aaebcb3 683 && expectlen > (STRLEN) OFFUNISKIP(uv)
2f8f112e
KW
684 && ! (flags & UTF8_ALLOW_LONG))
685 {
eb83ed87
KW
686 /* The overlong malformation has lower precedence than the others.
687 * Note that if this malformation is allowed, we return the actual
688 * value, instead of the replacement character. This is because this
689 * value is actually well-defined. */
690 if (! (flags & UTF8_CHECK_ONLY)) {
5aaebcb3 691 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
692 }
693 goto malformed;
694 }
695
1a89bb6c 696 /* Here, the input is considered to be well-formed, but it still could be a
eb83ed87
KW
697 * problematic code point that is not allowed by the input parameters. */
698 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
699 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
700 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
701 {
949cf498 702 if (UNICODE_IS_SURROGATE(uv)) {
ea5ced44
KW
703
704 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
705 * generation of the sv, since no warnings are raised under CHECK */
eb83ed87 706 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
54f4afef 707 && ckWARN_d(WARN_SURROGATE))
eb83ed87 708 {
111d382d 709 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
54f4afef 710 pack_warn = packWARN(WARN_SURROGATE);
949cf498
KW
711 }
712 if (flags & UTF8_DISALLOW_SURROGATE) {
713 goto disallowed;
714 }
715 }
949cf498 716 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87 717 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
ea5ced44 718 && ckWARN_d(WARN_NON_UNICODE))
eb83ed87 719 {
111d382d 720 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
54f4afef 721 pack_warn = packWARN(WARN_NON_UNICODE);
949cf498 722 }
ea5ced44
KW
723#ifndef EBCDIC /* EBCDIC always allows FE, FF */
724
725 /* The first byte being 0xFE or 0xFF is a subset of the SUPER code
726 * points. We test for these after the regular SUPER ones, and
727 * before possibly bailing out, so that the more dire warning
728 * overrides the regular one, if applicable */
729 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
730 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
731 {
732 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY))
733 == UTF8_WARN_FE_FF
734 && ckWARN_d(WARN_UTF8))
735 {
736 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%"UVXf" is not Unicode, and not portable", uv));
737 pack_warn = packWARN(WARN_UTF8);
738 }
739 if (flags & UTF8_DISALLOW_FE_FF) {
740 goto disallowed;
741 }
742 }
743#endif
949cf498
KW
744 if (flags & UTF8_DISALLOW_SUPER) {
745 goto disallowed;
746 }
747 }
4190d317
KW
748 else if (UNICODE_IS_NONCHAR(uv)) {
749 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
54f4afef 750 && ckWARN_d(WARN_NONCHAR))
4190d317
KW
751 {
752 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
54f4afef 753 pack_warn = packWARN(WARN_NONCHAR);
4190d317
KW
754 }
755 if (flags & UTF8_DISALLOW_NONCHAR) {
756 goto disallowed;
757 }
758 }
949cf498 759
eb83ed87 760 if (sv) {
de69f3af
KW
761 outlier_ret = uv; /* Note we don't bother to convert to native,
762 as all the outlier code points are the same
763 in both ASCII and EBCDIC */
eb83ed87
KW
764 goto do_warn;
765 }
766
949cf498
KW
767 /* Here, this is not considered a malformed character, so drop through
768 * to return it */
a0ed51b3 769 }
ba210ebe 770
de69f3af 771 return UNI_TO_NATIVE(uv);
ba210ebe 772
eb83ed87
KW
773 /* There are three cases which get to beyond this point. In all 3 cases:
774 * <sv> if not null points to a string to print as a warning.
775 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
776 * set.
777 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
778 * This is done by initializing it to 0, and changing it only
779 * for case 1).
780 * The 3 cases are:
781 * 1) The input is valid but problematic, and to be warned about. The
782 * return value is the resultant code point; <*retlen> is set to
783 * <curlen>, the number of bytes that comprise the code point.
784 * <pack_warn> contains the result of packWARN() for the warning
785 * types. The entry point for this case is the label <do_warn>;
786 * 2) The input is a valid code point but disallowed by the parameters to
787 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
788 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
789 * comprise the code point. <pack_warn> contains the result of
790 * packWARN() for the warning types. The entry point for this case is
791 * the label <disallowed>.
792 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
793 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
794 * bytes that comprise the malformation. All such malformations are
795 * assumed to be warning type <utf8>. The entry point for this case
796 * is the label <malformed>.
797 */
949cf498 798
ba210ebe
JH
799malformed:
800
eb83ed87
KW
801 if (sv && ckWARN_d(WARN_UTF8)) {
802 pack_warn = packWARN(WARN_UTF8);
803 }
804
805disallowed:
806
fcc8fcf6 807 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 808 if (retlen)
10edeb5d 809 *retlen = ((STRLEN) -1);
ba210ebe
JH
810 return 0;
811 }
812
eb83ed87 813do_warn:
5b311467 814
eb83ed87
KW
815 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
816 if warnings are to be raised. */
f555bc63 817 const char * const string = SvPVX_const(sv);
a0dbb045 818
f555bc63
KW
819 if (PL_op)
820 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
821 else
822 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
823 }
824
eb83ed87
KW
825 if (retlen) {
826 *retlen = curlen;
827 }
ba210ebe 828
eb83ed87 829 return outlier_ret;
a0ed51b3
LW
830}
831
8e84507e 832/*
ec5f19d0
KW
833=for apidoc utf8_to_uvchr_buf
834
835Returns the native code point of the first character in the string C<s> which
836is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 837C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 838
524080c4
KW
839If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
840enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
173db420
KW
841NULL) to -1. If those warnings are off, the computed value, if well-defined
842(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
843C<*retlen> is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is
844the next possible position in C<s> that could begin a non-malformed character.
de69f3af 845See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 846returned.
ec5f19d0
KW
847
848=cut
849*/
850
851
852UV
853Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
854{
ec5f19d0
KW
855 assert(s < send);
856
857 return utf8n_to_uvchr(s, send - s, retlen,
858 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
859}
860
27d6c58a 861/* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
3986bb7c 862 * there are no malformations in the input UTF-8 string C<s>. surrogates,
57b0056d 863 * non-character code points, and non-Unicode code points are allowed. */
27d6c58a
KW
864
865UV
866Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
867{
010ab96b
KW
868 UV expectlen = UTF8SKIP(s);
869 const U8* send = s + expectlen;
9ff2f0f7 870 UV uv = *s;
3986bb7c 871
27d6c58a
KW
872 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
873
010ab96b
KW
874 if (retlen) {
875 *retlen = expectlen;
876 }
877
878 /* An invariant is trivially returned */
879 if (expectlen == 1) {
9ff2f0f7 880 return uv;
010ab96b
KW
881 }
882
9ff2f0f7
KW
883#ifdef EBCDIC
884 uv = NATIVE_UTF8_TO_I8(uv);
885#endif
886
010ab96b
KW
887 /* Remove the leading bits that indicate the number of bytes, leaving just
888 * the bits that are part of the value */
889 uv &= UTF_START_MASK(expectlen);
890
891 /* Now, loop through the remaining bytes, accumulating each into the
892 * working total as we go. (I khw tried unrolling the loop for up to 4
893 * bytes, but there was no performance improvement) */
894 for (++s; s < send; s++) {
895 uv = UTF8_ACCUMULATE(uv, *s);
896 }
897
3986bb7c 898 return UNI_TO_NATIVE(uv);
010ab96b 899
27d6c58a
KW
900}
901
ec5f19d0 902/*
ec5f19d0
KW
903=for apidoc utf8_to_uvuni_buf
904
de69f3af
KW
905Only in very rare circumstances should code need to be dealing in Unicode
906(as opposed to native) code points. In those few cases, use
907C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
908
909Returns the Unicode (not-native) code point of the first character in the
910string C<s> which
ec5f19d0
KW
911is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
912C<retlen> will be set to the length, in bytes, of that character.
913
524080c4
KW
914If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
915enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
916NULL) to -1. If those warnings are off, the computed value if well-defined (or
917the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
918is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
919next possible position in C<s> that could begin a non-malformed character.
de69f3af 920See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
921
922=cut
923*/
924
925UV
926Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
927{
928 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
929
930 assert(send > s);
931
932 /* Call the low level routine asking for checks */
de69f3af
KW
933 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
934 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
ec5f19d0
KW
935}
936
b76347f2 937/*
87cea99e 938=for apidoc utf8_length
b76347f2
JH
939
940Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
941Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
942up past C<e>, croaks.
b76347f2
JH
943
944=cut
945*/
946
947STRLEN
35a4481c 948Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2 949{
97aff369 950 dVAR;
b76347f2
JH
951 STRLEN len = 0;
952
7918f24d
NC
953 PERL_ARGS_ASSERT_UTF8_LENGTH;
954
8850bf83
JH
955 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
956 * the bitops (especially ~) can create illegal UTF-8.
957 * In other words: in Perl UTF-8 is not just for Unicode. */
958
a3b680e6
AL
959 if (e < s)
960 goto warn_and_return;
b76347f2 961 while (s < e) {
4cbf4130 962 s += UTF8SKIP(s);
8e91ec7f
AV
963 len++;
964 }
965
966 if (e != s) {
967 len--;
968 warn_and_return:
9b387841
NC
969 if (PL_op)
970 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
971 "%s in %s", unees, OP_DESC(PL_op));
972 else
61a12c31 973 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
974 }
975
976 return len;
977}
978
b06226ff 979/*
87cea99e 980=for apidoc utf8_distance
b06226ff 981
1e54db1a 982Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
b06226ff
JH
983and C<b>.
984
985WARNING: use only if you *know* that the pointers point inside the
986same UTF-8 buffer.
987
37607a96
PK
988=cut
989*/
a0ed51b3 990
02eb7b47 991IV
35a4481c 992Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
a0ed51b3 993{
7918f24d
NC
994 PERL_ARGS_ASSERT_UTF8_DISTANCE;
995
bf1665bc 996 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
a0ed51b3
LW
997}
998
b06226ff 999/*
87cea99e 1000=for apidoc utf8_hop
b06226ff 1001
8850bf83
JH
1002Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1003forward or backward.
b06226ff
JH
1004
1005WARNING: do not use the following unless you *know* C<off> is within
8850bf83
JH
1006the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1007on the first byte of character or just after the last byte of a character.
b06226ff 1008
37607a96
PK
1009=cut
1010*/
a0ed51b3
LW
1011
1012U8 *
4373e329 1013Perl_utf8_hop(pTHX_ const U8 *s, I32 off)
a0ed51b3 1014{
7918f24d
NC
1015 PERL_ARGS_ASSERT_UTF8_HOP;
1016
96a5add6 1017 PERL_UNUSED_CONTEXT;
8850bf83
JH
1018 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1019 * the bitops (especially ~) can create illegal UTF-8.
1020 * In other words: in Perl UTF-8 is not just for Unicode. */
1021
a0ed51b3
LW
1022 if (off >= 0) {
1023 while (off--)
1024 s += UTF8SKIP(s);
1025 }
1026 else {
1027 while (off++) {
1028 s--;
8850bf83
JH
1029 while (UTF8_IS_CONTINUATION(*s))
1030 s--;
a0ed51b3
LW
1031 }
1032 }
4373e329 1033 return (U8 *)s;
a0ed51b3
LW
1034}
1035
6940069f 1036/*
fed3ba5d
NC
1037=for apidoc bytes_cmp_utf8
1038
a1433954 1039Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1040sequence of characters (stored as UTF-8)
1041in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1042equal, -1 or -2 if the first string is less than the second string, +1 or +2
1043if the first string is greater than the second string.
1044
1045-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1046longer string. -2 or +2 is returned if
1047there was a difference between characters
fed3ba5d
NC
1048within the strings.
1049
1050=cut
1051*/
1052
1053int
1054Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1055{
1056 const U8 *const bend = b + blen;
1057 const U8 *const uend = u + ulen;
1058
1059 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1060
1061 PERL_UNUSED_CONTEXT;
1062
1063 while (b < bend && u < uend) {
1064 U8 c = *u++;
1065 if (!UTF8_IS_INVARIANT(c)) {
1066 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1067 if (u < uend) {
1068 U8 c1 = *u++;
1069 if (UTF8_IS_CONTINUATION(c1)) {
94bb8c36 1070 c = TWO_BYTE_UTF8_TO_NATIVE(c, c1);
fed3ba5d
NC
1071 } else {
1072 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1073 "Malformed UTF-8 character "
1074 "(unexpected non-continuation byte 0x%02x"
1075 ", immediately after start byte 0x%02x)"
1076 /* Dear diag.t, it's in the pod. */
1077 "%s%s", c1, c,
1078 PL_op ? " in " : "",
1079 PL_op ? OP_DESC(PL_op) : "");
1080 return -2;
1081 }
1082 } else {
1083 if (PL_op)
1084 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1085 "%s in %s", unees, OP_DESC(PL_op));
1086 else
61a12c31 1087 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1088 return -2; /* Really want to return undef :-) */
1089 }
1090 } else {
1091 return -2;
1092 }
1093 }
1094 if (*b != c) {
1095 return *b < c ? -2 : +2;
1096 }
1097 ++b;
1098 }
1099
1100 if (b == bend && u == uend)
1101 return 0;
1102
1103 return b < bend ? +1 : -1;
1104}
1105
1106/*
87cea99e 1107=for apidoc utf8_to_bytes
6940069f 1108
2bbc8d55 1109Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1110Unlike L</bytes_to_utf8>, this over-writes the original string, and
1111updates C<len> to contain the new length.
67e989fb 1112Returns zero on failure, setting C<len> to -1.
6940069f 1113
a1433954 1114If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1115
6940069f
GS
1116=cut
1117*/
1118
1119U8 *
37607a96 1120Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1121{
d4c19fe8
AL
1122 U8 * const save = s;
1123 U8 * const send = s + *len;
6940069f 1124 U8 *d;
246fae53 1125
7918f24d
NC
1126 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1127
1e54db1a 1128 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1129 while (s < send) {
d59937ca
KW
1130 if (! UTF8_IS_INVARIANT(*s)) {
1131 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1132 *len = ((STRLEN) -1);
1133 return 0;
1134 }
1135 s++;
dcad2880 1136 }
d59937ca 1137 s++;
246fae53 1138 }
dcad2880
JH
1139
1140 d = s = save;
6940069f 1141 while (s < send) {
80e0b38f
KW
1142 U8 c = *s++;
1143 if (! UTF8_IS_INVARIANT(c)) {
1144 /* Then it is two-byte encoded */
1145 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1146 s++;
1147 }
1148 *d++ = c;
6940069f
GS
1149 }
1150 *d = '\0';
246fae53 1151 *len = d - save;
6940069f
GS
1152 return save;
1153}
1154
1155/*
87cea99e 1156=for apidoc bytes_from_utf8
f9a63242 1157
2bbc8d55 1158Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1159Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1160the newly-created string, and updates C<len> to contain the new
1161length. Returns the original string if no conversion occurs, C<len>
72d33970 1162is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55
SP
11630 if C<s> is converted or consisted entirely of characters that are invariant
1164in utf8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1165
37607a96
PK
1166=cut
1167*/
f9a63242
JH
1168
1169U8 *
e1ec3a88 1170Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1171{
f9a63242 1172 U8 *d;
e1ec3a88
AL
1173 const U8 *start = s;
1174 const U8 *send;
f9a63242
JH
1175 I32 count = 0;
1176
7918f24d
NC
1177 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1178
96a5add6 1179 PERL_UNUSED_CONTEXT;
f9a63242 1180 if (!*is_utf8)
73d840c0 1181 return (U8 *)start;
f9a63242 1182
1e54db1a 1183 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1184 for (send = s + *len; s < send;) {
d59937ca
KW
1185 if (! UTF8_IS_INVARIANT(*s)) {
1186 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1187 return (U8 *)start;
d59937ca
KW
1188 }
1189 count++;
1190 s++;
db42d148 1191 }
d59937ca 1192 s++;
f9a63242
JH
1193 }
1194
35da51f7 1195 *is_utf8 = FALSE;
f9a63242 1196
212542aa 1197 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1198 s = start; start = d;
f9a63242
JH
1199 while (s < send) {
1200 U8 c = *s++;
1a91c45d 1201 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1202 /* Then it is two-byte encoded */
1a91c45d
KW
1203 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1204 s++;
c4d5f83a
NIS
1205 }
1206 *d++ = c;
f9a63242
JH
1207 }
1208 *d = '\0';
1209 *len = d - start;
73d840c0 1210 return (U8 *)start;
f9a63242
JH
1211}
1212
1213/*
87cea99e 1214=for apidoc bytes_to_utf8
6940069f 1215
ff97e5cf
KW
1216Converts a string C<s> of length C<len> bytes from the native encoding into
1217UTF-8.
6662521e 1218Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1219reflect the new length in bytes.
6940069f 1220
75200dff 1221A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
1222
1223If you want to convert to UTF-8 from encodings other than
1224the native (Latin1 or EBCDIC),
a1433954 1225see L</sv_recode_to_utf8>().
c9ada85f 1226
497711e7 1227=cut
6940069f
GS
1228*/
1229
c682ebef
FC
1230/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1231 likewise need duplication. */
1232
6940069f 1233U8*
35a4481c 1234Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1235{
35a4481c 1236 const U8 * const send = s + (*len);
6940069f
GS
1237 U8 *d;
1238 U8 *dst;
7918f24d
NC
1239
1240 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1241 PERL_UNUSED_CONTEXT;
6940069f 1242
212542aa 1243 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1244 dst = d;
1245
1246 while (s < send) {
55d09dc8
KW
1247 append_utf8_from_native_byte(*s, &d);
1248 s++;
6940069f
GS
1249 }
1250 *d = '\0';
6662521e 1251 *len = d-dst;
6940069f
GS
1252 return dst;
1253}
1254
a0ed51b3 1255/*
dea0fc0b 1256 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1257 *
1258 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1259 * We optimize for native, for obvious reasons. */
1260
1261U8*
dea0fc0b 1262Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1263{
dea0fc0b
JH
1264 U8* pend;
1265 U8* dstart = d;
1266
7918f24d
NC
1267 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1268
dea0fc0b 1269 if (bytelen & 1)
f5992bc4 1270 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1271
1272 pend = p + bytelen;
1273
a0ed51b3 1274 while (p < pend) {
dea0fc0b
JH
1275 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1276 p += 2;
56d37426
KW
1277 if (UNI_IS_INVARIANT(uv)) {
1278 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
1279 continue;
1280 }
56d37426
KW
1281 if (uv <= MAX_UTF8_TWO_BYTE) {
1282 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1283 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
1284 continue;
1285 }
46956fad
KW
1286#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1287#define LAST_HIGH_SURROGATE 0xDBFF
1288#define FIRST_LOW_SURROGATE 0xDC00
1289#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
1290 if (uv >= FIRST_HIGH_SURROGATE && uv <= LAST_HIGH_SURROGATE) {
01ea242b 1291 if (p >= pend) {
dea0fc0b 1292 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
01ea242b
NC
1293 } else {
1294 UV low = (p[0] << 8) + p[1];
1295 p += 2;
46956fad 1296 if (low < FIRST_LOW_SURROGATE || low > LAST_LOW_SURROGATE)
01ea242b 1297 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
46956fad
KW
1298 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1299 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 1300 }
46956fad 1301 } else if (uv >= FIRST_LOW_SURROGATE && uv <= LAST_LOW_SURROGATE) {
dbde1951 1302 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
a0ed51b3 1303 }
56d37426
KW
1304#ifdef EBCDIC
1305 d = uvoffuni_to_utf8_flags(d, uv, 0);
1306#else
a0ed51b3 1307 if (uv < 0x10000) {
eb160463
GS
1308 *d++ = (U8)(( uv >> 12) | 0xe0);
1309 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1310 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1311 continue;
1312 }
1313 else {
eb160463
GS
1314 *d++ = (U8)(( uv >> 18) | 0xf0);
1315 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1316 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1317 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1318 continue;
1319 }
56d37426 1320#endif
a0ed51b3 1321 }
dea0fc0b 1322 *newlen = d - dstart;
a0ed51b3
LW
1323 return d;
1324}
1325
1326/* Note: this one is slightly destructive of the source. */
1327
1328U8*
dea0fc0b 1329Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1330{
1331 U8* s = (U8*)p;
d4c19fe8 1332 U8* const send = s + bytelen;
7918f24d
NC
1333
1334 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1335
e0ea5e2d
NC
1336 if (bytelen & 1)
1337 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1338 (UV)bytelen);
1339
a0ed51b3 1340 while (s < send) {
d4c19fe8 1341 const U8 tmp = s[0];
a0ed51b3
LW
1342 s[0] = s[1];
1343 s[1] = tmp;
1344 s += 2;
1345 }
dea0fc0b 1346 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1347}
1348
922e8cb4
KW
1349bool
1350Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1351{
1352 U8 tmpbuf[UTF8_MAXBYTES+1];
1353 uvchr_to_utf8(tmpbuf, c);
1354 return _is_utf8_FOO(classnum, tmpbuf);
1355}
1356
f9ae8fb6
JD
1357/* Internal function so we can deprecate the external one, and call
1358 this one from other deprecated functions in this file */
1359
f2645549
KW
1360bool
1361Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385
KW
1362{
1363 dVAR;
f2645549 1364 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
1365
1366 if (*p == '_')
1367 return TRUE;
f25ce844 1368 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
1369}
1370
5092f92a 1371bool
eba68aa0
KW
1372Perl__is_uni_perl_idcont(pTHX_ UV c)
1373{
1374 U8 tmpbuf[UTF8_MAXBYTES+1];
1375 uvchr_to_utf8(tmpbuf, c);
1376 return _is_utf8_perl_idcont(tmpbuf);
1377}
1378
1379bool
f91dcd13
KW
1380Perl__is_uni_perl_idstart(pTHX_ UV c)
1381{
1382 U8 tmpbuf[UTF8_MAXBYTES+1];
1383 uvchr_to_utf8(tmpbuf, c);
1384 return _is_utf8_perl_idstart(tmpbuf);
1385}
1386
3a4c58c9
KW
1387UV
1388Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1389{
1390 /* We have the latin1-range values compiled into the core, so just use
1391 * those, converting the result to utf8. The only difference between upper
1392 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1393 * either "SS" or "Ss". Which one to use is passed into the routine in
1394 * 'S_or_s' to avoid a test */
1395
1396 UV converted = toUPPER_LATIN1_MOD(c);
1397
1398 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1399
1400 assert(S_or_s == 'S' || S_or_s == 's');
1401
6f2d5cbc 1402 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 1403 characters in this range */
3a4c58c9
KW
1404 *p = (U8) converted;
1405 *lenp = 1;
1406 return converted;
1407 }
1408
1409 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1410 * which it maps to one of them, so as to only have to have one check for
1411 * it in the main case */
1412 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1413 switch (c) {
1414 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1415 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1416 break;
1417 case MICRO_SIGN:
1418 converted = GREEK_CAPITAL_LETTER_MU;
1419 break;
1420 case LATIN_SMALL_LETTER_SHARP_S:
1421 *(p)++ = 'S';
1422 *p = S_or_s;
1423 *lenp = 2;
1424 return 'S';
1425 default:
1426 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
118e2215 1427 assert(0); /* NOTREACHED */
3a4c58c9
KW
1428 }
1429 }
1430
1431 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1432 *p = UTF8_TWO_BYTE_LO(converted);
1433 *lenp = 2;
1434
1435 return converted;
1436}
1437
50bda2c3
KW
1438/* Call the function to convert a UTF-8 encoded character to the specified case.
1439 * Note that there may be more than one character in the result.
1440 * INP is a pointer to the first byte of the input character
1441 * OUTP will be set to the first byte of the string of changed characters. It
1442 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1443 * LENP will be set to the length in bytes of the string of changed characters
1444 *
1445 * The functions return the ordinal of the first character in the string of OUTP */
4a8240a3
KW
1446#define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "")
1447#define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "")
1448#define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "")
50bda2c3
KW
1449
1450/* This additionally has the input parameter SPECIALS, which if non-zero will
1451 * cause this to use the SPECIALS hash for folding (meaning get full case
1452 * folding); otherwise, when zero, this implies a simple case fold */
4a8240a3 1453#define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "" : NULL)
c3fd2246 1454
84afefe6
JH
1455UV
1456Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1457{
3a4c58c9
KW
1458 dVAR;
1459
a1433954
KW
1460 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1461 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1462 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1463 * the changed version may be longer than the original character.
1464 *
1465 * The ordinal of the first character of the changed version is returned
1466 * (but note, as explained above, that there may be more.) */
1467
7918f24d
NC
1468 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1469
3a4c58c9
KW
1470 if (c < 256) {
1471 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1472 }
1473
0ebc6274 1474 uvchr_to_utf8(p, c);
3a4c58c9 1475 return CALL_UPPER_CASE(p, p, lenp);
a0ed51b3
LW
1476}
1477
84afefe6
JH
1478UV
1479Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1480{
3a4c58c9
KW
1481 dVAR;
1482
7918f24d
NC
1483 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1484
3a4c58c9
KW
1485 if (c < 256) {
1486 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1487 }
1488
0ebc6274 1489 uvchr_to_utf8(p, c);
3a4c58c9 1490 return CALL_TITLE_CASE(p, p, lenp);
a0ed51b3
LW
1491}
1492
afc16117
KW
1493STATIC U8
1494S_to_lower_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp)
1495{
1496 /* We have the latin1-range values compiled into the core, so just use
1497 * those, converting the result to utf8. Since the result is always just
a1433954 1498 * one character, we allow <p> to be NULL */
afc16117
KW
1499
1500 U8 converted = toLOWER_LATIN1(c);
1501
1502 if (p != NULL) {
6f2d5cbc 1503 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
1504 *p = converted;
1505 *lenp = 1;
1506 }
1507 else {
1508 *p = UTF8_TWO_BYTE_HI(converted);
1509 *(p+1) = UTF8_TWO_BYTE_LO(converted);
1510 *lenp = 2;
1511 }
1512 }
1513 return converted;
1514}
1515
84afefe6
JH
1516UV
1517Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1518{
968c5e6a
KW
1519 dVAR;
1520
7918f24d
NC
1521 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1522
afc16117
KW
1523 if (c < 256) {
1524 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1525 }
1526
afc16117 1527 uvchr_to_utf8(p, c);
968c5e6a 1528 return CALL_LOWER_CASE(p, p, lenp);
a0ed51b3
LW
1529}
1530
84afefe6 1531UV
51910141 1532Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 1533{
51910141 1534 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 1535 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 1536 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
1537 *
1538 * Not to be used for locale folds
51910141 1539 */
f673fad4 1540
a1dde8de
KW
1541 UV converted;
1542
1543 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1544
1ca267a5
KW
1545 assert (! (flags & FOLD_FLAGS_LOCALE));
1546
a1dde8de
KW
1547 if (c == MICRO_SIGN) {
1548 converted = GREEK_SMALL_LETTER_MU;
1549 }
51910141 1550 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1ca267a5
KW
1551
1552 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1553 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1554 * under those circumstances. */
1555 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1556 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1557 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1558 p, *lenp, U8);
1559 return LATIN_SMALL_LETTER_LONG_S;
1560 }
1561 else {
4f489194
KW
1562 *(p)++ = 's';
1563 *p = 's';
1564 *lenp = 2;
1565 return 's';
1ca267a5 1566 }
a1dde8de
KW
1567 }
1568 else { /* In this range the fold of all other characters is their lower
1569 case */
1570 converted = toLOWER_LATIN1(c);
1571 }
1572
6f2d5cbc 1573 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
1574 *p = (U8) converted;
1575 *lenp = 1;
1576 }
1577 else {
1578 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1579 *p = UTF8_TWO_BYTE_LO(converted);
1580 *lenp = 2;
1581 }
1582
1583 return converted;
1584}
1585
1586UV
31f05a37 1587Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 1588{
4b593389 1589
a0270393
KW
1590 /* Not currently externally documented, and subject to change
1591 * <flags> bits meanings:
1592 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
1593 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1594 * locale are to be used.
a0270393
KW
1595 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1596 */
4b593389 1597
36bb2ab6 1598 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1599
31f05a37
KW
1600 /* Tread a UTF-8 locale as not being in locale at all */
1601 if (IN_UTF8_CTYPE_LOCALE) {
1602 flags &= ~FOLD_FLAGS_LOCALE;
1603 }
1604
a1dde8de 1605 if (c < 256) {
a0270393 1606 UV result = _to_fold_latin1((U8) c, p, lenp,
31f05a37 1607 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a0270393
KW
1608 /* It is illegal for the fold to cross the 255/256 boundary under
1609 * locale; in this case return the original */
1610 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1611 ? c
1612 : result;
a1dde8de
KW
1613 }
1614
a0270393
KW
1615 /* If no special needs, just use the macro */
1616 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1617 uvchr_to_utf8(p, c);
1618 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1619 }
1620 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1621 the special flags. */
1622 U8 utf8_c[UTF8_MAXBYTES + 1];
1623 uvchr_to_utf8(utf8_c, c);
445bf929 1624 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
a0270393 1625 }
84afefe6
JH
1626}
1627
26483009 1628PERL_STATIC_INLINE bool
5141f98e 1629S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 1630 const char *const swashname, SV* const invlist)
bde6a22d 1631{
ea317ccb
KW
1632 /* returns a boolean giving whether or not the UTF8-encoded character that
1633 * starts at <p> is in the swash indicated by <swashname>. <swash>
1634 * contains a pointer to where the swash indicated by <swashname>
1635 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
1636 * look at <*swash> and only generate a swash if it is not null. <invlist>
1637 * is NULL or an inversion list that defines the swash. If not null, it
1638 * saves time during initialization of the swash.
ea317ccb
KW
1639 *
1640 * Note that it is assumed that the buffer length of <p> is enough to
1641 * contain all the bytes that comprise the character. Thus, <*p> should
1642 * have been checked before this call for mal-formedness enough to assure
1643 * that. */
1644
97aff369 1645 dVAR;
7918f24d
NC
1646
1647 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1648
492a624f 1649 /* The API should have included a length for the UTF-8 character in <p>,
28123549 1650 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
1651 * as far as there being enough bytes available in it to accommodate the
1652 * character without reading beyond the end, and pass that number on to the
1653 * validating routine */
6302f837 1654 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
28123549
KW
1655 if (ckWARN_d(WARN_UTF8)) {
1656 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
9816f121 1657 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
28123549
KW
1658 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1659 what the malformation is */
1660 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1661 }
1662 }
1663 return FALSE;
1664 }
87367d5f
KW
1665 if (!*swash) {
1666 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
1667 *swash = _core_swash_init("utf8",
1668
1669 /* Only use the name if there is no inversion
1670 * list; otherwise will go out to disk */
1671 (invlist) ? "" : swashname,
1672
1673 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 1674 }
28123549 1675
bde6a22d
NC
1676 return swash_fetch(*swash, p, TRUE) != 0;
1677}
1678
1679bool
922e8cb4
KW
1680Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1681{
1682 dVAR;
1683
1684 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1685
1686 assert(classnum < _FIRST_NON_SWASH_CC);
1687
f25ce844
KW
1688 return is_utf8_common(p,
1689 &PL_utf8_swash_ptrs[classnum],
1690 swash_property_names[classnum],
1691 PL_XPosix_ptrs[classnum]);
922e8cb4
KW
1692}
1693
1694bool
f2645549 1695Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
a0ed51b3 1696{
97aff369 1697 dVAR;
f2645549 1698 SV* invlist = NULL;
7918f24d 1699
f2645549 1700 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
7918f24d 1701
f2645549
KW
1702 if (! PL_utf8_perl_idstart) {
1703 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1704 }
1705 return is_utf8_common(p, &PL_utf8_perl_idstart, "", invlist);
82686b01
JH
1706}
1707
1708bool
f2645549 1709Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943
KW
1710{
1711 dVAR;
1712
f2645549 1713 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
1714
1715 if (*p == '_')
1716 return TRUE;
f25ce844 1717 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
1718}
1719
1720bool
eba68aa0
KW
1721Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1722{
1723 dVAR;
b24b43f7 1724 SV* invlist = NULL;
eba68aa0
KW
1725
1726 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1727
b24b43f7
KW
1728 if (! PL_utf8_perl_idcont) {
1729 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1730 }
1731 return is_utf8_common(p, &PL_utf8_perl_idcont, "", invlist);
eba68aa0
KW
1732}
1733
eba68aa0 1734bool
f2645549 1735Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 1736{
97aff369 1737 dVAR;
7918f24d 1738
f2645549 1739 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 1740
f25ce844 1741 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
1742}
1743
1744bool
f2645549 1745Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943
KW
1746{
1747 dVAR;
1748
f2645549 1749 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 1750
f25ce844 1751 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
1752}
1753
1754bool
7dbf68d2
KW
1755Perl__is_utf8_mark(pTHX_ const U8 *p)
1756{
1757 dVAR;
1758
1759 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1760
f25ce844 1761 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
1762}
1763
6b5c0936 1764/*
87cea99e 1765=for apidoc to_utf8_case
6b5c0936 1766
6fae5207 1767C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
1768the character that is being converted. This routine assumes that the character
1769at C<p> is well-formed.
6b5c0936 1770
6fae5207
KW
1771C<ustrp> is a pointer to the character buffer to put the
1772conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
1773of the result.
1774
6fae5207 1775C<swashp> is a pointer to the swash to use.
6b5c0936 1776
a1433954 1777Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
6fae5207 1778and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 1779but not always, a multicharacter mapping), is tried first.
6b5c0936 1780
4a8240a3
KW
1781C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1782any special mappings; C<""> means to use the special mappings. Values other
1783than these two are treated as the name of the hash containing the special
1784mappings, like C<"utf8::ToSpecLower">.
6b5c0936 1785
6fae5207 1786C<normal> is a string like "ToLower" which means the swash
0134edef
JH
1787%utf8::ToLower.
1788
1789=cut */
6b5c0936 1790
2104c8d9 1791UV
9a957fbc
AL
1792Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1793 SV **swashp, const char *normal, const char *special)
a0ed51b3 1794{
97aff369 1795 dVAR;
0134edef 1796 STRLEN len = 0;
f4cd282c 1797 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
7918f24d
NC
1798
1799 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1800
9ae3ac1a
KW
1801 /* Note that swash_fetch() doesn't output warnings for these because it
1802 * assumes we will */
8457b38f 1803 if (uv1 >= UNICODE_SURROGATE_FIRST) {
9ae3ac1a 1804 if (uv1 <= UNICODE_SURROGATE_LAST) {
8457b38f
KW
1805 if (ckWARN_d(WARN_SURROGATE)) {
1806 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1807 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1808 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1809 }
9ae3ac1a
KW
1810 }
1811 else if (UNICODE_IS_SUPER(uv1)) {
8457b38f
KW
1812 if (ckWARN_d(WARN_NON_UNICODE)) {
1813 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1814 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1815 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1816 }
9ae3ac1a
KW
1817 }
1818
1819 /* Note that non-characters are perfectly legal, so no warning should
1820 * be given */
1821 }
1822
0134edef 1823 if (!*swashp) /* load on-demand */
5ab9d2ef 1824 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 1825
a6f87d8c 1826 if (special) {
0134edef 1827 /* It might be "special" (sometimes, but not always,
2a37f04d 1828 * a multicharacter mapping) */
4a8240a3 1829 HV *hv = NULL;
b08cf34e
JH
1830 SV **svp;
1831
4a8240a3
KW
1832 /* If passed in the specials name, use that; otherwise use any
1833 * given in the swash */
1834 if (*special != '\0') {
1835 hv = get_hv(special, 0);
1836 }
1837 else {
1838 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
1839 if (svp) {
1840 hv = MUTABLE_HV(SvRV(*svp));
1841 }
1842 }
1843
176fe009
KW
1844 if (hv
1845 && (svp = hv_fetch(hv, (const char*)p, UNISKIP(uv1), FALSE))
1846 && (*svp))
1847 {
cfd0369c 1848 const char *s;
47654450 1849
cfd0369c 1850 s = SvPV_const(*svp, len);
47654450 1851 if (len == 1)
f4cd282c 1852 /* EIGHTBIT */
c80e42f3 1853 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 1854 else {
d2dcd0fb 1855 Copy(s, ustrp, len, U8);
29e98929 1856 }
983ffd37 1857 }
0134edef
JH
1858 }
1859
1860 if (!len && *swashp) {
f4cd282c 1861 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is utf8 */);
d4c19fe8 1862
0134edef
JH
1863 if (uv2) {
1864 /* It was "normal" (a single character mapping). */
f4cd282c 1865 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
1866 }
1867 }
1feea2c7 1868
cbe07460
KW
1869 if (len) {
1870 if (lenp) {
1871 *lenp = len;
1872 }
1873 return valid_utf8_to_uvchr(ustrp, 0);
1874 }
1875
1876 /* Here, there was no mapping defined, which means that the code point maps
1877 * to itself. Return the inputs */
bfdf22ec 1878 len = UTF8SKIP(p);
ca9fab46
KW
1879 if (p != ustrp) { /* Don't copy onto itself */
1880 Copy(p, ustrp, len, U8);
1881 }
0134edef 1882
2a37f04d
JH
1883 if (lenp)
1884 *lenp = len;
1885
f4cd282c 1886 return uv1;
cbe07460 1887
a0ed51b3
LW
1888}
1889
051a06d4
KW
1890STATIC UV
1891S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
1892{
1893 /* This is called when changing the case of a utf8-encoded character above
31f05a37
KW
1894 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
1895 * result contains a character that crosses the 255/256 boundary, disallow
1896 * the change, and return the original code point. See L<perlfunc/lc> for
1897 * why;
051a06d4 1898 *
a1433954
KW
1899 * p points to the original string whose case was changed; assumed
1900 * by this routine to be well-formed
051a06d4
KW
1901 * result the code point of the first character in the changed-case string
1902 * ustrp points to the changed-case string (<result> represents its first char)
1903 * lenp points to the length of <ustrp> */
1904
1905 UV original; /* To store the first code point of <p> */
1906
1907 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
1908
a4f12ed7 1909 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
1910
1911 /* We know immediately if the first character in the string crosses the
1912 * boundary, so can skip */
1913 if (result > 255) {
1914
1915 /* Look at every character in the result; if any cross the
1916 * boundary, the whole thing is disallowed */
1917 U8* s = ustrp + UTF8SKIP(ustrp);
1918 U8* e = ustrp + *lenp;
1919 while (s < e) {
a4f12ed7 1920 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
1921 goto bad_crossing;
1922 }
1923 s += UTF8SKIP(s);
1924 }
1925
1926 /* Here, no characters crossed, result is ok as-is */
1927 return result;
1928 }
1929
1930bad_crossing:
1931
1932 /* Failed, have to return the original */
4b88fb76 1933 original = valid_utf8_to_uvchr(p, lenp);
051a06d4
KW
1934 Copy(p, ustrp, *lenp, char);
1935 return original;
1936}
1937
d3e79532 1938/*
87cea99e 1939=for apidoc to_utf8_upper
d3e79532 1940
1f607577 1941Instead use L</toUPPER_utf8>.
a1433954 1942
d3e79532
JH
1943=cut */
1944
051a06d4 1945/* Not currently externally documented, and subject to change:
31f05a37
KW
1946 * <flags> is set iff iff the rules from the current underlying locale are to
1947 * be used. */
051a06d4 1948
2104c8d9 1949UV
31f05a37 1950Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 1951{
97aff369 1952 dVAR;
7918f24d 1953
051a06d4
KW
1954 UV result;
1955
1956 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 1957
31f05a37
KW
1958 if (flags && IN_UTF8_CTYPE_LOCALE) {
1959 flags = FALSE;
1960 }
1961
3a4c58c9 1962 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
1963 if (flags) {
1964 result = toUPPER_LC(*p);
1965 }
1966 else {
81c6c7ce 1967 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 1968 }
3a4c58c9
KW
1969 }
1970 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 1971 if (flags) {
a6d8b88b 1972 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 1973 result = toUPPER_LC(c);
051a06d4
KW
1974 }
1975 else {
94bb8c36 1976 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 1977 ustrp, lenp, 'S');
051a06d4
KW
1978 }
1979 }
1980 else { /* utf8, ord above 255 */
1981 result = CALL_UPPER_CASE(p, ustrp, lenp);
1982
1983 if (flags) {
1984 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
1985 }
1986 return result;
1987 }
1988
1989 /* Here, used locale rules. Convert back to utf8 */
1990 if (UTF8_IS_INVARIANT(result)) {
1991 *ustrp = (U8) result;
1992 *lenp = 1;
1993 }
1994 else {
62cb07ea
KW
1995 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
1996 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 1997 *lenp = 2;
3a4c58c9 1998 }
baa60164 1999
051a06d4 2000 return result;
983ffd37 2001}
a0ed51b3 2002
d3e79532 2003/*
87cea99e 2004=for apidoc to_utf8_title
d3e79532 2005
1f607577 2006Instead use L</toTITLE_utf8>.
a1433954 2007
d3e79532
JH
2008=cut */
2009
051a06d4 2010/* Not currently externally documented, and subject to change:
31f05a37
KW
2011 * <flags> is set iff the rules from the current underlying locale are to be
2012 * used. Since titlecase is not defined in POSIX, for other than a
2013 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 2014 */
051a06d4 2015
983ffd37 2016UV
31f05a37 2017Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
983ffd37 2018{
97aff369 2019 dVAR;
7918f24d 2020
051a06d4
KW
2021 UV result;
2022
2023 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2024
31f05a37
KW
2025 if (flags && IN_UTF8_CTYPE_LOCALE) {
2026 flags = FALSE;
2027 }
2028
3a4c58c9 2029 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2030 if (flags) {
2031 result = toUPPER_LC(*p);
2032 }
2033 else {
81c6c7ce 2034 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2035 }
3a4c58c9
KW
2036 }
2037 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2038 if (flags) {
a6d8b88b 2039 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2040 result = toUPPER_LC(c);
051a06d4
KW
2041 }
2042 else {
94bb8c36 2043 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2044 ustrp, lenp, 's');
051a06d4
KW
2045 }
2046 }
2047 else { /* utf8, ord above 255 */
2048 result = CALL_TITLE_CASE(p, ustrp, lenp);
2049
2050 if (flags) {
2051 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2052 }
2053 return result;
2054 }
2055
2056 /* Here, used locale rules. Convert back to utf8 */
2057 if (UTF8_IS_INVARIANT(result)) {
2058 *ustrp = (U8) result;
2059 *lenp = 1;
2060 }
2061 else {
62cb07ea
KW
2062 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2063 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2064 *lenp = 2;
3a4c58c9
KW
2065 }
2066
051a06d4 2067 return result;
a0ed51b3
LW
2068}
2069
d3e79532 2070/*
87cea99e 2071=for apidoc to_utf8_lower
d3e79532 2072
1f607577 2073Instead use L</toLOWER_utf8>.
a1433954 2074
d3e79532
JH
2075=cut */
2076
051a06d4 2077/* Not currently externally documented, and subject to change:
31f05a37
KW
2078 * <flags> is set iff iff the rules from the current underlying locale are to
2079 * be used.
2080 */
051a06d4 2081
2104c8d9 2082UV
31f05a37 2083Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2084{
051a06d4
KW
2085 UV result;
2086
97aff369 2087 dVAR;
7918f24d 2088
051a06d4 2089 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2090
31f05a37
KW
2091 if (flags && IN_UTF8_CTYPE_LOCALE) {
2092 flags = FALSE;
2093 }
2094
968c5e6a 2095 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2096 if (flags) {
2097 result = toLOWER_LC(*p);
2098 }
2099 else {
81c6c7ce 2100 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2101 }
968c5e6a
KW
2102 }
2103 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2104 if (flags) {
a6d8b88b 2105 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2106 result = toLOWER_LC(c);
051a06d4
KW
2107 }
2108 else {
94bb8c36 2109 return to_lower_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2110 ustrp, lenp);
051a06d4 2111 }
968c5e6a 2112 }
051a06d4
KW
2113 else { /* utf8, ord above 255 */
2114 result = CALL_LOWER_CASE(p, ustrp, lenp);
2115
2116 if (flags) {
2117 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2118 }
968c5e6a 2119
051a06d4
KW
2120 return result;
2121 }
2122
2123 /* Here, used locale rules. Convert back to utf8 */
2124 if (UTF8_IS_INVARIANT(result)) {
2125 *ustrp = (U8) result;
2126 *lenp = 1;
2127 }
2128 else {
62cb07ea
KW
2129 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2130 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2131 *lenp = 2;
2132 }
2133
051a06d4 2134 return result;
b4e400f9
JH
2135}
2136
d3e79532 2137/*
87cea99e 2138=for apidoc to_utf8_fold
d3e79532 2139
1f607577 2140Instead use L</toFOLD_utf8>.
a1433954 2141
d3e79532
JH
2142=cut */
2143
051a06d4
KW
2144/* Not currently externally documented, and subject to change,
2145 * in <flags>
31f05a37
KW
2146 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2147 * locale are to be used.
051a06d4
KW
2148 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2149 * otherwise simple folds
a0270393
KW
2150 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2151 * prohibited
445bf929 2152 */
36bb2ab6 2153
b4e400f9 2154UV
445bf929 2155Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
b4e400f9 2156{
97aff369 2157 dVAR;
7918f24d 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
31f05a37
KW
2168 if (flags & FOLD_FLAGS_LOCALE && IN_UTF8_CTYPE_LOCALE) {
2169 flags &= ~FOLD_FLAGS_LOCALE;
2170 }
2171
a1dde8de 2172 if (UTF8_IS_INVARIANT(*p)) {
051a06d4 2173 if (flags & FOLD_FLAGS_LOCALE) {
d22b930b 2174 result = toFOLD_LC(*p);
051a06d4
KW
2175 }
2176 else {
81c6c7ce 2177 return _to_fold_latin1(*p, ustrp, lenp,
1ca267a5 2178 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2179 }
a1dde8de
KW
2180 }
2181 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2182 if (flags & FOLD_FLAGS_LOCALE) {
a6d8b88b 2183 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2184 result = toFOLD_LC(c);
051a06d4
KW
2185 }
2186 else {
94bb8c36 2187 return _to_fold_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
51910141 2188 ustrp, lenp,
1ca267a5 2189 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2190 }
a1dde8de 2191 }
051a06d4 2192 else { /* utf8, ord above 255 */
a0270393 2193 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2194
1ca267a5
KW
2195 if (flags & FOLD_FLAGS_LOCALE) {
2196
538e84ed
KW
2197 /* Special case these two characters, as what normally gets
2198 * returned under locale doesn't work */
1ca267a5
KW
2199 if (UTF8SKIP(p) == sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1
2200 && memEQ((char *) p, LATIN_CAPITAL_LETTER_SHARP_S_UTF8,
2201 sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1))
2202 {
2203 goto return_long_s;
2204 }
9fc2026f
KW
2205 else if (UTF8SKIP(p) == sizeof(LATIN_SMALL_LIGATURE_LONG_S_T) - 1
2206 && memEQ((char *) p, LATIN_SMALL_LIGATURE_LONG_S_T_UTF8,
2207 sizeof(LATIN_SMALL_LIGATURE_LONG_S_T_UTF8) - 1))
2208 {
2209 goto return_ligature_st;
2210 }
a0270393 2211 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2212 }
a0270393
KW
2213 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2214 return result;
2215 }
2216 else {
2217 /* This is called when changing the case of a utf8-encoded
9fc2026f
KW
2218 * character above the ASCII range, and the result should not
2219 * contain an ASCII character. */
a0270393
KW
2220
2221 UV original; /* To store the first code point of <p> */
2222
2223 /* Look at every character in the result; if any cross the
2224 * boundary, the whole thing is disallowed */
2225 U8* s = ustrp;
2226 U8* e = ustrp + *lenp;
2227 while (s < e) {
2228 if (isASCII(*s)) {
2229 /* Crossed, have to return the original */
2230 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 2231
9fc2026f 2232 /* But in these instances, there is an alternative we can
1ca267a5 2233 * return that is valid */
9fc2026f
KW
2234 if (original == LATIN_CAPITAL_LETTER_SHARP_S
2235 || original == LATIN_SMALL_LETTER_SHARP_S)
2236 {
1ca267a5
KW
2237 goto return_long_s;
2238 }
9fc2026f
KW
2239 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2240 goto return_ligature_st;
2241 }
a0270393
KW
2242 Copy(p, ustrp, *lenp, char);
2243 return original;
2244 }
2245 s += UTF8SKIP(s);
2246 }
051a06d4 2247
a0270393
KW
2248 /* Here, no characters crossed, result is ok as-is */
2249 return result;
2250 }
051a06d4
KW
2251 }
2252
2253 /* Here, used locale rules. Convert back to utf8 */
2254 if (UTF8_IS_INVARIANT(result)) {
2255 *ustrp = (U8) result;
2256 *lenp = 1;
2257 }
2258 else {
62cb07ea
KW
2259 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2260 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2261 *lenp = 2;
2262 }
2263
051a06d4 2264 return result;
1ca267a5
KW
2265
2266 return_long_s:
2267 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2268 * folds to a string of two of these characters. By returning this
2269 * instead, then, e.g.,
2270 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2271 * works. */
2272
2273 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2274 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2275 ustrp, *lenp, U8);
2276 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
2277
2278 return_ligature_st:
2279 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2280 * have the other one fold to it */
2281
2282 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2283 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2284 return LATIN_SMALL_LIGATURE_ST;
a0ed51b3
LW
2285}
2286
711a919c 2287/* Note:
f90a9a02 2288 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2289 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2290 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2291 */
c4a5db0c 2292
a0ed51b3 2293SV*
7fc63493 2294Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2295{
c4a5db0c
KW
2296 PERL_ARGS_ASSERT_SWASH_INIT;
2297
2298 /* Returns a copy of a swash initiated by the called function. This is the
2299 * public interface, and returning a copy prevents others from doing
2300 * mischief on the original */
2301
5d3d13d1 2302 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
2303}
2304
2305SV*
5d3d13d1 2306Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c
KW
2307{
2308 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
2309 * by calling utf8_heavy.pl in the general case. The returned value may be
2310 * the swash's inversion list instead if the input parameters allow it.
2311 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
2312 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2313 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
2314 *
2315 * This interface should only be used by functions that won't destroy or
2316 * adversely change the swash, as doing so affects all other uses of the
2317 * swash in the program; the general public should use 'Perl_swash_init'
2318 * instead.
2319 *
2320 * pkg is the name of the package that <name> should be in.
2321 * name is the name of the swash to find. Typically it is a Unicode
2322 * property name, including user-defined ones
2323 * listsv is a string to initialize the swash with. It must be of the form
2324 * documented as the subroutine return value in
2325 * L<perlunicode/User-Defined Character Properties>
2326 * minbits is the number of bits required to represent each data element.
2327 * It is '1' for binary properties.
2328 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 2329 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
2330 * flags_p if non-NULL is the address of various input and output flag bits
2331 * to the routine, as follows: ('I' means is input to the routine;
2332 * 'O' means output from the routine. Only flags marked O are
2333 * meaningful on return.)
2334 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2335 * came from a user-defined property. (I O)
5d3d13d1
KW
2336 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2337 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
2338 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2339 * return of an inversion list instead of a swash hash if this routine
2340 * thinks that would result in faster execution of swash_fetch() later
2341 * on. (I)
9a53f6cf
KW
2342 *
2343 * Thus there are three possible inputs to find the swash: <name>,
2344 * <listsv>, and <invlist>. At least one must be specified. The result
2345 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
2346 * actions can intersect, etc. what <name> gives. To avoid going out to
2347 * disk at all, <invlist> should specify completely what the swash should
2348 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
2349 *
2350 * <invlist> is only valid for binary properties */
c4a5db0c 2351
27da23d5 2352 dVAR;
c4a5db0c 2353 SV* retval = &PL_sv_undef;
83199d38 2354 HV* swash_hv = NULL;
87367d5f
KW
2355 const int invlist_swash_boundary =
2356 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2357 ? 512 /* Based on some benchmarking, but not extensive, see commit
2358 message */
2359 : -1; /* Never return just an inversion list */
9a53f6cf
KW
2360
2361 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2362 assert(! invlist || minbits == 1);
2363
2364 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2365 * so */
2366 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2367 dSP;
2368 const size_t pkg_len = strlen(pkg);
2369 const size_t name_len = strlen(name);
2370 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2371 SV* errsv_save;
2372 GV *method;
2373
2374 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2375
2376 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2377 ENTER;
69794297
KW
2378 SAVEHINTS();
2379 save_re_context();
650f067c
JL
2380 /* We might get here via a subroutine signature which uses a utf8
2381 * parameter name, at which point PL_subname will have been set
2382 * but not yet used. */
2383 save_item(PL_subname);
69794297
KW
2384 if (PL_parser && PL_parser->error_count)
2385 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2386 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2387 if (!method) { /* demand load utf8 */
2388 ENTER;
db2c6cb3
FC
2389 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2390 GvSV(PL_errgv) = NULL;
69794297
KW
2391 /* It is assumed that callers of this routine are not passing in
2392 * any user derived data. */
2393 /* Need to do this after save_re_context() as it will set
2394 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2395 * in Perl_magic_get). Even line to create errsv_save can turn on
2396 * PL_tainted. */
284167a5
SM
2397#ifndef NO_TAINT_SUPPORT
2398 SAVEBOOL(TAINT_get);
2399 TAINT_NOT;
2400#endif
69794297
KW
2401 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2402 NULL);
eed484f9 2403 {
db2c6cb3
FC
2404 /* Not ERRSV, as there is no need to vivify a scalar we are
2405 about to discard. */
2406 SV * const errsv = GvSV(PL_errgv);
2407 if (!SvTRUE(errsv)) {
2408 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2409 SvREFCNT_dec(errsv);
2410 }
eed484f9 2411 }
69794297
KW
2412 LEAVE;
2413 }
2414 SPAGAIN;
2415 PUSHMARK(SP);
2416 EXTEND(SP,5);
2417 mPUSHp(pkg, pkg_len);
2418 mPUSHp(name, name_len);
2419 PUSHs(listsv);
2420 mPUSHi(minbits);
2421 mPUSHi(none);
2422 PUTBACK;
db2c6cb3
FC
2423 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2424 GvSV(PL_errgv) = NULL;
69794297
KW
2425 /* If we already have a pointer to the method, no need to use
2426 * call_method() to repeat the lookup. */
c41800a8
KW
2427 if (method
2428 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
2429 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2430 {
2431 retval = *PL_stack_sp--;
2432 SvREFCNT_inc(retval);
2433 }
eed484f9 2434 {
db2c6cb3
FC
2435 /* Not ERRSV. See above. */
2436 SV * const errsv = GvSV(PL_errgv);
2437 if (!SvTRUE(errsv)) {
2438 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2439 SvREFCNT_dec(errsv);
2440 }
eed484f9 2441 }
ce3b816e 2442 LEAVE;
69794297
KW
2443 POPSTACK;
2444 if (IN_PERL_COMPILETIME) {
2445 CopHINTS_set(PL_curcop, PL_hints);
2446 }
2447 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2448 if (SvPOK(retval))
2449
2450 /* If caller wants to handle missing properties, let them */
5d3d13d1 2451 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
69794297
KW
2452 return NULL;
2453 }
2454 Perl_croak(aTHX_
2455 "Can't find Unicode property definition \"%"SVf"\"",
2456 SVfARG(retval));
2457 Perl_croak(aTHX_ "SWASHNEW didn't return an HV ref");
2458 }
9a53f6cf 2459 } /* End of calling the module to find the swash */
36eb48b4 2460
83199d38
KW
2461 /* If this operation fetched a swash, and we will need it later, get it */
2462 if (retval != &PL_sv_undef
2463 && (minbits == 1 || (flags_p
2464 && ! (*flags_p
2465 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2466 {
2467 swash_hv = MUTABLE_HV(SvRV(retval));
2468
2469 /* If we don't already know that there is a user-defined component to
2470 * this swash, and the user has indicated they wish to know if there is
2471 * one (by passing <flags_p>), find out */
2472 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2473 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2474 if (user_defined && SvUV(*user_defined)) {
2475 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2476 }
2477 }
2478 }
2479
36eb48b4
KW
2480 /* Make sure there is an inversion list for binary properties */
2481 if (minbits == 1) {
2482 SV** swash_invlistsvp = NULL;
2483 SV* swash_invlist = NULL;
9a53f6cf 2484 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
2485 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2486 an unclaimed reference count */
36eb48b4 2487
9a53f6cf 2488 /* If this operation fetched a swash, get its already existing
83199d38 2489 * inversion list, or create one for it */
36eb48b4 2490
83199d38 2491 if (swash_hv) {
5c9f4bd2 2492 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
2493 if (swash_invlistsvp) {
2494 swash_invlist = *swash_invlistsvp;
2495 invlist_in_swash_is_valid = TRUE;
2496 }
2497 else {
36eb48b4 2498 swash_invlist = _swash_to_invlist(retval);
02c85471 2499 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
2500 }
2501 }
2502
2503 /* If an inversion list was passed in, have to include it */
2504 if (invlist) {
2505
2506 /* Any fetched swash will by now have an inversion list in it;
2507 * otherwise <swash_invlist> will be NULL, indicating that we
2508 * didn't fetch a swash */
2509 if (swash_invlist) {
2510
2511 /* Add the passed-in inversion list, which invalidates the one
2512 * already stored in the swash */
2513 invlist_in_swash_is_valid = FALSE;
2514 _invlist_union(invlist, swash_invlist, &swash_invlist);
2515 }
2516 else {
2517
87367d5f
KW
2518 /* Here, there is no swash already. Set up a minimal one, if
2519 * we are going to return a swash */
2520 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 2521 swash_hv = newHV();
4aca0fe6 2522 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 2523 }
9a53f6cf
KW
2524 swash_invlist = invlist;
2525 }
9a53f6cf
KW
2526 }
2527
2528 /* Here, we have computed the union of all the passed-in data. It may
2529 * be that there was an inversion list in the swash which didn't get
538e84ed 2530 * touched; otherwise save the computed one */
87367d5f
KW
2531 if (! invlist_in_swash_is_valid
2532 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2533 {
5c9f4bd2 2534 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
2535 {
2536 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2537 }
cc34d8c5
FC
2538 /* We just stole a reference count. */
2539 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2540 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 2541 }
87367d5f 2542
dbfdbd26
KW
2543 SvREADONLY_on(swash_invlist);
2544
c41800a8 2545 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
2546 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2547 SvREFCNT_dec(retval);
02c85471
FC
2548 if (!swash_invlist_unclaimed)
2549 SvREFCNT_inc_simple_void_NN(swash_invlist);
2550 retval = newRV_noinc(swash_invlist);
87367d5f 2551 }
36eb48b4
KW
2552 }
2553
a0ed51b3
LW
2554 return retval;
2555}
2556
035d37be
JH
2557
2558/* This API is wrong for special case conversions since we may need to
2559 * return several Unicode characters for a single Unicode character
2560 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2561 * the lower-level routine, and it is similarly broken for returning
38684baa
KW
2562 * multiple values. --jhi
2563 * For those, you should use to_utf8_case() instead */
b0e3252e 2564/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
2565
2566/* Note:
2567 * Returns the value of property/mapping C<swash> for the first character
2568 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3d0f8846
KW
2569 * assumed to be in well-formed utf8. If C<do_utf8> is false, the string C<ptr>
2570 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
2571 *
2572 * A "swash" is a hash which contains initially the keys/values set up by
2573 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2574 * property for all possible code points. Things are stored in a compact form
2575 * (see utf8_heavy.pl) so that calculation is required to find the actual
2576 * property value for a given code point. As code points are looked up, new
2577 * key/value pairs are added to the hash, so that the calculation doesn't have
2578 * to ever be re-done. Further, each calculation is done, not just for the
2579 * desired one, but for a whole block of code points adjacent to that one.
2580 * For binary properties on ASCII machines, the block is usually for 64 code
2581 * points, starting with a code point evenly divisible by 64. Thus if the
2582 * property value for code point 257 is requested, the code goes out and
2583 * calculates the property values for all 64 code points between 256 and 319,
2584 * and stores these as a single 64-bit long bit vector, called a "swatch",
2585 * under the key for code point 256. The key is the UTF-8 encoding for code
2586 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2587 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2588 * for code point 258 is then requested, this code realizes that it would be
2589 * stored under the key for 256, and would find that value and extract the
2590 * relevant bit, offset from 256.
2591 *
2592 * Non-binary properties are stored in as many bits as necessary to represent
2593 * their values (32 currently, though the code is more general than that), not
2594 * as single bits, but the principal is the same: the value for each key is a
2595 * vector that encompasses the property values for all code points whose UTF-8
2596 * representations are represented by the key. That is, for all code points
2597 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2598 * bytes of that.
680c470c 2599 */
a0ed51b3 2600UV
680c470c 2601Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 2602{
27da23d5 2603 dVAR;
ef8f7699 2604 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
2605 U32 klen;
2606 U32 off;
9b56a019 2607 STRLEN slen = 0;
7d85a32c 2608 STRLEN needents;
cfd0369c 2609 const U8 *tmps = NULL;
979f2922 2610 SV *swatch;
08fb1ac5 2611 const U8 c = *ptr;
3568d838 2612
7918f24d
NC
2613 PERL_ARGS_ASSERT_SWASH_FETCH;
2614
87367d5f
KW
2615 /* If it really isn't a hash, it isn't really swash; must be an inversion
2616 * list */
2617 if (SvTYPE(hv) != SVt_PVHV) {
2618 return _invlist_contains_cp((SV*)hv,
2619 (do_utf8)
2620 ? valid_utf8_to_uvchr(ptr, NULL)
2621 : c);
2622 }
2623
08fb1ac5
KW
2624 /* We store the values in a "swatch" which is a vec() value in a swash
2625 * hash. Code points 0-255 are a single vec() stored with key length
2626 * (klen) 0. All other code points have a UTF-8 representation
2627 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2628 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2629 * length for them is the length of the encoded char - 1. ptr[klen] is the
2630 * final byte in the sequence representing the character */
2631 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2632 klen = 0;
2633 needents = 256;
2634 off = c;
3568d838 2635 }
08fb1ac5
KW
2636 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2637 klen = 0;
2638 needents = 256;
2639 off = TWO_BYTE_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
2640 }
2641 else {
08fb1ac5
KW
2642 klen = UTF8SKIP(ptr) - 1;
2643
2644 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2645 * the vec is the final byte in the sequence. (In EBCDIC this is
2646 * converted to I8 to get consecutive values.) To help you visualize
2647 * all this:
2648 * Straight 1047 After final byte
2649 * UTF-8 UTF-EBCDIC I8 transform
2650 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2651 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2652 * ...
2653 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2654 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2655 * ...
2656 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2657 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2658 * ...
2659 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2660 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2661 * ...
2662 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2663 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2664 *
2665 * (There are no discontinuities in the elided (...) entries.)
2666 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2667 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2668 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2669 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2670 * index into the vec() swatch (after subtracting 0x80, which we
2671 * actually do with an '&').
2672 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2673 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2674 * dicontinuities which go away by transforming it into I8, and we
2675 * effectively subtract 0xA0 to get the index. */
979f2922 2676 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 2677 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 2678 }
7d85a32c 2679
a0ed51b3
LW
2680 /*
2681 * This single-entry cache saves about 1/3 of the utf8 overhead in test
2682 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2683 * it's nothing to sniff at.) Pity we usually come through at least
2684 * two function calls to get here...
2685 *
2686 * NB: this code assumes that swatches are never modified, once generated!
2687 */
2688
3568d838 2689 if (hv == PL_last_swash_hv &&
a0ed51b3 2690 klen == PL_last_swash_klen &&
27da23d5 2691 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
2692 {
2693 tmps = PL_last_swash_tmps;
2694 slen = PL_last_swash_slen;
2695 }
2696 else {
2697 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 2698 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 2699
b0e3252e 2700 /* If not cached, generate it via swatch_get */
979f2922 2701 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
2702 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2703 {
2704 if (klen) {
2705 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2706 swatch = swatch_get(swash,
2707 code_point & ~((UV)needents - 1),
2708 needents);
2709 }
2710 else { /* For the first 256 code points, the swatch has a key of
2711 length 0 */
2712 swatch = swatch_get(swash, 0, needents);
2713 }
979f2922 2714
923e4eb5 2715 if (IN_PERL_COMPILETIME)
623e6609 2716 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 2717
979f2922 2718 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 2719
979f2922
ST
2720 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2721 || (slen << 3) < needents)
5637ef5b
NC
2722 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2723 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2724 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
2725 }
2726
2727 PL_last_swash_hv = hv;
16d8f38a 2728 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 2729 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
2730 /* FIXME change interpvar.h? */
2731 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
2732 PL_last_swash_slen = slen;
2733 if (klen)
2734 Copy(ptr, PL_last_swash_key, klen, U8);
2735 }
2736
9faf8d75 2737 switch ((int)((slen << 3) / needents)) {
a0ed51b3 2738 case 1:
e7aca353 2739 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
a0ed51b3 2740 case 8:
e7aca353 2741 return ((UV) tmps[off]);
a0ed51b3
LW
2742 case 16:
2743 off <<= 1;
e7aca353
JH
2744 return
2745 ((UV) tmps[off ] << 8) +
2746 ((UV) tmps[off + 1]);
a0ed51b3
LW
2747 case 32:
2748 off <<= 2;
e7aca353
JH
2749 return
2750 ((UV) tmps[off ] << 24) +
2751 ((UV) tmps[off + 1] << 16) +
2752 ((UV) tmps[off + 2] << 8) +
2753 ((UV) tmps[off + 3]);
a0ed51b3 2754 }
5637ef5b
NC
2755 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2756 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 2757 NORETURN_FUNCTION_END;
a0ed51b3 2758}
2b9d42f0 2759
319009ee
KW
2760/* Read a single line of the main body of the swash input text. These are of
2761 * the form:
2762 * 0053 0056 0073
2763 * where each number is hex. The first two numbers form the minimum and
2764 * maximum of a range, and the third is the value associated with the range.
2765 * Not all swashes should have a third number
2766 *
2767 * On input: l points to the beginning of the line to be examined; it points
2768 * to somewhere in the string of the whole input text, and is
2769 * terminated by a \n or the null string terminator.
2770 * lend points to the null terminator of that string
2771 * wants_value is non-zero if the swash expects a third number
2772 * typestr is the name of the swash's mapping, like 'ToLower'
2773 * On output: *min, *max, and *val are set to the values read from the line.
2774 * returns a pointer just beyond the line examined. If there was no
2775 * valid min number on the line, returns lend+1
2776 */
2777
2778STATIC U8*
2779S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2780 const bool wants_value, const U8* const typestr)
2781{
2782 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2783 STRLEN numlen; /* Length of the number */
02470786
KW
2784 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2785 | PERL_SCAN_DISALLOW_PREFIX
2786 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2787
2788 /* nl points to the next \n in the scan */
2789 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2790
95543e92
KW
2791 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
2792
319009ee
KW
2793 /* Get the first number on the line: the range minimum */
2794 numlen = lend - l;
2795 *min = grok_hex((char *)l, &numlen, &flags, NULL);
c88850db 2796 *max = *min; /* So can never return without setting max */
319009ee
KW
2797 if (numlen) /* If found a hex number, position past it */
2798 l += numlen;
2799 else if (nl) { /* Else, go handle next line, if any */
2800 return nl + 1; /* 1 is length of "\n" */
2801 }
2802 else { /* Else, no next line */
2803 return lend + 1; /* to LIST's end at which \n is not found */
2804 }
2805
2806 /* The max range value follows, separated by a BLANK */
2807 if (isBLANK(*l)) {
2808 ++l;
02470786
KW
2809 flags = PERL_SCAN_SILENT_ILLDIGIT
2810 | PERL_SCAN_DISALLOW_PREFIX
2811 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2812 numlen = lend - l;
2813 *max = grok_hex((char *)l, &numlen, &flags, NULL);
2814 if (numlen)
2815 l += numlen;
2816 else /* If no value here, it is a single element range */
2817 *max = *min;
2818
2819 /* Non-binary tables have a third entry: what the first element of the
24303724 2820 * range maps to. The map for those currently read here is in hex */
319009ee
KW
2821 if (wants_value) {
2822 if (isBLANK(*l)) {
2823 ++l;
f2a7d0fc
KW
2824 flags = PERL_SCAN_SILENT_ILLDIGIT
2825 | PERL_SCAN_DISALLOW_PREFIX
2826 | PERL_SCAN_SILENT_NON_PORTABLE;
2827 numlen = lend - l;
2828 *val = grok_hex((char *)l, &numlen, &flags, NULL);
2829 if (numlen)
2830 l += numlen;
2831 else
2832 *val = 0;
319009ee
KW
2833 }
2834 else {
2835 *val = 0;
2836 if (typeto) {
dcbac5bb 2837 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
2838 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
2839 typestr, l);
2840 }
2841 }
2842 }
2843 else
2844 *val = 0; /* bits == 1, then any val should be ignored */
2845 }
2846 else { /* Nothing following range min, should be single element with no
2847 mapping expected */
319009ee
KW
2848 if (wants_value) {
2849 *val = 0;
2850 if (typeto) {
dcbac5bb 2851 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
2852 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
2853 }
2854 }
2855 else
2856 *val = 0; /* bits == 1, then val should be ignored */
2857 }
2858
2859 /* Position to next line if any, or EOF */
2860 if (nl)
2861 l = nl + 1;
2862 else
2863 l = lend;
2864
2865 return l;
2866}
2867
979f2922
ST
2868/* Note:
2869 * Returns a swatch (a bit vector string) for a code point sequence
2870 * that starts from the value C<start> and comprises the number C<span>.
2871 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
2872 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
2873 */
2874STATIC SV*
b0e3252e 2875S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
2876{
2877 SV *swatch;
77f9f126 2878 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 2879 STRLEN lcur, xcur, scur;
ef8f7699 2880 HV *const hv = MUTABLE_HV(SvRV(swash));
5c9f4bd2 2881 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
36eb48b4 2882
88d45d28
KW
2883 SV** listsvp = NULL; /* The string containing the main body of the table */
2884 SV** extssvp = NULL;
2885 SV** invert_it_svp = NULL;
2886 U8* typestr = NULL;
786861f5
KW
2887 STRLEN bits;
2888 STRLEN octets; /* if bits == 1, then octets == 0 */
2889 UV none;
2890 UV end = start + span;
972dd592 2891
36eb48b4 2892 if (invlistsvp == NULL) {
786861f5
KW
2893 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
2894 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
2895 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
2896 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
2897 listsvp = hv_fetchs(hv, "LIST", FALSE);
2898 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
2899
2900 bits = SvUV(*bitssvp);
2901 none = SvUV(*nonesvp);
2902 typestr = (U8*)SvPV_nolen(*typesvp);
2903 }
36eb48b4
KW
2904 else {
2905 bits = 1;
2906 none = 0;
2907 }
786861f5 2908 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 2909
b0e3252e 2910 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 2911
979f2922 2912 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 2913 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 2914 (UV)bits);
979f2922
ST
2915 }
2916
84ea5ef6
KW
2917 /* If overflowed, use the max possible */
2918 if (end < start) {
2919 end = UV_MAX;
2920 span = end - start;
2921 }
2922
979f2922 2923 /* create and initialize $swatch */
979f2922 2924 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
2925 swatch = newSV(scur);
2926 SvPOK_on(swatch);
979f2922
ST
2927 s = (U8*)SvPVX(swatch);
2928 if (octets && none) {
0bd48802 2929 const U8* const e = s + scur;
979f2922
ST
2930 while (s < e) {
2931 if (bits == 8)
2932 *s++ = (U8)(none & 0xff);
2933 else if (bits == 16) {
2934 *s++ = (U8)((none >> 8) & 0xff);
2935 *s++ = (U8)( none & 0xff);
2936 }
2937 else if (bits == 32) {
2938 *s++ = (U8)((none >> 24) & 0xff);
2939 *s++ = (U8)((none >> 16) & 0xff);
2940 *s++ = (U8)((none >> 8) & 0xff);
2941 *s++ = (U8)( none & 0xff);
2942 }
2943 }
2944 *s = '\0';
2945 }
2946 else {
2947 (void)memzero((U8*)s, scur + 1);
2948 }
2949 SvCUR_set(swatch, scur);
2950 s = (U8*)SvPVX(swatch);
2951
36eb48b4
KW
2952 if (invlistsvp) { /* If has an inversion list set up use that */
2953 _invlist_populate_swatch(*invlistsvp, start, end, s);
2954 return swatch;
2955 }
2956
2957 /* read $swash->{LIST} */
979f2922
ST
2958 l = (U8*)SvPV(*listsvp, lcur);
2959 lend = l + lcur;
2960 while (l < lend) {
8ed25d53 2961 UV min, max, val, upper;
95543e92
KW
2962 l = swash_scan_list_line(l, lend, &min, &max, &val,
2963 cBOOL(octets), typestr);
319009ee 2964 if (l > lend) {
979f2922
ST
2965 break;
2966 }
2967
972dd592 2968 /* If looking for something beyond this range, go try the next one */
979f2922
ST
2969 if (max < start)
2970 continue;
2971
8ed25d53
KW
2972 /* <end> is generally 1 beyond where we want to set things, but at the
2973 * platform's infinity, where we can't go any higher, we want to
2974 * include the code point at <end> */
2975 upper = (max < end)
2976 ? max
2977 : (max != UV_MAX || end != UV_MAX)
2978 ? end - 1
2979 : end;
2980
979f2922 2981 if (octets) {
35da51f7 2982 UV key;
979f2922
ST
2983 if (min < start) {
2984 if (!none || val < none) {
2985 val += start - min;
2986 }
2987 min = start;
2988 }
8ed25d53 2989 for (key = min; key <= upper; key++) {
979f2922 2990 STRLEN offset;
979f2922
ST
2991 /* offset must be non-negative (start <= min <= key < end) */
2992 offset = octets * (key - start);
2993 if (bits == 8)
2994 s[offset] = (U8)(val & 0xff);
2995 else if (bits == 16) {
2996 s[offset ] = (U8)((val >> 8) & 0xff);
2997 s[offset + 1] = (U8)( val & 0xff);
2998 }
2999 else if (bits == 32) {
3000 s[offset ] = (U8)((val >> 24) & 0xff);
3001 s[offset + 1] = (U8)((val >> 16) & 0xff);
3002 s[offset + 2] = (U8)((val >> 8) & 0xff);
3003 s[offset + 3] = (U8)( val & 0xff);
3004 }
3005
3006 if (!none || val < none)
3007 ++val;
3008 }
3009 }
711a919c 3010 else { /* bits == 1, then val should be ignored */
35da51f7 3011 UV key;
979f2922
ST
3012 if (min < start)
3013 min = start;
6cb05c12 3014
8ed25d53 3015 for (key = min; key <= upper; key++) {
0bd48802 3016 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3017 s[offset >> 3] |= 1 << (offset & 7);
3018 }
3019 }
3020 } /* while */
979f2922 3021
9479a769 3022 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3023 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3024
3025 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3026 * be 0, and their inversion should also be 0, as we don't succeed any
3027 * Unicode property matches for non-Unicode code points */
3028 if (start <= PERL_UNICODE_MAX) {
3029
3030 /* The code below assumes that we never cross the
3031 * Unicode/above-Unicode boundary in a range, as otherwise we would
3032 * have to figure out where to stop flipping the bits. Since this
3033 * boundary is divisible by a large power of 2, and swatches comes
3034 * in small powers of 2, this should be a valid assumption */
3035 assert(start + span - 1 <= PERL_UNICODE_MAX);
3036
507a8485
KW
3037 send = s + scur;
3038 while (s < send) {
3039 *s = ~(*s);
3040 s++;
3041 }
0bda3001 3042 }
77f9f126
KW
3043 }
3044
d73c39c5
KW
3045 /* read $swash->{EXTRAS}
3046 * This code also copied to swash_to_invlist() below */
979f2922
ST
3047 x = (U8*)SvPV(*extssvp, xcur);
3048 xend = x + xcur;
3049 while (x < xend) {
3050 STRLEN namelen;
3051 U8 *namestr;
3052 SV** othersvp;
3053 HV* otherhv;
3054 STRLEN otherbits;
3055 SV **otherbitssvp, *other;
711a919c 3056 U8 *s, *o, *nl;
979f2922
ST
3057 STRLEN slen, olen;
3058
35da51f7 3059 const U8 opc = *x++;
979f2922
ST
3060 if (opc == '\n')
3061 continue;
3062
3063 nl = (U8*)memchr(x, '\n', xend - x);
3064
3065 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3066 if (nl) {
3067 x = nl + 1; /* 1 is length of "\n" */
3068 continue;
3069 }
3070 else {
3071 x = xend; /* to EXTRAS' end at which \n is not found */
3072 break;
3073 }
3074 }
3075
3076 namestr = x;
3077 if (nl) {
3078 namelen = nl - namestr;
3079 x = nl + 1;
3080 }
3081 else {
3082 namelen = xend - namestr;
3083 x = xend;
3084 }
3085
3086 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3087 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3088 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3089 otherbits = (STRLEN)SvUV(*otherbitssvp);
3090 if (bits < otherbits)
5637ef5b
NC
3091 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3092 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3093
3094 /* The "other" swatch must be destroyed after. */
b0e3252e 3095 other = swatch_get(*othersvp, start, span);
979f2922
ST
3096 o = (U8*)SvPV(other, olen);
3097
3098 if (!olen)
b0e3252e 3099 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3100
3101 s = (U8*)SvPV(swatch, slen);
3102 if (bits == 1 && otherbits == 1) {
3103 if (slen != olen)
5637ef5b
NC
3104 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3105 "mismatch, slen=%"UVuf", olen=%"UVuf,
3106 (UV)slen, (UV)olen);
979f2922
ST
3107
3108 switch (opc) {
3109 case '+':
3110 while (slen--)
3111 *s++ |= *o++;
3112 break;
3113 case '!':
3114 while (slen--)
3115 *s++ |= ~*o++;
3116 break;
3117 case '-':
3118 while (slen--)
3119 *s++ &= ~*o++;
3120 break;
3121 case '&':
3122 while (slen--)
3123 *s++ &= *o++;
3124 break;
3125 default:
3126 break;
3127 }
3128 }
711a919c 3129 else {
979f2922
ST
3130 STRLEN otheroctets = otherbits >> 3;
3131 STRLEN offset = 0;
35da51f7 3132 U8* const send = s + slen;
979f2922
ST
3133
3134 while (s < send) {
3135 UV otherval = 0;
3136
3137 if (otherbits == 1) {
3138 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3139 ++offset;
3140 }
3141 else {
3142 STRLEN vlen = otheroctets;
3143 otherval = *o++;
3144 while (--vlen) {
3145 otherval <<= 8;
3146 otherval |= *o++;
3147 }
3148 }
3149
711a919c 3150 if (opc == '+' && otherval)
6f207bd3 3151 NOOP; /* replace with otherval */
979f2922
ST
3152 else if (opc == '!' && !otherval)
3153 otherval = 1;
3154 else if (opc == '-' && otherval)
3155 otherval = 0;
3156 else if (opc == '&' && !otherval)
3157 otherval = 0;
3158 else {
711a919c 3159 s += octets; /* no replacement */
979f2922
ST
3160 continue;
3161 }
3162
3163 if (bits == 8)
3164 *s++ = (U8)( otherval & 0xff);
3165 else if (bits == 16) {
3166 *s++ = (U8)((otherval >> 8) & 0xff);
3167 *s++ = (U8)( otherval & 0xff);
3168 }
3169 else if (bits == 32) {
3170 *s++ = (U8)((otherval >> 24) & 0xff);
3171 *s++ = (U8)((otherval >> 16) & 0xff);
3172 *s++ = (U8)((otherval >> 8) & 0xff);
3173 *s++ = (U8)( otherval & 0xff);
3174 }
3175 }
3176 }
3177 sv_free(other); /* through with it! */
3178 } /* while */
3179 return swatch;
3180}
3181
064c021d 3182HV*
4c2e1131 3183Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3184{
3185
79a2a0e8 3186 /* Subject to change or removal. For use only in regcomp.c and regexec.c
5662e334
KW
3187 * Can't be used on a property that is subject to user override, as it
3188 * relies on the value of SPECIALS in the swash which would be set by
3189 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3190 * for overridden properties
064c021d
KW
3191 *
3192 * Returns a hash which is the inversion and closure of a swash mapping.
3193 * For example, consider the input lines:
3194 * 004B 006B
3195 * 004C 006C
3196 * 212A 006B
3197 *
3198 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3199 * 006C. The value for each key is an array. For 006C, the array would
17d5c697 3200 * have two elements, the utf8 for itself, and for 004C. For 006B, there
064c021d
KW
3201 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3202 *
538e84ed
KW
3203 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3204 * keys are only code points that are folded-to, so it isn't a full closure.
3205 *
064c021d
KW
3206 * Essentially, for any code point, it gives all the code points that map to
3207 * it, or the list of 'froms' for that point.
3208 *
5662e334
KW
3209 * Currently it ignores any additions or deletions from other swashes,
3210 * looking at just the main body of the swash, and if there are SPECIALS
3211 * in the swash, at that hash
3212 *
3213 * The specials hash can be extra code points, and most likely consists of
3214 * maps from single code points to multiple ones (each expressed as a string
3215 * of utf8 characters). This function currently returns only 1-1 mappings.
3216 * However consider this possible input in the specials hash:
3217 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3218 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3219 *
3220 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3221 * currently handle. But it also means that FB05 and FB06 are equivalent in
3222 * a 1-1 mapping which we should handle, and this relationship may not be in
3223 * the main table. Therefore this function examines all the multi-char
3224 * sequences and adds the 1-1 mappings that come out of that. */
064c021d
KW
3225
3226 U8 *l, *lend;
3227 STRLEN lcur;
3228 HV *const hv = MUTABLE_HV(SvRV(swash));
3229
923b6d4e
KW
3230 /* The string containing the main body of the table. This will have its
3231 * assertion fail if the swash has been converted to its inversion list */
064c021d
KW
3232 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3233
3234 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3235 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3236 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3237 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3238 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3239 const STRLEN bits = SvUV(*bitssvp);
3240 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3241 const UV none = SvUV(*nonesvp);
5662e334 3242 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
064c021d
KW
3243
3244 HV* ret = newHV();
3245
3246 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3247
3248 /* Must have at least 8 bits to get the mappings */
3249 if (bits != 8 && bits != 16 && bits != 32) {
3250 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3251 (UV)bits);
3252 }
3253
5662e334
KW
3254 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3255 mapping to more than one character */
3256
3257 /* Construct an inverse mapping hash for the specials */
3258 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3259 HV * specials_inverse = newHV();
3260 char *char_from; /* the lhs of the map */
3261 I32 from_len; /* its byte length */
3262 char *char_to; /* the rhs of the map */
3263 I32 to_len; /* its byte length */
3264 SV *sv_to; /* and in a sv */
3265 AV* from_list; /* list of things that map to each 'to' */
3266
3267 hv_iterinit(specials_hv);
3268
3269 /* The keys are the characters (in utf8) that map to the corresponding
3270 * utf8 string value. Iterate through the list creating the inverse
3271 * list. */
3272 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3273 SV** listp;
3274 if (! SvPOK(sv_to)) {
5637ef5b
NC
3275 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3276 "unexpectedly is not a string, flags=%lu",
3277 (unsigned long)SvFLAGS(sv_to));
5662e334 3278 }
4b88fb76 3279 /*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
3280
3281 /* Each key in the inverse list is a mapped-to value, and the key's
3282 * hash value is a list of the strings (each in utf8) that map to
3283 * it. Those strings are all one character long */
3284 if ((listp = hv_fetch(specials_inverse,
3285 SvPVX(sv_to),
3286 SvCUR(sv_to), 0)))
3287 {
3288 from_list = (AV*) *listp;
3289 }
3290 else { /* No entry yet for it: create one */
3291 from_list = newAV();
3292 if (! hv_store(specials_inverse,
3293 SvPVX(sv_to),
3294 SvCUR(sv_to),
3295 (SV*) from_list, 0))
3296 {
3297 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3298 }
3299 }
3300
3301 /* Here have the list associated with this 'to' (perhaps newly
3302 * created and empty). Just add to it. Note that we ASSUME that
3303 * the input is guaranteed to not have duplications, so we don't
3304 * check for that. Duplications just slow down execution time. */
3305 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3306 }
3307
3308 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3309 * it looking for cases like the FB05/FB06 examples above. There would
3310 * be an entry in the hash like
3311 * 'st' => [ FB05, FB06 ]
3312 * In this example we will create two lists that get stored in the
3313 * returned hash, 'ret':
3314 * FB05 => [ FB05, FB06 ]
3315 * FB06 => [ FB05, FB06 ]
3316 *
3317 * Note that there is nothing to do if the array only has one element.
3318 * (In the normal 1-1 case handled below, we don't have to worry about
3319 * two lists, as everything gets tied to the single list that is
3320 * generated for the single character 'to'. But here, we are omitting
3321 * that list, ('st' in the example), so must have multiple lists.) */
3322 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3323 &char_to, &to_len)))
3324 {
b9f2b683 3325 if (av_tindex(from_list) > 0) {
c70927a6 3326 SSize_t i;
5662e334
KW
3327
3328 /* We iterate over all combinations of i,j to place each code
3329 * point on each list */
b9f2b683 3330 for (i = 0; i <= av_tindex(from_list); i++) {
c70927a6 3331 SSize_t j;
5662e334
KW
3332 AV* i_list = newAV();
3333 SV** entryp = av_fetch(from_list, i, FALSE);
3334 if (entryp == NULL) {
3335 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3336 }
3337 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3338 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3339 }
3340 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3341 (SV*) i_list, FALSE))
3342 {
3343 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3344 }
3345
538e84ed 3346 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
b9f2b683 3347 for (j = 0; j <= av_tindex(from_list); j++) {
5662e334
KW
3348 entryp = av_fetch(from_list, j, FALSE);
3349 if (entryp == NULL) {
3350 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3351 }
3352
3353 /* When i==j this adds itself to the list */
4b88fb76
KW
3354 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3355 (U8*) SvPVX(*entryp),
3356 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3357 0)));
4637d003 3358 /*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
3359 }
3360 }
3361 }
3362 }
3363 SvREFCNT_dec(specials_inverse); /* done with it */
3364 } /* End of specials */
3365
064c021d
KW
3366 /* read $swash->{LIST} */
3367 l = (U8*)SvPV(*listsvp, lcur);
3368 lend = l + lcur;
3369
3370 /* Go through each input line */
3371 while (l < lend) {
3372 UV min, max, val;
3373 UV inverse;
95543e92
KW
3374 l = swash_scan_list_line(l, lend, &min, &max, &val,
3375 cBOOL(octets), typestr);
064c021d
KW
3376 if (l > lend) {
3377 break;
3378 }
3379
3380 /* Each element in the range is to be inverted */
3381 for (inverse = min; inverse <= max; inverse++) {
3382 AV* list;
064c021d
KW
3383 SV** listp;
3384 IV i;
3385 bool found_key = FALSE;
5662e334 3386 bool found_inverse = FALSE;
064c021d
KW
3387
3388 /* The key is the inverse mapping */
3389 char key[UTF8_MAXBYTES+1];
c80e42f3 3390 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
064c021d
KW
3391 STRLEN key_len = key_end - key;
3392
064c021d
KW
3393 /* Get the list for the map */
3394 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3395 list = (AV*) *listp;
3396 }
3397 else { /* No entry yet for it: create one */
3398 list = newAV();
3399 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3400 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3401 }
3402 }
3403
5662e334
KW
3404 /* Look through list to see if this inverse mapping already is
3405 * listed, or if there is a mapping to itself already */
b9f2b683 3406 for (i = 0; i <= av_tindex(list); i++) {
064c021d
KW
3407 SV** entryp = av_fetch(list, i, FALSE);
3408 SV* entry;
3409 if (entryp == NULL) {
3410 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3411 }
3412 entry = *entryp;
5662e334 3413 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, SvUV(entry)));*/
56ca34ca 3414 if (SvUV(entry) == val) {
064c021d 3415 found_key = TRUE;
5662e334
KW
3416 }
3417 if (SvUV(entry) == inverse) {
3418 found_inverse = TRUE;
3419 }
3420
3421 /* No need to continue searching if found everything we are
3422 * looking for */
3423 if (found_key && found_inverse) {
064c021d
KW
3424 break;
3425 }
3426 }
56ca34ca
KW
3427
3428 /* Make sure there is a mapping to itself on the list */
064c021d 3429 if (! found_key) {
d397ff6a 3430 av_push(list, newSVuv(val));
4637d003 3431 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
064c021d
KW
3432 }
3433
3434
3435 /* Simply add the value to the list */
5662e334
KW
3436 if (! found_inverse) {
3437 av_push(list, newSVuv(inverse));
4637d003 3438 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
5662e334 3439 }
064c021d 3440
b0e3252e 3441 /* swatch_get() increments the value of val for each element in the
064c021d
KW
3442 * range. That makes more compact tables possible. You can
3443 * express the capitalization, for example, of all consecutive
3444 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3445 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
bd3f2f94 3446 * and it's not documented; it appears to be used only in
b0e3252e 3447 * implementing tr//; I copied the semantics from swatch_get(), just
bd3f2f94 3448 * in case */
064c021d
KW
3449 if (!none || val < none) {
3450 ++val;
3451 }
3452 }
3453 }
3454
3455 return ret;
3456}
3457
a25abddc 3458SV*
d764b54e
KW
3459Perl__swash_to_invlist(pTHX_ SV* const swash)
3460{
3461
ed92f1b3
KW
3462 /* Subject to change or removal. For use only in one place in regcomp.c.
3463 * Ownership is given to one reference count in the returned SV* */
d764b54e
KW
3464
3465 U8 *l, *lend;
3466 char *loc;
3467 STRLEN lcur;
3468 HV *const hv = MUTABLE_HV(SvRV(swash));
3469 UV elements = 0; /* Number of elements in the inversion list */
b443038a 3470 U8 empty[] = "";
923b6d4e
KW
3471 SV** listsvp;
3472 SV** typesvp;
3473 SV** bitssvp;
3474 SV** extssvp;
3475 SV** invert_it_svp;
d764b54e 3476
923b6d4e
KW
3477 U8* typestr;
3478 STRLEN bits;
3479 STRLEN octets; /* if bits == 1, then octets == 0 */
d73c39c5
KW
3480 U8 *x, *xend;
3481 STRLEN xcur;
d764b54e 3482
a25abddc 3483 SV* invlist;
d764b54e 3484
b81740c0
KW
3485 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3486
923b6d4e
KW
3487 /* If not a hash, it must be the swash's inversion list instead */
3488 if (SvTYPE(hv) != SVt_PVHV) {
ed92f1b3 3489 return SvREFCNT_inc_simple_NN((SV*) hv);
923b6d4e
KW
3490 }
3491
3492 /* The string containing the main body of the table */
3493 listsvp = hv_fetchs(hv, "LIST", FALSE);
3494 typesvp = hv_fetchs(hv, "TYPE", FALSE);
3495 bitssvp = hv_fetchs(hv, "BITS", FALSE);
3496 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3497 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3498
3499 typestr = (U8*)SvPV_nolen(*typesvp);
3500 bits = SvUV(*bitssvp);
3501 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3502
d764b54e 3503 /* read $swash->{LIST} */
b443038a
KW
3504 if (SvPOK(*listsvp)) {
3505 l = (U8*)SvPV(*listsvp, lcur);
3506 }
3507 else {
3508 /* LIST legitimately doesn't contain a string during compilation phases
3509 * of Perl itself, before the Unicode tables are generated. In this
3510 * case, just fake things up by creating an empty list */
3511 l = empty;
3512 lcur = 0;
3513 }
d764b54e
KW
3514 loc = (char *) l;
3515 lend = l + lcur;
3516
31aa6e0b
KW
3517 if (*l == 'V') { /* Inversion list format */
3518 char *after_strtol = (char *) lend;
3519 UV element0;
3520 UV* other_elements_ptr;
3521
3522 /* The first number is a count of the rest */
3523 l++;
3524 elements = Strtoul((char *)l, &after_strtol, 10);
eb092534
KW
3525 if (elements == 0) {
3526 invlist = _new_invlist(0);
3527 }
3528 else {
31aa6e0b 3529 l = (U8 *) after_strtol;
1f9f7d4c
KW
3530
3531 /* Get the 0th element, which is needed to setup the inversion list */
3532 element0 = (UV) Strtoul((char *)l, &after_strtol, 10);
3533 l = (U8 *) after_strtol;
3534 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
3535 elements--;
3536
3537 /* Then just populate the rest of the input */
3538 while (elements-- > 0) {
3539 if (l > lend) {
3540 Perl_croak(aTHX_ "panic: Expecting %"UVuf" more elements than available", elements);
3541 }
3542 *other_elements_ptr++ = (UV) Strtoul((char *)l, &after_strtol, 10);
3543 l = (U8 *) after_strtol;
3544 }
eb092534 3545 }
31aa6e0b
KW
3546 }
3547 else {
3548
1784d2f9
KW
3549 /* Scan the input to count the number of lines to preallocate array
3550 * size based on worst possible case, which is each line in the input
3551 * creates 2 elements in the inversion list: 1) the beginning of a
3552 * range in the list; 2) the beginning of a range not in the list. */
3553 while ((loc = (strchr(loc, '\n'))) != NULL) {
3554 elements += 2;
3555 loc++;
3556 }
d764b54e 3557
1784d2f9
KW
3558 /* If the ending is somehow corrupt and isn't a new line, add another
3559 * element for the final range that isn't in the inversion list */
3560 if (! (*lend == '\n'
3561 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
3562 {
3563 elements++;
3564 }
d764b54e 3565
1784d2f9 3566 invlist = _new_invlist(elements);
d764b54e 3567
1784d2f9
KW
3568 /* Now go through the input again, adding each range to the list */
3569 while (l < lend) {
3570 UV start, end;
3571 UV val; /* Not used by this function */
d764b54e 3572
95543e92
KW
3573 l = swash_scan_list_line(l, lend, &start, &end, &val,
3574 cBOOL(octets), typestr);
d764b54e 3575
1784d2f9
KW
3576 if (l > lend) {
3577 break;
3578 }
3579
3580 invlist = _add_range_to_invlist(invlist, start, end);
3581 }
31aa6e0b 3582 }
d764b54e 3583
77f9f126
KW
3584 /* Invert if the data says it should be */
3585 if (invert_it_svp && SvUV(*invert_it_svp)) {
25151030 3586 _invlist_invert(invlist);
77f9f126
KW
3587 }
3588
b0e3252e 3589 /* This code is copied from swatch_get()
d73c39c5
KW
3590 * read $swash->{EXTRAS} */
3591 x = (U8*)SvPV(*extssvp, xcur);
3592 xend = x + xcur;
3593 while (x < xend) {
3594 STRLEN namelen;
3595 U8 *namestr;
3596 SV** othersvp;
3597 HV* otherhv;
3598 STRLEN otherbits;
3599 SV **otherbitssvp, *other;
3600 U8 *nl;
3601
3602 const U8 opc = *x++;
3603 if (opc == '\n')
3604 continue;
3605
3606 nl = (U8*)memchr(x, '\n', xend - x);
3607
3608 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3609 if (nl) {
3610 x = nl + 1; /* 1 is length of "\n" */
3611 continue;
3612 }
3613 else {
3614 x = xend; /* to EXTRAS' end at which \n is not found */
3615 break;
3616 }
3617 }
3618
3619 namestr = x;
3620 if (nl) {
3621 namelen = nl - namestr;
3622 x = nl + 1;
3623 }
3624 else {
3625 namelen = xend - namestr;
3626 x = xend;
3627 }
3628
3629 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3630 otherhv = MUTABLE_HV(SvRV(*othersvp));
3631 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3632 otherbits = (STRLEN)SvUV(*otherbitssvp);
3633
3634 if (bits != otherbits || bits != 1) {
5637ef5b
NC
3635 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
3636 "properties, bits=%"UVuf", otherbits=%"UVuf,
3637 (UV)bits, (UV)otherbits);
d73c39c5
KW
3638 }
3639
3640 /* The "other" swatch must be destroyed after. */
3641 other = _swash_to_invlist((SV *)*othersvp);
3642
b0e3252e 3643 /* End of code copied from swatch_get() */
d73c39c5
KW
3644 switch (opc) {
3645 case '+':
3646 _invlist_union(invlist, other, &invlist);
3647 break;
3648 case '!':
6c46377d 3649 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
d73c39c5
KW
3650 break;
3651 case '-':
3652 _invlist_subtract(invlist, other, &invlist);
3653 break;
3654 case '&':
3655 _invlist_intersection(invlist, other, &invlist);
3656 break;
3657 default:
3658 break;
3659 }
3660 sv_free(other); /* through with it! */
3661 }
3662
dbfdbd26 3663 SvREADONLY_on(invlist);
d764b54e
KW
3664 return invlist;
3665}
3666
3fdfee00
KW
3667SV*
3668Perl__get_swash_invlist(pTHX_ SV* const swash)
3669{
872dd7e0 3670 SV** ptr;
3fdfee00
KW
3671
3672 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
3673
87367d5f 3674 if (! SvROK(swash)) {
872dd7e0
KW
3675 return NULL;
3676 }
3677
87367d5f
KW
3678 /* If it really isn't a hash, it isn't really swash; must be an inversion
3679 * list */
3680 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
3681 return SvRV(swash);
3682 }
872dd7e0 3683
87367d5f 3684 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
3fdfee00
KW
3685 if (! ptr) {
3686 return NULL;
3687 }
3688
3689 return *ptr;
3690}
3691
0876b9a0 3692bool
5aaab254 3693Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
0876b9a0
KW
3694{
3695 /* May change: warns if surrogates, non-character code points, or
af2af982
KW
3696 * non-Unicode code points are in s which has length len bytes. Returns
3697 * TRUE if none found; FALSE otherwise. The only other validity check is
3698 * to make sure that this won't exceed the string's length */
0876b9a0
KW
3699
3700 const U8* const e = s + len;
3701 bool ok = TRUE;
3702
3703 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
3704
3705 while (s < e) {
3706 if (UTF8SKIP(s) > len) {
3707 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
3708 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");