This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
utf8.c: Add some function parameter assertions
[perl5.git] / utf8.c
CommitLineData
a0ed51b3
LW
1/* utf8.c
2 *
1129b882 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
b94e2f88 4 * by Larry Wall and others
a0ed51b3
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550
TC
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
15 *
cdad3b53 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
a0ed51b3
LW
17 *
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
4ac71550 20 * as is the custom in the West, if you wish to be answered?'
cdad3b53 21 * --Gandalf, addressing Théoden's door wardens
4ac71550
TC
22 *
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
24 *
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
4ac71550
TC
27 *
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
29 */
30
31#include "EXTERN.h"
864dbfa3 32#define PERL_IN_UTF8_C
a0ed51b3 33#include "perl.h"
81e983c1 34#include "inline_invlist.c"
b24b43f7 35#include "charclass_invlists.h"
a0ed51b3 36
27da23d5
JH
37static const char unees[] =
38 "Malformed UTF-8 character (unexpected end of string)";
901b21bf 39
48ef279e 40/*
ccfc67b7 41=head1 Unicode Support
7fefc6c1 42These are various utility functions for manipulating UTF8-encoded
72d33970 43strings. For the uninitiated, this is a method of representing arbitrary
61296642 44Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
45characters in the ASCII range are unmodified, and a zero byte never appears
46within non-zero characters.
166f8a29 47
eaf7a4d2
CS
48=cut
49*/
50
51/*
7bbfa158 52=for apidoc is_invariant_string
eaf7a4d2 53
b8ed1ca9
KW
54Returns true iff the first C<len> bytes of the string C<s> are the same
55regardless of the UTF-8 encoding of the string (or UTF-EBCDIC encoding on
56EBCDIC machines). That is, if they are UTF-8 invariant. On ASCII-ish
57machines, all the ASCII characters and only the ASCII characters fit this
7bbfa158 58definition. On EBCDIC machines, the ASCII-range characters are invariant, but
b8ed1ca9
KW
59so also are the C1 controls and C<\c?> (which isn't in the ASCII range on
60EBCDIC).
eaf7a4d2 61
75200dff
KW
62If C<len> is 0, it will be calculated using C<strlen(s)>, (which means if you
63use this option, that C<s> can't have embedded C<NUL> characters and has to
64have a terminating C<NUL> byte).
9f7e3d64 65
a1433954 66See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
eaf7a4d2
CS
67
68=cut
69*/
70
71bool
7bbfa158 72Perl_is_invariant_string(const U8 *s, STRLEN len)
eaf7a4d2
CS
73{
74 const U8* const send = s + (len ? len : strlen((const char *)s));
75 const U8* x = s;
76
7bbfa158 77 PERL_ARGS_ASSERT_IS_INVARIANT_STRING;
eaf7a4d2
CS
78
79 for (; x < send; ++x) {
80 if (!UTF8_IS_INVARIANT(*x))
81 break;
82 }
83
84 return x == send;
85}
86
87/*
378516de 88=for apidoc uvoffuni_to_utf8_flags
eebe1485 89
a27992cc 90THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
91Instead, B<Almost all code should use L</uvchr_to_utf8> or
92L</uvchr_to_utf8_flags>>.
a27992cc 93
de69f3af
KW
94This function is like them, but the input is a strict Unicode
95(as opposed to native) code point. Only in very rare circumstances should code
96not be using the native code point.
949cf498 97
de69f3af 98For details, see the description for L</uvchr_to_utf8_flags>>.
949cf498 99
eebe1485
SC
100=cut
101*/
102
dfe13c55 103U8 *
378516de 104Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 105{
378516de 106 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 107
d9432125
KW
108 if (UNI_IS_INVARIANT(uv)) {
109 *d++ = (U8) LATIN1_TO_NATIVE(uv);
110 return d;
111 }
112
9415f659
KW
113#ifdef EBCDIC
114 /* Not representable in UTF-EBCDIC */
115 flags |= UNICODE_DISALLOW_FE_FF;
116#endif
117
979f77b6
KW
118 /* The first problematic code point is the first surrogate */
119 if (uv >= UNICODE_SURROGATE_FIRST
0cfa64bf 120 && ckWARN3_d(WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
979f77b6 121 {
949cf498
KW
122 if (UNICODE_IS_SURROGATE(uv)) {
123 if (flags & UNICODE_WARN_SURROGATE) {
8457b38f 124 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
949cf498
KW
125 "UTF-16 surrogate U+%04"UVXf, uv);
126 }
127 if (flags & UNICODE_DISALLOW_SURROGATE) {
128 return NULL;
129 }
130 }
131 else if (UNICODE_IS_SUPER(uv)) {
132 if (flags & UNICODE_WARN_SUPER
133 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
134 {
8457b38f 135 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
949cf498
KW
136 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
137 }
138 if (flags & UNICODE_DISALLOW_SUPER
139 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
140 {
9415f659
KW
141#ifdef EBCDIC
142 Perl_die(aTHX_ "Can't represent character for Ox%"UVXf" on this platform", uv);
e5964223 143 NOT_REACHED;
9415f659 144#endif
949cf498
KW
145 return NULL;
146 }
147 }
148 else if (UNICODE_IS_NONCHAR(uv)) {
149 if (flags & UNICODE_WARN_NONCHAR) {
8457b38f 150 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
949cf498
KW
151 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
152 uv);
153 }
154 if (flags & UNICODE_DISALLOW_NONCHAR) {
155 return NULL;
156 }
157 }
507b9800 158 }
d9432125 159
2d331972 160#if defined(EBCDIC)
d9432125 161 {
5aaebcb3 162 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
163 U8 *p = d+len-1;
164 while (p > d) {
bc3632a8 165 *p-- = (U8) I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
166 uv >>= UTF_ACCUMULATION_SHIFT;
167 }
bc3632a8 168 *p = (U8) I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
169 return d+len;
170 }
171#else /* Non loop style */
a0ed51b3 172 if (uv < 0x800) {
eb160463
GS
173 *d++ = (U8)(( uv >> 6) | 0xc0);
174 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
175 return d;
176 }
177 if (uv < 0x10000) {
eb160463
GS
178 *d++ = (U8)(( uv >> 12) | 0xe0);
179 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
180 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
181 return d;
182 }
183 if (uv < 0x200000) {
eb160463
GS
184 *d++ = (U8)(( uv >> 18) | 0xf0);
185 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
186 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
187 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
188 return d;
189 }
190 if (uv < 0x4000000) {
eb160463
GS
191 *d++ = (U8)(( uv >> 24) | 0xf8);
192 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
193 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
194 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
195 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
196 return d;
197 }
198 if (uv < 0x80000000) {
eb160463
GS
199 *d++ = (U8)(( uv >> 30) | 0xfc);
200 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
201 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
202 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
203 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
204 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
205 return d;
206 }
6588300d 207#ifdef UTF8_QUAD_MAX
d7578b48 208 if (uv < UTF8_QUAD_MAX)
a0ed51b3
LW
209#endif
210 {
eb160463
GS
211 *d++ = 0xfe; /* Can't match U+FEFF! */
212 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
213 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
214 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
215 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
217 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
218 return d;
219 }
6588300d 220#ifdef UTF8_QUAD_MAX
a0ed51b3 221 {
eb160463
GS
222 *d++ = 0xff; /* Can't match U+FFFE! */
223 *d++ = 0x80; /* 6 Reserved bits */
224 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
225 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
226 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
227 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
228 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
229 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
230 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
231 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
232 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
233 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
234 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
235 return d;
236 }
237#endif
537124e4 238#endif /* Non loop style */
a0ed51b3 239}
646ca15d 240/*
07693fe6
KW
241=for apidoc uvchr_to_utf8
242
bcb1a2d4 243Adds the UTF-8 representation of the native code point C<uv> to the end
c749c9fd
KW
244of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
245C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
246the byte after the end of the new character. In other words,
07693fe6
KW
247
248 d = uvchr_to_utf8(d, uv);
249
250is the recommended wide native character-aware way of saying
251
252 *(d++) = uv;
253
de69f3af
KW
254This function accepts any UV as input. To forbid or warn on non-Unicode code
255points, or those that may be problematic, see L</uvchr_to_utf8_flags>.
256
07693fe6
KW
257=cut
258*/
259
de69f3af
KW
260/* This is also a macro */
261PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
262
07693fe6
KW
263U8 *
264Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
265{
de69f3af 266 return uvchr_to_utf8(d, uv);
07693fe6
KW
267}
268
de69f3af
KW
269/*
270=for apidoc uvchr_to_utf8_flags
271
272Adds the UTF-8 representation of the native code point C<uv> to the end
c749c9fd
KW
273of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
274C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
275the byte after the end of the new character. In other words,
de69f3af
KW
276
277 d = uvchr_to_utf8_flags(d, uv, flags);
278
279or, in most cases,
280
281 d = uvchr_to_utf8_flags(d, uv, 0);
282
283This is the Unicode-aware way of saying
284
285 *(d++) = uv;
286
287This function will convert to UTF-8 (and not warn) even code points that aren't
288legal Unicode or are problematic, unless C<flags> contains one or more of the
289following flags:
290
291If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
292the function will raise a warning, provided UTF8 warnings are enabled. If instead
293UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
294If both flags are set, the function will both warn and return NULL.
295
4c3cfd5d 296The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags
de69f3af 297affect how the function handles a Unicode non-character. And likewise, the
4c3cfd5d 298UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags affect the handling of
de69f3af
KW
299code points that are
300above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
301even less portable) can be warned and/or disallowed even if other above-Unicode
302code points are accepted, by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
303flags.
304
305And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
306above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
307DISALLOW flags.
308
309=cut
310*/
311
312/* This is also a macro */
313PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
314
07693fe6
KW
315U8 *
316Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
317{
de69f3af 318 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
319}
320
321/*
492a624f
KW
322=for apidoc is_utf8_char_buf
323
3cedd9d9 324This is identical to the macro L</isUTF8_CHAR>.
492a624f
KW
325
326=cut */
327
328STRLEN
329Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
330{
331
492a624f
KW
332 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
333
6302f837 334 return isUTF8_CHAR(buf, buf_end);
492a624f
KW
335}
336
337/*
87cea99e 338=for apidoc is_utf8_string
6662521e 339
a1433954 340Returns true if the first C<len> bytes of string C<s> form a valid
9f7e3d64 341UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
75200dff
KW
342using C<strlen(s)> (which means if you use this option, that C<s> can't have
343embedded C<NUL> characters and has to have a terminating C<NUL> byte). Note
344that all characters being ASCII constitute 'a valid UTF-8 string'.
6662521e 345
7bbfa158 346See also L</is_invariant_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
768c67ee 347
6662521e
GS
348=cut
349*/
350
8e84507e 351bool
668b6d8d 352Perl_is_utf8_string(const U8 *s, STRLEN len)
6662521e 353{
35da51f7 354 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 355 const U8* x = s;
067a85ef 356
7918f24d 357 PERL_ARGS_ASSERT_IS_UTF8_STRING;
1aa99e6b 358
6662521e 359 while (x < send) {
6302f837
KW
360 STRLEN len = isUTF8_CHAR(x, send);
361 if (UNLIKELY(! len)) {
362 return FALSE;
363 }
364 x += len;
6662521e 365 }
768c67ee 366
067a85ef 367 return TRUE;
6662521e
GS
368}
369
67e989fb 370/*
814fafa7
NC
371Implemented as a macro in utf8.h
372
87cea99e 373=for apidoc is_utf8_string_loc
814fafa7 374
a1433954
KW
375Like L</is_utf8_string> but stores the location of the failure (in the
376case of "utf8ness failure") or the location C<s>+C<len> (in the case of
814fafa7
NC
377"utf8ness success") in the C<ep>.
378
a1433954 379See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
814fafa7 380
87cea99e 381=for apidoc is_utf8_string_loclen
81cd54e3 382
a1433954
KW
383Like L</is_utf8_string>() but stores the location of the failure (in the
384case of "utf8ness failure") or the location C<s>+C<len> (in the case of
768c67ee
JH
385"utf8ness success") in the C<ep>, and the number of UTF-8
386encoded characters in the C<el>.
387
a1433954 388See also L</is_utf8_string_loc>() and L</is_utf8_string>().
81cd54e3
JH
389
390=cut
391*/
392
393bool
668b6d8d 394Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
81cd54e3 395{
35da51f7 396 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 397 const U8* x = s;
3ebfea28 398 STRLEN outlen = 0;
7918f24d
NC
399
400 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
81cd54e3 401
81cd54e3 402 while (x < send) {
6302f837
KW
403 STRLEN len = isUTF8_CHAR(x, send);
404 if (UNLIKELY(! len)) {
405 goto out;
406 }
407 x += len;
408 outlen++;
81cd54e3 409 }
768c67ee
JH
410
411 out:
3ebfea28
AL
412 if (el)
413 *el = outlen;
414
768c67ee
JH
415 if (ep)
416 *ep = x;
3ebfea28 417 return (x == send);
81cd54e3
JH
418}
419
420/*
768c67ee 421
de69f3af 422=for apidoc utf8n_to_uvchr
378516de
KW
423
424THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 425Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 426
9041c2e3 427Bottom level UTF-8 decode routine.
de69f3af 428Returns the native code point value of the first character in the string C<s>,
746afd53
KW
429which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
430C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
431the length, in bytes, of that character.
949cf498
KW
432
433The value of C<flags> determines the behavior when C<s> does not point to a
434well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
435zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
436next possible position in C<s> that could begin a non-malformed character.
437Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
438
439Various ALLOW flags can be set in C<flags> to allow (and not warn on)
440individual types of malformations, such as the sequence being overlong (that
441is, when there is a shorter sequence that can express the same code point;
442overlong sequences are expressly forbidden in the UTF-8 standard due to
443potential security issues). Another malformation example is the first byte of
444a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
445flags. For allowed 0 length strings, this function returns 0; for allowed
446overlong sequences, the computed code point is returned; for all other allowed
447malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
448determinable reasonable value.
949cf498
KW
449
450The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
451flags) malformation is found. If this flag is set, the routine assumes that
452the caller will raise a warning, and this function will silently just set
d088425d
KW
453C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
454
75200dff 455Note that this API requires disambiguation between successful decoding a C<NUL>
d088425d
KW
456character, and an error return (unless the UTF8_CHECK_ONLY flag is set), as
457in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
75200dff
KW
458first byte of C<s> is 0 as well. If so, the input was a C<NUL>; if not, the
459input had an error.
949cf498
KW
460
461Certain code points are considered problematic. These are Unicode surrogates,
746afd53 462Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 463By default these are considered regular code points, but certain situations
5eafe189 464warrant special handling for them. If C<flags> contains
949cf498
KW
465UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
466malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
467UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
468maximum) can be set to disallow these categories individually.
469
470The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
471UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
472for their respective categories, but otherwise the code points are considered
473valid (not malformations). To get a category to both be treated as a
474malformation and raise a warning, specify both the WARN and DISALLOW flags.
475(But note that warnings are not raised if lexically disabled nor if
476UTF8_CHECK_ONLY is also specified.)
477
478Very large code points (above 0x7FFF_FFFF) are considered more problematic than
479the others that are above the Unicode legal maximum. There are several
eb83ed87
KW
480reasons: they requre at least 32 bits to represent them on ASCII platforms, are
481not representable at all on EBCDIC platforms, and the original UTF-8
482specification never went above this number (the current 0x10FFFF limit was
483imposed later). (The smaller ones, those that fit into 32 bits, are
484representable by a UV on ASCII platforms, but not by an IV, which means that
485the number of operations that can be performed on them is quite restricted.)
486The UTF-8 encoding on ASCII platforms for these large code points begins with a
487byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
488be treated as malformations, while allowing smaller above-Unicode code points.
489(Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
72d33970
FC
490including these, as malformations.)
491Similarly, UTF8_WARN_FE_FF acts just like
eb83ed87 492the other WARN flags, but applies just to these code points.
949cf498
KW
493
494All other code points corresponding to Unicode characters, including private
495use and those yet to be assigned, are never considered malformed and never
496warn.
67e989fb 497
37607a96
PK
498=cut
499*/
67e989fb 500
a0ed51b3 501UV
de69f3af 502Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 503{
d4c19fe8 504 const U8 * const s0 = s;
eb83ed87 505 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 506 U8 * send;
eb83ed87
KW
507 UV uv = *s;
508 STRLEN expectlen;
949cf498 509 SV* sv = NULL;
eb83ed87
KW
510 UV outlier_ret = 0; /* return value when input is in error or problematic
511 */
512 UV pack_warn = 0; /* Save result of packWARN() for later */
513 bool unexpected_non_continuation = FALSE;
514 bool overflowed = FALSE;
2f8f112e 515 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 516
eb83ed87 517 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 518
de69f3af 519 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
a0dbb045 520
eb83ed87
KW
521 /* The order of malformation tests here is important. We should consume as
522 * few bytes as possible in order to not skip any valid character. This is
523 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
524 * http://unicode.org/reports/tr36 for more discussion as to why. For
525 * example, once we've done a UTF8SKIP, we can tell the expected number of
526 * bytes, and could fail right off the bat if the input parameters indicate
527 * that there are too few available. But it could be that just that first
528 * byte is garbled, and the intended character occupies fewer bytes. If we
529 * blindly assumed that the first byte is correct, and skipped based on
530 * that number, we could skip over a valid input character. So instead, we
531 * always examine the sequence byte-by-byte.
532 *
533 * We also should not consume too few bytes, otherwise someone could inject
534 * things. For example, an input could be deliberately designed to
535 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 536 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
537 * which is actually part of of the overflowing sequence), that could look
538 * legitimate to the caller, which could discard the initial partial
539 * sequence and process the rest, inappropriately */
540
541 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 542 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
543 if (retlen) {
544 *retlen = 0;
545 }
a0dbb045 546
eb83ed87
KW
547 if (flags & UTF8_ALLOW_EMPTY) {
548 return 0;
549 }
550 if (! (flags & UTF8_CHECK_ONLY)) {
551 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
552 }
0c443dc2
JH
553 goto malformed;
554 }
555
eb83ed87
KW
556 expectlen = UTF8SKIP(s);
557
558 /* A well-formed UTF-8 character, as the vast majority of calls to this
559 * function will be for, has this expected length. For efficiency, set
560 * things up here to return it. It will be overriden only in those rare
561 * cases where a malformation is found */
562 if (retlen) {
563 *retlen = expectlen;
564 }
565
566 /* An invariant is trivially well-formed */
1d72bdf6 567 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 568 return uv;
a0ed51b3 569 }
67e989fb 570
eb83ed87 571 /* A continuation character can't start a valid sequence */
b5b9af04 572 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
573 if (flags & UTF8_ALLOW_CONTINUATION) {
574 if (retlen) {
575 *retlen = 1;
576 }
577 return UNICODE_REPLACEMENT;
578 }
ba210ebe 579
eb83ed87
KW
580 if (! (flags & UTF8_CHECK_ONLY)) {
581 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
582 }
583 curlen = 1;
ba210ebe
JH
584 goto malformed;
585 }
9041c2e3 586
dcd27b3c
KW
587 /* Here is not a continuation byte, nor an invariant. The only thing left
588 * is a start byte (possibly for an overlong) */
589
1d72bdf6 590#ifdef EBCDIC
bc3632a8 591 uv = NATIVE_UTF8_TO_I8(uv);
1d72bdf6
NIS
592#endif
593
eb83ed87
KW
594 /* Remove the leading bits that indicate the number of bytes in the
595 * character's whole UTF-8 sequence, leaving just the bits that are part of
596 * the value */
597 uv &= UTF_START_MASK(expectlen);
ba210ebe 598
eb83ed87
KW
599 /* Now, loop through the remaining bytes in the character's sequence,
600 * accumulating each into the working value as we go. Be sure to not look
601 * past the end of the input string */
0b8d30e8
KW
602 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
603
eb83ed87 604 for (s = s0 + 1; s < send; s++) {
b5b9af04 605 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
606#ifndef EBCDIC /* Can't overflow in EBCDIC */
607 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
608
609 /* The original implementors viewed this malformation as more
610 * serious than the others (though I, khw, don't understand
611 * why, since other malformations also give very very wrong
612 * results), so there is no way to turn off checking for it.
613 * Set a flag, but keep going in the loop, so that we absorb
614 * the rest of the bytes that comprise the character. */
615 overflowed = TRUE;
616 overflow_byte = *s; /* Save for warning message's use */
617 }
618#endif
8850bf83 619 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
620 }
621 else {
622 /* Here, found a non-continuation before processing all expected
623 * bytes. This byte begins a new character, so quit, even if
624 * allowing this malformation. */
625 unexpected_non_continuation = TRUE;
626 break;
627 }
628 } /* End of loop through the character's bytes */
629
630 /* Save how many bytes were actually in the character */
631 curlen = s - s0;
632
633 /* The loop above finds two types of malformations: non-continuation and/or
634 * overflow. The non-continuation malformation is really a too-short
635 * malformation, as it means that the current character ended before it was
636 * expected to (being terminated prematurely by the beginning of the next
637 * character, whereas in the too-short malformation there just are too few
638 * bytes available to hold the character. In both cases, the check below
639 * that we have found the expected number of bytes would fail if executed.)
640 * Thus the non-continuation malformation is really unnecessary, being a
641 * subset of the too-short malformation. But there may be existing
642 * applications that are expecting the non-continuation type, so we retain
643 * it, and return it in preference to the too-short malformation. (If this
644 * code were being written from scratch, the two types might be collapsed
645 * into one.) I, khw, am also giving priority to returning the
646 * non-continuation and too-short malformations over overflow when multiple
647 * ones are present. I don't know of any real reason to prefer one over
648 * the other, except that it seems to me that multiple-byte errors trumps
649 * errors from a single byte */
b5b9af04 650 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
651 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
652 if (! (flags & UTF8_CHECK_ONLY)) {
653 if (curlen == 1) {
654 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
655 }
656 else {
657 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
658 }
659 }
eb83ed87
KW
660 goto malformed;
661 }
662 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
663
664 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
665 * as what the original expectations were. */
666 do_overlong_test = FALSE;
eb83ed87
KW
667 if (retlen) {
668 *retlen = curlen;
669 }
670 }
b5b9af04 671 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
672 if (! (flags & UTF8_ALLOW_SHORT)) {
673 if (! (flags & UTF8_CHECK_ONLY)) {
674 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 675 }
eb83ed87
KW
676 goto malformed;
677 }
678 uv = UNICODE_REPLACEMENT;
2f8f112e 679 do_overlong_test = FALSE;
eb83ed87
KW
680 if (retlen) {
681 *retlen = curlen;
682 }
683 }
684
ea5ced44 685#ifndef EBCDIC /* EBCDIC can't overflow */
b5b9af04 686 if (UNLIKELY(overflowed)) {
eb83ed87 687 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 688 goto malformed;
eb83ed87
KW
689 }
690#endif
691
2f8f112e 692 if (do_overlong_test
5aaebcb3 693 && expectlen > (STRLEN) OFFUNISKIP(uv)
2f8f112e
KW
694 && ! (flags & UTF8_ALLOW_LONG))
695 {
eb83ed87
KW
696 /* The overlong malformation has lower precedence than the others.
697 * Note that if this malformation is allowed, we return the actual
698 * value, instead of the replacement character. This is because this
699 * value is actually well-defined. */
700 if (! (flags & UTF8_CHECK_ONLY)) {
5aaebcb3 701 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
702 }
703 goto malformed;
704 }
705
1a89bb6c 706 /* Here, the input is considered to be well-formed, but it still could be a
eb83ed87
KW
707 * problematic code point that is not allowed by the input parameters. */
708 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
709 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
710 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
711 {
949cf498 712 if (UNICODE_IS_SURROGATE(uv)) {
ea5ced44
KW
713
714 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
715 * generation of the sv, since no warnings are raised under CHECK */
eb83ed87 716 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
54f4afef 717 && ckWARN_d(WARN_SURROGATE))
eb83ed87 718 {
111d382d 719 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
54f4afef 720 pack_warn = packWARN(WARN_SURROGATE);
949cf498
KW
721 }
722 if (flags & UTF8_DISALLOW_SURROGATE) {
723 goto disallowed;
724 }
725 }
949cf498 726 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87 727 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
ea5ced44 728 && ckWARN_d(WARN_NON_UNICODE))
eb83ed87 729 {
111d382d 730 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
54f4afef 731 pack_warn = packWARN(WARN_NON_UNICODE);
949cf498 732 }
ea5ced44
KW
733#ifndef EBCDIC /* EBCDIC always allows FE, FF */
734
735 /* The first byte being 0xFE or 0xFF is a subset of the SUPER code
736 * points. We test for these after the regular SUPER ones, and
737 * before possibly bailing out, so that the more dire warning
738 * overrides the regular one, if applicable */
739 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
740 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
741 {
742 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY))
743 == UTF8_WARN_FE_FF
744 && ckWARN_d(WARN_UTF8))
745 {
746 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%"UVXf" is not Unicode, and not portable", uv));
747 pack_warn = packWARN(WARN_UTF8);
748 }
749 if (flags & UTF8_DISALLOW_FE_FF) {
750 goto disallowed;
751 }
752 }
753#endif
949cf498
KW
754 if (flags & UTF8_DISALLOW_SUPER) {
755 goto disallowed;
756 }
757 }
4190d317
KW
758 else if (UNICODE_IS_NONCHAR(uv)) {
759 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
54f4afef 760 && ckWARN_d(WARN_NONCHAR))
4190d317
KW
761 {
762 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
54f4afef 763 pack_warn = packWARN(WARN_NONCHAR);
4190d317
KW
764 }
765 if (flags & UTF8_DISALLOW_NONCHAR) {
766 goto disallowed;
767 }
768 }
949cf498 769
eb83ed87 770 if (sv) {
de69f3af
KW
771 outlier_ret = uv; /* Note we don't bother to convert to native,
772 as all the outlier code points are the same
773 in both ASCII and EBCDIC */
eb83ed87
KW
774 goto do_warn;
775 }
776
949cf498
KW
777 /* Here, this is not considered a malformed character, so drop through
778 * to return it */
a0ed51b3 779 }
ba210ebe 780
de69f3af 781 return UNI_TO_NATIVE(uv);
ba210ebe 782
eb83ed87
KW
783 /* There are three cases which get to beyond this point. In all 3 cases:
784 * <sv> if not null points to a string to print as a warning.
785 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
786 * set.
787 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
788 * This is done by initializing it to 0, and changing it only
789 * for case 1).
790 * The 3 cases are:
791 * 1) The input is valid but problematic, and to be warned about. The
792 * return value is the resultant code point; <*retlen> is set to
793 * <curlen>, the number of bytes that comprise the code point.
794 * <pack_warn> contains the result of packWARN() for the warning
795 * types. The entry point for this case is the label <do_warn>;
796 * 2) The input is a valid code point but disallowed by the parameters to
797 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
798 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
799 * comprise the code point. <pack_warn> contains the result of
800 * packWARN() for the warning types. The entry point for this case is
801 * the label <disallowed>.
802 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
803 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
804 * bytes that comprise the malformation. All such malformations are
805 * assumed to be warning type <utf8>. The entry point for this case
806 * is the label <malformed>.
807 */
949cf498 808
ba210ebe
JH
809malformed:
810
eb83ed87
KW
811 if (sv && ckWARN_d(WARN_UTF8)) {
812 pack_warn = packWARN(WARN_UTF8);
813 }
814
815disallowed:
816
fcc8fcf6 817 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 818 if (retlen)
10edeb5d 819 *retlen = ((STRLEN) -1);
ba210ebe
JH
820 return 0;
821 }
822
eb83ed87 823do_warn:
5b311467 824
eb83ed87
KW
825 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
826 if warnings are to be raised. */
f555bc63 827 const char * const string = SvPVX_const(sv);
a0dbb045 828
f555bc63
KW
829 if (PL_op)
830 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
831 else
832 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
833 }
834
eb83ed87
KW
835 if (retlen) {
836 *retlen = curlen;
837 }
ba210ebe 838
eb83ed87 839 return outlier_ret;
a0ed51b3
LW
840}
841
8e84507e 842/*
ec5f19d0
KW
843=for apidoc utf8_to_uvchr_buf
844
845Returns the native code point of the first character in the string C<s> which
846is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 847C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 848
524080c4
KW
849If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
850enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
173db420
KW
851NULL) to -1. If those warnings are off, the computed value, if well-defined
852(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
853C<*retlen> is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is
854the next possible position in C<s> that could begin a non-malformed character.
de69f3af 855See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 856returned.
ec5f19d0
KW
857
858=cut
859*/
860
861
862UV
863Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
864{
ec5f19d0
KW
865 assert(s < send);
866
867 return utf8n_to_uvchr(s, send - s, retlen,
868 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
869}
870
27d6c58a 871/* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
3986bb7c 872 * there are no malformations in the input UTF-8 string C<s>. surrogates,
57b0056d 873 * non-character code points, and non-Unicode code points are allowed. */
27d6c58a
KW
874
875UV
876Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
877{
010ab96b
KW
878 UV expectlen = UTF8SKIP(s);
879 const U8* send = s + expectlen;
9ff2f0f7 880 UV uv = *s;
3986bb7c 881
27d6c58a 882 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
81611534 883 PERL_UNUSED_CONTEXT;
27d6c58a 884
010ab96b
KW
885 if (retlen) {
886 *retlen = expectlen;
887 }
888
889 /* An invariant is trivially returned */
890 if (expectlen == 1) {
9ff2f0f7 891 return uv;
010ab96b
KW
892 }
893
9ff2f0f7
KW
894#ifdef EBCDIC
895 uv = NATIVE_UTF8_TO_I8(uv);
896#endif
897
010ab96b
KW
898 /* Remove the leading bits that indicate the number of bytes, leaving just
899 * the bits that are part of the value */
900 uv &= UTF_START_MASK(expectlen);
901
902 /* Now, loop through the remaining bytes, accumulating each into the
903 * working total as we go. (I khw tried unrolling the loop for up to 4
904 * bytes, but there was no performance improvement) */
905 for (++s; s < send; s++) {
906 uv = UTF8_ACCUMULATE(uv, *s);
907 }
908
3986bb7c 909 return UNI_TO_NATIVE(uv);
010ab96b 910
27d6c58a
KW
911}
912
ec5f19d0 913/*
ec5f19d0
KW
914=for apidoc utf8_to_uvuni_buf
915
de69f3af
KW
916Only in very rare circumstances should code need to be dealing in Unicode
917(as opposed to native) code points. In those few cases, use
918C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
919
920Returns the Unicode (not-native) code point of the first character in the
921string C<s> which
ec5f19d0
KW
922is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
923C<retlen> will be set to the length, in bytes, of that character.
924
524080c4
KW
925If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
926enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
927NULL) to -1. If those warnings are off, the computed value if well-defined (or
928the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
929is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
930next possible position in C<s> that could begin a non-malformed character.
de69f3af 931See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
932
933=cut
934*/
935
936UV
937Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
938{
939 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
940
941 assert(send > s);
942
943 /* Call the low level routine asking for checks */
de69f3af
KW
944 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
945 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
ec5f19d0
KW
946}
947
b76347f2 948/*
87cea99e 949=for apidoc utf8_length
b76347f2
JH
950
951Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
952Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
953up past C<e>, croaks.
b76347f2
JH
954
955=cut
956*/
957
958STRLEN
35a4481c 959Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
960{
961 STRLEN len = 0;
962
7918f24d
NC
963 PERL_ARGS_ASSERT_UTF8_LENGTH;
964
8850bf83
JH
965 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
966 * the bitops (especially ~) can create illegal UTF-8.
967 * In other words: in Perl UTF-8 is not just for Unicode. */
968
a3b680e6
AL
969 if (e < s)
970 goto warn_and_return;
b76347f2 971 while (s < e) {
4cbf4130 972 s += UTF8SKIP(s);
8e91ec7f
AV
973 len++;
974 }
975
976 if (e != s) {
977 len--;
978 warn_and_return:
9b387841
NC
979 if (PL_op)
980 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
981 "%s in %s", unees, OP_DESC(PL_op));
982 else
61a12c31 983 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
984 }
985
986 return len;
987}
988
b06226ff 989/*
87cea99e 990=for apidoc utf8_distance
b06226ff 991
1e54db1a 992Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
b06226ff
JH
993and C<b>.
994
995WARNING: use only if you *know* that the pointers point inside the
996same UTF-8 buffer.
997
37607a96
PK
998=cut
999*/
a0ed51b3 1000
02eb7b47 1001IV
35a4481c 1002Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
a0ed51b3 1003{
7918f24d
NC
1004 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1005
bf1665bc 1006 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
a0ed51b3
LW
1007}
1008
b06226ff 1009/*
87cea99e 1010=for apidoc utf8_hop
b06226ff 1011
8850bf83
JH
1012Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1013forward or backward.
b06226ff
JH
1014
1015WARNING: do not use the following unless you *know* C<off> is within
8850bf83
JH
1016the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1017on the first byte of character or just after the last byte of a character.
b06226ff 1018
37607a96
PK
1019=cut
1020*/
a0ed51b3
LW
1021
1022U8 *
ddeaf645 1023Perl_utf8_hop(const U8 *s, I32 off)
a0ed51b3 1024{
7918f24d
NC
1025 PERL_ARGS_ASSERT_UTF8_HOP;
1026
8850bf83
JH
1027 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1028 * the bitops (especially ~) can create illegal UTF-8.
1029 * In other words: in Perl UTF-8 is not just for Unicode. */
1030
a0ed51b3
LW
1031 if (off >= 0) {
1032 while (off--)
1033 s += UTF8SKIP(s);
1034 }
1035 else {
1036 while (off++) {
1037 s--;
8850bf83
JH
1038 while (UTF8_IS_CONTINUATION(*s))
1039 s--;
a0ed51b3
LW
1040 }
1041 }
4373e329 1042 return (U8 *)s;
a0ed51b3
LW
1043}
1044
6940069f 1045/*
fed3ba5d
NC
1046=for apidoc bytes_cmp_utf8
1047
a1433954 1048Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1049sequence of characters (stored as UTF-8)
1050in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1051equal, -1 or -2 if the first string is less than the second string, +1 or +2
1052if the first string is greater than the second string.
1053
1054-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1055longer string. -2 or +2 is returned if
1056there was a difference between characters
fed3ba5d
NC
1057within the strings.
1058
1059=cut
1060*/
1061
1062int
1063Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1064{
1065 const U8 *const bend = b + blen;
1066 const U8 *const uend = u + ulen;
1067
1068 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1069
1070 while (b < bend && u < uend) {
1071 U8 c = *u++;
1072 if (!UTF8_IS_INVARIANT(c)) {
1073 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1074 if (u < uend) {
1075 U8 c1 = *u++;
1076 if (UTF8_IS_CONTINUATION(c1)) {
94bb8c36 1077 c = TWO_BYTE_UTF8_TO_NATIVE(c, c1);
fed3ba5d
NC
1078 } else {
1079 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1080 "Malformed UTF-8 character "
1081 "(unexpected non-continuation byte 0x%02x"
1082 ", immediately after start byte 0x%02x)"
1083 /* Dear diag.t, it's in the pod. */
1084 "%s%s", c1, c,
1085 PL_op ? " in " : "",
1086 PL_op ? OP_DESC(PL_op) : "");
1087 return -2;
1088 }
1089 } else {
1090 if (PL_op)
1091 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1092 "%s in %s", unees, OP_DESC(PL_op));
1093 else
61a12c31 1094 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1095 return -2; /* Really want to return undef :-) */
1096 }
1097 } else {
1098 return -2;
1099 }
1100 }
1101 if (*b != c) {
1102 return *b < c ? -2 : +2;
1103 }
1104 ++b;
1105 }
1106
1107 if (b == bend && u == uend)
1108 return 0;
1109
1110 return b < bend ? +1 : -1;
1111}
1112
1113/*
87cea99e 1114=for apidoc utf8_to_bytes
6940069f 1115
2bbc8d55 1116Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1117Unlike L</bytes_to_utf8>, this over-writes the original string, and
1118updates C<len> to contain the new length.
67e989fb 1119Returns zero on failure, setting C<len> to -1.
6940069f 1120
a1433954 1121If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1122
6940069f
GS
1123=cut
1124*/
1125
1126U8 *
37607a96 1127Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1128{
d4c19fe8
AL
1129 U8 * const save = s;
1130 U8 * const send = s + *len;
6940069f 1131 U8 *d;
246fae53 1132
7918f24d 1133 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 1134 PERL_UNUSED_CONTEXT;
7918f24d 1135
1e54db1a 1136 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1137 while (s < send) {
d59937ca
KW
1138 if (! UTF8_IS_INVARIANT(*s)) {
1139 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1140 *len = ((STRLEN) -1);
1141 return 0;
1142 }
1143 s++;
dcad2880 1144 }
d59937ca 1145 s++;
246fae53 1146 }
dcad2880
JH
1147
1148 d = s = save;
6940069f 1149 while (s < send) {
80e0b38f
KW
1150 U8 c = *s++;
1151 if (! UTF8_IS_INVARIANT(c)) {
1152 /* Then it is two-byte encoded */
1153 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1154 s++;
1155 }
1156 *d++ = c;
6940069f
GS
1157 }
1158 *d = '\0';
246fae53 1159 *len = d - save;
6940069f
GS
1160 return save;
1161}
1162
1163/*
87cea99e 1164=for apidoc bytes_from_utf8
f9a63242 1165
2bbc8d55 1166Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1167Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1168the newly-created string, and updates C<len> to contain the new
1169length. Returns the original string if no conversion occurs, C<len>
72d33970 1170is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55
SP
11710 if C<s> is converted or consisted entirely of characters that are invariant
1172in utf8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1173
37607a96
PK
1174=cut
1175*/
f9a63242
JH
1176
1177U8 *
e1ec3a88 1178Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1179{
f9a63242 1180 U8 *d;
e1ec3a88
AL
1181 const U8 *start = s;
1182 const U8 *send;
f9a63242
JH
1183 I32 count = 0;
1184
7918f24d 1185 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 1186 PERL_UNUSED_CONTEXT;
f9a63242 1187 if (!*is_utf8)
73d840c0 1188 return (U8 *)start;
f9a63242 1189
1e54db1a 1190 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1191 for (send = s + *len; s < send;) {
d59937ca
KW
1192 if (! UTF8_IS_INVARIANT(*s)) {
1193 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1194 return (U8 *)start;
d59937ca
KW
1195 }
1196 count++;
1197 s++;
db42d148 1198 }
d59937ca 1199 s++;
f9a63242
JH
1200 }
1201
35da51f7 1202 *is_utf8 = FALSE;
f9a63242 1203
212542aa 1204 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1205 s = start; start = d;
f9a63242
JH
1206 while (s < send) {
1207 U8 c = *s++;
1a91c45d 1208 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1209 /* Then it is two-byte encoded */
1a91c45d
KW
1210 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1211 s++;
c4d5f83a
NIS
1212 }
1213 *d++ = c;
f9a63242
JH
1214 }
1215 *d = '\0';
1216 *len = d - start;
73d840c0 1217 return (U8 *)start;
f9a63242
JH
1218}
1219
1220/*
87cea99e 1221=for apidoc bytes_to_utf8
6940069f 1222
ff97e5cf
KW
1223Converts a string C<s> of length C<len> bytes from the native encoding into
1224UTF-8.
6662521e 1225Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1226reflect the new length in bytes.
6940069f 1227
75200dff 1228A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
1229
1230If you want to convert to UTF-8 from encodings other than
1231the native (Latin1 or EBCDIC),
a1433954 1232see L</sv_recode_to_utf8>().
c9ada85f 1233
497711e7 1234=cut
6940069f
GS
1235*/
1236
c682ebef
FC
1237/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1238 likewise need duplication. */
1239
6940069f 1240U8*
35a4481c 1241Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1242{
35a4481c 1243 const U8 * const send = s + (*len);
6940069f
GS
1244 U8 *d;
1245 U8 *dst;
7918f24d
NC
1246
1247 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1248 PERL_UNUSED_CONTEXT;
6940069f 1249
212542aa 1250 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1251 dst = d;
1252
1253 while (s < send) {
55d09dc8
KW
1254 append_utf8_from_native_byte(*s, &d);
1255 s++;
6940069f
GS
1256 }
1257 *d = '\0';
6662521e 1258 *len = d-dst;
6940069f
GS
1259 return dst;
1260}
1261
a0ed51b3 1262/*
dea0fc0b 1263 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1264 *
1265 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1266 * We optimize for native, for obvious reasons. */
1267
1268U8*
dea0fc0b 1269Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1270{
dea0fc0b
JH
1271 U8* pend;
1272 U8* dstart = d;
1273
7918f24d
NC
1274 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1275
dea0fc0b 1276 if (bytelen & 1)
f5992bc4 1277 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1278
1279 pend = p + bytelen;
1280
a0ed51b3 1281 while (p < pend) {
dea0fc0b
JH
1282 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1283 p += 2;
56d37426
KW
1284 if (UNI_IS_INVARIANT(uv)) {
1285 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
1286 continue;
1287 }
56d37426
KW
1288 if (uv <= MAX_UTF8_TWO_BYTE) {
1289 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1290 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
1291 continue;
1292 }
46956fad
KW
1293#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1294#define LAST_HIGH_SURROGATE 0xDBFF
1295#define FIRST_LOW_SURROGATE 0xDC00
1296#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
1297 if (uv >= FIRST_HIGH_SURROGATE && uv <= LAST_HIGH_SURROGATE) {
01ea242b 1298 if (p >= pend) {
dea0fc0b 1299 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
01ea242b
NC
1300 } else {
1301 UV low = (p[0] << 8) + p[1];
1302 p += 2;
46956fad 1303 if (low < FIRST_LOW_SURROGATE || low > LAST_LOW_SURROGATE)
01ea242b 1304 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
46956fad
KW
1305 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1306 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 1307 }
46956fad 1308 } else if (uv >= FIRST_LOW_SURROGATE && uv <= LAST_LOW_SURROGATE) {
dbde1951 1309 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
a0ed51b3 1310 }
56d37426
KW
1311#ifdef EBCDIC
1312 d = uvoffuni_to_utf8_flags(d, uv, 0);
1313#else
a0ed51b3 1314 if (uv < 0x10000) {
eb160463
GS
1315 *d++ = (U8)(( uv >> 12) | 0xe0);
1316 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1317 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1318 continue;
1319 }
1320 else {
eb160463
GS
1321 *d++ = (U8)(( uv >> 18) | 0xf0);
1322 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1323 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1324 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1325 continue;
1326 }
56d37426 1327#endif
a0ed51b3 1328 }
dea0fc0b 1329 *newlen = d - dstart;
a0ed51b3
LW
1330 return d;
1331}
1332
1333/* Note: this one is slightly destructive of the source. */
1334
1335U8*
dea0fc0b 1336Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1337{
1338 U8* s = (U8*)p;
d4c19fe8 1339 U8* const send = s + bytelen;
7918f24d
NC
1340
1341 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1342
e0ea5e2d
NC
1343 if (bytelen & 1)
1344 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1345 (UV)bytelen);
1346
a0ed51b3 1347 while (s < send) {
d4c19fe8 1348 const U8 tmp = s[0];
a0ed51b3
LW
1349 s[0] = s[1];
1350 s[1] = tmp;
1351 s += 2;
1352 }
dea0fc0b 1353 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1354}
1355
922e8cb4
KW
1356bool
1357Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1358{
1359 U8 tmpbuf[UTF8_MAXBYTES+1];
1360 uvchr_to_utf8(tmpbuf, c);
1361 return _is_utf8_FOO(classnum, tmpbuf);
1362}
1363
f9ae8fb6
JD
1364/* Internal function so we can deprecate the external one, and call
1365 this one from other deprecated functions in this file */
1366
f2645549
KW
1367bool
1368Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 1369{
f2645549 1370 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
1371
1372 if (*p == '_')
1373 return TRUE;
f25ce844 1374 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
1375}
1376
5092f92a 1377bool
eba68aa0
KW
1378Perl__is_uni_perl_idcont(pTHX_ UV c)
1379{
1380 U8 tmpbuf[UTF8_MAXBYTES+1];
1381 uvchr_to_utf8(tmpbuf, c);
1382 return _is_utf8_perl_idcont(tmpbuf);
1383}
1384
1385bool
f91dcd13
KW
1386Perl__is_uni_perl_idstart(pTHX_ UV c)
1387{
1388 U8 tmpbuf[UTF8_MAXBYTES+1];
1389 uvchr_to_utf8(tmpbuf, c);
1390 return _is_utf8_perl_idstart(tmpbuf);
1391}
1392
3a4c58c9
KW
1393UV
1394Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1395{
1396 /* We have the latin1-range values compiled into the core, so just use
1397 * those, converting the result to utf8. The only difference between upper
1398 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1399 * either "SS" or "Ss". Which one to use is passed into the routine in
1400 * 'S_or_s' to avoid a test */
1401
1402 UV converted = toUPPER_LATIN1_MOD(c);
1403
1404 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1405
1406 assert(S_or_s == 'S' || S_or_s == 's');
1407
6f2d5cbc 1408 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 1409 characters in this range */
3a4c58c9
KW
1410 *p = (U8) converted;
1411 *lenp = 1;
1412 return converted;
1413 }
1414
1415 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1416 * which it maps to one of them, so as to only have to have one check for
1417 * it in the main case */
1418 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1419 switch (c) {
1420 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1421 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1422 break;
1423 case MICRO_SIGN:
1424 converted = GREEK_CAPITAL_LETTER_MU;
1425 break;
1426 case LATIN_SMALL_LETTER_SHARP_S:
1427 *(p)++ = 'S';
1428 *p = S_or_s;
1429 *lenp = 2;
1430 return 'S';
1431 default:
1432 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 1433 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
1434 }
1435 }
1436
1437 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1438 *p = UTF8_TWO_BYTE_LO(converted);
1439 *lenp = 2;
1440
1441 return converted;
1442}
1443
50bda2c3
KW
1444/* Call the function to convert a UTF-8 encoded character to the specified case.
1445 * Note that there may be more than one character in the result.
1446 * INP is a pointer to the first byte of the input character
1447 * OUTP will be set to the first byte of the string of changed characters. It
1448 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1449 * LENP will be set to the length in bytes of the string of changed characters
1450 *
1451 * The functions return the ordinal of the first character in the string of OUTP */
4a8240a3
KW
1452#define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "")
1453#define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "")
1454#define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "")
50bda2c3
KW
1455
1456/* This additionally has the input parameter SPECIALS, which if non-zero will
1457 * cause this to use the SPECIALS hash for folding (meaning get full case
1458 * folding); otherwise, when zero, this implies a simple case fold */
4a8240a3 1459#define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "" : NULL)
c3fd2246 1460
84afefe6
JH
1461UV
1462Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1463{
a1433954
KW
1464 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1465 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1466 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1467 * the changed version may be longer than the original character.
1468 *
1469 * The ordinal of the first character of the changed version is returned
1470 * (but note, as explained above, that there may be more.) */
1471
7918f24d
NC
1472 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1473
3a4c58c9
KW
1474 if (c < 256) {
1475 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1476 }
1477
0ebc6274 1478 uvchr_to_utf8(p, c);
3a4c58c9 1479 return CALL_UPPER_CASE(p, p, lenp);
a0ed51b3
LW
1480}
1481
84afefe6
JH
1482UV
1483Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1484{
7918f24d
NC
1485 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1486
3a4c58c9
KW
1487 if (c < 256) {
1488 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1489 }
1490
0ebc6274 1491 uvchr_to_utf8(p, c);
3a4c58c9 1492 return CALL_TITLE_CASE(p, p, lenp);
a0ed51b3
LW
1493}
1494
afc16117 1495STATIC U8
81611534 1496S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp)
afc16117
KW
1497{
1498 /* We have the latin1-range values compiled into the core, so just use
1499 * those, converting the result to utf8. Since the result is always just
a1433954 1500 * one character, we allow <p> to be NULL */
afc16117
KW
1501
1502 U8 converted = toLOWER_LATIN1(c);
1503
1504 if (p != NULL) {
6f2d5cbc 1505 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
1506 *p = converted;
1507 *lenp = 1;
1508 }
1509 else {
430c9760
KW
1510 /* Result is known to always be < 256, so can use the EIGHT_BIT
1511 * macros */
1512 *p = UTF8_EIGHT_BIT_HI(converted);
1513 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
1514 *lenp = 2;
1515 }
1516 }
1517 return converted;
1518}
1519
84afefe6
JH
1520UV
1521Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1522{
7918f24d
NC
1523 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1524
afc16117
KW
1525 if (c < 256) {
1526 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1527 }
1528
afc16117 1529 uvchr_to_utf8(p, c);
968c5e6a 1530 return CALL_LOWER_CASE(p, p, lenp);
a0ed51b3
LW
1531}
1532
84afefe6 1533UV
51910141 1534Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 1535{
51910141 1536 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 1537 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 1538 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
1539 *
1540 * Not to be used for locale folds
51910141 1541 */
f673fad4 1542
a1dde8de
KW
1543 UV converted;
1544
1545 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 1546 PERL_UNUSED_CONTEXT;
a1dde8de 1547
1ca267a5
KW
1548 assert (! (flags & FOLD_FLAGS_LOCALE));
1549
a1dde8de
KW
1550 if (c == MICRO_SIGN) {
1551 converted = GREEK_SMALL_LETTER_MU;
1552 }
51910141 1553 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1ca267a5
KW
1554
1555 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1556 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1557 * under those circumstances. */
1558 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1559 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1560 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1561 p, *lenp, U8);
1562 return LATIN_SMALL_LETTER_LONG_S;
1563 }
1564 else {
4f489194
KW
1565 *(p)++ = 's';
1566 *p = 's';
1567 *lenp = 2;
1568 return 's';
1ca267a5 1569 }
a1dde8de
KW
1570 }
1571 else { /* In this range the fold of all other characters is their lower
1572 case */
1573 converted = toLOWER_LATIN1(c);
1574 }
1575
6f2d5cbc 1576 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
1577 *p = (U8) converted;
1578 *lenp = 1;
1579 }
1580 else {
1581 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1582 *p = UTF8_TWO_BYTE_LO(converted);
1583 *lenp = 2;
1584 }
1585
1586 return converted;
1587}
1588
1589UV
31f05a37 1590Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 1591{
4b593389 1592
a0270393
KW
1593 /* Not currently externally documented, and subject to change
1594 * <flags> bits meanings:
1595 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
1596 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1597 * locale are to be used.
a0270393
KW
1598 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1599 */
4b593389 1600
36bb2ab6 1601 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1602
780fcc9f
KW
1603 if (flags & FOLD_FLAGS_LOCALE) {
1604 /* Treat a UTF-8 locale as not being in locale at all */
1605 if (IN_UTF8_CTYPE_LOCALE) {
1606 flags &= ~FOLD_FLAGS_LOCALE;
1607 }
1608 else {
1609 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
1610 }
31f05a37
KW
1611 }
1612
a1dde8de 1613 if (c < 256) {
a0270393 1614 UV result = _to_fold_latin1((U8) c, p, lenp,
31f05a37 1615 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a0270393
KW
1616 /* It is illegal for the fold to cross the 255/256 boundary under
1617 * locale; in this case return the original */
1618 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1619 ? c
1620 : result;
a1dde8de
KW
1621 }
1622
2f306ab9 1623 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
1624 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1625 uvchr_to_utf8(p, c);
1626 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1627 }
1628 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1629 the special flags. */
1630 U8 utf8_c[UTF8_MAXBYTES + 1];
1631 uvchr_to_utf8(utf8_c, c);
445bf929 1632 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
a0270393 1633 }
84afefe6
JH
1634}
1635
26483009 1636PERL_STATIC_INLINE bool
5141f98e 1637S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 1638 const char *const swashname, SV* const invlist)
bde6a22d 1639{
ea317ccb
KW
1640 /* returns a boolean giving whether or not the UTF8-encoded character that
1641 * starts at <p> is in the swash indicated by <swashname>. <swash>
1642 * contains a pointer to where the swash indicated by <swashname>
1643 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
1644 * look at <*swash> and only generate a swash if it is not null. <invlist>
1645 * is NULL or an inversion list that defines the swash. If not null, it
1646 * saves time during initialization of the swash.
ea317ccb
KW
1647 *
1648 * Note that it is assumed that the buffer length of <p> is enough to
1649 * contain all the bytes that comprise the character. Thus, <*p> should
1650 * have been checked before this call for mal-formedness enough to assure
1651 * that. */
1652
7918f24d
NC
1653 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1654
492a624f 1655 /* The API should have included a length for the UTF-8 character in <p>,
28123549 1656 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
1657 * as far as there being enough bytes available in it to accommodate the
1658 * character without reading beyond the end, and pass that number on to the
1659 * validating routine */
6302f837 1660 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
28123549
KW
1661 if (ckWARN_d(WARN_UTF8)) {
1662 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
9816f121 1663 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
28123549
KW
1664 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1665 what the malformation is */
1666 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1667 }
1668 }
1669 return FALSE;
1670 }
87367d5f
KW
1671 if (!*swash) {
1672 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
1673 *swash = _core_swash_init("utf8",
1674
1675 /* Only use the name if there is no inversion
1676 * list; otherwise will go out to disk */
1677 (invlist) ? "" : swashname,
1678
1679 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 1680 }
28123549 1681
bde6a22d
NC
1682 return swash_fetch(*swash, p, TRUE) != 0;
1683}
1684
1685bool
922e8cb4
KW
1686Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1687{
922e8cb4
KW
1688 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1689
1690 assert(classnum < _FIRST_NON_SWASH_CC);
1691
f25ce844
KW
1692 return is_utf8_common(p,
1693 &PL_utf8_swash_ptrs[classnum],
1694 swash_property_names[classnum],
1695 PL_XPosix_ptrs[classnum]);
922e8cb4
KW
1696}
1697
1698bool
f2645549 1699Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
a0ed51b3 1700{
f2645549 1701 SV* invlist = NULL;
7918f24d 1702
f2645549 1703 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
7918f24d 1704
f2645549
KW
1705 if (! PL_utf8_perl_idstart) {
1706 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1707 }
60071a22 1708 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
82686b01
JH
1709}
1710
1711bool
f2645549 1712Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 1713{
f2645549 1714 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
1715
1716 if (*p == '_')
1717 return TRUE;
f25ce844 1718 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
1719}
1720
1721bool
eba68aa0
KW
1722Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1723{
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 }
60071a22 1731 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
eba68aa0
KW
1732}
1733
eba68aa0 1734bool
f2645549 1735Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 1736{
f2645549 1737 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 1738
f25ce844 1739 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
1740}
1741
1742bool
f2645549 1743Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 1744{
f2645549 1745 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 1746
f25ce844 1747 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
1748}
1749
1750bool
7dbf68d2
KW
1751Perl__is_utf8_mark(pTHX_ const U8 *p)
1752{
7dbf68d2
KW
1753 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1754
f25ce844 1755 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
1756}
1757
6b5c0936 1758/*
87cea99e 1759=for apidoc to_utf8_case
6b5c0936 1760
6fae5207 1761C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
1762the character that is being converted. This routine assumes that the character
1763at C<p> is well-formed.
6b5c0936 1764
6fae5207
KW
1765C<ustrp> is a pointer to the character buffer to put the
1766conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
1767of the result.
1768
6fae5207 1769C<swashp> is a pointer to the swash to use.
6b5c0936 1770
a1433954 1771Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
6fae5207 1772and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 1773but not always, a multicharacter mapping), is tried first.
6b5c0936 1774
4a8240a3
KW
1775C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1776any special mappings; C<""> means to use the special mappings. Values other
1777than these two are treated as the name of the hash containing the special
1778mappings, like C<"utf8::ToSpecLower">.
6b5c0936 1779
6fae5207 1780C<normal> is a string like "ToLower" which means the swash
0134edef
JH
1781%utf8::ToLower.
1782
1783=cut */
6b5c0936 1784
2104c8d9 1785UV
9a957fbc
AL
1786Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1787 SV **swashp, const char *normal, const char *special)
a0ed51b3 1788{
0134edef 1789 STRLEN len = 0;
f4cd282c 1790 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
7918f24d
NC
1791
1792 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1793
9ae3ac1a
KW
1794 /* Note that swash_fetch() doesn't output warnings for these because it
1795 * assumes we will */
8457b38f 1796 if (uv1 >= UNICODE_SURROGATE_FIRST) {
9ae3ac1a 1797 if (uv1 <= UNICODE_SURROGATE_LAST) {
8457b38f
KW
1798 if (ckWARN_d(WARN_SURROGATE)) {
1799 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1800 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1801 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1802 }
9ae3ac1a
KW
1803 }
1804 else if (UNICODE_IS_SUPER(uv1)) {
8457b38f
KW
1805 if (ckWARN_d(WARN_NON_UNICODE)) {
1806 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1807 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1808 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1809 }
9ae3ac1a
KW
1810 }
1811
1812 /* Note that non-characters are perfectly legal, so no warning should
1813 * be given */
1814 }
1815
0134edef 1816 if (!*swashp) /* load on-demand */
5ab9d2ef 1817 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 1818
a6f87d8c 1819 if (special) {
0134edef 1820 /* It might be "special" (sometimes, but not always,
2a37f04d 1821 * a multicharacter mapping) */
4a8240a3 1822 HV *hv = NULL;
b08cf34e
JH
1823 SV **svp;
1824
4a8240a3
KW
1825 /* If passed in the specials name, use that; otherwise use any
1826 * given in the swash */
1827 if (*special != '\0') {
1828 hv = get_hv(special, 0);
1829 }
1830 else {
1831 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
1832 if (svp) {
1833 hv = MUTABLE_HV(SvRV(*svp));
1834 }
1835 }
1836
176fe009
KW
1837 if (hv
1838 && (svp = hv_fetch(hv, (const char*)p, UNISKIP(uv1), FALSE))
1839 && (*svp))
1840 {
cfd0369c 1841 const char *s;
47654450 1842
cfd0369c 1843 s = SvPV_const(*svp, len);
47654450 1844 if (len == 1)
f4cd282c 1845 /* EIGHTBIT */
c80e42f3 1846 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 1847 else {
d2dcd0fb 1848 Copy(s, ustrp, len, U8);
29e98929 1849 }
983ffd37 1850 }
0134edef
JH
1851 }
1852
1853 if (!len && *swashp) {
f4cd282c 1854 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is utf8 */);
d4c19fe8 1855
0134edef
JH
1856 if (uv2) {
1857 /* It was "normal" (a single character mapping). */
f4cd282c 1858 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
1859 }
1860 }
1feea2c7 1861
cbe07460
KW
1862 if (len) {
1863 if (lenp) {
1864 *lenp = len;
1865 }
1866 return valid_utf8_to_uvchr(ustrp, 0);
1867 }
1868
1869 /* Here, there was no mapping defined, which means that the code point maps
1870 * to itself. Return the inputs */
bfdf22ec 1871 len = UTF8SKIP(p);
ca9fab46
KW
1872 if (p != ustrp) { /* Don't copy onto itself */
1873 Copy(p, ustrp, len, U8);
1874 }
0134edef 1875
2a37f04d
JH
1876 if (lenp)
1877 *lenp = len;
1878
f4cd282c 1879 return uv1;
cbe07460 1880
a0ed51b3
LW
1881}
1882
051a06d4 1883STATIC UV
ab0b796c 1884S_check_locale_boundary_crossing(pTHX_ const char * const func_name, const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
051a06d4
KW
1885{
1886 /* This is called when changing the case of a utf8-encoded character above
31f05a37
KW
1887 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
1888 * result contains a character that crosses the 255/256 boundary, disallow
1889 * the change, and return the original code point. See L<perlfunc/lc> for
1890 * why;
051a06d4 1891 *
a1433954
KW
1892 * p points to the original string whose case was changed; assumed
1893 * by this routine to be well-formed
051a06d4
KW
1894 * result the code point of the first character in the changed-case string
1895 * ustrp points to the changed-case string (<result> represents its first char)
1896 * lenp points to the length of <ustrp> */
1897
1898 UV original; /* To store the first code point of <p> */
1899
1900 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
1901
a4f12ed7 1902 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
1903
1904 /* We know immediately if the first character in the string crosses the
1905 * boundary, so can skip */
1906 if (result > 255) {
1907
1908 /* Look at every character in the result; if any cross the
1909 * boundary, the whole thing is disallowed */
1910 U8* s = ustrp + UTF8SKIP(ustrp);
1911 U8* e = ustrp + *lenp;
1912 while (s < e) {
a4f12ed7 1913 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
1914 goto bad_crossing;
1915 }
1916 s += UTF8SKIP(s);
1917 }
1918
1919 /* Here, no characters crossed, result is ok as-is */
1920 return result;
1921 }
1922
1923bad_crossing:
1924
1925 /* Failed, have to return the original */
4b88fb76 1926 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
1927
1928 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
1929 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1930 "Can't do %s(\"\\x{%"UVXf"}\") on non-UTF-8 locale; "
1931 "resolved to \"\\x{%"UVXf"}\".",
1932 func_name,
1933 original,
1934 original);
051a06d4
KW
1935 Copy(p, ustrp, *lenp, char);
1936 return original;
1937}
1938
d3e79532 1939/*
87cea99e 1940=for apidoc to_utf8_upper
d3e79532 1941
1f607577 1942Instead use L</toUPPER_utf8>.
a1433954 1943
d3e79532
JH
1944=cut */
1945
051a06d4 1946/* Not currently externally documented, and subject to change:
31f05a37
KW
1947 * <flags> is set iff iff the rules from the current underlying locale are to
1948 * be used. */
051a06d4 1949
2104c8d9 1950UV
31f05a37 1951Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 1952{
051a06d4
KW
1953 UV result;
1954
1955 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 1956
780fcc9f
KW
1957 if (flags) {
1958 /* Treat a UTF-8 locale as not being in locale at all */
1959 if (IN_UTF8_CTYPE_LOCALE) {
1960 flags = FALSE;
1961 }
1962 else {
1963 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
1964 }
31f05a37
KW
1965 }
1966
3a4c58c9 1967 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
1968 if (flags) {
1969 result = toUPPER_LC(*p);
1970 }
1971 else {
81c6c7ce 1972 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 1973 }
3a4c58c9
KW
1974 }
1975 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 1976 if (flags) {
a6d8b88b 1977 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 1978 result = toUPPER_LC(c);
051a06d4
KW
1979 }
1980 else {
94bb8c36 1981 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 1982 ustrp, lenp, 'S');
051a06d4
KW
1983 }
1984 }
1985 else { /* utf8, ord above 255 */
1986 result = CALL_UPPER_CASE(p, ustrp, lenp);
1987
1988 if (flags) {
ab0b796c 1989 result = check_locale_boundary_crossing("uc", p, result, ustrp, lenp);
051a06d4
KW
1990 }
1991 return result;
1992 }
1993
1994 /* Here, used locale rules. Convert back to utf8 */
1995 if (UTF8_IS_INVARIANT(result)) {
1996 *ustrp = (U8) result;
1997 *lenp = 1;
1998 }
1999 else {
62cb07ea
KW
2000 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2001 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2002 *lenp = 2;
3a4c58c9 2003 }
baa60164 2004
051a06d4 2005 return result;
983ffd37 2006}
a0ed51b3 2007
d3e79532 2008/*
87cea99e 2009=for apidoc to_utf8_title
d3e79532 2010
1f607577 2011Instead use L</toTITLE_utf8>.
a1433954 2012
d3e79532
JH
2013=cut */
2014
051a06d4 2015/* Not currently externally documented, and subject to change:
31f05a37
KW
2016 * <flags> is set iff the rules from the current underlying locale are to be
2017 * used. Since titlecase is not defined in POSIX, for other than a
2018 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 2019 */
051a06d4 2020
983ffd37 2021UV
31f05a37 2022Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
983ffd37 2023{
051a06d4
KW
2024 UV result;
2025
2026 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2027
780fcc9f
KW
2028 if (flags) {
2029 /* Treat a UTF-8 locale as not being in locale at all */
2030 if (IN_UTF8_CTYPE_LOCALE) {
2031 flags = FALSE;
2032 }
2033 else {
2034 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2035 }
31f05a37
KW
2036 }
2037
3a4c58c9 2038 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2039 if (flags) {
2040 result = toUPPER_LC(*p);
2041 }
2042 else {
81c6c7ce 2043 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2044 }
3a4c58c9
KW
2045 }
2046 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2047 if (flags) {
a6d8b88b 2048 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2049 result = toUPPER_LC(c);
051a06d4
KW
2050 }
2051 else {
94bb8c36 2052 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2053 ustrp, lenp, 's');
051a06d4
KW
2054 }
2055 }
2056 else { /* utf8, ord above 255 */
2057 result = CALL_TITLE_CASE(p, ustrp, lenp);
2058
2059 if (flags) {
ab0b796c 2060 result = check_locale_boundary_crossing("ucfirst", p, result, ustrp, lenp);
051a06d4
KW
2061 }
2062 return result;
2063 }
2064
2065 /* Here, used locale rules. Convert back to utf8 */
2066 if (UTF8_IS_INVARIANT(result)) {
2067 *ustrp = (U8) result;
2068 *lenp = 1;
2069 }
2070 else {
62cb07ea
KW
2071 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2072 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2073 *lenp = 2;
3a4c58c9
KW
2074 }
2075
051a06d4 2076 return result;
a0ed51b3
LW
2077}
2078
d3e79532 2079/*
87cea99e 2080=for apidoc to_utf8_lower
d3e79532 2081
1f607577 2082Instead use L</toLOWER_utf8>.
a1433954 2083
d3e79532
JH
2084=cut */
2085
051a06d4 2086/* Not currently externally documented, and subject to change:
31f05a37
KW
2087 * <flags> is set iff iff the rules from the current underlying locale are to
2088 * be used.
2089 */
051a06d4 2090
2104c8d9 2091UV
31f05a37 2092Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2093{
051a06d4
KW
2094 UV result;
2095
051a06d4 2096 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2097
780fcc9f
KW
2098 if (flags) {
2099 /* Treat a UTF-8 locale as not being in locale at all */
2100 if (IN_UTF8_CTYPE_LOCALE) {
2101 flags = FALSE;
2102 }
2103 else {
2104 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2105 }
31f05a37
KW
2106 }
2107
968c5e6a 2108 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2109 if (flags) {
2110 result = toLOWER_LC(*p);
2111 }
2112 else {
81c6c7ce 2113 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2114 }
968c5e6a
KW
2115 }
2116 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2117 if (flags) {
a6d8b88b 2118 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2119 result = toLOWER_LC(c);
051a06d4
KW
2120 }
2121 else {
94bb8c36 2122 return to_lower_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2123 ustrp, lenp);
051a06d4 2124 }
968c5e6a 2125 }
051a06d4
KW
2126 else { /* utf8, ord above 255 */
2127 result = CALL_LOWER_CASE(p, ustrp, lenp);
2128
2129 if (flags) {
ab0b796c 2130 result = check_locale_boundary_crossing("lc", p, result, ustrp, lenp);
051a06d4 2131 }
968c5e6a 2132
051a06d4
KW
2133 return result;
2134 }
2135
2136 /* Here, used locale rules. Convert back to utf8 */
2137 if (UTF8_IS_INVARIANT(result)) {
2138 *ustrp = (U8) result;
2139 *lenp = 1;
2140 }
2141 else {
62cb07ea
KW
2142 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2143 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2144 *lenp = 2;
2145 }
2146
051a06d4 2147 return result;
b4e400f9
JH
2148}
2149
d3e79532 2150/*
87cea99e 2151=for apidoc to_utf8_fold
d3e79532 2152
1f607577 2153Instead use L</toFOLD_utf8>.
a1433954 2154
d3e79532
JH
2155=cut */
2156
051a06d4
KW
2157/* Not currently externally documented, and subject to change,
2158 * in <flags>
31f05a37
KW
2159 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2160 * locale are to be used.
051a06d4
KW
2161 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2162 * otherwise simple folds
a0270393
KW
2163 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2164 * prohibited
445bf929 2165 */
36bb2ab6 2166
b4e400f9 2167UV
445bf929 2168Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
b4e400f9 2169{
051a06d4
KW
2170 UV result;
2171
36bb2ab6 2172 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2173
a0270393
KW
2174 /* These are mutually exclusive */
2175 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2176
50ba90ff
KW
2177 assert(p != ustrp); /* Otherwise overwrites */
2178
780fcc9f
KW
2179 if (flags & FOLD_FLAGS_LOCALE) {
2180 /* Treat a UTF-8 locale as not being in locale at all */
2181 if (IN_UTF8_CTYPE_LOCALE) {
2182 flags &= ~FOLD_FLAGS_LOCALE;
2183 }
2184 else {
2185 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2186 }
31f05a37
KW
2187 }
2188
a1dde8de 2189 if (UTF8_IS_INVARIANT(*p)) {
051a06d4 2190 if (flags & FOLD_FLAGS_LOCALE) {
d22b930b 2191 result = toFOLD_LC(*p);
051a06d4
KW
2192 }
2193 else {
81c6c7ce 2194 return _to_fold_latin1(*p, ustrp, lenp,
1ca267a5 2195 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2196 }
a1dde8de
KW
2197 }
2198 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2199 if (flags & FOLD_FLAGS_LOCALE) {
a6d8b88b 2200 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2201 result = toFOLD_LC(c);
051a06d4
KW
2202 }
2203 else {
94bb8c36 2204 return _to_fold_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
51910141 2205 ustrp, lenp,
1ca267a5 2206 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2207 }
a1dde8de 2208 }
051a06d4 2209 else { /* utf8, ord above 255 */
a0270393 2210 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2211
1ca267a5
KW
2212 if (flags & FOLD_FLAGS_LOCALE) {
2213
76f2ffcd
KW
2214# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
2215# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
2216
2217 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
2218 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
2219
538e84ed
KW
2220 /* Special case these two characters, as what normally gets
2221 * returned under locale doesn't work */
76f2ffcd
KW
2222 if (UTF8SKIP(p) == cap_sharp_s_len
2223 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 2224 {
ab0b796c
KW
2225 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2226 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2227 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
2228 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
2229 goto return_long_s;
2230 }
76f2ffcd
KW
2231 else if (UTF8SKIP(p) == long_s_t_len
2232 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 2233 {
ab0b796c
KW
2234 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2235 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2236 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
2237 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
2238 goto return_ligature_st;
2239 }
ab0b796c 2240 return check_locale_boundary_crossing("fc", p, result, ustrp, lenp);
051a06d4 2241 }
a0270393
KW
2242 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2243 return result;
2244 }
2245 else {
2246 /* This is called when changing the case of a utf8-encoded
9fc2026f
KW
2247 * character above the ASCII range, and the result should not
2248 * contain an ASCII character. */
a0270393
KW
2249
2250 UV original; /* To store the first code point of <p> */
2251
2252 /* Look at every character in the result; if any cross the
2253 * boundary, the whole thing is disallowed */
2254 U8* s = ustrp;
2255 U8* e = ustrp + *lenp;
2256 while (s < e) {
2257 if (isASCII(*s)) {
2258 /* Crossed, have to return the original */
2259 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 2260
9fc2026f 2261 /* But in these instances, there is an alternative we can
1ca267a5 2262 * return that is valid */
9fc2026f
KW
2263 if (original == LATIN_CAPITAL_LETTER_SHARP_S
2264 || original == LATIN_SMALL_LETTER_SHARP_S)
2265 {
1ca267a5
KW
2266 goto return_long_s;
2267 }
9fc2026f
KW
2268 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2269 goto return_ligature_st;
2270 }
a0270393
KW
2271 Copy(p, ustrp, *lenp, char);
2272 return original;
2273 }
2274 s += UTF8SKIP(s);
2275 }
051a06d4 2276
a0270393
KW
2277 /* Here, no characters crossed, result is ok as-is */
2278 return result;
2279 }
051a06d4
KW
2280 }
2281
2282 /* Here, used locale rules. Convert back to utf8 */
2283 if (UTF8_IS_INVARIANT(result)) {
2284 *ustrp = (U8) result;
2285 *lenp = 1;
2286 }
2287 else {
62cb07ea
KW
2288 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2289 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2290 *lenp = 2;
2291 }
2292
051a06d4 2293 return result;
1ca267a5
KW
2294
2295 return_long_s:
2296 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2297 * folds to a string of two of these characters. By returning this
2298 * instead, then, e.g.,
2299 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2300 * works. */
2301
2302 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2303 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2304 ustrp, *lenp, U8);
2305 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
2306
2307 return_ligature_st:
2308 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2309 * have the other one fold to it */
2310
2311 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2312 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2313 return LATIN_SMALL_LIGATURE_ST;
a0ed51b3
LW
2314}
2315
711a919c 2316/* Note:
f90a9a02 2317 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2318 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2319 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2320 */
c4a5db0c 2321
a0ed51b3 2322SV*
7fc63493 2323Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2324{
c4a5db0c
KW
2325 PERL_ARGS_ASSERT_SWASH_INIT;
2326
2327 /* Returns a copy of a swash initiated by the called function. This is the
2328 * public interface, and returning a copy prevents others from doing
2329 * mischief on the original */
2330
5d3d13d1 2331 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
2332}
2333
2334SV*
5d3d13d1 2335Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c 2336{
2c1f00b9
YO
2337
2338 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
2339 * use the following define */
2340
2341#define CORE_SWASH_INIT_RETURN(x) \
2342 PL_curpm= old_PL_curpm; \
2343 return x
2344
c4a5db0c 2345 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
2346 * by calling utf8_heavy.pl in the general case. The returned value may be
2347 * the swash's inversion list instead if the input parameters allow it.
2348 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
2349 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2350 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
2351 *
2352 * This interface should only be used by functions that won't destroy or
2353 * adversely change the swash, as doing so affects all other uses of the
2354 * swash in the program; the general public should use 'Perl_swash_init'
2355 * instead.
2356 *
2357 * pkg is the name of the package that <name> should be in.
2358 * name is the name of the swash to find. Typically it is a Unicode
2359 * property name, including user-defined ones
2360 * listsv is a string to initialize the swash with. It must be of the form
2361 * documented as the subroutine return value in
2362 * L<perlunicode/User-Defined Character Properties>
2363 * minbits is the number of bits required to represent each data element.
2364 * It is '1' for binary properties.
2365 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 2366 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
2367 * flags_p if non-NULL is the address of various input and output flag bits
2368 * to the routine, as follows: ('I' means is input to the routine;
2369 * 'O' means output from the routine. Only flags marked O are
2370 * meaningful on return.)
2371 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2372 * came from a user-defined property. (I O)
5d3d13d1
KW
2373 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2374 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
2375 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2376 * return of an inversion list instead of a swash hash if this routine
2377 * thinks that would result in faster execution of swash_fetch() later
2378 * on. (I)
9a53f6cf
KW
2379 *
2380 * Thus there are three possible inputs to find the swash: <name>,
2381 * <listsv>, and <invlist>. At least one must be specified. The result
2382 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
2383 * actions can intersect, etc. what <name> gives. To avoid going out to
2384 * disk at all, <invlist> should specify completely what the swash should
2385 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
2386 *
2387 * <invlist> is only valid for binary properties */
c4a5db0c 2388
2c1f00b9
YO
2389 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
2390
c4a5db0c 2391 SV* retval = &PL_sv_undef;
83199d38 2392 HV* swash_hv = NULL;
87367d5f
KW
2393 const int invlist_swash_boundary =
2394 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2395 ? 512 /* Based on some benchmarking, but not extensive, see commit
2396 message */
2397 : -1; /* Never return just an inversion list */
9a53f6cf
KW
2398
2399 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2400 assert(! invlist || minbits == 1);
2401
2c1f00b9
YO
2402 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
2403 that triggered the swash init and the swash init perl logic itself.
2404 See perl #122747 */
2405
9a53f6cf
KW
2406 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2407 * so */
2408 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2409 dSP;
2410 const size_t pkg_len = strlen(pkg);
2411 const size_t name_len = strlen(name);
2412 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2413 SV* errsv_save;
2414 GV *method;
2415
2416 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2417
2418 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2419 ENTER;
69794297 2420 SAVEHINTS();
650f067c
JL
2421 /* We might get here via a subroutine signature which uses a utf8
2422 * parameter name, at which point PL_subname will have been set
2423 * but not yet used. */
2424 save_item(PL_subname);
69794297
KW
2425 if (PL_parser && PL_parser->error_count)
2426 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2427 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2428 if (!method) { /* demand load utf8 */
2429 ENTER;
db2c6cb3
FC
2430 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2431 GvSV(PL_errgv) = NULL;
1a419e6b 2432#ifndef NO_TAINT_SUPPORT
69794297
KW
2433 /* It is assumed that callers of this routine are not passing in
2434 * any user derived data. */
284167a5
SM
2435 SAVEBOOL(TAINT_get);
2436 TAINT_NOT;
2437#endif
69794297
KW
2438 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2439 NULL);
eed484f9 2440 {
db2c6cb3
FC
2441 /* Not ERRSV, as there is no need to vivify a scalar we are
2442 about to discard. */
2443 SV * const errsv = GvSV(PL_errgv);
2444 if (!SvTRUE(errsv)) {
2445 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2446 SvREFCNT_dec(errsv);
2447 }
eed484f9 2448 }
69794297
KW
2449 LEAVE;
2450 }
2451 SPAGAIN;
2452 PUSHMARK(SP);
2453 EXTEND(SP,5);
2454 mPUSHp(pkg, pkg_len);
2455 mPUSHp(name, name_len);
2456 PUSHs(listsv);
2457 mPUSHi(minbits);
2458 mPUSHi(none);
2459 PUTBACK;
db2c6cb3
FC
2460 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2461 GvSV(PL_errgv) = NULL;
69794297
KW
2462 /* If we already have a pointer to the method, no need to use
2463 * call_method() to repeat the lookup. */
c41800a8
KW
2464 if (method
2465 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
2466 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2467 {
2468 retval = *PL_stack_sp--;
2469 SvREFCNT_inc(retval);
2470 }
eed484f9 2471 {
db2c6cb3
FC
2472 /* Not ERRSV. See above. */
2473 SV * const errsv = GvSV(PL_errgv);
2474 if (!SvTRUE(errsv)) {
2475 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2476 SvREFCNT_dec(errsv);
2477 }
eed484f9 2478 }
ce3b816e 2479 LEAVE;
69794297
KW
2480 POPSTACK;
2481 if (IN_PERL_COMPILETIME) {
2482 CopHINTS_set(PL_curcop, PL_hints);
2483 }
2484 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2485 if (SvPOK(retval))
2486
2487 /* If caller wants to handle missing properties, let them */
5d3d13d1 2488 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2c1f00b9 2489 CORE_SWASH_INIT_RETURN(NULL);
69794297
KW
2490 }
2491 Perl_croak(aTHX_
2492 "Can't find Unicode property definition \"%"SVf"\"",
2493 SVfARG(retval));
a25b5927 2494 NOT_REACHED; /* NOTREACHED */
69794297 2495 }
9a53f6cf 2496 } /* End of calling the module to find the swash */
36eb48b4 2497
83199d38
KW
2498 /* If this operation fetched a swash, and we will need it later, get it */
2499 if (retval != &PL_sv_undef
2500 && (minbits == 1 || (flags_p
2501 && ! (*flags_p
2502 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2503 {
2504 swash_hv = MUTABLE_HV(SvRV(retval));
2505
2506 /* If we don't already know that there is a user-defined component to
2507 * this swash, and the user has indicated they wish to know if there is
2508 * one (by passing <flags_p>), find out */
2509 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2510 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2511 if (user_defined && SvUV(*user_defined)) {
2512 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2513 }
2514 }
2515 }
2516
36eb48b4
KW
2517 /* Make sure there is an inversion list for binary properties */
2518 if (minbits == 1) {
2519 SV** swash_invlistsvp = NULL;
2520 SV* swash_invlist = NULL;
9a53f6cf 2521 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
2522 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2523 an unclaimed reference count */
36eb48b4 2524
9a53f6cf 2525 /* If this operation fetched a swash, get its already existing
83199d38 2526 * inversion list, or create one for it */
36eb48b4 2527
83199d38 2528 if (swash_hv) {
5c9f4bd2 2529 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
2530 if (swash_invlistsvp) {
2531 swash_invlist = *swash_invlistsvp;
2532 invlist_in_swash_is_valid = TRUE;
2533 }
2534 else {
36eb48b4 2535 swash_invlist = _swash_to_invlist(retval);
02c85471 2536 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
2537 }
2538 }
2539
2540 /* If an inversion list was passed in, have to include it */
2541 if (invlist) {
2542
2543 /* Any fetched swash will by now have an inversion list in it;
2544 * otherwise <swash_invlist> will be NULL, indicating that we
2545 * didn't fetch a swash */
2546 if (swash_invlist) {
2547
2548 /* Add the passed-in inversion list, which invalidates the one
2549 * already stored in the swash */
2550 invlist_in_swash_is_valid = FALSE;
2551 _invlist_union(invlist, swash_invlist, &swash_invlist);
2552 }
2553 else {
2554
87367d5f
KW
2555 /* Here, there is no swash already. Set up a minimal one, if
2556 * we are going to return a swash */
2557 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 2558 swash_hv = newHV();
4aca0fe6 2559 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 2560 }
9a53f6cf
KW
2561 swash_invlist = invlist;
2562 }
9a53f6cf
KW
2563 }
2564
2565 /* Here, we have computed the union of all the passed-in data. It may
2566 * be that there was an inversion list in the swash which didn't get
538e84ed 2567 * touched; otherwise save the computed one */
87367d5f
KW
2568 if (! invlist_in_swash_is_valid
2569 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2570 {
5c9f4bd2 2571 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
2572 {
2573 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2574 }
cc34d8c5
FC
2575 /* We just stole a reference count. */
2576 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2577 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 2578 }
87367d5f 2579
dbfdbd26
KW
2580 SvREADONLY_on(swash_invlist);
2581
c41800a8 2582 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
2583 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2584 SvREFCNT_dec(retval);
02c85471
FC
2585 if (!swash_invlist_unclaimed)
2586 SvREFCNT_inc_simple_void_NN(swash_invlist);
2587 retval = newRV_noinc(swash_invlist);
87367d5f 2588 }
36eb48b4
KW
2589 }
2590
2c1f00b9
YO
2591 CORE_SWASH_INIT_RETURN(retval);
2592#undef CORE_SWASH_INIT_RETURN
a0ed51b3
LW
2593}
2594
035d37be
JH
2595
2596/* This API is wrong for special case conversions since we may need to
2597 * return several Unicode characters for a single Unicode character
2598 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2599 * the lower-level routine, and it is similarly broken for returning
38684baa
KW
2600 * multiple values. --jhi
2601 * For those, you should use to_utf8_case() instead */
b0e3252e 2602/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
2603
2604/* Note:
2605 * Returns the value of property/mapping C<swash> for the first character
2606 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3d0f8846
KW
2607 * assumed to be in well-formed utf8. If C<do_utf8> is false, the string C<ptr>
2608 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
2609 *
2610 * A "swash" is a hash which contains initially the keys/values set up by
2611 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2612 * property for all possible code points. Things are stored in a compact form
2613 * (see utf8_heavy.pl) so that calculation is required to find the actual
2614 * property value for a given code point. As code points are looked up, new
2615 * key/value pairs are added to the hash, so that the calculation doesn't have
2616 * to ever be re-done. Further, each calculation is done, not just for the
2617 * desired one, but for a whole block of code points adjacent to that one.
2618 * For binary properties on ASCII machines, the block is usually for 64 code
2619 * points, starting with a code point evenly divisible by 64. Thus if the
2620 * property value for code point 257 is requested, the code goes out and
2621 * calculates the property values for all 64 code points between 256 and 319,
2622 * and stores these as a single 64-bit long bit vector, called a "swatch",
2623 * under the key for code point 256. The key is the UTF-8 encoding for code
2624 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2625 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2626 * for code point 258 is then requested, this code realizes that it would be
2627 * stored under the key for 256, and would find that value and extract the
2628 * relevant bit, offset from 256.
2629 *
2630 * Non-binary properties are stored in as many bits as necessary to represent
2631 * their values (32 currently, though the code is more general than that), not
2632 * as single bits, but the principal is the same: the value for each key is a
2633 * vector that encompasses the property values for all code points whose UTF-8
2634 * representations are represented by the key. That is, for all code points
2635 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2636 * bytes of that.
680c470c 2637 */
a0ed51b3 2638UV
680c470c 2639Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 2640{
ef8f7699 2641 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
2642 U32 klen;
2643 U32 off;
9b56a019 2644 STRLEN slen = 0;
7d85a32c 2645 STRLEN needents;
cfd0369c 2646 const U8 *tmps = NULL;
979f2922 2647 SV *swatch;
08fb1ac5 2648 const U8 c = *ptr;
3568d838 2649
7918f24d
NC
2650 PERL_ARGS_ASSERT_SWASH_FETCH;
2651
87367d5f
KW
2652 /* If it really isn't a hash, it isn't really swash; must be an inversion
2653 * list */
2654 if (SvTYPE(hv) != SVt_PVHV) {
2655 return _invlist_contains_cp((SV*)hv,
2656 (do_utf8)
2657 ? valid_utf8_to_uvchr(ptr, NULL)
2658 : c);
2659 }
2660
08fb1ac5
KW
2661 /* We store the values in a "swatch" which is a vec() value in a swash
2662 * hash. Code points 0-255 are a single vec() stored with key length
2663 * (klen) 0. All other code points have a UTF-8 representation
2664 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2665 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2666 * length for them is the length of the encoded char - 1. ptr[klen] is the
2667 * final byte in the sequence representing the character */
2668 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2669 klen = 0;
2670 needents = 256;
2671 off = c;
3568d838 2672 }
08fb1ac5
KW
2673 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2674 klen = 0;
2675 needents = 256;
2676 off = TWO_BYTE_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
2677 }
2678 else {
08fb1ac5
KW
2679 klen = UTF8SKIP(ptr) - 1;
2680
2681 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2682 * the vec is the final byte in the sequence. (In EBCDIC this is
2683 * converted to I8 to get consecutive values.) To help you visualize
2684 * all this:
2685 * Straight 1047 After final byte
2686 * UTF-8 UTF-EBCDIC I8 transform
2687 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2688 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2689 * ...
2690 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2691 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2692 * ...
2693 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2694 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2695 * ...
2696 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2697 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2698 * ...
2699 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2700 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2701 *
2702 * (There are no discontinuities in the elided (...) entries.)
2703 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2704 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2705 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2706 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2707 * index into the vec() swatch (after subtracting 0x80, which we
2708 * actually do with an '&').
2709 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2710 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2711 * dicontinuities which go away by transforming it into I8, and we
2712 * effectively subtract 0xA0 to get the index. */
979f2922 2713 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 2714 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 2715 }
7d85a32c 2716
a0ed51b3
LW
2717 /*
2718 * This single-entry cache saves about 1/3 of the utf8 overhead in test
2719 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2720 * it's nothing to sniff at.) Pity we usually come through at least
2721 * two function calls to get here...
2722 *
2723 * NB: this code assumes that swatches are never modified, once generated!
2724 */
2725
3568d838 2726 if (hv == PL_last_swash_hv &&
a0ed51b3 2727 klen == PL_last_swash_klen &&
27da23d5 2728 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
2729 {
2730 tmps = PL_last_swash_tmps;
2731 slen = PL_last_swash_slen;
2732 }
2733 else {
2734 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 2735 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 2736
b0e3252e 2737 /* If not cached, generate it via swatch_get */
979f2922 2738 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
2739 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2740 {
2741 if (klen) {
2742 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2743 swatch = swatch_get(swash,
2744 code_point & ~((UV)needents - 1),
2745 needents);
2746 }
2747 else { /* For the first 256 code points, the swatch has a key of
2748 length 0 */
2749 swatch = swatch_get(swash, 0, needents);
2750 }
979f2922 2751
923e4eb5 2752 if (IN_PERL_COMPILETIME)
623e6609 2753 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 2754
979f2922 2755 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 2756
979f2922
ST
2757 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2758 || (slen << 3) < needents)
5637ef5b
NC
2759 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2760 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2761 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
2762 }
2763
2764 PL_last_swash_hv = hv;
16d8f38a 2765 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 2766 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
2767 /* FIXME change interpvar.h? */
2768 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
2769 PL_last_swash_slen = slen;
2770 if (klen)
2771 Copy(ptr, PL_last_swash_key, klen, U8);
2772 }
2773
9faf8d75 2774 switch ((int)((slen << 3) / needents)) {
a0ed51b3 2775 case 1:
e7aca353 2776 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
a0ed51b3 2777 case 8:
e7aca353 2778 return ((UV) tmps[off]);
a0ed51b3
LW
2779 case 16:
2780 off <<= 1;
e7aca353
JH
2781 return
2782 ((UV) tmps[off ] << 8) +
2783 ((UV) tmps[off + 1]);
a0ed51b3
LW
2784 case 32:
2785 off <<= 2;
e7aca353
JH
2786 return
2787 ((UV) tmps[off ] << 24) +
2788 ((UV) tmps[off + 1] << 16) +
2789 ((UV) tmps[off + 2] << 8) +
2790 ((UV) tmps[off + 3]);
a0ed51b3 2791 }
5637ef5b
NC
2792 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2793 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 2794 NORETURN_FUNCTION_END;
a0ed51b3 2795}
2b9d42f0 2796
319009ee
KW
2797/* Read a single line of the main body of the swash input text. These are of
2798 * the form:
2799 * 0053 0056 0073
2800 * where each number is hex. The first two numbers form the minimum and
2801 * maximum of a range, and the third is the value associated with the range.
2802 * Not all swashes should have a third number
2803 *
2804 * On input: l points to the beginning of the line to be examined; it points
2805 * to somewhere in the string of the whole input text, and is
2806 * terminated by a \n or the null string terminator.
2807 * lend points to the null terminator of that string
2808 * wants_value is non-zero if the swash expects a third number
2809 * typestr is the name of the swash's mapping, like 'ToLower'
2810 * On output: *min, *max, and *val are set to the values read from the line.
2811 * returns a pointer just beyond the line examined. If there was no
2812 * valid min number on the line, returns lend+1
2813 */
2814
2815STATIC U8*
2816S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2817 const bool wants_value, const U8* const typestr)
2818{
2819 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2820 STRLEN numlen; /* Length of the number */
02470786
KW
2821 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2822 | PERL_SCAN_DISALLOW_PREFIX
2823 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2824
2825 /* nl points to the next \n in the scan */
2826 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2827
95543e92
KW
2828 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
2829
319009ee
KW
2830 /* Get the first number on the line: the range minimum */
2831 numlen = lend - l;
2832 *min = grok_hex((char *)l, &numlen, &flags, NULL);
c88850db 2833 *max = *min; /* So can never return without setting max */
319009ee
KW
2834 if (numlen) /* If found a hex number, position past it */
2835 l += numlen;
2836 else if (nl) { /* Else, go handle next line, if any */
2837 return nl + 1; /* 1 is length of "\n" */
2838 }
2839 else { /* Else, no next line */
2840 return lend + 1; /* to LIST's end at which \n is not found */
2841 }
2842
2843 /* The max range value follows, separated by a BLANK */
2844 if (isBLANK(*l)) {
2845 ++l;
02470786
KW
2846 flags = PERL_SCAN_SILENT_ILLDIGIT
2847 | PERL_SCAN_DISALLOW_PREFIX
2848 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2849 numlen = lend - l;
2850 *max = grok_hex((char *)l, &numlen, &flags, NULL);
2851 if (numlen)
2852 l += numlen;
2853 else /* If no value here, it is a single element range */
2854 *max = *min;
2855
2856 /* Non-binary tables have a third entry: what the first element of the
24303724 2857 * range maps to. The map for those currently read here is in hex */
319009ee
KW
2858 if (wants_value) {
2859 if (isBLANK(*l)) {
2860 ++l;
f2a7d0fc
KW
2861 flags = PERL_SCAN_SILENT_ILLDIGIT
2862 | PERL_SCAN_DISALLOW_PREFIX
2863 | PERL_SCAN_SILENT_NON_PORTABLE;
2864 numlen = lend - l;
2865 *val = grok_hex((char *)l, &numlen, &flags, NULL);
2866 if (numlen)
2867 l += numlen;
2868 else
2869 *val = 0;
319009ee
KW
2870 }
2871 else {
2872 *val = 0;
2873 if (typeto) {
dcbac5bb 2874 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
2875 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
2876 typestr, l);
2877 }
2878 }
2879 }
2880 else
2881 *val = 0; /* bits == 1, then any val should be ignored */
2882 }
2883 else { /* Nothing following range min, should be single element with no
2884 mapping expected */
319009ee
KW
2885 if (wants_value) {
2886 *val = 0;
2887 if (typeto) {
dcbac5bb 2888 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
2889 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
2890 }
2891 }
2892 else
2893 *val = 0; /* bits == 1, then val should be ignored */
2894 }
2895
2896 /* Position to next line if any, or EOF */
2897 if (nl)
2898 l = nl + 1;
2899 else
2900 l = lend;
2901
2902 return l;
2903}
2904
979f2922
ST
2905/* Note:
2906 * Returns a swatch (a bit vector string) for a code point sequence
2907 * that starts from the value C<start> and comprises the number C<span>.
2908 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
2909 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
2910 */
2911STATIC SV*
b0e3252e 2912S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
2913{
2914 SV *swatch;
77f9f126 2915 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 2916 STRLEN lcur, xcur, scur;
ef8f7699 2917 HV *const hv = MUTABLE_HV(SvRV(swash));
5c9f4bd2 2918 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
36eb48b4 2919
88d45d28
KW
2920 SV** listsvp = NULL; /* The string containing the main body of the table */
2921 SV** extssvp = NULL;
2922 SV** invert_it_svp = NULL;
2923 U8* typestr = NULL;
786861f5
KW
2924 STRLEN bits;
2925 STRLEN octets; /* if bits == 1, then octets == 0 */
2926 UV none;
2927 UV end = start + span;
972dd592 2928
36eb48b4 2929 if (invlistsvp == NULL) {
786861f5
KW
2930 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
2931 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
2932 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
2933 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
2934 listsvp = hv_fetchs(hv, "LIST", FALSE);
2935 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
2936
2937 bits = SvUV(*bitssvp);
2938 none = SvUV(*nonesvp);
2939 typestr = (U8*)SvPV_nolen(*typesvp);
2940 }
36eb48b4
KW
2941 else {
2942 bits = 1;
2943 none = 0;
2944 }
786861f5 2945 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 2946
b0e3252e 2947 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 2948
979f2922 2949 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 2950 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 2951 (UV)bits);
979f2922
ST
2952 }
2953
84ea5ef6
KW
2954 /* If overflowed, use the max possible */
2955 if (end < start) {
2956 end = UV_MAX;
2957 span = end - start;
2958 }
2959
979f2922 2960 /* create and initialize $swatch */
979f2922 2961 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
2962 swatch = newSV(scur);
2963 SvPOK_on(swatch);
979f2922
ST
2964 s = (U8*)SvPVX(swatch);
2965 if (octets && none) {
0bd48802 2966 const U8* const e = s + scur;
979f2922
ST
2967 while (s < e) {
2968 if (bits == 8)
2969 *s++ = (U8)(none & 0xff);
2970 else if (bits == 16) {
2971 *s++ = (U8)((none >> 8) & 0xff);
2972 *s++ = (U8)( none & 0xff);
2973 }
2974 else if (bits == 32) {
2975 *s++ = (U8)((none >> 24) & 0xff);
2976 *s++ = (U8)((none >> 16) & 0xff);
2977 *s++ = (U8)((none >> 8) & 0xff);
2978 *s++ = (U8)( none & 0xff);
2979 }
2980 }
2981 *s = '\0';
2982 }
2983 else {
2984 (void)memzero((U8*)s, scur + 1);
2985 }
2986 SvCUR_set(swatch, scur);
2987 s = (U8*)SvPVX(swatch);
2988
36eb48b4
KW
2989 if (invlistsvp) { /* If has an inversion list set up use that */
2990 _invlist_populate_swatch(*invlistsvp, start, end, s);
2991 return swatch;
2992 }
2993
2994 /* read $swash->{LIST} */
979f2922
ST
2995 l = (U8*)SvPV(*listsvp, lcur);
2996 lend = l + lcur;
2997 while (l < lend) {
8ed25d53 2998 UV min, max, val, upper;
95543e92
KW
2999 l = swash_scan_list_line(l, lend, &min, &max, &val,
3000 cBOOL(octets), typestr);
319009ee 3001 if (l > lend) {
979f2922
ST
3002 break;
3003 }
3004
972dd592 3005 /* If looking for something beyond this range, go try the next one */
979f2922
ST
3006 if (max < start)
3007 continue;
3008
8ed25d53
KW
3009 /* <end> is generally 1 beyond where we want to set things, but at the
3010 * platform's infinity, where we can't go any higher, we want to
3011 * include the code point at <end> */
3012 upper = (max < end)
3013 ? max
3014 : (max != UV_MAX || end != UV_MAX)
3015 ? end - 1
3016 : end;
3017
979f2922 3018 if (octets) {
35da51f7 3019 UV key;
979f2922
ST
3020 if (min < start) {
3021 if (!none || val < none) {
3022 val += start - min;
3023 }
3024 min = start;
3025 }
8ed25d53 3026 for (key = min; key <= upper; key++) {
979f2922 3027 STRLEN offset;
979f2922
ST
3028 /* offset must be non-negative (start <= min <= key < end) */
3029 offset = octets * (key - start);
3030 if (bits == 8)
3031 s[offset] = (U8)(val & 0xff);
3032 else if (bits == 16) {
3033 s[offset ] = (U8)((val >> 8) & 0xff);
3034 s[offset + 1] = (U8)( val & 0xff);
3035 }
3036 else if (bits == 32) {
3037 s[offset ] = (U8)((val >> 24) & 0xff);
3038 s[offset + 1] = (U8)((val >> 16) & 0xff);
3039 s[offset + 2] = (U8)((val >> 8) & 0xff);
3040 s[offset + 3] = (U8)( val & 0xff);
3041 }
3042
3043 if (!none || val < none)
3044 ++val;
3045 }
3046 }
711a919c 3047 else { /* bits == 1, then val should be ignored */
35da51f7 3048 UV key;
979f2922
ST
3049 if (min < start)
3050 min = start;
6cb05c12 3051
8ed25d53 3052 for (key = min; key <= upper; key++) {
0bd48802 3053 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3054 s[offset >> 3] |= 1 << (offset & 7);
3055 }
3056 }
3057 } /* while */
979f2922 3058
9479a769 3059 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3060 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3061
3062 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3063 * be 0, and their inversion should also be 0, as we don't succeed any
3064 * Unicode property matches for non-Unicode code points */
3065 if (start <= PERL_UNICODE_MAX) {
3066
3067 /* The code below assumes that we never cross the
3068 * Unicode/above-Unicode boundary in a range, as otherwise we would
3069 * have to figure out where to stop flipping the bits. Since this
3070 * boundary is divisible by a large power of 2, and swatches comes
3071 * in small powers of 2, this should be a valid assumption */
3072 assert(start + span - 1 <= PERL_UNICODE_MAX);
3073
507a8485
KW
3074 send = s + scur;
3075 while (s < send) {
3076 *s = ~(*s);
3077 s++;
3078 }
0bda3001 3079 }
77f9f126
KW
3080 }
3081
d73c39c5
KW
3082 /* read $swash->{EXTRAS}
3083 * This code also copied to swash_to_invlist() below */
979f2922
ST
3084 x = (U8*)SvPV(*extssvp, xcur);
3085 xend = x + xcur;
3086 while (x < xend) {
3087 STRLEN namelen;
3088 U8 *namestr;
3089 SV** othersvp;
3090 HV* otherhv;
3091 STRLEN otherbits;
3092 SV **otherbitssvp, *other;
711a919c 3093 U8 *s, *o, *nl;
979f2922
ST
3094 STRLEN slen, olen;
3095
35da51f7 3096 const U8 opc = *x++;
979f2922
ST
3097 if (opc == '\n')
3098 continue;
3099
3100 nl = (U8*)memchr(x, '\n', xend - x);
3101
3102 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3103 if (nl) {
3104 x = nl + 1; /* 1 is length of "\n" */
3105 continue;
3106 }
3107 else {
3108 x = xend; /* to EXTRAS' end at which \n is not found */
3109 break;
3110 }
3111 }
3112
3113 namestr = x;
3114 if (nl) {
3115 namelen = nl - namestr;
3116 x = nl + 1;
3117 }
3118 else {
3119 namelen = xend - namestr;
3120 x = xend;
3121 }
3122
3123 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3124 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3125 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3126 otherbits = (STRLEN)SvUV(*otherbitssvp);
3127 if (bits < otherbits)
5637ef5b
NC
3128 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3129 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3130
3131 /* The "other" swatch must be destroyed after. */
b0e3252e 3132 other = swatch_get(*othersvp, start, span);
979f2922
ST
3133 o = (U8*)SvPV(other, olen);
3134
3135 if (!olen)
b0e3252e 3136 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3137
3138 s = (U8*)SvPV(swatch, slen);
3139 if (bits == 1 && otherbits == 1) {
3140 if (slen != olen)
5637ef5b
NC
3141 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3142 "mismatch, slen=%"UVuf", olen=%"UVuf,
3143 (UV)slen, (UV)olen);
979f2922
ST
3144
3145 switch (opc) {
3146 case '+':
3147 while (slen--)
3148 *s++ |= *o++;
3149 break;
3150 case '!':
3151 while (slen--)
3152 *s++ |= ~*o++;
3153 break;
3154 case '-':
3155 while (slen--)
3156 *s++ &= ~*o++;
3157 break;
3158 case '&':
3159 while (slen--)
3160 *s++ &= *o++;
3161 break;
3162 default:
3163 break;
3164 }
3165 }
711a919c 3166 else {
979f2922
ST
3167 STRLEN otheroctets = otherbits >> 3;
3168 STRLEN offset = 0;
35da51f7 3169 U8* const send = s + slen;
979f2922
ST
3170
3171 while (s < send) {
3172 UV otherval = 0;
3173
3174 if (otherbits == 1) {
3175 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3176 ++offset;
3177 }
3178 else {
3179 STRLEN vlen = otheroctets;
3180 otherval = *o++;
3181 while (--vlen) {
3182 otherval <<= 8;
3183 otherval |= *o++;
3184 }
3185 }
3186
711a919c 3187 if (opc == '+' && otherval)
6f207bd3 3188 NOOP; /* replace with otherval */
979f2922
ST
3189 else if (opc == '!' && !otherval)
3190 otherval = 1;
3191 else if (opc == '-' && otherval)
3192 otherval = 0;
3193 else if (opc == '&' && !otherval)
3194 otherval = 0;
3195 else {
711a919c 3196 s += octets; /* no replacement */
979f2922
ST
3197 continue;
3198 }
3199
3200 if (bits == 8)
3201 *s++ = (U8)( otherval & 0xff);
3202 else if (bits == 16) {
3203 *s++ = (U8)((otherval >> 8) & 0xff);
3204 *s++ = (U8)( otherval & 0xff);
3205 }
3206 else if (bits == 32) {
3207 *s++ = (U8)((otherval >> 24) & 0xff);
3208 *s++ = (U8)((otherval >> 16) & 0xff);
3209 *s++ = (U8)((otherval >> 8) & 0xff);
3210 *s++ = (U8)( otherval & 0xff);
3211 }
3212 }
3213 }
3214 sv_free(other); /* through with it! */
3215 } /* while */
3216 return swatch;
3217}
3218
064c021d 3219HV*
4c2e1131 3220Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3221{
3222
79a2a0e8 3223 /* Subject to change or removal. For use only in regcomp.c and regexec.c
5662e334
KW
3224 * Can't be used on a property that is subject to user override, as it
3225 * relies on the value of SPECIALS in the swash which would be set by
3226 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3227 * for overridden properties
064c021d
KW
3228 *
3229 * Returns a hash which is the inversion and closure of a swash mapping.
3230 * For example, consider the input lines:
3231 * 004B 006B
3232 * 004C 006C
3233 * 212A 006B
3234 *
3235 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3236 * 006C. The value for each key is an array. For 006C, the array would
17d5c697 3237 * have two elements, the utf8 for itself, and for 004C. For 006B, there
064c021d
KW
3238 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3239 *
538e84ed
KW
3240 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3241 * keys are only code points that are folded-to, so it isn't a full closure.
3242 *
064c021d
KW
3243 * Essentially, for any code point, it gives all the code points that map to
3244 * it, or the list of 'froms' for that point.
3245 *
5662e334
KW
3246 * Currently it ignores any additions or deletions from other swashes,
3247 * looking at just the main body of the swash, and if there are SPECIALS
3248 * in the swash, at that hash
3249 *
3250 * The specials hash can be extra code points, and most likely consists of
3251 * maps from single code points to multiple ones (each expressed as a string
3252 * of utf8 characters). This function currently returns only 1-1 mappings.
3253 * However consider this possible input in the specials hash:
3254 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3255 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3256 *
3257 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3258 * currently handle. But it also means that FB05 and FB06 are equivalent in
3259 * a 1-1 mapping which we should handle, and this relationship may not be in
3260 * the main table. Therefore this function examines all the multi-char
3261 * sequences and adds the 1-1 mappings that come out of that. */
064c021d
KW
3262
3263 U8 *l, *lend;
3264 STRLEN lcur;
3265 HV *const hv = MUTABLE_HV(SvRV(swash));
3266
923b6d4e
KW
3267 /* The string containing the main body of the table. This will have its
3268 * assertion fail if the swash has been converted to its inversion list */
064c021d
KW
3269 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3270
3271 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3272 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3273 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3274 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3275 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3276 const STRLEN bits = SvUV(*bitssvp);
3277 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3278 const UV none = SvUV(*nonesvp);
5662e334 3279 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
064c021d
KW
3280
3281 HV* ret = newHV();
3282
3283 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3284
3285 /* Must have at least 8 bits to get the mappings */
3286 if (bits != 8 && bits != 16 && bits != 32) {
3287 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3288 (UV)bits);
3289 }
3290
5662e334
KW
3291 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3292 mapping to more than one character */
3293
3294 /* Construct an inverse mapping hash for the specials */
3295 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3296 HV * specials_inverse = newHV();
3297 char *char_from; /* the lhs of the map */
3298 I32 from_len; /* its byte length */
3299 char *char_to; /* the rhs of the map */
3300 I32 to_len; /* its byte length */
3301 SV *sv_to; /* and in a sv */
3302 AV* from_list; /* list of things that map to each 'to' */
3303
3304 hv_iterinit(specials_hv);
3305
3306 /* The keys are the characters (in utf8) that map to the corresponding
3307 * utf8 string value. Iterate through the list creating the inverse
3308 * list. */
3309 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3310 SV** listp;
3311 if (! SvPOK(sv_to)) {
5637ef5b
NC
3312 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3313 "unexpectedly is not a string, flags=%lu",
3314 (unsigned long)SvFLAGS(sv_to));
5662e334 3315 }
4b88fb76 3316 /*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
3317
3318 /* Each key in the inverse list is a mapped-to value, and the key's
3319 * hash value is a list of the strings (each in utf8) that map to
3320 * it. Those strings are all one character long */
3321 if ((listp = hv_fetch(specials_inverse,
3322 SvPVX(sv_to),
3323 SvCUR(sv_to), 0)))
3324 {
3325 from_list = (AV*) *listp;
3326 }
3327 else { /* No entry yet for it: create one */
3328 from_list = newAV();
3329 if (! hv_store(specials_inverse,
3330 SvPVX(sv_to),
3331 SvCUR(sv_to),
3332 (SV*) from_list, 0))
3333 {
3334 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3335 }
3336 }
3337
3338 /* Here have the list associated with this 'to' (perhaps newly
3339 * created and empty). Just add to it. Note that we ASSUME that
3340 * the input is guaranteed to not have duplications, so we don't
3341 * check for that. Duplications just slow down execution time. */
3342 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3343 }
3344
3345 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3346 * it looking for cases like the FB05/FB06 examples above. There would
3347 * be an entry in the hash like
3348 * 'st' => [ FB05, FB06 ]
3349 * In this example we will create two lists that get stored in the
3350 * returned hash, 'ret':
3351 * FB05 => [ FB05, FB06 ]
3352 * FB06 => [ FB05, FB06 ]
3353 *
3354 * Note that there is nothing to do if the array only has one element.
3355 * (In the normal 1-1 case handled below, we don't have to worry about
3356 * two lists, as everything gets tied to the single list that is
3357 * generated for the single character 'to'. But here, we are omitting
3358 * that list, ('st' in the example), so must have multiple lists.) */
3359 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3360 &char_to, &to_len)))
3361 {
b9f2b683 3362 if (av_tindex(from_list) > 0) {
c70927a6 3363 SSize_t i;
5662e334
KW
3364
3365 /* We iterate over all combinations of i,j to place each code
3366 * point on each list */
b9f2b683 3367 for (i = 0; i <= av_tindex(from_list); i++) {
c70927a6 3368 SSize_t j;
5662e334
KW
3369 AV* i_list = newAV();
3370 SV** entryp = av_fetch(from_list, i, FALSE);
3371 if (entryp == NULL) {
3372 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3373 }
3374 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3375 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3376 }
3377 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3378 (SV*) i_list, FALSE))
3379 {
3380 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3381 }
3382
538e84ed 3383 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
b9f2b683 3384 for (j = 0; j <= av_tindex(from_list); j++) {
5662e334
KW
3385 entryp = av_fetch(from_list, j, FALSE);
3386 if (entryp == NULL) {
3387 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3388 }
3389
3390 /* When i==j this adds itself to the list */
4b88fb76
KW
3391 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3392 (U8*) SvPVX(*entryp),
3393 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3394 0)));
4637d003 3395 /*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
3396 }
3397 }
3398 }
3399 }
3400 SvREFCNT_dec(specials_inverse); /* done with it */
3401 } /* End of specials */
3402
064c021d
KW
3403 /* read $swash->{LIST} */
3404 l = (U8*)SvPV(*listsvp, lcur);
3405 lend = l + lcur;
3406
3407 /* Go through each input line */
3408 while (l < lend) {
3409 UV min, max, val;
3410 UV inverse;
95543e92
KW
3411 l = swash_scan_list_line(l, lend, &min, &max, &val,
3412 cBOOL(octets), typestr);
064c021d
KW
3413 if (l > lend) {
3414 break;
3415 }
3416
3417 /* Each element in the range is to be inverted */
3418 for (inverse = min; inverse <= max; inverse++) {
3419 AV* list;
064c021d
KW
3420 SV** listp;
3421 IV i;
3422 bool found_key = FALSE;
5662e334 3423 bool found_inverse = FALSE;
064c021d
KW
3424
3425 /* The key is the inverse mapping */
3426 char key[UTF8_MAXBYTES+1];
c80e42f3 3427 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
064c021d
KW
3428 STRLEN key_len = key_end - key;
3429
064c021d
KW
3430 /* Get the list for the map */
3431 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3432 list = (AV*) *listp;
3433 }
3434 else { /* No entry yet for it: create one */
3435 list = newAV();
3436 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3437 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3438 }
3439 }
3440
5662e334
KW
3441 /* Look through list to see if this inverse mapping already is
3442 * listed, or if there is a mapping to itself already */
b9f2b683 3443 for (i = 0; i <= av_tindex(list); i++) {
064c021d
KW
3444 SV** entryp = av_fetch(list, i, FALSE);
3445 SV* entry;
414db8a1 3446 UV uv;
064c021d
KW
3447 if (entryp == NULL) {
3448 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3449 }
3450 entry = *entryp;
414db8a1
DM
3451 uv = SvUV(entry);
3452 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, uv));*/
3453 if (uv == val) {
064c021d 3454 found_key = TRUE;
5662e334 3455 }
414db8a1 3456 if (uv == inverse) {
5662e334
KW
3457 found_inverse = TRUE;
3458 }
3459
3460 /* No need to continue searching if found everything we are
3461 * looking for */
3462 if (found_key && found_inverse) {
064c021d
KW
3463 break;
3464 }
3465 }
56ca34ca
KW
3466
3467 /* Make sure there is a mapping to itself on the list */
064c021d 3468 if (! found_key) {
d397ff6a 3469 av_push(list, newSVuv(val));
4637d003 3470 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
064c021d
KW
3471 }
3472
3473
3474 /* Simply add the value to the list */
5662e334
KW
3475 if (! found_inverse) {
3476 av_push(list, newSVuv(inverse));
4637d003 3477 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
5662e334 3478 }
064c021d 3479
b0e3252e 3480 /* swatch_get() increments the value of val for each element in the
064c021d
KW
3481 * range. That makes more compact tables possible. You can
3482 * express the capitalization, for example, of all consecutive
3483 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3484 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
bd3f2f94 3485 * and it's not documented; it appears to be used only in
b0e3252e 3486 * implementing tr//; I copied the semantics from swatch_get(), just
bd3f2f94 3487 * in case */
064c021d
KW
3488 if (!none || val < none) {
3489 ++val;
3490 }
3491 }
3492 }
3493
3494 return ret;
3495}
3496
a25abddc 3497SV*
d764b54e
KW
3498Perl__swash_to_invlist(pTHX_ SV* const swash)
3499{
3500
ed92f1b3
KW
3501 /* Subject to change or removal. For use only in one place in regcomp.c.
3502 * Ownership is given to one reference count in the returned SV* */
d764b54e
KW
3503
3504 U8 *l, *lend;
3505 char *loc;
3506 STRLEN lcur;
3507 HV *const hv = MUTABLE_HV(SvRV(swash));
3508 UV elements = 0; /* Number of elements in the inversion list */
b443038a 3509 U8 empty[] = "";
923b6d4e
KW
3510 SV** listsvp;
3511 SV** typesvp;
3512 SV** bitssvp;
3513 SV** extssvp;
3514 SV** invert_it_svp;
d764b54e 3515
923b6d4e
KW
3516 U8* typestr;
3517 STRLEN bits;
3518 STRLEN octets; /* if bits == 1, then octets == 0 */
d73c39c5
KW
3519 U8 *x, *xend;
3520 STRLEN xcur;
d764b54e 3521
a25abddc 3522 SV* invlist;
d764b54e 3523
b81740c0
KW
3524 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3525
923b6d4e
KW
3526 /* If not a hash, it must be the swash's inversion list instead */
3527 if (SvTYPE(hv) != SVt_PVHV) {
ed92f1b3 3528 return SvREFCNT_inc_simple_NN((SV*) hv);
923b6d4e
KW
3529 }
3530
3531 /* The string containing the main body of the table */
3532 listsvp = hv_fetchs(hv, "LIST", FALSE);
3533 typesvp = hv_fetchs(hv, "TYPE", FALSE);
3534 bitssvp = hv_fetchs(hv, "BITS", FALSE);
3535 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3536 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3537
3538 typestr = (U8*)SvPV_nolen(*typesvp);
3539 bits = SvUV(*bitssvp);
3540 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3541
d764b54e 3542 /* read $swash->{LIST} */
b443038a
KW
3543 if (SvPOK(*listsvp)) {
3544 l = (U8*)SvPV(*listsvp, lcur);
3545 }
3546 else {
3547 /* LIST legitimately doesn't contain a string during compilation phases
3548 * of Perl itself, before the Unicode tables are generated. In this
3549 * case, just fake things up by creating an empty list */
3550 l = empty;
3551 lcur = 0;
3552 }
d764b54e
KW
3553 loc = (char *) l;
3554 lend = l + lcur;
3555
31aa6e0b 3556 if (*l == 'V') { /* Inversion list format */
99944878 3557 const char *after_atou = (char *) lend;
31aa6e0b
KW
3558 UV element0;
3559 UV* other_elements_ptr;
3560
3561 /* The first number is a count of the rest */
3562 l++;
99944878 3563 elements = grok_atou((const char *)l, &after_atou);
eb092534
KW
3564 if (elements == 0) {
3565 invlist = _new_invlist(0);
3566 }
3567 else {
99944878
JH
3568 while (isSPACE(*l)) l++;
3569 l = (U8 *) after_atou;
1f9f7d4c
KW
3570
3571 /* Get the 0th element, which is needed to setup the inversion list */
99944878
JH
3572 while (isSPACE(*l)) l++;
3573 element0 = (UV) grok_atou((const char *)l, &after_atou);
3574 l = (U8 *) after_atou;
1f9f7d4c
KW
3575 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
3576 elements--;
3577
3578 /* Then just populate the rest of the input */
3579 while (elements-- > 0) {
3580 if (l > lend) {
3581 Perl_croak(aTHX_ "panic: Expecting %"UVuf" more elements than available", elements);
3582 }
99944878
JH
3583 while (isSPACE(*l)) l++;
3584 *other_elements_ptr++ = (UV) grok_atou((const char *)l, &after_atou);
3585 l = (U8 *) after_atou;
1f9f7d4c 3586 }
eb092534 3587 }
31aa6e0b
KW
3588 }
3589 else {
3590
1784d2f9
KW
3591 /* Scan the input to count the number of lines to preallocate array
3592 * size based on worst possible case, which is each line in the input
3593 * creates 2 elements in the inversion list: 1) the beginning of a
3594 * range in the list; 2) the beginning of a range not in the list. */
3595 while ((loc = (strchr(loc, '\n'))) != NULL) {
3596 elements += 2;
3597 loc++;
3598 }
d764b54e 3599
1784d2f9
KW
3600 /* If the ending is somehow corrupt and isn't a new line, add another
3601 * element for the final range that isn't in the inversion list */
3602 if (! (*lend == '\n'
3603 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
3604 {
3605 elements++;
3606 }
d764b54e 3607
1784d2f9 3608 invlist = _new_invlist(elements);
d764b54e 3609
1784d2f9
KW
3610 /* Now go through the input again, adding each range to the list */
3611 while (l < lend) {
3612 UV start, end;
3613 UV val; /* Not used by this function */
d764b54e 3614
95543e92
KW
3615 l = swash_scan_list_line(l, lend, &start, &end, &val,
3616 cBOOL(octets), typestr);
d764b54e 3617
1784d2f9
KW
3618 if (l > lend) {
3619 break;
3620 }
3621
3622 invlist = _add_range_to_invlist(invlist, start, end);
3623 }
31aa6e0b 3624 }
d764b54e 3625
77f9f126
KW
3626 /* Invert if the data says it should be */
3627 if (invert_it_svp && SvUV(*invert_it_svp)) {
25151030 3628 _invlist_invert(invlist);
77f9f126
KW
3629 }
3630
b0e3252e 3631 /* This code is copied from swatch_get()
d73c39c5
KW
3632 * read $swash->{EXTRAS} */
3633 x = (U8*)SvPV(*extssvp, xcur);
3634 xend = x + xcur;
3635 while (x < xend) {
3636 STRLEN namelen;
3637 U8 *namestr;
3638 SV** othersvp;
3639 HV* otherhv;
3640 STRLEN otherbits;
3641 SV **otherbitssvp, *other;
3642 U8 *nl;
3643
3644 const U8 opc = *x++;
3645 if (opc == '\n')
3646 continue;
3647
3648 nl = (U8*)memchr(x, '\n', xend - x);
3649
3650 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3651 if (nl) {
3652 x = nl + 1; /* 1 is length of "\n" */
3653 continue;
3654 }
3655 else {
3656 x = xend; /* to EXTRAS' end at which \n is not found */
3657 break;
3658 }
3659 }
3660
3661 namestr = x;
3662 if (nl) {
3663 namelen = nl - namestr;
3664 x = nl + 1;
3665 }
3666 else {
3667 namelen = xend - namestr;
3668 x = xend;
3669 }
3670
3671 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3672 otherhv = MUTABLE_HV(SvRV(*othersvp));
3673 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3674 otherbits = (STRLEN)SvUV(*otherbitssvp);
3675
3676 if (bits != otherbits || bits != 1) {
5637ef5b
NC
3677 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
3678 "properties, bits=%"UVuf", otherbits=%"UVuf,
3679 (UV)bits, (UV)otherbits);
d73c39c5
KW
3680 }
3681