Commit | Line | Data |
---|---|---|
a0ed51b3 LW |
1 | /* utf8.c |
2 | * | |
1129b882 | 3 | * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 |
b94e2f88 | 4 | * by Larry Wall and others |
a0ed51b3 LW |
5 | * |
6 | * You may distribute under the terms of either the GNU General Public | |
7 | * License or the Artistic License, as specified in the README file. | |
8 | * | |
9 | */ | |
10 | ||
11 | /* | |
4ac71550 TC |
12 | * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever |
13 | * heard of that we don't want to see any closer; and that's the one place | |
14 | * we're trying to get to! And that's just where we can't get, nohow.' | |
15 | * | |
cdad3b53 | 16 | * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"] |
a0ed51b3 LW |
17 | * |
18 | * 'Well do I understand your speech,' he answered in the same language; | |
19 | * 'yet few strangers do so. Why then do you not speak in the Common Tongue, | |
4ac71550 | 20 | * as is the custom in the West, if you wish to be answered?' |
cdad3b53 | 21 | * --Gandalf, addressing Théoden's door wardens |
4ac71550 TC |
22 | * |
23 | * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"] | |
a0ed51b3 LW |
24 | * |
25 | * ...the travellers perceived that the floor was paved with stones of many | |
26 | * hues; branching runes and strange devices intertwined beneath their feet. | |
4ac71550 TC |
27 | * |
28 | * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"] | |
a0ed51b3 LW |
29 | */ |
30 | ||
31 | #include "EXTERN.h" | |
864dbfa3 | 32 | #define PERL_IN_UTF8_C |
a0ed51b3 | 33 | #include "perl.h" |
b992490d | 34 | #include "invlist_inline.h" |
a0ed51b3 | 35 | |
806547a7 | 36 | static const char malformed_text[] = "Malformed UTF-8 character"; |
27da23d5 | 37 | static const char unees[] = |
806547a7 | 38 | "Malformed UTF-8 character (unexpected end of string)"; |
fb7e7255 KW |
39 | |
40 | /* Be sure to synchronize this message with the similar one in regcomp.c */ | |
760c7c2f | 41 | static const char cp_above_legal_max[] = |
d22ec717 KW |
42 | "Use of code point 0x%" UVXf " is not allowed; the" |
43 | " permissible max is 0x%" UVXf; | |
760c7c2f | 44 | |
48ef279e | 45 | /* |
ccfc67b7 | 46 | =head1 Unicode Support |
7fefc6c1 | 47 | These are various utility functions for manipulating UTF8-encoded |
72d33970 | 48 | strings. For the uninitiated, this is a method of representing arbitrary |
61296642 | 49 | Unicode characters as a variable number of bytes, in such a way that |
56da48f7 DM |
50 | characters in the ASCII range are unmodified, and a zero byte never appears |
51 | within non-zero characters. | |
166f8a29 | 52 | |
eaf7a4d2 CS |
53 | =cut |
54 | */ | |
55 | ||
9cbfb8ab KW |
56 | void |
57 | Perl__force_out_malformed_utf8_message(pTHX_ | |
58 | const U8 *const p, /* First byte in UTF-8 sequence */ | |
59 | const U8 * const e, /* Final byte in sequence (may include | |
60 | multiple chars */ | |
61 | const U32 flags, /* Flags to pass to utf8n_to_uvchr(), | |
62 | usually 0, or some DISALLOW flags */ | |
63 | const bool die_here) /* If TRUE, this function does not return */ | |
64 | { | |
65 | /* This core-only function is to be called when a malformed UTF-8 character | |
66 | * is found, in order to output the detailed information about the | |
67 | * malformation before dieing. The reason it exists is for the occasions | |
68 | * when such a malformation is fatal, but warnings might be turned off, so | |
69 | * that normally they would not be actually output. This ensures that they | |
70 | * do get output. Because a sequence may be malformed in more than one | |
71 | * way, multiple messages may be generated, so we can't make them fatal, as | |
72 | * that would cause the first one to die. | |
73 | * | |
74 | * Instead we pretend -W was passed to perl, then die afterwards. The | |
75 | * flexibility is here to return to the caller so they can finish up and | |
76 | * die themselves */ | |
77 | U32 errors; | |
78 | ||
79 | PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE; | |
80 | ||
81 | ENTER; | |
c15a80f3 | 82 | SAVEI8(PL_dowarn); |
9cbfb8ab KW |
83 | SAVESPTR(PL_curcop); |
84 | ||
85 | PL_dowarn = G_WARN_ALL_ON|G_WARN_ON; | |
86 | if (PL_curcop) { | |
87 | PL_curcop->cop_warnings = pWARN_ALL; | |
88 | } | |
89 | ||
90 | (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors); | |
91 | ||
92 | LEAVE; | |
93 | ||
94 | if (! errors) { | |
95 | Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should" | |
96 | " be called only when there are errors found"); | |
97 | } | |
98 | ||
99 | if (die_here) { | |
100 | Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)"); | |
101 | } | |
102 | } | |
103 | ||
bb07812e KW |
104 | STATIC HV * |
105 | S_new_msg_hv(pTHX_ const char * const message, /* The message text */ | |
106 | U32 categories, /* Packed warning categories */ | |
107 | U32 flag) /* Flag associated with this message */ | |
108 | { | |
109 | /* Creates, populates, and returns an HV* that describes an error message | |
110 | * for the translators between UTF8 and code point */ | |
111 | ||
112 | SV* msg_sv = newSVpv(message, 0); | |
113 | SV* category_sv = newSVuv(categories); | |
114 | SV* flag_bit_sv = newSVuv(flag); | |
115 | ||
116 | HV* msg_hv = newHV(); | |
117 | ||
118 | PERL_ARGS_ASSERT_NEW_MSG_HV; | |
119 | ||
2b672cf5 KW |
120 | (void) hv_stores(msg_hv, "text", msg_sv); |
121 | (void) hv_stores(msg_hv, "warn_categories", category_sv); | |
122 | (void) hv_stores(msg_hv, "flag_bit", flag_bit_sv); | |
bb07812e KW |
123 | |
124 | return msg_hv; | |
125 | } | |
126 | ||
eaf7a4d2 | 127 | /* |
378516de | 128 | =for apidoc uvoffuni_to_utf8_flags |
eebe1485 | 129 | |
a27992cc | 130 | THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. |
de69f3af KW |
131 | Instead, B<Almost all code should use L</uvchr_to_utf8> or |
132 | L</uvchr_to_utf8_flags>>. | |
a27992cc | 133 | |
de69f3af KW |
134 | This function is like them, but the input is a strict Unicode |
135 | (as opposed to native) code point. Only in very rare circumstances should code | |
136 | not be using the native code point. | |
949cf498 | 137 | |
efa9cd84 | 138 | For details, see the description for L</uvchr_to_utf8_flags>. |
949cf498 | 139 | |
eebe1485 SC |
140 | =cut |
141 | */ | |
142 | ||
33f38593 KW |
143 | U8 * |
144 | Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags) | |
145 | { | |
146 | PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS; | |
147 | ||
148 | return uvoffuni_to_utf8_flags_msgs(d, uv, flags, NULL); | |
149 | } | |
150 | ||
c94c2f39 KW |
151 | /* All these formats take a single UV code point argument */ |
152 | const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf; | |
153 | const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf | |
154 | " is not recommended for open interchange"; | |
155 | const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode," | |
156 | " may not be portable"; | |
57ff5f59 KW |
157 | const char perl_extended_cp_format[] = "Code point 0x%" UVXf " is not" \ |
158 | " Unicode, requires a Perl extension," \ | |
159 | " and so is not portable"; | |
c94c2f39 | 160 | |
33f38593 | 161 | #define HANDLE_UNICODE_SURROGATE(uv, flags, msgs) \ |
8ee1cdcb KW |
162 | STMT_START { \ |
163 | if (flags & UNICODE_WARN_SURROGATE) { \ | |
33f38593 KW |
164 | U32 category = packWARN(WARN_SURROGATE); \ |
165 | const char * format = surrogate_cp_format; \ | |
166 | if (msgs) { \ | |
167 | *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \ | |
168 | category, \ | |
169 | UNICODE_GOT_SURROGATE); \ | |
170 | } \ | |
171 | else { \ | |
172 | Perl_ck_warner_d(aTHX_ category, format, uv); \ | |
173 | } \ | |
8ee1cdcb KW |
174 | } \ |
175 | if (flags & UNICODE_DISALLOW_SURROGATE) { \ | |
176 | return NULL; \ | |
177 | } \ | |
178 | } STMT_END; | |
179 | ||
33f38593 | 180 | #define HANDLE_UNICODE_NONCHAR(uv, flags, msgs) \ |
8ee1cdcb KW |
181 | STMT_START { \ |
182 | if (flags & UNICODE_WARN_NONCHAR) { \ | |
33f38593 KW |
183 | U32 category = packWARN(WARN_NONCHAR); \ |
184 | const char * format = nonchar_cp_format; \ | |
185 | if (msgs) { \ | |
186 | *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \ | |
187 | category, \ | |
188 | UNICODE_GOT_NONCHAR); \ | |
189 | } \ | |
190 | else { \ | |
191 | Perl_ck_warner_d(aTHX_ category, format, uv); \ | |
192 | } \ | |
8ee1cdcb KW |
193 | } \ |
194 | if (flags & UNICODE_DISALLOW_NONCHAR) { \ | |
195 | return NULL; \ | |
196 | } \ | |
197 | } STMT_END; | |
198 | ||
ba6ed43c KW |
199 | /* Use shorter names internally in this file */ |
200 | #define SHIFT UTF_ACCUMULATION_SHIFT | |
201 | #undef MARK | |
202 | #define MARK UTF_CONTINUATION_MARK | |
203 | #define MASK UTF_CONTINUATION_MASK | |
204 | ||
33f38593 KW |
205 | /* |
206 | =for apidoc uvchr_to_utf8_flags_msgs | |
207 | ||
208 | THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. | |
209 | ||
210 | Most code should use C<L</uvchr_to_utf8_flags>()> rather than call this directly. | |
211 | ||
212 | This function is for code that wants any warning and/or error messages to be | |
213 | returned to the caller rather than be displayed. All messages that would have | |
884a31ee | 214 | been displayed if all lexical warnings are enabled will be returned. |
33f38593 KW |
215 | |
216 | It is just like C<L</uvchr_to_utf8_flags>> but it takes an extra parameter | |
217 | placed after all the others, C<msgs>. If this parameter is 0, this function | |
218 | behaves identically to C<L</uvchr_to_utf8_flags>>. Otherwise, C<msgs> should | |
219 | be a pointer to an C<HV *> variable, in which this function creates a new HV to | |
220 | contain any appropriate messages. The hash has three key-value pairs, as | |
221 | follows: | |
222 | ||
223 | =over 4 | |
224 | ||
225 | =item C<text> | |
226 | ||
227 | The text of the message as a C<SVpv>. | |
228 | ||
229 | =item C<warn_categories> | |
230 | ||
231 | The warning category (or categories) packed into a C<SVuv>. | |
232 | ||
233 | =item C<flag> | |
234 | ||
235 | A single flag bit associated with this message, in a C<SVuv>. | |
236 | The bit corresponds to some bit in the C<*errors> return value, | |
237 | such as C<UNICODE_GOT_SURROGATE>. | |
238 | ||
239 | =back | |
240 | ||
241 | It's important to note that specifying this parameter as non-null will cause | |
242 | any warnings this function would otherwise generate to be suppressed, and | |
243 | instead be placed in C<*msgs>. The caller can check the lexical warnings state | |
244 | (or not) when choosing what to do with the returned messages. | |
245 | ||
246 | The caller, of course, is responsible for freeing any returned HV. | |
247 | ||
248 | =cut | |
249 | */ | |
250 | ||
251 | /* Undocumented; we don't want people using this. Instead they should use | |
252 | * uvchr_to_utf8_flags_msgs() */ | |
dfe13c55 | 253 | U8 * |
33f38593 | 254 | Perl_uvoffuni_to_utf8_flags_msgs(pTHX_ U8 *d, UV uv, const UV flags, HV** msgs) |
a0ed51b3 | 255 | { |
33f38593 KW |
256 | PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS_MSGS; |
257 | ||
258 | if (msgs) { | |
259 | *msgs = NULL; | |
260 | } | |
7918f24d | 261 | |
2d1545e5 | 262 | if (OFFUNI_IS_INVARIANT(uv)) { |
4c8cd605 | 263 | *d++ = LATIN1_TO_NATIVE(uv); |
d9432125 KW |
264 | return d; |
265 | } | |
facc1dc2 | 266 | |
3ea68d71 | 267 | if (uv <= MAX_UTF8_TWO_BYTE) { |
facc1dc2 KW |
268 | *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2)); |
269 | *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK); | |
3ea68d71 KW |
270 | return d; |
271 | } | |
d9432125 | 272 | |
ba6ed43c KW |
273 | /* Not 2-byte; test for and handle 3-byte result. In the test immediately |
274 | * below, the 16 is for start bytes E0-EF (which are all the possible ones | |
275 | * for 3 byte characters). The 2 is for 2 continuation bytes; these each | |
276 | * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000 | |
277 | * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC; | |
278 | * 0x800-0xFFFF on ASCII */ | |
279 | if (uv < (16 * (1U << (2 * SHIFT)))) { | |
280 | *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3)); | |
281 | *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK); | |
282 | *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK); | |
283 | ||
284 | #ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so | |
285 | aren't tested here */ | |
286 | /* The most likely code points in this range are below the surrogates. | |
287 | * Do an extra test to quickly exclude those. */ | |
288 | if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) { | |
289 | if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv) | |
290 | || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) | |
291 | { | |
33f38593 | 292 | HANDLE_UNICODE_NONCHAR(uv, flags, msgs); |
8ee1cdcb KW |
293 | } |
294 | else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) { | |
33f38593 | 295 | HANDLE_UNICODE_SURROGATE(uv, flags, msgs); |
760c7c2f | 296 | } |
ba6ed43c KW |
297 | } |
298 | #endif | |
299 | return d; | |
300 | } | |
301 | ||
302 | /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII | |
303 | * platforms, and 0x4000 on EBCDIC. There are problematic cases that can | |
304 | * happen starting with 4-byte characters on ASCII platforms. We unify the | |
305 | * code for these with EBCDIC, even though some of them require 5-bytes on | |
306 | * those, because khw believes the code saving is worth the very slight | |
307 | * performance hit on these high EBCDIC code points. */ | |
308 | ||
309 | if (UNLIKELY(UNICODE_IS_SUPER(uv))) { | |
40606899 KW |
310 | if (UNLIKELY(uv > MAX_LEGAL_CP)) { |
311 | Perl_croak(aTHX_ cp_above_legal_max, uv, MAX_LEGAL_CP); | |
a5bf80e0 | 312 | } |
33f38593 KW |
313 | if ( (flags & UNICODE_WARN_SUPER) |
314 | || ( (flags & UNICODE_WARN_PERL_EXTENDED) | |
0a8a1a5b | 315 | && UNICODE_IS_PERL_EXTENDED(uv))) |
a5bf80e0 | 316 | { |
33f38593 KW |
317 | const char * format = super_cp_format; |
318 | U32 category = packWARN(WARN_NON_UNICODE); | |
319 | U32 flag = UNICODE_GOT_SUPER; | |
320 | ||
321 | /* Choose the more dire applicable warning */ | |
322 | if (UNICODE_IS_PERL_EXTENDED(uv)) { | |
323 | format = perl_extended_cp_format; | |
324 | if (flags & (UNICODE_WARN_PERL_EXTENDED | |
325 | |UNICODE_DISALLOW_PERL_EXTENDED)) | |
326 | { | |
327 | flag = UNICODE_GOT_PERL_EXTENDED; | |
328 | } | |
329 | } | |
a5bf80e0 | 330 | |
33f38593 KW |
331 | if (msgs) { |
332 | *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), | |
333 | category, flag); | |
334 | } | |
335 | else { | |
336 | Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE), format, uv); | |
337 | } | |
a5bf80e0 | 338 | } |
56576a04 | 339 | if ( (flags & UNICODE_DISALLOW_SUPER) |
0a8a1a5b KW |
340 | || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED) |
341 | && UNICODE_IS_PERL_EXTENDED(uv))) | |
a5bf80e0 KW |
342 | { |
343 | return NULL; | |
344 | } | |
345 | } | |
ba6ed43c | 346 | else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) { |
33f38593 | 347 | HANDLE_UNICODE_NONCHAR(uv, flags, msgs); |
507b9800 | 348 | } |
d9432125 | 349 | |
ba6ed43c KW |
350 | /* Test for and handle 4-byte result. In the test immediately below, the |
351 | * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte | |
352 | * characters). The 3 is for 3 continuation bytes; these each contribute | |
353 | * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on | |
354 | * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC; | |
355 | * 0x1_0000-0x1F_FFFF on ASCII */ | |
356 | if (uv < (8 * (1U << (3 * SHIFT)))) { | |
357 | *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4)); | |
358 | *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK); | |
359 | *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK); | |
360 | *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK); | |
361 | ||
362 | #ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte | |
363 | characters. The end-plane non-characters for EBCDIC were | |
364 | handled just above */ | |
365 | if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) { | |
33f38593 | 366 | HANDLE_UNICODE_NONCHAR(uv, flags, msgs); |
d528804a | 367 | } |
ba6ed43c | 368 | else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) { |
33f38593 | 369 | HANDLE_UNICODE_SURROGATE(uv, flags, msgs); |
ba6ed43c KW |
370 | } |
371 | #endif | |
372 | ||
373 | return d; | |
374 | } | |
375 | ||
376 | /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII | |
377 | * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop | |
378 | * format. The unrolled version above turns out to not save all that much | |
379 | * time, and at these high code points (well above the legal Unicode range | |
380 | * on ASCII platforms, and well above anything in common use in EBCDIC), | |
381 | * khw believes that less code outweighs slight performance gains. */ | |
382 | ||
d9432125 | 383 | { |
5aaebcb3 | 384 | STRLEN len = OFFUNISKIP(uv); |
1d72bdf6 NIS |
385 | U8 *p = d+len-1; |
386 | while (p > d) { | |
957a9e81 KW |
387 | *p-- = I8_TO_NATIVE_UTF8((uv & MASK) | MARK); |
388 | uv >>= SHIFT; | |
1d72bdf6 | 389 | } |
4c8cd605 | 390 | *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len)); |
1d72bdf6 NIS |
391 | return d+len; |
392 | } | |
a0ed51b3 | 393 | } |
a5bf80e0 | 394 | |
646ca15d | 395 | /* |
07693fe6 KW |
396 | =for apidoc uvchr_to_utf8 |
397 | ||
bcb1a2d4 | 398 | Adds the UTF-8 representation of the native code point C<uv> to the end |
f2fc1b45 | 399 | of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to |
c749c9fd KW |
400 | C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to |
401 | the byte after the end of the new character. In other words, | |
07693fe6 KW |
402 | |
403 | d = uvchr_to_utf8(d, uv); | |
404 | ||
405 | is the recommended wide native character-aware way of saying | |
406 | ||
407 | *(d++) = uv; | |
408 | ||
d22ec717 KW |
409 | This function accepts any code point from 0..C<IV_MAX> as input. |
410 | C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word. | |
760c7c2f KW |
411 | |
412 | It is possible to forbid or warn on non-Unicode code points, or those that may | |
413 | be problematic by using L</uvchr_to_utf8_flags>. | |
de69f3af | 414 | |
07693fe6 KW |
415 | =cut |
416 | */ | |
417 | ||
de69f3af KW |
418 | /* This is also a macro */ |
419 | PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv); | |
420 | ||
07693fe6 KW |
421 | U8 * |
422 | Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv) | |
423 | { | |
de69f3af | 424 | return uvchr_to_utf8(d, uv); |
07693fe6 KW |
425 | } |
426 | ||
de69f3af KW |
427 | /* |
428 | =for apidoc uvchr_to_utf8_flags | |
429 | ||
430 | Adds the UTF-8 representation of the native code point C<uv> to the end | |
f2fc1b45 | 431 | of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to |
c749c9fd KW |
432 | C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to |
433 | the byte after the end of the new character. In other words, | |
de69f3af KW |
434 | |
435 | d = uvchr_to_utf8_flags(d, uv, flags); | |
436 | ||
437 | or, in most cases, | |
438 | ||
439 | d = uvchr_to_utf8_flags(d, uv, 0); | |
440 | ||
441 | This is the Unicode-aware way of saying | |
442 | ||
443 | *(d++) = uv; | |
444 | ||
d22ec717 KW |
445 | If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as |
446 | input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word. | |
760c7c2f KW |
447 | |
448 | Specifying C<flags> can further restrict what is allowed and not warned on, as | |
449 | follows: | |
de69f3af | 450 | |
796b6530 | 451 | If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set, |
7ee537e6 KW |
452 | the function will raise a warning, provided UTF8 warnings are enabled. If |
453 | instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return | |
454 | NULL. If both flags are set, the function will both warn and return NULL. | |
de69f3af | 455 | |
760c7c2f KW |
456 | Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags |
457 | affect how the function handles a Unicode non-character. | |
93e6dbd6 | 458 | |
760c7c2f KW |
459 | And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags |
460 | affect the handling of code points that are above the Unicode maximum of | |
461 | 0x10FFFF. Languages other than Perl may not be able to accept files that | |
462 | contain these. | |
93e6dbd6 KW |
463 | |
464 | The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of | |
465 | the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all | |
ecc1615f KW |
466 | three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the |
467 | allowed inputs to the strict UTF-8 traditionally defined by Unicode. | |
468 | Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and | |
469 | C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the | |
470 | above-Unicode and surrogate flags, but not the non-character ones, as | |
471 | defined in | |
472 | L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>. | |
473 | See L<perlunicode/Noncharacter code points>. | |
93e6dbd6 | 474 | |
57ff5f59 KW |
475 | Extremely high code points were never specified in any standard, and require an |
476 | extension to UTF-8 to express, which Perl does. It is likely that programs | |
477 | written in something other than Perl would not be able to read files that | |
478 | contain these; nor would Perl understand files written by something that uses a | |
479 | different extension. For these reasons, there is a separate set of flags that | |
480 | can warn and/or disallow these extremely high code points, even if other | |
481 | above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED> | |
482 | and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see | |
483 | L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will | |
484 | treat all above-Unicode code points, including these, as malformations. (Note | |
485 | that the Unicode standard considers anything above 0x10FFFF to be illegal, but | |
486 | there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1)) | |
487 | ||
488 | A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is | |
489 | retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly, | |
490 | C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named | |
7c4a22ed KW |
491 | C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because on EBCDIC |
492 | platforms,these flags can apply to code points that actually do fit in 31 bits. | |
493 | The new names accurately describe the situation in all cases. | |
de69f3af | 494 | |
de69f3af KW |
495 | =cut |
496 | */ | |
497 | ||
498 | /* This is also a macro */ | |
499 | PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags); | |
500 | ||
07693fe6 KW |
501 | U8 * |
502 | Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags) | |
503 | { | |
de69f3af | 504 | return uvchr_to_utf8_flags(d, uv, flags); |
07693fe6 KW |
505 | } |
506 | ||
57ff5f59 KW |
507 | #ifndef UV_IS_QUAD |
508 | ||
e050c007 KW |
509 | STATIC int |
510 | S_is_utf8_cp_above_31_bits(const U8 * const s, | |
511 | const U8 * const e, | |
512 | const bool consider_overlongs) | |
83dc0f42 KW |
513 | { |
514 | /* Returns TRUE if the first code point represented by the Perl-extended- | |
515 | * UTF-8-encoded string starting at 's', and looking no further than 'e - | |
516 | * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31. | |
517 | * | |
518 | * The function handles the case where the input bytes do not include all | |
519 | * the ones necessary to represent a full character. That is, they may be | |
520 | * the intial bytes of the representation of a code point, but possibly | |
521 | * the final ones necessary for the complete representation may be beyond | |
522 | * 'e - 1'. | |
523 | * | |
e050c007 KW |
524 | * The function also can handle the case where the input is an overlong |
525 | * sequence. If 'consider_overlongs' is 0, the function assumes the | |
526 | * input is not overlong, without checking, and will return based on that | |
527 | * assumption. If this parameter is 1, the function will go to the trouble | |
528 | * of figuring out if it actually evaluates to above or below 31 bits. | |
83dc0f42 | 529 | * |
e050c007 | 530 | * The sequence is otherwise assumed to be well-formed, without checking. |
83dc0f42 KW |
531 | */ |
532 | ||
e050c007 KW |
533 | const STRLEN len = e - s; |
534 | int is_overlong; | |
535 | ||
536 | PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS; | |
537 | ||
538 | assert(! UTF8_IS_INVARIANT(*s) && e > s); | |
539 | ||
83dc0f42 KW |
540 | #ifdef EBCDIC |
541 | ||
e050c007 | 542 | PERL_UNUSED_ARG(consider_overlongs); |
83dc0f42 | 543 | |
e050c007 KW |
544 | /* On the EBCDIC code pages we handle, only the native start byte 0xFE can |
545 | * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can | |
546 | * also be the start byte for a 31-bit code point; we need at least 2 | |
547 | * bytes, and maybe up through 8 bytes, to determine that. (It can also be | |
548 | * the start byte for an overlong sequence, but for 30-bit or smaller code | |
549 | * points, so we don't have to worry about overlongs on EBCDIC.) */ | |
550 | if (*s != 0xFE) { | |
551 | return 0; | |
552 | } | |
83dc0f42 | 553 | |
e050c007 KW |
554 | if (len == 1) { |
555 | return -1; | |
556 | } | |
83dc0f42 | 557 | |
e050c007 | 558 | #else |
83dc0f42 | 559 | |
e050c007 KW |
560 | /* On ASCII, FE and FF are the only start bytes that can evaluate to |
561 | * needing more than 31 bits. */ | |
562 | if (LIKELY(*s < 0xFE)) { | |
563 | return 0; | |
564 | } | |
83dc0f42 | 565 | |
e050c007 KW |
566 | /* What we have left are FE and FF. Both of these require more than 31 |
567 | * bits unless they are for overlongs. */ | |
568 | if (! consider_overlongs) { | |
569 | return 1; | |
570 | } | |
83dc0f42 | 571 | |
e050c007 KW |
572 | /* Here, we have FE or FF. If the input isn't overlong, it evaluates to |
573 | * above 31 bits. But we need more than one byte to discern this, so if | |
574 | * passed just the start byte, it could be an overlong evaluating to | |
575 | * smaller */ | |
576 | if (len == 1) { | |
577 | return -1; | |
578 | } | |
83dc0f42 | 579 | |
e050c007 KW |
580 | /* Having excluded len==1, and knowing that FE and FF are both valid start |
581 | * bytes, we can call the function below to see if the sequence is | |
582 | * overlong. (We don't need the full generality of the called function, | |
583 | * but for these huge code points, speed shouldn't be a consideration, and | |
584 | * the compiler does have enough information, since it's static to this | |
585 | * file, to optimize to just the needed parts.) */ | |
586 | is_overlong = is_utf8_overlong_given_start_byte_ok(s, len); | |
83dc0f42 | 587 | |
e050c007 KW |
588 | /* If it isn't overlong, more than 31 bits are required. */ |
589 | if (is_overlong == 0) { | |
590 | return 1; | |
591 | } | |
83dc0f42 | 592 | |
e050c007 KW |
593 | /* If it is indeterminate if it is overlong, return that */ |
594 | if (is_overlong < 0) { | |
595 | return -1; | |
596 | } | |
597 | ||
598 | /* Here is overlong. Such a sequence starting with FE is below 31 bits, as | |
599 | * the max it can be is 2**31 - 1 */ | |
600 | if (*s == 0xFE) { | |
601 | return 0; | |
83dc0f42 KW |
602 | } |
603 | ||
e050c007 KW |
604 | #endif |
605 | ||
606 | /* Here, ASCII and EBCDIC rejoin: | |
607 | * On ASCII: We have an overlong sequence starting with FF | |
608 | * On EBCDIC: We have a sequence starting with FE. */ | |
609 | ||
610 | { /* For C89, use a block so the declaration can be close to its use */ | |
611 | ||
612 | #ifdef EBCDIC | |
613 | ||
5f995336 KW |
614 | /* U+7FFFFFFF (2 ** 31 - 1) |
615 | * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13 | |
616 | * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73 | |
617 | * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72 | |
618 | * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75 | |
619 | * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF | |
620 | * U+80000000 (2 ** 31): | |
621 | * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41 | |
622 | * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41 | |
623 | * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41 | |
624 | * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0 | |
e050c007 KW |
625 | * |
626 | * and since we know that *s = \xfe, any continuation sequcence | |
627 | * following it that is gt the below is above 31 bits | |
628 | [0] [1] [2] [3] [4] [5] [6] */ | |
629 | const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42"; | |
630 | ||
631 | #else | |
632 | ||
633 | /* FF overlong for U+7FFFFFFF (2 ** 31 - 1) | |
634 | * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF | |
635 | * FF overlong for U+80000000 (2 ** 31): | |
636 | * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80 | |
637 | * and since we know that *s = \xff, any continuation sequcence | |
638 | * following it that is gt the below is above 30 bits | |
639 | [0] [1] [2] [3] [4] [5] [6] */ | |
640 | const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81"; | |
5f995336 | 641 | |
83dc0f42 KW |
642 | |
643 | #endif | |
e050c007 KW |
644 | const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1; |
645 | const STRLEN cmp_len = MIN(conts_len, len - 1); | |
646 | ||
647 | /* Now compare the continuation bytes in s with the ones we have | |
648 | * compiled in that are for the largest 30 bit code point. If we have | |
649 | * enough bytes available to determine the answer, or the bytes we do | |
650 | * have differ from them, we can compare the two to get a definitive | |
651 | * answer (Note that in UTF-EBCDIC, the two lowest possible | |
652 | * continuation bytes are \x41 and \x42.) */ | |
653 | if (cmp_len >= conts_len || memNE(s + 1, | |
654 | conts_for_highest_30_bit, | |
655 | cmp_len)) | |
656 | { | |
657 | return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len)); | |
658 | } | |
83dc0f42 | 659 | |
e050c007 KW |
660 | /* Here, all the bytes we have are the same as the highest 30-bit code |
661 | * point, but we are missing so many bytes that we can't make the | |
662 | * determination */ | |
663 | return -1; | |
664 | } | |
83dc0f42 KW |
665 | } |
666 | ||
57ff5f59 KW |
667 | #endif |
668 | ||
d6be65ae | 669 | PERL_STATIC_INLINE int |
12a4bed3 KW |
670 | S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len) |
671 | { | |
d6be65ae KW |
672 | /* Returns an int indicating whether or not the UTF-8 sequence from 's' to |
673 | * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if | |
674 | * it isn't, and -1 if there isn't enough information to tell. This last | |
675 | * return value can happen if the sequence is incomplete, missing some | |
676 | * trailing bytes that would form a complete character. If there are | |
677 | * enough bytes to make a definitive decision, this function does so. | |
678 | * Usually 2 bytes sufficient. | |
679 | * | |
680 | * Overlongs can occur whenever the number of continuation bytes changes. | |
681 | * That means whenever the number of leading 1 bits in a start byte | |
682 | * increases from the next lower start byte. That happens for start bytes | |
683 | * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal | |
684 | * start bytes have already been excluded, so don't need to be tested here; | |
12a4bed3 KW |
685 | * ASCII platforms: C0, C1 |
686 | * EBCDIC platforms C0, C1, C2, C3, C4, E0 | |
d6be65ae | 687 | */ |
12a4bed3 KW |
688 | |
689 | const U8 s0 = NATIVE_UTF8_TO_I8(s[0]); | |
690 | const U8 s1 = NATIVE_UTF8_TO_I8(s[1]); | |
691 | ||
692 | PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK; | |
693 | assert(len > 1 && UTF8_IS_START(*s)); | |
694 | ||
695 | /* Each platform has overlongs after the start bytes given above (expressed | |
696 | * in I8 for EBCDIC). What constitutes an overlong varies by platform, but | |
697 | * the logic is the same, except the E0 overlong has already been excluded | |
698 | * on EBCDIC platforms. The values below were found by manually | |
699 | * inspecting the UTF-8 patterns. See the tables in utf8.h and | |
700 | * utfebcdic.h. */ | |
701 | ||
702 | # ifdef EBCDIC | |
703 | # define F0_ABOVE_OVERLONG 0xB0 | |
704 | # define F8_ABOVE_OVERLONG 0xA8 | |
705 | # define FC_ABOVE_OVERLONG 0xA4 | |
706 | # define FE_ABOVE_OVERLONG 0xA2 | |
707 | # define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41" | |
708 | /* I8(0xfe) is FF */ | |
709 | # else | |
710 | ||
711 | if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) { | |
d6be65ae | 712 | return 1; |
12a4bed3 KW |
713 | } |
714 | ||
715 | # define F0_ABOVE_OVERLONG 0x90 | |
716 | # define F8_ABOVE_OVERLONG 0x88 | |
717 | # define FC_ABOVE_OVERLONG 0x84 | |
718 | # define FE_ABOVE_OVERLONG 0x82 | |
719 | # define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80" | |
720 | # endif | |
721 | ||
722 | ||
723 | if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG)) | |
724 | || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG)) | |
725 | || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG)) | |
726 | || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG))) | |
727 | { | |
d6be65ae | 728 | return 1; |
12a4bed3 KW |
729 | } |
730 | ||
b0b342d4 | 731 | /* Check for the FF overlong */ |
d6be65ae | 732 | return isFF_OVERLONG(s, len); |
b0b342d4 KW |
733 | } |
734 | ||
8d6204cc | 735 | PERL_STATIC_INLINE int |
b0b342d4 KW |
736 | S_isFF_OVERLONG(const U8 * const s, const STRLEN len) |
737 | { | |
8d6204cc KW |
738 | /* Returns an int indicating whether or not the UTF-8 sequence from 's' to |
739 | * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if | |
740 | * it isn't, and -1 if there isn't enough information to tell. This last | |
741 | * return value can happen if the sequence is incomplete, missing some | |
742 | * trailing bytes that would form a complete character. If there are | |
743 | * enough bytes to make a definitive decision, this function does so. */ | |
744 | ||
b0b342d4 | 745 | PERL_ARGS_ASSERT_ISFF_OVERLONG; |
12a4bed3 | 746 | |
8d6204cc KW |
747 | /* To be an FF overlong, all the available bytes must match */ |
748 | if (LIKELY(memNE(s, FF_OVERLONG_PREFIX, | |
749 | MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1)))) | |
750 | { | |
751 | return 0; | |
752 | } | |
753 | ||
754 | /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must | |
755 | * be there; what comes after them doesn't matter. See tables in utf8.h, | |
b0b342d4 | 756 | * utfebcdic.h. */ |
8d6204cc KW |
757 | if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) { |
758 | return 1; | |
759 | } | |
12a4bed3 | 760 | |
8d6204cc KW |
761 | /* The missing bytes could cause the result to go one way or the other, so |
762 | * the result is indeterminate */ | |
763 | return -1; | |
12a4bed3 KW |
764 | } |
765 | ||
d22ec717 | 766 | #if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */ |
a77c906e KW |
767 | # ifdef EBCDIC /* Actually is I8 */ |
768 | # define HIGHEST_REPRESENTABLE_UTF8 \ | |
d22ec717 | 769 | "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF" |
a77c906e KW |
770 | # else |
771 | # define HIGHEST_REPRESENTABLE_UTF8 \ | |
d22ec717 | 772 | "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF" |
a77c906e KW |
773 | # endif |
774 | #endif | |
775 | ||
c285bbc4 | 776 | PERL_STATIC_INLINE int |
e050c007 KW |
777 | S_does_utf8_overflow(const U8 * const s, |
778 | const U8 * e, | |
779 | const bool consider_overlongs) | |
a77c906e | 780 | { |
c285bbc4 | 781 | /* Returns an int indicating whether or not the UTF-8 sequence from 's' to |
d22ec717 KW |
782 | * 'e' - 1 would overflow an IV on this platform; that is if it represents |
783 | * a code point larger than the highest representable code point. It | |
784 | * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't | |
785 | * enough information to tell. This last return value can happen if the | |
786 | * sequence is incomplete, missing some trailing bytes that would form a | |
787 | * complete character. If there are enough bytes to make a definitive | |
788 | * decision, this function does so. | |
c285bbc4 | 789 | * |
e050c007 KW |
790 | * If 'consider_overlongs' is TRUE, the function checks for the possibility |
791 | * that the sequence is an overlong that doesn't overflow. Otherwise, it | |
792 | * assumes the sequence is not an overlong. This can give different | |
793 | * results only on ASCII 32-bit platforms. | |
794 | * | |
c285bbc4 KW |
795 | * (For ASCII platforms, we could use memcmp() because we don't have to |
796 | * convert each byte to I8, but it's very rare input indeed that would | |
797 | * approach overflow, so the loop below will likely only get executed once.) | |
798 | * | |
799 | * 'e' - 1 must not be beyond a full character. */ | |
a77c906e | 800 | |
a77c906e KW |
801 | |
802 | PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW; | |
803 | assert(s <= e && s + UTF8SKIP(s) >= e); | |
804 | ||
d22ec717 KW |
805 | #if ! defined(UV_IS_QUAD) |
806 | ||
807 | return is_utf8_cp_above_31_bits(s, e, consider_overlongs); | |
808 | ||
809 | #else | |
810 | ||
811 | PERL_UNUSED_ARG(consider_overlongs); | |
812 | ||
813 | { | |
814 | const STRLEN len = e - s; | |
815 | const U8 *x; | |
816 | const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8; | |
817 | ||
818 | for (x = s; x < e; x++, y++) { | |
819 | ||
820 | if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) { | |
821 | continue; | |
822 | } | |
823 | ||
824 | /* If this byte is larger than the corresponding highest UTF-8 | |
825 | * byte, the sequence overflow; otherwise the byte is less than, | |
826 | * and so the sequence doesn't overflow */ | |
827 | return NATIVE_UTF8_TO_I8(*x) > *y; | |
828 | ||
829 | } | |
830 | ||
831 | /* Got to the end and all bytes are the same. If the input is a whole | |
832 | * character, it doesn't overflow. And if it is a partial character, | |
833 | * there's not enough information to tell */ | |
834 | if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) { | |
835 | return -1; | |
836 | } | |
837 | ||
838 | return 0; | |
839 | } | |
840 | ||
841 | #endif | |
842 | ||
843 | } | |
844 | ||
845 | #if 0 | |
846 | ||
847 | /* This is the portions of the above function that deal with UV_MAX instead of | |
848 | * IV_MAX. They are left here in case we want to combine them so that internal | |
849 | * uses can have larger code points. The only logic difference is that the | |
850 | * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has | |
851 | * different logic. | |
852 | */ | |
853 | ||
854 | /* Anything larger than this will overflow the word if it were converted into a UV */ | |
855 | #if defined(UV_IS_QUAD) | |
856 | # ifdef EBCDIC /* Actually is I8 */ | |
857 | # define HIGHEST_REPRESENTABLE_UTF8 \ | |
858 | "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF" | |
859 | # else | |
860 | # define HIGHEST_REPRESENTABLE_UTF8 \ | |
861 | "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF" | |
862 | # endif | |
863 | #else /* 32-bit */ | |
864 | # ifdef EBCDIC | |
865 | # define HIGHEST_REPRESENTABLE_UTF8 \ | |
866 | "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF" | |
867 | # else | |
868 | # define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF" | |
869 | # endif | |
870 | #endif | |
871 | ||
a77c906e KW |
872 | #if ! defined(UV_IS_QUAD) && ! defined(EBCDIC) |
873 | ||
874 | /* On 32 bit ASCII machines, many overlongs that start with FF don't | |
875 | * overflow */ | |
e050c007 | 876 | if (consider_overlongs && isFF_OVERLONG(s, len) > 0) { |
c285bbc4 KW |
877 | |
878 | /* To be such an overlong, the first bytes of 's' must match | |
879 | * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we | |
880 | * don't have any additional bytes available, the sequence, when | |
881 | * completed might or might not fit in 32 bits. But if we have that | |
882 | * next byte, we can tell for sure. If it is <= 0x83, then it does | |
883 | * fit. */ | |
884 | if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) { | |
885 | return -1; | |
886 | } | |
887 | ||
888 | return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83; | |
a77c906e KW |
889 | } |
890 | ||
d22ec717 KW |
891 | /* Starting with the #else, the rest of the function is identical except |
892 | * 1. we need to move the 'len' declaration to be global to the function | |
893 | * 2. the endif move to just after the UNUSED_ARG. | |
894 | * An empty endif is given just below to satisfy the preprocessor | |
895 | */ | |
a77c906e KW |
896 | #endif |
897 | ||
d22ec717 | 898 | #endif |
a77c906e | 899 | |
12a4bed3 KW |
900 | #undef F0_ABOVE_OVERLONG |
901 | #undef F8_ABOVE_OVERLONG | |
902 | #undef FC_ABOVE_OVERLONG | |
903 | #undef FE_ABOVE_OVERLONG | |
904 | #undef FF_OVERLONG_PREFIX | |
905 | ||
35f8c9bd | 906 | STRLEN |
edc2c47a | 907 | Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags) |
35f8c9bd | 908 | { |
2b479609 | 909 | STRLEN len; |
12a4bed3 | 910 | const U8 *x; |
35f8c9bd | 911 | |
2b479609 KW |
912 | /* A helper function that should not be called directly. |
913 | * | |
914 | * This function returns non-zero if the string beginning at 's' and | |
915 | * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a | |
916 | * code point; otherwise it returns 0. The examination stops after the | |
917 | * first code point in 's' is validated, not looking at the rest of the | |
918 | * input. If 'e' is such that there are not enough bytes to represent a | |
919 | * complete code point, this function will return non-zero anyway, if the | |
920 | * bytes it does have are well-formed UTF-8 as far as they go, and aren't | |
921 | * excluded by 'flags'. | |
922 | * | |
923 | * A non-zero return gives the number of bytes required to represent the | |
924 | * code point. Be aware that if the input is for a partial character, the | |
925 | * return will be larger than 'e - s'. | |
926 | * | |
927 | * This function assumes that the code point represented is UTF-8 variant. | |
56576a04 KW |
928 | * The caller should have excluded the possibility of it being invariant |
929 | * before calling this function. | |
2b479609 KW |
930 | * |
931 | * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags | |
932 | * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return | |
933 | * 0 if the code point represented is well-formed Perl-extended-UTF-8, but | |
934 | * disallowed by the flags. If the input is only for a partial character, | |
935 | * the function will return non-zero if there is any sequence of | |
936 | * well-formed UTF-8 that, when appended to the input sequence, could | |
937 | * result in an allowed code point; otherwise it returns 0. Non characters | |
938 | * cannot be determined based on partial character input. But many of the | |
939 | * other excluded types can be determined with just the first one or two | |
940 | * bytes. | |
941 | * | |
942 | */ | |
943 | ||
944 | PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER; | |
945 | ||
946 | assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE | |
d044b7a7 | 947 | |UTF8_DISALLOW_PERL_EXTENDED))); |
2b479609 | 948 | assert(! UTF8_IS_INVARIANT(*s)); |
35f8c9bd | 949 | |
2b479609 | 950 | /* A variant char must begin with a start byte */ |
35f8c9bd KW |
951 | if (UNLIKELY(! UTF8_IS_START(*s))) { |
952 | return 0; | |
953 | } | |
954 | ||
edc2c47a KW |
955 | /* Examine a maximum of a single whole code point */ |
956 | if (e - s > UTF8SKIP(s)) { | |
957 | e = s + UTF8SKIP(s); | |
958 | } | |
959 | ||
2b479609 KW |
960 | len = e - s; |
961 | ||
962 | if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) { | |
963 | const U8 s0 = NATIVE_UTF8_TO_I8(s[0]); | |
35f8c9bd | 964 | |
56576a04 KW |
965 | /* Here, we are disallowing some set of largish code points, and the |
966 | * first byte indicates the sequence is for a code point that could be | |
967 | * in the excluded set. We generally don't have to look beyond this or | |
968 | * the second byte to see if the sequence is actually for one of the | |
969 | * excluded classes. The code below is derived from this table: | |
970 | * | |
2b479609 KW |
971 | * UTF-8 UTF-EBCDIC I8 |
972 | * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate | |
973 | * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate | |
974 | * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode | |
975 | * | |
56576a04 KW |
976 | * Keep in mind that legal continuation bytes range between \x80..\xBF |
977 | * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't | |
978 | * continuation bytes. Hence, we don't have to test the upper edge | |
979 | * because if any of those is encountered, the sequence is malformed, | |
980 | * and would fail elsewhere in this function. | |
981 | * | |
982 | * The code here likewise assumes that there aren't other | |
983 | * malformations; again the function should fail elsewhere because of | |
984 | * these. For example, an overlong beginning with FC doesn't actually | |
985 | * have to be a super; it could actually represent a small code point, | |
986 | * even U+0000. But, since overlongs (and other malformations) are | |
987 | * illegal, the function should return FALSE in either case. | |
2b479609 KW |
988 | */ |
989 | ||
990 | #ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */ | |
991 | # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA | |
19794540 | 992 | # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2) |
2b479609 | 993 | |
19794540 KW |
994 | # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \ |
995 | /* B6 and B7 */ \ | |
996 | && ((s1) & 0xFE ) == 0xB6) | |
57ff5f59 | 997 | # define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF)) |
2b479609 KW |
998 | #else |
999 | # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5 | |
19794540 KW |
1000 | # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90) |
1001 | # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0) | |
57ff5f59 | 1002 | # define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE) |
2b479609 KW |
1003 | #endif |
1004 | ||
1005 | if ( (flags & UTF8_DISALLOW_SUPER) | |
ddb65933 KW |
1006 | && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER)) |
1007 | { | |
2b479609 KW |
1008 | return 0; /* Above Unicode */ |
1009 | } | |
1010 | ||
d044b7a7 | 1011 | if ( (flags & UTF8_DISALLOW_PERL_EXTENDED) |
57ff5f59 | 1012 | && UNLIKELY(isUTF8_PERL_EXTENDED(s))) |
2b479609 | 1013 | { |
57ff5f59 | 1014 | return 0; |
2b479609 KW |
1015 | } |
1016 | ||
1017 | if (len > 1) { | |
1018 | const U8 s1 = NATIVE_UTF8_TO_I8(s[1]); | |
1019 | ||
1020 | if ( (flags & UTF8_DISALLOW_SUPER) | |
19794540 | 1021 | && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1))) |
2b479609 KW |
1022 | { |
1023 | return 0; /* Above Unicode */ | |
1024 | } | |
1025 | ||
1026 | if ( (flags & UTF8_DISALLOW_SURROGATE) | |
19794540 | 1027 | && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1))) |
2b479609 KW |
1028 | { |
1029 | return 0; /* Surrogate */ | |
1030 | } | |
1031 | ||
1032 | if ( (flags & UTF8_DISALLOW_NONCHAR) | |
1033 | && UNLIKELY(UTF8_IS_NONCHAR(s, e))) | |
1034 | { | |
1035 | return 0; /* Noncharacter code point */ | |
1036 | } | |
1037 | } | |
1038 | } | |
1039 | ||
1040 | /* Make sure that all that follows are continuation bytes */ | |
35f8c9bd KW |
1041 | for (x = s + 1; x < e; x++) { |
1042 | if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) { | |
1043 | return 0; | |
1044 | } | |
1045 | } | |
1046 | ||
af13dd8a | 1047 | /* Here is syntactically valid. Next, make sure this isn't the start of an |
12a4bed3 | 1048 | * overlong. */ |
d6be65ae | 1049 | if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) { |
12a4bed3 | 1050 | return 0; |
af13dd8a KW |
1051 | } |
1052 | ||
12a4bed3 KW |
1053 | /* And finally, that the code point represented fits in a word on this |
1054 | * platform */ | |
e050c007 KW |
1055 | if (0 < does_utf8_overflow(s, e, |
1056 | 0 /* Don't consider overlongs */ | |
1057 | )) | |
1058 | { | |
12a4bed3 | 1059 | return 0; |
35f8c9bd KW |
1060 | } |
1061 | ||
2b479609 | 1062 | return UTF8SKIP(s); |
35f8c9bd KW |
1063 | } |
1064 | ||
7e2f38b2 | 1065 | char * |
63ab03b3 | 1066 | Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format) |
7cf8d05d KW |
1067 | { |
1068 | /* Returns a mortalized C string that is a displayable copy of the 'len' | |
63ab03b3 | 1069 | * bytes starting at 'start'. 'format' gives how to display each byte. |
7e2f38b2 KW |
1070 | * Currently, there are only two formats, so it is currently a bool: |
1071 | * 0 \xab | |
1072 | * 1 ab (that is a space between two hex digit bytes) | |
1073 | */ | |
7cf8d05d KW |
1074 | |
1075 | const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a | |
1076 | trailing NUL */ | |
63ab03b3 KW |
1077 | const U8 * s = start; |
1078 | const U8 * const e = start + len; | |
7cf8d05d KW |
1079 | char * output; |
1080 | char * d; | |
1081 | ||
1082 | PERL_ARGS_ASSERT__BYTE_DUMP_STRING; | |
1083 | ||
1084 | Newx(output, output_len, char); | |
1085 | SAVEFREEPV(output); | |
1086 | ||
1087 | d = output; | |
63ab03b3 | 1088 | for (s = start; s < e; s++) { |
7cf8d05d KW |
1089 | const unsigned high_nibble = (*s & 0xF0) >> 4; |
1090 | const unsigned low_nibble = (*s & 0x0F); | |
1091 | ||
7e2f38b2 | 1092 | if (format) { |
63ab03b3 KW |
1093 | if (s > start) { |
1094 | *d++ = ' '; | |
1095 | } | |
7e2f38b2 KW |
1096 | } |
1097 | else { | |
1098 | *d++ = '\\'; | |
1099 | *d++ = 'x'; | |
1100 | } | |
7cf8d05d KW |
1101 | |
1102 | if (high_nibble < 10) { | |
1103 | *d++ = high_nibble + '0'; | |
1104 | } | |
1105 | else { | |
1106 | *d++ = high_nibble - 10 + 'a'; | |
1107 | } | |
1108 | ||
1109 | if (low_nibble < 10) { | |
1110 | *d++ = low_nibble + '0'; | |
1111 | } | |
1112 | else { | |
1113 | *d++ = low_nibble - 10 + 'a'; | |
1114 | } | |
1115 | } | |
1116 | ||
1117 | *d = '\0'; | |
1118 | return output; | |
1119 | } | |
1120 | ||
806547a7 | 1121 | PERL_STATIC_INLINE char * |
7cf8d05d KW |
1122 | S_unexpected_non_continuation_text(pTHX_ const U8 * const s, |
1123 | ||
421da25c | 1124 | /* Max number of bytes to print */ |
3cc6a05e | 1125 | STRLEN print_len, |
7cf8d05d KW |
1126 | |
1127 | /* Which one is the non-continuation */ | |
1128 | const STRLEN non_cont_byte_pos, | |
1129 | ||
1130 | /* How many bytes should there be? */ | |
1131 | const STRLEN expect_len) | |
806547a7 KW |
1132 | { |
1133 | /* Return the malformation warning text for an unexpected continuation | |
1134 | * byte. */ | |
1135 | ||
7cf8d05d | 1136 | const char * const where = (non_cont_byte_pos == 1) |
806547a7 | 1137 | ? "immediately" |
7cf8d05d KW |
1138 | : Perl_form(aTHX_ "%d bytes", |
1139 | (int) non_cont_byte_pos); | |
421da25c KW |
1140 | const U8 * x = s + non_cont_byte_pos; |
1141 | const U8 * e = s + print_len; | |
806547a7 KW |
1142 | |
1143 | PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT; | |
1144 | ||
7cf8d05d KW |
1145 | /* We don't need to pass this parameter, but since it has already been |
1146 | * calculated, it's likely faster to pass it; verify under DEBUGGING */ | |
1147 | assert(expect_len == UTF8SKIP(s)); | |
1148 | ||
421da25c KW |
1149 | /* As a defensive coding measure, don't output anything past a NUL. Such |
1150 | * bytes shouldn't be in the middle of a malformation, and could mark the | |
1151 | * end of the allocated string, and what comes after is undefined */ | |
1152 | for (; x < e; x++) { | |
1153 | if (*x == '\0') { | |
1154 | x++; /* Output this particular NUL */ | |
1155 | break; | |
1156 | } | |
1157 | } | |
1158 | ||
7cf8d05d KW |
1159 | return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x," |
1160 | " %s after start byte 0x%02x; need %d bytes, got %d)", | |
1161 | malformed_text, | |
421da25c | 1162 | _byte_dump_string(s, x - s, 0), |
7cf8d05d KW |
1163 | *(s + non_cont_byte_pos), |
1164 | where, | |
1165 | *s, | |
1166 | (int) expect_len, | |
1167 | (int) non_cont_byte_pos); | |
806547a7 KW |
1168 | } |
1169 | ||
35f8c9bd KW |
1170 | /* |
1171 | ||
de69f3af | 1172 | =for apidoc utf8n_to_uvchr |
378516de KW |
1173 | |
1174 | THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. | |
de69f3af | 1175 | Most code should use L</utf8_to_uvchr_buf>() rather than call this directly. |
67e989fb | 1176 | |
9041c2e3 | 1177 | Bottom level UTF-8 decode routine. |
de69f3af | 1178 | Returns the native code point value of the first character in the string C<s>, |
746afd53 KW |
1179 | which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than |
1180 | C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to | |
1181 | the length, in bytes, of that character. | |
949cf498 KW |
1182 | |
1183 | The value of C<flags> determines the behavior when C<s> does not point to a | |
2b5e7bc2 KW |
1184 | well-formed UTF-8 character. If C<flags> is 0, encountering a malformation |
1185 | causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) | |
1186 | is the next possible position in C<s> that could begin a non-malformed | |
1187 | character. Also, if UTF-8 warnings haven't been lexically disabled, a warning | |
1188 | is raised. Some UTF-8 input sequences may contain multiple malformations. | |
1189 | This function tries to find every possible one in each call, so multiple | |
56576a04 | 1190 | warnings can be raised for the same sequence. |
949cf498 KW |
1191 | |
1192 | Various ALLOW flags can be set in C<flags> to allow (and not warn on) | |
1193 | individual types of malformations, such as the sequence being overlong (that | |
1194 | is, when there is a shorter sequence that can express the same code point; | |
1195 | overlong sequences are expressly forbidden in the UTF-8 standard due to | |
1196 | potential security issues). Another malformation example is the first byte of | |
1197 | a character not being a legal first byte. See F<utf8.h> for the list of such | |
94953955 KW |
1198 | flags. Even if allowed, this function generally returns the Unicode |
1199 | REPLACEMENT CHARACTER when it encounters a malformation. There are flags in | |
1200 | F<utf8.h> to override this behavior for the overlong malformations, but don't | |
1201 | do that except for very specialized purposes. | |
949cf498 | 1202 | |
796b6530 | 1203 | The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other |
949cf498 KW |
1204 | flags) malformation is found. If this flag is set, the routine assumes that |
1205 | the caller will raise a warning, and this function will silently just set | |
d088425d KW |
1206 | C<retlen> to C<-1> (cast to C<STRLEN>) and return zero. |
1207 | ||
75200dff | 1208 | Note that this API requires disambiguation between successful decoding a C<NUL> |
796b6530 | 1209 | character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as |
111fa700 KW |
1210 | in both cases, 0 is returned, and, depending on the malformation, C<retlen> may |
1211 | be set to 1. To disambiguate, upon a zero return, see if the first byte of | |
1212 | C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an | |
f9380377 | 1213 | error. Or you can use C<L</utf8n_to_uvchr_error>>. |
949cf498 KW |
1214 | |
1215 | Certain code points are considered problematic. These are Unicode surrogates, | |
746afd53 | 1216 | Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF. |
949cf498 | 1217 | By default these are considered regular code points, but certain situations |
ecc1615f KW |
1218 | warrant special handling for them, which can be specified using the C<flags> |
1219 | parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all | |
1220 | three classes are treated as malformations and handled as such. The flags | |
1221 | C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and | |
1222 | C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to | |
1223 | disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE> | |
1224 | restricts the allowed inputs to the strict UTF-8 traditionally defined by | |
1225 | Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness | |
1226 | definition given by | |
1227 | L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>. | |
1228 | The difference between traditional strictness and C9 strictness is that the | |
1229 | latter does not forbid non-character code points. (They are still discouraged, | |
1230 | however.) For more discussion see L<perlunicode/Noncharacter code points>. | |
1231 | ||
1232 | The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>, | |
1233 | C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>, | |
796b6530 KW |
1234 | C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be |
1235 | raised for their respective categories, but otherwise the code points are | |
1236 | considered valid (not malformations). To get a category to both be treated as | |
1237 | a malformation and raise a warning, specify both the WARN and DISALLOW flags. | |
949cf498 | 1238 | (But note that warnings are not raised if lexically disabled nor if |
796b6530 | 1239 | C<UTF8_CHECK_ONLY> is also specified.) |
949cf498 | 1240 | |
57ff5f59 KW |
1241 | Extremely high code points were never specified in any standard, and require an |
1242 | extension to UTF-8 to express, which Perl does. It is likely that programs | |
1243 | written in something other than Perl would not be able to read files that | |
1244 | contain these; nor would Perl understand files written by something that uses a | |
1245 | different extension. For these reasons, there is a separate set of flags that | |
1246 | can warn and/or disallow these extremely high code points, even if other | |
1247 | above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and | |
1248 | C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see | |
1249 | L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all | |
1250 | above-Unicode code points, including these, as malformations. | |
1251 | (Note that the Unicode standard considers anything above 0x10FFFF to be | |
1252 | illegal, but there are standards predating it that allow up to 0x7FFF_FFFF | |
1253 | (2**31 -1)) | |
1254 | ||
1255 | A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is | |
1256 | retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly, | |
1257 | C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named | |
1258 | C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags | |
1259 | can apply to code points that actually do fit in 31 bits. This happens on | |
1260 | EBCDIC platforms, and sometimes when the L<overlong | |
1261 | malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately | |
1262 | describe the situation in all cases. | |
1263 | ||
ab8e6d41 | 1264 | |
949cf498 KW |
1265 | All other code points corresponding to Unicode characters, including private |
1266 | use and those yet to be assigned, are never considered malformed and never | |
1267 | warn. | |
67e989fb | 1268 | |
37607a96 | 1269 | =cut |
f9380377 KW |
1270 | |
1271 | Also implemented as a macro in utf8.h | |
1272 | */ | |
1273 | ||
1274 | UV | |
e6a4ffc3 KW |
1275 | Perl_utf8n_to_uvchr(const U8 *s, |
1276 | STRLEN curlen, | |
1277 | STRLEN *retlen, | |
1278 | const U32 flags) | |
f9380377 KW |
1279 | { |
1280 | PERL_ARGS_ASSERT_UTF8N_TO_UVCHR; | |
1281 | ||
1282 | return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL); | |
1283 | } | |
1284 | ||
1285 | /* | |
1286 | ||
1287 | =for apidoc utf8n_to_uvchr_error | |
1288 | ||
1289 | THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. | |
1290 | Most code should use L</utf8_to_uvchr_buf>() rather than call this directly. | |
1291 | ||
1292 | This function is for code that needs to know what the precise malformation(s) | |
37657a5b KW |
1293 | are when an error is found. If you also need to know the generated warning |
1294 | messages, use L</utf8n_to_uvchr_msgs>() instead. | |
f9380377 KW |
1295 | |
1296 | It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after | |
1297 | all the others, C<errors>. If this parameter is 0, this function behaves | |
1298 | identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer | |
1299 | to a C<U32> variable, which this function sets to indicate any errors found. | |
1300 | Upon return, if C<*errors> is 0, there were no errors found. Otherwise, | |
1301 | C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some | |
1302 | of these bits will be set if a malformation is found, even if the input | |
7a65503b | 1303 | C<flags> parameter indicates that the given malformation is allowed; those |
f9380377 KW |
1304 | exceptions are noted: |
1305 | ||
1306 | =over 4 | |
1307 | ||
57ff5f59 | 1308 | =item C<UTF8_GOT_PERL_EXTENDED> |
f9380377 | 1309 | |
57ff5f59 KW |
1310 | The input sequence is not standard UTF-8, but a Perl extension. This bit is |
1311 | set only if the input C<flags> parameter contains either the | |
1312 | C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags. | |
1313 | ||
1314 | Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard, | |
1315 | and so some extension must be used to express them. Perl uses a natural | |
1316 | extension to UTF-8 to represent the ones up to 2**36-1, and invented a further | |
1317 | extension to represent even higher ones, so that any code point that fits in a | |
1318 | 64-bit word can be represented. Text using these extensions is not likely to | |
1319 | be portable to non-Perl code. We lump both of these extensions together and | |
1320 | refer to them as Perl extended UTF-8. There exist other extensions that people | |
1321 | have invented, incompatible with Perl's. | |
1322 | ||
1323 | On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing | |
1324 | extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower | |
1325 | than on ASCII. Prior to that, code points 2**31 and higher were simply | |
1326 | unrepresentable, and a different, incompatible method was used to represent | |
1327 | code points between 2**30 and 2**31 - 1. | |
1328 | ||
1329 | On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if | |
1330 | Perl extended UTF-8 is used. | |
1331 | ||
1332 | In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still | |
1333 | may use for backward compatibility. That name is misleading, as this flag may | |
1334 | be set when the code point actually does fit in 31 bits. This happens on | |
1335 | EBCDIC platforms, and sometimes when the L<overlong | |
1336 | malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately | |
1337 | describes the situation in all cases. | |
f9380377 KW |
1338 | |
1339 | =item C<UTF8_GOT_CONTINUATION> | |
1340 | ||
1341 | The input sequence was malformed in that the first byte was a a UTF-8 | |
1342 | continuation byte. | |
1343 | ||
1344 | =item C<UTF8_GOT_EMPTY> | |
1345 | ||
1346 | The input C<curlen> parameter was 0. | |
1347 | ||
1348 | =item C<UTF8_GOT_LONG> | |
1349 | ||
1350 | The input sequence was malformed in that there is some other sequence that | |
1351 | evaluates to the same code point, but that sequence is shorter than this one. | |
1352 | ||
fecaf136 KW |
1353 | Until Unicode 3.1, it was legal for programs to accept this malformation, but |
1354 | it was discovered that this created security issues. | |
1355 | ||
f9380377 KW |
1356 | =item C<UTF8_GOT_NONCHAR> |
1357 | ||
1358 | The code point represented by the input UTF-8 sequence is for a Unicode | |
1359 | non-character code point. | |
1360 | This bit is set only if the input C<flags> parameter contains either the | |
1361 | C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags. | |
1362 | ||
1363 | =item C<UTF8_GOT_NON_CONTINUATION> | |
1364 | ||
1365 | The input sequence was malformed in that a non-continuation type byte was found | |
00d976bb KW |
1366 | in a position where only a continuation type one should be. See also |
1367 | L</C<UTF8_GOT_SHORT>>. | |
f9380377 KW |
1368 | |
1369 | =item C<UTF8_GOT_OVERFLOW> | |
1370 | ||
1371 | The input sequence was malformed in that it is for a code point that is not | |
d22ec717 | 1372 | representable in the number of bits available in an IV on the current platform. |
f9380377 KW |
1373 | |
1374 | =item C<UTF8_GOT_SHORT> | |
1375 | ||
1376 | The input sequence was malformed in that C<curlen> is smaller than required for | |
1377 | a complete sequence. In other words, the input is for a partial character | |
1378 | sequence. | |
1379 | ||
00d976bb KW |
1380 | |
1381 | C<UTF8_GOT_SHORT> and C<UTF8_GOT_NON_CONTINUATION> both indicate a too short | |
1382 | sequence. The difference is that C<UTF8_GOT_NON_CONTINUATION> indicates always | |
1383 | that there is an error, while C<UTF8_GOT_SHORT> means that an incomplete | |
1384 | sequence was looked at. If no other flags are present, it means that the | |
1385 | sequence was valid as far as it went. Depending on the application, this could | |
1386 | mean one of three things: | |
1387 | ||
1388 | =over | |
1389 | ||
1390 | =item * | |
1391 | ||
1392 | The C<curlen> length parameter passed in was too small, and the function was | |
1393 | prevented from examining all the necessary bytes. | |
1394 | ||
1395 | =item * | |
1396 | ||
1397 | The buffer being looked at is based on reading data, and the data received so | |
1398 | far stopped in the middle of a character, so that the next read will | |
1399 | read the remainder of this character. (It is up to the caller to deal with the | |
1400 | split bytes somehow.) | |
1401 | ||
1402 | =item * | |
1403 | ||
1404 | This is a real error, and the partial sequence is all we're going to get. | |
1405 | ||
1406 | =back | |
1407 | ||
f9380377 KW |
1408 | =item C<UTF8_GOT_SUPER> |
1409 | ||
1410 | The input sequence was malformed in that it is for a non-Unicode code point; | |
1411 | that is, one above the legal Unicode maximum. | |
1412 | This bit is set only if the input C<flags> parameter contains either the | |
1413 | C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags. | |
1414 | ||
1415 | =item C<UTF8_GOT_SURROGATE> | |
1416 | ||
1417 | The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate | |
1418 | code point. | |
1419 | This bit is set only if the input C<flags> parameter contains either the | |
1420 | C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags. | |
1421 | ||
1422 | =back | |
1423 | ||
133551d8 KW |
1424 | To do your own error handling, call this function with the C<UTF8_CHECK_ONLY> |
1425 | flag to suppress any warnings, and then examine the C<*errors> return. | |
1426 | ||
f9380377 | 1427 | =cut |
37657a5b KW |
1428 | |
1429 | Also implemented as a macro in utf8.h | |
37607a96 | 1430 | */ |
67e989fb | 1431 | |
a0ed51b3 | 1432 | UV |
e6a4ffc3 | 1433 | Perl_utf8n_to_uvchr_error(const U8 *s, |
37657a5b KW |
1434 | STRLEN curlen, |
1435 | STRLEN *retlen, | |
1436 | const U32 flags, | |
1437 | U32 * errors) | |
1438 | { | |
1439 | PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR; | |
1440 | ||
1441 | return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL); | |
1442 | } | |
1443 | ||
1444 | /* | |
1445 | ||
1446 | =for apidoc utf8n_to_uvchr_msgs | |
1447 | ||
1448 | THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. | |
1449 | Most code should use L</utf8_to_uvchr_buf>() rather than call this directly. | |
1450 | ||
1451 | This function is for code that needs to know what the precise malformation(s) | |
1452 | are when an error is found, and wants the corresponding warning and/or error | |
1453 | messages to be returned to the caller rather than be displayed. All messages | |
1454 | that would have been displayed if all lexcial warnings are enabled will be | |
1455 | returned. | |
1456 | ||
1457 | It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter | |
1458 | placed after all the others, C<msgs>. If this parameter is 0, this function | |
1459 | behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should | |
1460 | be a pointer to an C<AV *> variable, in which this function creates a new AV to | |
1461 | contain any appropriate messages. The elements of the array are ordered so | |
1462 | that the first message that would have been displayed is in the 0th element, | |
1463 | and so on. Each element is a hash with three key-value pairs, as follows: | |
1464 | ||
1465 | =over 4 | |
1466 | ||
1467 | =item C<text> | |
1468 | ||
1469 | The text of the message as a C<SVpv>. | |
1470 | ||
1471 | =item C<warn_categories> | |
1472 | ||
1473 | The warning category (or categories) packed into a C<SVuv>. | |
1474 | ||
1475 | =item C<flag> | |
1476 | ||
1477 | A single flag bit associated with this message, in a C<SVuv>. | |
1478 | The bit corresponds to some bit in the C<*errors> return value, | |
1479 | such as C<UTF8_GOT_LONG>. | |
1480 | ||
1481 | =back | |
1482 | ||
1483 | It's important to note that specifying this parameter as non-null will cause | |
1484 | any warnings this function would otherwise generate to be suppressed, and | |
1485 | instead be placed in C<*msgs>. The caller can check the lexical warnings state | |
1486 | (or not) when choosing what to do with the returned messages. | |
1487 | ||
1488 | If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence | |
1489 | no AV is created. | |
1490 | ||
1491 | The caller, of course, is responsible for freeing any returned AV. | |
1492 | ||
1493 | =cut | |
1494 | */ | |
1495 | ||
1496 | UV | |
e6a4ffc3 | 1497 | Perl__utf8n_to_uvchr_msgs_helper(const U8 *s, |
37657a5b KW |
1498 | STRLEN curlen, |
1499 | STRLEN *retlen, | |
1500 | const U32 flags, | |
1501 | U32 * errors, | |
1502 | AV ** msgs) | |
a0ed51b3 | 1503 | { |
d4c19fe8 | 1504 | const U8 * const s0 = s; |
2b9519f0 | 1505 | const U8 * send = s0 + curlen; |
5af9f822 KW |
1506 | U32 possible_problems; /* A bit is set here for each potential problem |
1507 | found as we go along */ | |
1508 | UV uv; | |
1509 | STRLEN expectlen; /* How long should this sequence be? */ | |
1510 | STRLEN avail_len; /* When input is too short, gives what that is */ | |
1511 | U32 discard_errors; /* Used to save branches when 'errors' is NULL; this | |
1512 | gets set and discarded */ | |
a0dbb045 | 1513 | |
2b5e7bc2 KW |
1514 | /* The below are used only if there is both an overlong malformation and a |
1515 | * too short one. Otherwise the first two are set to 's0' and 'send', and | |
1516 | * the third not used at all */ | |
5af9f822 | 1517 | U8 * adjusted_s0; |
e9f2c446 KW |
1518 | U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this |
1519 | routine; see [perl #130921] */ | |
5af9f822 | 1520 | UV uv_so_far; |
e6a4ffc3 | 1521 | dTHX; |
5af9f822 | 1522 | |
e6a4ffc3 | 1523 | PERL_ARGS_ASSERT__UTF8N_TO_UVCHR_MSGS_HELPER; |
5af9f822 KW |
1524 | |
1525 | /* Here, is one of: a) malformed; b) a problematic code point (surrogate, | |
1526 | * non-unicode, or nonchar); or c) on ASCII platforms, one of the Hangul | |
1527 | * syllables that the dfa doesn't properly handle. Quickly dispose of the | |
1528 | * final case. */ | |
1529 | ||
1530 | #ifndef EBCDIC | |
1531 | ||
1532 | /* Each of the affected Hanguls starts with \xED */ | |
1533 | ||
1534 | if (is_HANGUL_ED_utf8_safe(s0, send)) { | |
1535 | if (retlen) { | |
1536 | *retlen = 3; | |
1537 | } | |
1538 | if (errors) { | |
1539 | *errors = 0; | |
1540 | } | |
1541 | if (msgs) { | |
1542 | *msgs = NULL; | |
1543 | } | |
1544 | ||
1545 | return ((0xED & UTF_START_MASK(3)) << (2 * UTF_ACCUMULATION_SHIFT)) | |
1546 | | ((s0[1] & UTF_CONTINUATION_MASK) << UTF_ACCUMULATION_SHIFT) | |
1547 | | (s0[2] & UTF_CONTINUATION_MASK); | |
1548 | } | |
1549 | ||
1550 | #endif | |
1551 | ||
1552 | /* In conjunction with the exhaustive tests that can be enabled in | |
1553 | * APItest/t/utf8_warn_base.pl, this can make sure the dfa does precisely | |
1554 | * what it is intended to do, and that no flaws in it are masked by | |
1555 | * dropping down and executing the code below | |
1556 | assert(! isUTF8_CHAR(s0, send) | |
1557 | || UTF8_IS_SURROGATE(s0, send) | |
1558 | || UTF8_IS_SUPER(s0, send) | |
1559 | || UTF8_IS_NONCHAR(s0,send)); | |
1560 | */ | |
1561 | ||
1562 | s = s0; | |
1563 | uv = *s0; | |
1564 | possible_problems = 0; | |
1565 | expectlen = 0; | |
1566 | avail_len = 0; | |
1567 | discard_errors = 0; | |
1568 | adjusted_s0 = (U8 *) s0; | |
1569 | uv_so_far = 0; | |
1570 | ||
f9380377 KW |
1571 | if (errors) { |
1572 | *errors = 0; | |
1573 | } | |
1574 | else { | |
1575 | errors = &discard_errors; | |
1576 | } | |
a0dbb045 | 1577 | |
eb83ed87 KW |
1578 | /* The order of malformation tests here is important. We should consume as |
1579 | * few bytes as possible in order to not skip any valid character. This is | |
1580 | * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also | |
1581 | * http://unicode.org/reports/tr36 for more discussion as to why. For | |
1582 | * example, once we've done a UTF8SKIP, we can tell the expected number of | |
1583 | * bytes, and could fail right off the bat if the input parameters indicate | |
1584 | * that there are too few available. But it could be that just that first | |
1585 | * byte is garbled, and the intended character occupies fewer bytes. If we | |
1586 | * blindly assumed that the first byte is correct, and skipped based on | |
1587 | * that number, we could skip over a valid input character. So instead, we | |
1588 | * always examine the sequence byte-by-byte. | |
1589 | * | |
1590 | * We also should not consume too few bytes, otherwise someone could inject | |
1591 | * things. For example, an input could be deliberately designed to | |
1592 | * overflow, and if this code bailed out immediately upon discovering that, | |
e2660c54 | 1593 | * returning to the caller C<*retlen> pointing to the very next byte (one |
eb83ed87 KW |
1594 | * which is actually part of of the overflowing sequence), that could look |
1595 | * legitimate to the caller, which could discard the initial partial | |
2b5e7bc2 KW |
1596 | * sequence and process the rest, inappropriately. |
1597 | * | |
1598 | * Some possible input sequences are malformed in more than one way. This | |
1599 | * function goes to lengths to try to find all of them. This is necessary | |
1600 | * for correctness, as the inputs may allow one malformation but not | |
1601 | * another, and if we abandon searching for others after finding the | |
1602 | * allowed one, we could allow in something that shouldn't have been. | |
1603 | */ | |
eb83ed87 | 1604 | |
b5b9af04 | 1605 | if (UNLIKELY(curlen == 0)) { |
2b5e7bc2 KW |
1606 | possible_problems |= UTF8_GOT_EMPTY; |
1607 | curlen = 0; | |
5a48568d | 1608 | uv = UNICODE_REPLACEMENT; |
2b5e7bc2 | 1609 | goto ready_to_handle_errors; |
0c443dc2 JH |
1610 | } |
1611 | ||
eb83ed87 KW |
1612 | expectlen = UTF8SKIP(s); |
1613 | ||
1614 | /* A well-formed UTF-8 character, as the vast majority of calls to this | |
1615 | * function will be for, has this expected length. For efficiency, set | |
1616 | * things up here to return it. It will be overriden only in those rare | |
1617 | * cases where a malformation is found */ | |
1618 | if (retlen) { | |
1619 | *retlen = expectlen; | |
1620 | } | |
1621 | ||
eb83ed87 | 1622 | /* A continuation character can't start a valid sequence */ |
b5b9af04 | 1623 | if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) { |
2b5e7bc2 KW |
1624 | possible_problems |= UTF8_GOT_CONTINUATION; |
1625 | curlen = 1; | |
1626 | uv = UNICODE_REPLACEMENT; | |
1627 | goto ready_to_handle_errors; | |
ba210ebe | 1628 | } |
9041c2e3 | 1629 | |
dcd27b3c | 1630 | /* Here is not a continuation byte, nor an invariant. The only thing left |
ddb65933 KW |
1631 | * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START |
1632 | * because it excludes start bytes like \xC0 that always lead to | |
1633 | * overlongs.) */ | |
dcd27b3c | 1634 | |
534752c1 KW |
1635 | /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits |
1636 | * that indicate the number of bytes in the character's whole UTF-8 | |
1637 | * sequence, leaving just the bits that are part of the value. */ | |
1638 | uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen); | |
ba210ebe | 1639 | |
e308b348 KW |
1640 | /* Setup the loop end point, making sure to not look past the end of the |
1641 | * input string, and flag it as too short if the size isn't big enough. */ | |
e308b348 KW |
1642 | if (UNLIKELY(curlen < expectlen)) { |
1643 | possible_problems |= UTF8_GOT_SHORT; | |
1644 | avail_len = curlen; | |
e308b348 KW |
1645 | } |
1646 | else { | |
2b9519f0 | 1647 | send = (U8*) s0 + expectlen; |
e308b348 | 1648 | } |
e308b348 | 1649 | |
eb83ed87 | 1650 | /* Now, loop through the remaining bytes in the character's sequence, |
e308b348 | 1651 | * accumulating each into the working value as we go. */ |
eb83ed87 | 1652 | for (s = s0 + 1; s < send; s++) { |
b5b9af04 | 1653 | if (LIKELY(UTF8_IS_CONTINUATION(*s))) { |
8850bf83 | 1654 | uv = UTF8_ACCUMULATE(uv, *s); |
2b5e7bc2 KW |
1655 | continue; |
1656 | } | |
1657 | ||
1658 | /* Here, found a non-continuation before processing all expected bytes. | |
1659 | * This byte indicates the beginning of a new character, so quit, even | |
1660 | * if allowing this malformation. */ | |
2b5e7bc2 | 1661 | possible_problems |= UTF8_GOT_NON_CONTINUATION; |
e308b348 | 1662 | break; |
eb83ed87 KW |
1663 | } /* End of loop through the character's bytes */ |
1664 | ||
1665 | /* Save how many bytes were actually in the character */ | |
1666 | curlen = s - s0; | |
1667 | ||
2b5e7bc2 KW |
1668 | /* Note that there are two types of too-short malformation. One is when |
1669 | * there is actual wrong data before the normal termination of the | |
1670 | * sequence. The other is that the sequence wasn't complete before the end | |
1671 | * of the data we are allowed to look at, based on the input 'curlen'. | |
1672 | * This means that we were passed data for a partial character, but it is | |
1673 | * valid as far as we saw. The other is definitely invalid. This | |
1674 | * distinction could be important to a caller, so the two types are kept | |
15b010f0 KW |
1675 | * separate. |
1676 | * | |
1677 | * A convenience macro that matches either of the too-short conditions. */ | |
1678 | # define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION) | |
1679 | ||
1680 | if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) { | |
1681 | uv_so_far = uv; | |
1682 | uv = UNICODE_REPLACEMENT; | |
1683 | } | |
2b5e7bc2 | 1684 | |
08e73697 KW |
1685 | /* Check for overflow. The algorithm requires us to not look past the end |
1686 | * of the current character, even if partial, so the upper limit is 's' */ | |
e050c007 KW |
1687 | if (UNLIKELY(0 < does_utf8_overflow(s0, s, |
1688 | 1 /* Do consider overlongs */ | |
1689 | ))) | |
1690 | { | |
2b5e7bc2 KW |
1691 | possible_problems |= UTF8_GOT_OVERFLOW; |
1692 | uv = UNICODE_REPLACEMENT; | |
eb83ed87 | 1693 | } |
eb83ed87 | 1694 | |
2b5e7bc2 KW |
1695 | /* Check for overlong. If no problems so far, 'uv' is the correct code |
1696 | * point value. Simply see if it is expressible in fewer bytes. Otherwise | |
1697 | * we must look at the UTF-8 byte sequence itself to see if it is for an | |
1698 | * overlong */ | |
1699 | if ( ( LIKELY(! possible_problems) | |
1700 | && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv))) | |
56576a04 | 1701 | || ( UNLIKELY(possible_problems) |
2b5e7bc2 KW |
1702 | && ( UNLIKELY(! UTF8_IS_START(*s0)) |
1703 | || ( curlen > 1 | |
d6be65ae | 1704 | && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0, |
08e73697 | 1705 | s - s0)))))) |
2f8f112e | 1706 | { |
2b5e7bc2 KW |
1707 | possible_problems |= UTF8_GOT_LONG; |
1708 | ||
abc28b54 | 1709 | if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT) |
56576a04 | 1710 | |
abc28b54 KW |
1711 | /* The calculation in the 'true' branch of this 'if' |
1712 | * below won't work if overflows, and isn't needed | |
1713 | * anyway. Further below we handle all overflow | |
1714 | * cases */ | |
1715 | && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))) | |
1716 | { | |
2b5e7bc2 KW |
1717 | UV min_uv = uv_so_far; |
1718 | STRLEN i; | |
1719 | ||
1720 | /* Here, the input is both overlong and is missing some trailing | |
1721 | * bytes. There is no single code point it could be for, but there | |
1722 | * may be enough information present to determine if what we have | |
1723 | * so far is for an unallowed code point, such as for a surrogate. | |
56576a04 KW |
1724 | * The code further below has the intelligence to determine this, |
1725 | * but just for non-overlong UTF-8 sequences. What we do here is | |
1726 | * calculate the smallest code point the input could represent if | |
1727 | * there were no too short malformation. Then we compute and save | |
1728 | * the UTF-8 for that, which is what the code below looks at | |
1729 | * instead of the raw input. It turns out that the smallest such | |
1730 | * code point is all we need. */ | |
2b5e7bc2 KW |
1731 | for (i = curlen; i < expectlen; i++) { |
1732 | min_uv = UTF8_ACCUMULATE(min_uv, | |
1733 | I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK)); | |
1734 | } | |
1735 | ||
e9f2c446 | 1736 | adjusted_s0 = temp_char_buf; |
57ff5f59 | 1737 | (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0); |
2b5e7bc2 | 1738 | } |
eb83ed87 KW |
1739 | } |
1740 | ||
56576a04 KW |
1741 | /* Here, we have found all the possible problems, except for when the input |
1742 | * is for a problematic code point not allowed by the input parameters. */ | |
1743 | ||
06188866 KW |
1744 | /* uv is valid for overlongs */ |
1745 | if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG)) | |
1746 | ||
1747 | /* isn't problematic if < this */ | |
1748 | && uv >= UNICODE_SURROGATE_FIRST) | |
2b5e7bc2 | 1749 | || ( UNLIKELY(possible_problems) |
d60baaa7 KW |
1750 | |
1751 | /* if overflow, we know without looking further | |
1752 | * precisely which of the problematic types it is, | |
1753 | * and we deal with those in the overflow handling | |
1754 | * code */ | |
1755 | && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)) | |
57ff5f59 KW |
1756 | && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0) |
1757 | || UNLIKELY(isUTF8_PERL_EXTENDED(s0))))) | |
760c7c2f KW |
1758 | && ((flags & ( UTF8_DISALLOW_NONCHAR |
1759 | |UTF8_DISALLOW_SURROGATE | |
1760 | |UTF8_DISALLOW_SUPER | |
d044b7a7 | 1761 | |UTF8_DISALLOW_PERL_EXTENDED |
760c7c2f KW |
1762 | |UTF8_WARN_NONCHAR |
1763 | |UTF8_WARN_SURROGATE | |
1764 | |UTF8_WARN_SUPER | |
d22ec717 | 1765 | |UTF8_WARN_PERL_EXTENDED)))) |
eb83ed87 | 1766 | { |
2b5e7bc2 KW |
1767 | /* If there were no malformations, or the only malformation is an |
1768 | * overlong, 'uv' is valid */ | |
1769 | if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) { | |
1770 | if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) { | |
1771 | possible_problems |= UTF8_GOT_SURROGATE; | |
1772 | } | |
1773 | else if (UNLIKELY(uv > PERL_UNICODE_MAX)) { | |
1774 | possible_problems |= UTF8_GOT_SUPER; | |
1775 | } | |
1776 | else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) { | |
1777 | possible_problems |= UTF8_GOT_NONCHAR; | |
1778 | } | |
1779 | } | |
1780 | else { /* Otherwise, need to look at the source UTF-8, possibly | |
1781 | adjusted to be non-overlong */ | |
1782 | ||
1783 | if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0) | |
1784 | >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER)) | |
ea5ced44 | 1785 | { |
2b5e7bc2 KW |
1786 | possible_problems |= UTF8_GOT_SUPER; |
1787 | } | |
1788 | else if (curlen > 1) { | |
1789 | if (UNLIKELY(IS_UTF8_2_BYTE_SUPER( | |
1790 | NATIVE_UTF8_TO_I8(*adjusted_s0), | |
1791 | NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1))))) | |
ea5ced44 | 1792 | { |
2b5e7bc2 | 1793 | possible_problems |= UTF8_GOT_SUPER; |
ea5ced44 | 1794 | } |
2b5e7bc2 KW |
1795 | else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE( |
1796 | NATIVE_UTF8_TO_I8(*adjusted_s0), | |
1797 | NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1))))) | |
1798 | { | |
1799 | possible_problems |= UTF8_GOT_SURROGATE; | |
ea5ced44 KW |
1800 | } |
1801 | } | |
c0236afe | 1802 | |
2b5e7bc2 KW |
1803 | /* We need a complete well-formed UTF-8 character to discern |
1804 | * non-characters, so can't look for them here */ | |
1805 | } | |
1806 | } | |
949cf498 | 1807 | |
2b5e7bc2 KW |
1808 | ready_to_handle_errors: |
1809 | ||
1810 | /* At this point: | |
1811 | * curlen contains the number of bytes in the sequence that | |
1812 | * this call should advance the input by. | |
e308b348 KW |
1813 | * avail_len gives the available number of bytes passed in, but |
1814 | * only if this is less than the expected number of | |
1815 | * bytes, based on the code point's start byte. | |
2b5e7bc2 KW |
1816 | * possible_problems' is 0 if there weren't any problems; otherwise a bit |
1817 | * is set in it for each potential problem found. | |
1818 | * uv contains the code point the input sequence | |
1819 | * represents; or if there is a problem that prevents | |
1820 | * a well-defined value from being computed, it is | |
1821 | * some subsitute value, typically the REPLACEMENT | |
1822 | * CHARACTER. | |
1823 | * s0 points to the first byte of the character | |
56576a04 KW |
1824 | * s points to just after were we left off processing |
1825 | * the character | |
1826 | * send points to just after where that character should | |
1827 | * end, based on how many bytes the start byte tells | |
1828 | * us should be in it, but no further than s0 + | |
1829 | * avail_len | |
2b5e7bc2 | 1830 | */ |
eb83ed87 | 1831 | |
2b5e7bc2 KW |
1832 | if (UNLIKELY(possible_problems)) { |
1833 | bool disallowed = FALSE; | |
1834 | const U32 orig_problems = possible_problems; | |
1835 | ||
37657a5b KW |
1836 | if (msgs) { |
1837 | *msgs = NULL; | |
1838 | } | |
1839 | ||
2b5e7bc2 KW |
1840 | while (possible_problems) { /* Handle each possible problem */ |
1841 | UV pack_warn = 0; | |
1842 | char * message = NULL; | |
37657a5b | 1843 | U32 this_flag_bit = 0; |
2b5e7bc2 KW |
1844 | |
1845 | /* Each 'if' clause handles one problem. They are ordered so that | |
1846 | * the first ones' messages will be displayed before the later | |
6c64cd9d KW |
1847 | * ones; this is kinda in decreasing severity order. But the |
1848 | * overlong must come last, as it changes 'uv' looked at by the | |
1849 | * others */ | |
2b5e7bc2 KW |
1850 | if (possible_problems & UTF8_GOT_OVERFLOW) { |
1851 | ||
56576a04 KW |
1852 | /* Overflow means also got a super and are using Perl's |
1853 | * extended UTF-8, but we handle all three cases here */ | |
2b5e7bc2 | 1854 | possible_problems |
d044b7a7 | 1855 | &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED); |
f9380377 KW |
1856 | *errors |= UTF8_GOT_OVERFLOW; |
1857 | ||
1858 | /* But the API says we flag all errors found */ | |
1859 | if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) { | |
1860 | *errors |= UTF8_GOT_SUPER; | |
1861 | } | |
ddb65933 | 1862 | if (flags |
d044b7a7 | 1863 | & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED)) |
ddb65933 | 1864 | { |
d044b7a7 | 1865 | *errors |= UTF8_GOT_PERL_EXTENDED; |
f9380377 | 1866 | } |
2b5e7bc2 | 1867 | |
d60baaa7 | 1868 | /* Disallow if any of the three categories say to */ |
56576a04 | 1869 | if ( ! (flags & UTF8_ALLOW_OVERFLOW) |
d60baaa7 | 1870 | || (flags & ( UTF8_DISALLOW_SUPER |
d044b7a7 | 1871 | |UTF8_DISALLOW_PERL_EXTENDED))) |
d60baaa7 KW |
1872 | { |
1873 | disallowed = TRUE; | |
1874 | } | |
1875 | ||
d22ec717 KW |
1876 | /* Likewise, warn if any say to */ |
1877 | if ( ! (flags & UTF8_ALLOW_OVERFLOW) | |
1878 | || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED))) | |
d60baaa7 | 1879 | { |
2b5e7bc2 | 1880 | |
ddb65933 KW |
1881 | /* The warnings code explicitly says it doesn't handle the |
1882 | * case of packWARN2 and two categories which have | |
1883 | * parent-child relationship. Even if it works now to | |
1884 | * raise the warning if either is enabled, it wouldn't | |
1885 | * necessarily do so in the future. We output (only) the | |
56576a04 | 1886 | * most dire warning */ |
ddb65933 | 1887 | if (! (flags & UTF8_CHECK_ONLY)) { |
37657a5b | 1888 | if (msgs || ckWARN_d(WARN_UTF8)) { |
ddb65933 KW |
1889 | pack_warn = packWARN(WARN_UTF8); |
1890 | } | |
37657a5b | 1891 | else if (msgs || ckWARN_d(WARN_NON_UNICODE)) { |
ddb65933 KW |
1892 | pack_warn = packWARN(WARN_NON_UNICODE); |
1893 | } | |
1894 | if (pack_warn) { | |
1895 | message = Perl_form(aTHX_ "%s: %s (overflows)", | |
1896 | malformed_text, | |
05b9033b | 1897 | _byte_dump_string(s0, curlen, 0)); |
37657a5b | 1898 | this_flag_bit = UTF8_GOT_OVERFLOW; |
ddb65933 | 1899 | } |
2b5e7bc2 KW |
1900 | } |
1901 | } | |
1902 | } | |
1903 | else if (possible_problems & UTF8_GOT_EMPTY) { | |
1904 | possible_problems &= ~UTF8_GOT_EMPTY; | |
f9380377 | 1905 | *errors |= UTF8_GOT_EMPTY; |
2b5e7bc2 KW |
1906 | |
1907 | if (! (flags & UTF8_ALLOW_EMPTY)) { | |
d1f8d421 KW |
1908 | |
1909 | /* This so-called malformation is now treated as a bug in | |
1910 | * the caller. If you have nothing to decode, skip calling | |
1911 | * this function */ | |
1912 | assert(0); | |
1913 | ||
2b5e7bc2 | 1914 | disallowed = TRUE; |
37657a5b KW |
1915 | if ( (msgs |
1916 | || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY)) | |
1917 | { | |
2b5e7bc2 KW |
1918 | pack_warn = packWARN(WARN_UTF8); |
1919 | message = Perl_form(aTHX_ "%s (empty string)", | |
1920 | malformed_text); | |
37657a5b | 1921 | this_flag_bit = UTF8_GOT_EMPTY; |
2b5e7bc2 KW |
1922 | } |
1923 | } | |
1924 | } | |
1925 | else if (possible_problems & UTF8_GOT_CONTINUATION) { | |
1926 | possible_problems &= ~UTF8_GOT_CONTINUATION; | |
f9380377 | 1927 | *errors |= UTF8_GOT_CONTINUATION; |
2b5e7bc2 KW |
1928 | |
1929 | if (! (flags & UTF8_ALLOW_CONTINUATION)) { | |
1930 | disallowed = TRUE; | |
37657a5b KW |
1931 | if (( msgs |
1932 | || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY)) | |
1933 | { | |
2b5e7bc2 KW |
1934 | pack_warn = packWARN(WARN_UTF8); |
1935 | message = Perl_form(aTHX_ | |
1936 | "%s: %s (unexpected continuation byte 0x%02x," | |
1937 | " with no preceding start byte)", | |
1938 | malformed_text, | |
7e2f38b2 | 1939 | _byte_dump_string(s0, 1, 0), *s0); |
37657a5b | 1940 | this_flag_bit = UTF8_GOT_CONTINUATION; |
2b5e7bc2 KW |
1941 | } |
1942 | } | |
1943 | } | |
2b5e7bc2 KW |
1944 | else if (possible_problems & UTF8_GOT_SHORT) { |
1945 | possible_problems &= ~UTF8_GOT_SHORT; | |
f9380377 | 1946 | *errors |= UTF8_GOT_SHORT; |
2b5e7bc2 KW |
1947 | |
1948 | if (! (flags & UTF8_ALLOW_SHORT)) { | |
1949 | disallowed = TRUE; | |
37657a5b KW |
1950 | if (( msgs |
1951 | || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY)) | |
1952 | { | |
2b5e7bc2 KW |
1953 | pack_warn = packWARN(WARN_UTF8); |
1954 | message = Perl_form(aTHX_ | |
56576a04 KW |
1955 | "%s: %s (too short; %d byte%s available, need %d)", |
1956 | malformed_text, | |
1957 | _byte_dump_string(s0, send - s0, 0), | |
1958 | (int)avail_len, | |
1959 | avail_len == 1 ? "" : "s", | |
1960 | (int)expectlen); | |
37657a5b | 1961 | this_flag_bit = UTF8_GOT_SHORT; |
2b5e7bc2 KW |
1962 | } |
1963 | } | |
ba210ebe | 1964 | |
2b5e7bc2 | 1965 | } |
e308b348 KW |
1966 | else if (possible_problems & UTF8_GOT_NON_CONTINUATION) { |
1967 | possible_problems &= ~UTF8_GOT_NON_CONTINUATION; | |
1968 | *errors |= UTF8_GOT_NON_CONTINUATION; | |
1969 | ||
1970 | if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) { | |
1971 | disallowed = TRUE; | |
37657a5b KW |
1972 | if (( msgs |
1973 | || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY)) | |
1974 | { | |
99a765e9 KW |
1975 | |
1976 | /* If we don't know for sure that the input length is | |
1977 | * valid, avoid as much as possible reading past the | |
1978 | * end of the buffer */ | |
1979 | int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN) | |
1980 | ? s - s0 | |
1981 | : send - s0; | |
e308b348 KW |
1982 | pack_warn = packWARN(WARN_UTF8); |
1983 | message = Perl_form(aTHX_ "%s", | |
1984 | unexpected_non_continuation_text(s0, | |
99a765e9 | 1985 | printlen, |
e308b348 KW |
1986 | s - s0, |
1987 | (int) expectlen)); | |
37657a5b | 1988 | this_flag_bit = UTF8_GOT_NON_CONTINUATION; |
e308b348 KW |
1989 | } |
1990 | } | |
1991 | } | |
2b5e7bc2 KW |
1992 | else if (possible_problems & UTF8_GOT_SURROGATE) { |
1993 | possible_problems &= ~UTF8_GOT_SURROGATE; | |
1994 | ||
f9380377 KW |
1995 | if (flags & UTF8_WARN_SURROGATE) { |
1996 | *errors |= UTF8_GOT_SURROGATE; | |
1997 | ||
1998 | if ( ! (flags & UTF8_CHECK_ONLY) | |
37657a5b | 1999 | && (msgs || ckWARN_d(WARN_SURROGATE))) |
f9380377 | 2000 | { |
2b5e7bc2 KW |
2001 | pack_warn = packWARN(WARN_SURROGATE); |
2002 | ||
2003 | /* These are the only errors that can occur with a | |
2004 | * surrogate when the 'uv' isn't valid */ | |
2005 | if (orig_problems & UTF8_GOT_TOO_SHORT) { | |
2006 | message = Perl_form(aTHX_ | |
2007 | "UTF-16 surrogate (any UTF-8 sequence that" | |
2008 | " starts with \"%s\" is for a surrogate)", | |
7e2f38b2 | 2009 | _byte_dump_string(s0, curlen, 0)); |
2b5e7bc2 KW |
2010 | } |
2011 | else { | |
c94c2f39 | 2012 | message = Perl_form(aTHX_ surrogate_cp_format, uv); |
2b5e7bc2 | 2013 | } |
37657a5b | 2014 | this_flag_bit = UTF8_GOT_SURROGATE; |
f9380377 | 2015 | } |
2b5e7bc2 | 2016 | } |
ba210ebe | 2017 | |
2b5e7bc2 KW |
2018 | if (flags & UTF8_DISALLOW_SURROGATE) { |
2019 | disallowed = TRUE; | |
f9380377 | 2020 | *errors |= UTF8_GOT_SURROGATE; |
2b5e7bc2 KW |
2021 | } |
2022 | } | |
2023 | else if (possible_problems & UTF8_GOT_SUPER) { | |
2024 | possible_problems &= ~UTF8_GOT_SUPER; | |
949cf498 | 2025 | |
f9380377 KW |
2026 | if (flags & UTF8_WARN_SUPER) { |
2027 | *errors |= UTF8_GOT_SUPER; | |
2028 | ||
2029 | if ( ! (flags & UTF8_CHECK_ONLY) | |
37657a5b | 2030 | && (msgs || ckWARN_d(WARN_NON_UNICODE))) |
f9380377 | 2031 | { |
2b5e7bc2 KW |
2032 | pack_warn = packWARN(WARN_NON_UNICODE); |
2033 | ||
2034 | if (orig_problems & UTF8_GOT_TOO_SHORT) { | |
2035 | message = Perl_form(aTHX_ | |
2036 | "Any UTF-8 sequence that starts with" | |
2037 | " \"%s\" is for a non-Unicode code point," | |
2038 | " may not be portable", | |
7e2f38b2 | 2039 | _byte_dump_string(s0, curlen, 0)); |
2b5e7bc2 KW |
2040 | } |
2041 | else { | |
c94c2f39 | 2042 | message = Perl_form(aTHX_ super_cp_format, uv); |
2b5e7bc2 | 2043 | } |
37657a5b | 2044 | this_flag_bit = UTF8_GOT_SUPER; |
f9380377 | 2045 | } |
2b5e7bc2 | 2046 | } |
ba210ebe | 2047 | |
57ff5f59 KW |
2048 | /* Test for Perl's extended UTF-8 after the regular SUPER ones, |
2049 | * and before possibly bailing out, so that the more dire | |
2050 | * warning will override the regular one. */ | |
2051 | if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) { | |
2b5e7bc2 | 2052 | if ( ! (flags & UTF8_CHECK_ONLY) |
d044b7a7 | 2053 | && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER)) |
37657a5b | 2054 | && (msgs || ckWARN_d(WARN_NON_UNICODE))) |
2b5e7bc2 | 2055 | { |
db0f09e6 | 2056 | pack_warn = packWARN(WARN_NON_UNICODE); |
2b5e7bc2 | 2057 | |
57ff5f59 KW |
2058 | /* If it is an overlong that evaluates to a code point |
2059 | * that doesn't have to use the Perl extended UTF-8, it | |
2060 | * still used it, and so we output a message that | |
2061 | * doesn't refer to the code point. The same is true | |
2062 | * if there was a SHORT malformation where the code | |
2063 | * point is not valid. In that case, 'uv' will have | |
2064 | * been set to the REPLACEMENT CHAR, and the message | |
2065 | * below without the code point in it will be selected | |
2066 | * */ | |
2067 | if (UNICODE_IS_PERL_EXTENDED(uv)) { | |
2b5e7bc2 | 2068 | message = Perl_form(aTHX_ |
57ff5f59 | 2069 | perl_extended_cp_format, uv); |
2b5e7bc2 KW |
2070 | } |
2071 | else { | |
2072 | message = Perl_form(aTHX_ | |
57ff5f59 KW |
2073 | "Any UTF-8 sequence that starts with" |
2074 | " \"%s\" is a Perl extension, and" | |
2075 | " so is not portable", | |
2076 | _byte_dump_string(s0, curlen, 0)); | |
2b5e7bc2 | 2077 | } |
37657a5b | 2078 | this_flag_bit = UTF8_GOT_PERL_EXTENDED; |
2b5e7bc2 KW |
2079 | } |
2080 | ||
d044b7a7 KW |
2081 | if (flags & ( UTF8_WARN_PERL_EXTENDED |
2082 | |UTF8_DISALLOW_PERL_EXTENDED)) | |
ddb65933 | 2083 | { |
d044b7a7 | 2084 | *errors |= UTF8_GOT_PERL_EXTENDED; |
f9380377 | 2085 | |
d044b7a7 | 2086 | if (flags & UTF8_DISALLOW_PERL_EXTENDED) { |
f9380377 KW |
2087 | disallowed = TRUE; |
2088 | } | |
2b5e7bc2 KW |
2089 | } |
2090 | } | |
eb83ed87 | 2091 | |
2b5e7bc2 | 2092 | if (flags & UTF8_DISALLOW_SUPER) { |
f9380377 | 2093 | *errors |= UTF8_GOT_SUPER; |
2b5e7bc2 KW |
2094 | disallowed = TRUE; |
2095 | } | |
2b5e7bc2 KW |
2096 | } |
2097 | else if (possible_problems & UTF8_GOT_NONCHAR) { | |
2098 | possible_problems &= ~UTF8_GOT_NONCHAR; | |
ba210ebe | 2099 | |
f9380377 KW |
2100 | if (flags & UTF8_WARN_NONCHAR) { |
2101 | *errors |= UTF8_GOT_NONCHAR; | |
2102 | ||
2103 | if ( ! (flags & UTF8_CHECK_ONLY) | |
37657a5b | 2104 | && (msgs || ckWARN_d(WARN_NONCHAR))) |
f9380377 | 2105 | { |
2b5e7bc2 KW |
2106 | /* The code above should have guaranteed that we don't |
2107 | * get here with errors other than overlong */ | |
2108 | assert (! (orig_problems | |
2109 | & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR))); | |
2110 | ||
2111 | pack_warn = packWARN(WARN_NONCHAR); | |
c94c2f39 | 2112 | message = Perl_form(aTHX_ nonchar_cp_format, uv); |
37657a5b | 2113 | this_flag_bit = UTF8_GOT_NONCHAR; |
f9380377 | 2114 | } |
2b5e7bc2 | 2115 | } |
5b311467 | 2116 | |
2b5e7bc2 KW |
2117 | if (flags & UTF8_DISALLOW_NONCHAR) { |
2118 | disallowed = TRUE; | |
f9380377 | 2119 | *errors |= UTF8_GOT_NONCHAR; |
2b5e7bc2 | 2120 | } |
6c64cd9d KW |
2121 | } |
2122 | else if (possible_problems & UTF8_GOT_LONG) { | |
2123 | possible_problems &= ~UTF8_GOT_LONG; | |
2124 | *errors |= UTF8_GOT_LONG; | |
2125 | ||
2126 | if (flags & UTF8_ALLOW_LONG) { | |
2127 | ||
2128 | /* We don't allow the actual overlong value, unless the | |
2129 | * special extra bit is also set */ | |
2130 | if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE | |
2131 | & ~UTF8_ALLOW_LONG))) | |
2132 | { | |
2133 | uv = UNICODE_REPLACEMENT; | |
2134 | } | |
2135 | } | |
2136 | else { | |
2137 | disallowed = TRUE; | |
2138 | ||
37657a5b KW |
2139 | if (( msgs |
2140 | || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY)) | |
2141 | { | |
6c64cd9d KW |
2142 | pack_warn = packWARN(WARN_UTF8); |
2143 | ||
2144 | /* These error types cause 'uv' to be something that | |
2145 | * isn't what was intended, so can't use it in the | |
2146 | * message. The other error types either can't | |
2147 | * generate an overlong, or else the 'uv' is valid */ | |
2148 | if (orig_problems & | |
2149 | (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW)) | |
2150 | { | |
2151 | message = Perl_form(aTHX_ | |
2152 | "%s: %s (any UTF-8 sequence that starts" | |
2153 | " with \"%s\" is overlong which can and" | |
2154 | " should be represented with a" | |
2155 | " different, shorter sequence)", | |
2156 | malformed_text, | |
2157 | _byte_dump_string(s0, send - s0, 0), | |
2158 | _byte_dump_string(s0, curlen, 0)); | |
2159 | } | |
2160 | else { | |
2161 | U8 tmpbuf[UTF8_MAXBYTES+1]; | |
1be62ab9 KW |
2162 | const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf, |
2163 | uv, 0); | |
d819dc50 KW |
2164 | /* Don't use U+ for non-Unicode code points, which |
2165 | * includes those in the Latin1 range */ | |
2166 | const char * preface = ( uv > PERL_UNICODE_MAX | |
2167 | #ifdef EBCDIC | |
2168 | || uv <= 0xFF | |
2169 | #endif | |
2170 | ) | |
2171 | ? "0x" | |
2172 | : "U+"; | |
6c64cd9d KW |
2173 | message = Perl_form(aTHX_ |
2174 | "%s: %s (overlong; instead use %s to represent" | |
2175 | " %s%0*" UVXf ")", | |
2176 | malformed_text, | |
2177 | _byte_dump_string(s0, send - s0, 0), | |
2178 | _byte_dump_string(tmpbuf, e - tmpbuf, 0), | |
2179 | preface, | |
2180 | ((uv < 256) ? 2 : 4), /* Field width of 2 for | |
2181 | small code points */ | |
1be62ab9 | 2182 | UNI_TO_NATIVE(uv)); |
6c64cd9d | 2183 | } |
37657a5b | 2184 | this_flag_bit = UTF8_GOT_LONG; |
6c64cd9d KW |
2185 | } |
2186 | } | |
2b5e7bc2 KW |
2187 | } /* End of looking through the possible flags */ |
2188 | ||
2189 | /* Display the message (if any) for the problem being handled in | |
2190 | * this iteration of the loop */ | |
2191 | if (message) { | |
37657a5b | 2192 | if (msgs) { |
37657a5b KW |
2193 | assert(this_flag_bit); |
2194 | ||
2195 | if (*msgs == NULL) { | |
2196 | *msgs = newAV(); | |
2197 | } | |
2198 | ||
bb07812e KW |
2199 | av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message, |
2200 | pack_warn, | |
2201 | this_flag_bit))); | |
37657a5b KW |
2202 | } |
2203 | else if (PL_op) | |
2b5e7bc2 KW |
2204 | Perl_warner(aTHX_ pack_warn, "%s in %s", message, |
2205 | OP_DESC(PL_op)); | |
2206 | else | |
2207 | Perl_warner(aTHX_ pack_warn, "%s", message); | |
2208 | } | |
ddb65933 | 2209 | } /* End of 'while (possible_problems)' */ |
a0dbb045 | 2210 | |
2b5e7bc2 KW |
2211 | /* Since there was a possible problem, the returned length may need to |
2212 | * be changed from the one stored at the beginning of this function. | |
2213 | * Instead of trying to figure out if that's needed, just do it. */ | |
2214 | if (retlen) { | |
2215 | *retlen = curlen; | |
2216 | } | |
a0dbb045 | 2217 | |
2b5e7bc2 KW |
2218 | if (disallowed) { |
2219 | if (flags & UTF8_CHECK_ONLY && retlen) { | |
2220 | *retlen = ((STRLEN) -1); | |
2221 | } | |
2222 | return 0; | |
2223 | } | |
eb83ed87 | 2224 | } |
ba210ebe | 2225 | |
2b5e7bc2 | 2226 | return UNI_TO_NATIVE(uv); |
a0ed51b3 LW |
2227 | } |
2228 | ||
8e84507e | 2229 | /* |
ec5f19d0 KW |
2230 | =for apidoc utf8_to_uvchr_buf |
2231 | ||
2232 | Returns the native code point of the first character in the string C<s> which | |
2233 | is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>. | |
524080c4 | 2234 | C<*retlen> will be set to the length, in bytes, of that character. |
ec5f19d0 | 2235 | |
524080c4 KW |
2236 | If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are |
2237 | enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't | |
796b6530 | 2238 | C<NULL>) to -1. If those warnings are off, the computed value, if well-defined |
173db420 | 2239 | (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and |
796b6530 | 2240 | C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is |
173db420 | 2241 | the next possible position in C<s> that could begin a non-malformed character. |
de69f3af | 2242 | See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is |
173db420 | 2243 | returned. |
ec5f19d0 KW |
2244 | |
2245 | =cut | |
52be2536 KW |
2246 | |
2247 | Also implemented as a macro in utf8.h | |
2248 | ||
ec5f19d0 KW |
2249 | */ |
2250 | ||
2251 | ||
2252 | UV | |
2253 | Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen) | |
2254 | { | |
7f974d7e KW |
2255 | PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF; |
2256 | ||
ec5f19d0 KW |
2257 | assert(s < send); |
2258 | ||
2259 | return utf8n_to_uvchr(s, send - s, retlen, | |
ddb65933 | 2260 | ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY); |
ec5f19d0 KW |
2261 | } |
2262 | ||
52be2536 KW |
2263 | /* This is marked as deprecated |
2264 | * | |
ec5f19d0 KW |
2265 | =for apidoc utf8_to_uvuni_buf |
2266 | ||
de69f3af KW |
2267 | Only in very rare circumstances should code need to be dealing in Unicode |
2268 | (as opposed to native) code points. In those few cases, use | |
07dfe0a4 KW |
2269 | C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead. If you |
2270 | are not absolutely sure this is one of those cases, then assume it isn't and | |
2271 | use plain C<utf8_to_uvchr_buf> instead. | |
4f83cdcd KW |
2272 | |
2273 | Returns the Unicode (not-native) code point of the first character in the | |
2274 | string C<s> which | |
ec5f19d0 KW |
2275 | is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>. |
2276 | C<retlen> will be set to the length, in bytes, of that character. | |
2277 | ||
524080c4 KW |
2278 | If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are |
2279 | enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't | |
2280 | NULL) to -1. If those warnings are off, the computed value if well-defined (or | |
2281 | the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen> | |
2282 | is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the | |
2283 | next possible position in C<s> that could begin a non-malformed character. | |
de69f3af | 2284 | See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned. |
ec5f19d0 KW |
2285 | |
2286 | =cut | |
2287 | */ | |
2288 | ||
2289 | UV | |
2290 | Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen) | |
2291 | { | |
2292 | PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF; | |
2293 | ||
2294 | assert(send > s); | |
2295 | ||
5962d97e | 2296 | return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen)); |
ec5f19d0 KW |
2297 | } |
2298 | ||
b76347f2 | 2299 | /* |
87cea99e | 2300 | =for apidoc utf8_length |
b76347f2 | 2301 | |
b2e7ed74 KW |
2302 | Returns the number of characters in the sequence of UTF-8-encoded bytes starting |
2303 | at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the | |
2304 | same place, it returns 0 with no warning raised. | |
2305 | ||
2306 | If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning | |
2307 | and returns the number of valid characters. | |
b76347f2 JH |
2308 | |
2309 | =cut | |
2310 | */ | |
2311 | ||
2312 | STRLEN | |
35a4481c | 2313 | Perl_utf8_length(pTHX_ const U8 *s, const U8 *e) |
b76347f2 JH |
2314 | { |
2315 | STRLEN len = 0; | |
2316 | ||
7918f24d NC |
2317 | PERL_ARGS_ASSERT_UTF8_LENGTH; |
2318 | ||
8850bf83 JH |
2319 | /* Note: cannot use UTF8_IS_...() too eagerly here since e.g. |
2320 | * the bitops (especially ~) can create illegal UTF-8. | |
2321 | * In other words: in Perl UTF-8 is not just for Unicode. */ | |
2322 | ||
a3b680e6 AL |
2323 | if (e < s) |
2324 | goto warn_and_return; | |
b76347f2 | 2325 | while (s < e) { |
4cbf4130 | 2326 | s += UTF8SKIP(s); |
8e91ec7f AV |
2327 | len++; |
2328 | } | |
2329 | ||
2330 | if (e != s) { | |
2331 | len--; | |
2332 | warn_and_return: | |
9b387841 NC |
2333 | if (PL_op) |
2334 | Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), | |
2335 | "%s in %s", unees, OP_DESC(PL_op)); | |
2336 | else | |
61a12c31 | 2337 | Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees); |
b76347f2 JH |
2338 | } |
2339 | ||
2340 | return len; | |
2341 | } | |
2342 | ||
b06226ff | 2343 | /* |
fed3ba5d NC |
2344 | =for apidoc bytes_cmp_utf8 |
2345 | ||
a1433954 | 2346 | Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the |
72d33970 FC |
2347 | sequence of characters (stored as UTF-8) |
2348 | in C<u>, C<ulen>. Returns 0 if they are | |
fed3ba5d NC |
2349 | equal, -1 or -2 if the first string is less than the second string, +1 or +2 |
2350 | if the first string is greater than the second string. | |
2351 | ||
2352 | -1 or +1 is returned if the shorter string was identical to the start of the | |
72d33970 FC |
2353 | longer string. -2 or +2 is returned if |
2354 | there was a difference between characters | |
fed3ba5d NC |
2355 | within the strings. |
2356 | ||
2357 | =cut | |
2358 | */ | |
2359 | ||
2360 | int | |
2361 | Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen) | |
2362 | { | |
2363 | const U8 *const bend = b + blen; | |
2364 | const U8 *const uend = u + ulen; | |
2365 | ||
2366 | PERL_ARGS_ASSERT_BYTES_CMP_UTF8; | |
fed3ba5d NC |
2367 | |
2368 | while (b < bend && u < uend) { | |
2369 | U8 c = *u++; | |
2370 | if (!UTF8_IS_INVARIANT(c)) { | |
2371 | if (UTF8_IS_DOWNGRADEABLE_START(c)) { | |
2372 | if (u < uend) { | |
2373 | U8 c1 = *u++; | |
2374 | if (UTF8_IS_CONTINUATION(c1)) { | |
a62b247b | 2375 | c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1); |
fed3ba5d | 2376 | } else { |
2b5e7bc2 | 2377 | /* diag_listed_as: Malformed UTF-8 character%s */ |
fed3ba5d | 2378 | Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), |
56576a04 KW |
2379 | "%s %s%s", |
2380 | unexpected_non_continuation_text(u - 2, 2, 1, 2), | |
2381 | PL_op ? " in " : "", | |
2382 | PL_op ? OP_DESC(PL_op) : ""); | |
fed3ba5d NC |
2383 | return -2; |
2384 | } | |
2385 | } else { | |
2386 | if (PL_op) | |
2387 | Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), | |
2388 | "%s in %s", unees, OP_DESC(PL_op)); | |
2389 | else | |
61a12c31 | 2390 | Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees); |
fed3ba5d NC |
2391 | return -2; /* Really want to return undef :-) */ |
2392 | } | |
2393 | } else { | |
2394 | return -2; | |
2395 | } | |
2396 | } | |
2397 | if (*b != c) { | |
2398 | return *b < c ? -2 : +2; | |
2399 | } | |
2400 | ++b; | |
2401 | } | |
2402 | ||
2403 | if (b == bend && u == uend) | |
2404 | return 0; | |
2405 | ||
2406 | return b < bend ? +1 : -1; | |
2407 | } | |
2408 | ||
2409 | /* | |
87cea99e | 2410 | =for apidoc utf8_to_bytes |
6940069f | 2411 | |
3bc0c78c | 2412 | Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding. |
a1433954 | 2413 | Unlike L</bytes_to_utf8>, this over-writes the original string, and |
09af0336 | 2414 | updates C<*lenp> to contain the new length. |
3bc0c78c KW |
2415 | Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1. |
2416 | ||
2417 | Upon successful return, the number of variants in the string can be computed by | |
23b37b12 KW |
2418 | having saved the value of C<*lenp> before the call, and subtracting the |
2419 | after-call value of C<*lenp> from it. | |
6940069f | 2420 | |
a1433954 | 2421 | If you need a copy of the string, see L</bytes_from_utf8>. |
95be277c | 2422 | |
6940069f GS |
2423 | =cut |
2424 | */ | |
2425 | ||
2426 | U8 * | |
09af0336 | 2427 | Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp) |
6940069f | 2428 | { |
9fe0d3c2 | 2429 | U8 * first_variant; |
246fae53 | 2430 | |
7918f24d | 2431 | PERL_ARGS_ASSERT_UTF8_TO_BYTES; |
81611534 | 2432 | PERL_UNUSED_CONTEXT; |
7918f24d | 2433 | |
9fe0d3c2 | 2434 | /* This is a no-op if no variants at all in the input */ |
09af0336 | 2435 | if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) { |
9fe0d3c2 KW |
2436 | return s; |
2437 | } | |
2438 | ||
2439 | { | |
3c5aa262 | 2440 | U8 * const save = s; |
09af0336 | 2441 | U8 * const send = s + *lenp; |
3c5aa262 KW |
2442 | U8 * d; |
2443 | ||
2444 | /* Nothing before the first variant needs to be changed, so start the real | |
2445 | * work there */ | |
2446 | s = first_variant; | |
2447 | while (s < send) { | |
2448 | if (! UTF8_IS_INVARIANT(*s)) { | |
2449 | if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) { | |
09af0336 | 2450 | *lenp = ((STRLEN) -1); |
3c5aa262 KW |
2451 | return 0; |
2452 | } | |
2453 | s++; | |
d59937ca KW |
2454 | } |
2455 | s++; | |
dcad2880 | 2456 | } |
dcad2880 | 2457 | |
3c5aa262 KW |
2458 | /* Is downgradable, so do it */ |
2459 | d = s = first_variant; | |
2460 | while (s < send) { | |
2461 | U8 c = *s++; | |
2462 | if (! UVCHR_IS_INVARIANT(c)) { | |
2463 | /* Then it is two-byte encoded */ | |
2464 | c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s); | |
2465 | s++; | |
2466 | } | |
2467 | *d++ = c; | |
2468 | } | |
2469 | *d = '\0'; | |
09af0336 | 2470 | *lenp = d - save; |
3c5aa262 KW |
2471 | |
2472 | return save; | |
9fe0d3c2 | 2473 | } |
6940069f GS |
2474 | } |
2475 | ||
2476 | /* | |
87cea99e | 2477 | =for apidoc bytes_from_utf8 |
f9a63242 | 2478 | |
09af0336 | 2479 | Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native |
41ae6089 | 2480 | byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is |
4f3d592d KW |
2481 | actually encoded in UTF-8. |
2482 | ||
2483 | Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of | |
2484 | the input string. | |
2485 | ||
41ae6089 KW |
2486 | Do nothing if C<*is_utf8p> is 0, or if there are code points in the string |
2487 | not expressible in native byte encoding. In these cases, C<*is_utf8p> and | |
09af0336 | 2488 | C<*lenp> are unchanged, and the return value is the original C<s>. |
4f3d592d | 2489 | |
41ae6089 | 2490 | Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a |
4f3d592d | 2491 | newly created string containing a downgraded copy of C<s>, and whose length is |
9ff99fb3 KW |
2492 | returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The |
2493 | caller is responsible for arranging for the memory used by this string to get | |
2494 | freed. | |
f9a63242 | 2495 | |
3bc0c78c | 2496 | Upon successful return, the number of variants in the string can be computed by |
23b37b12 KW |
2497 | having saved the value of C<*lenp> before the call, and subtracting the |
2498 | after-call value of C<*lenp> from it. | |
3bc0c78c | 2499 | |
37607a96 | 2500 | =cut |
976c1b08 KW |
2501 | |
2502 | There is a macro that avoids this function call, but this is retained for | |
2503 | anyone who calls it with the Perl_ prefix */ | |
f9a63242 JH |
2504 | |
2505 | U8 * | |
41ae6089 | 2506 | Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p) |
f9a63242 | 2507 | { |
7918f24d | 2508 | PERL_ARGS_ASSERT_BYTES_FROM_UTF8; |
96a5add6 | 2509 | PERL_UNUSED_CONTEXT; |
f9a63242 | 2510 | |
976c1b08 KW |
2511 | return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL); |
2512 | } | |
2513 | ||
2514 | /* | |
2515 | No = here because currently externally undocumented | |
2516 | for apidoc bytes_from_utf8_loc | |
2517 | ||
2518 | Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where | |
2519 | to store the location of the first character in C<"s"> that cannot be | |
2520 | converted to non-UTF8. | |
2521 | ||
2522 | If that parameter is C<NULL>, this function behaves identically to | |
2523 | C<bytes_from_utf8>. | |
2524 | ||
2525 | Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to | |
2526 | C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>. | |
2527 | ||
2528 | Otherwise, the function returns a newly created C<NUL>-terminated string | |
2529 | containing the non-UTF8 equivalent of the convertible first portion of | |
2530 | C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>. | |
2531 | If the entire input string was converted, C<*is_utf8p> is set to a FALSE value, | |
2532 | and C<*first_non_downgradable> is set to C<NULL>. | |
2533 | ||
2534 | Otherwise, C<*first_non_downgradable> set to point to the first byte of the | |
2535 | first character in the original string that wasn't converted. C<*is_utf8p> is | |
2536 | unchanged. Note that the new string may have length 0. | |
2537 | ||
2538 | Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and | |
2539 | C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and | |
2540 | converts as many characters in it as possible stopping at the first one it | |
385b74be | 2541 | finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is |
976c1b08 KW |
2542 | set to point to that. The function returns the portion that could be converted |
2543 | in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length, | |
2544 | not including the terminating C<NUL>. If the very first character in the | |
2545 | original could not be converted, C<*lenp> will be 0, and the new string will | |
2546 | contain just a single C<NUL>. If the entire input string was converted, | |
2547 | C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>. | |
2548 | ||
2549 | Upon successful return, the number of variants in the converted portion of the | |
2550 | string can be computed by having saved the value of C<*lenp> before the call, | |
2551 | and subtracting the after-call value of C<*lenp> from it. | |
2552 | ||
2553 | =cut | |
2554 | ||
2555 | ||
2556 | */ | |
2557 | ||
2558 | U8 * | |
2559 | Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted) | |
2560 | { | |
2561 | U8 *d; | |
2562 | const U8 *original = s; | |
2563 | U8 *converted_start; | |
2564 | const U8 *send = s + *lenp; | |
f9a63242 | 2565 | |
976c1b08 | 2566 | PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC; |
170a1c22 | 2567 | |
976c1b08 KW |
2568 | if (! *is_utf8p) { |
2569 | if (first_unconverted) { | |
2570 | *first_unconverted = NULL; | |
2571 | } | |
2572 | ||
2573 | return (U8 *) original; | |
2574 | } | |
2575 | ||
2576 | Newx(d, (*lenp) + 1, U8); | |
2577 | ||
2578 | converted_start = d; | |
7299a045 KW |
2579 | while (s < send) { |
2580 | U8 c = *s++; | |
2581 | if (! UTF8_IS_INVARIANT(c)) { | |
976c1b08 KW |
2582 | |
2583 | /* Then it is multi-byte encoded. If the code point is above 0xFF, | |
2584 | * have to stop now */ | |
2585 | if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) { | |
2586 | if (first_unconverted) { | |
2587 | *first_unconverted = s - 1; | |
2588 | goto finish_and_return; | |
2589 | } | |
2590 | else { | |
2591 | Safefree(converted_start); | |
2592 | return (U8 *) original; | |
2593 | } | |
2594 | } | |
2595 | ||
7299a045 KW |
2596 | c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s); |
2597 | s++; | |
38af28cf | 2598 | } |
7299a045 KW |
2599 | *d++ = c; |
2600 | } | |
170a1c22 | 2601 | |
976c1b08 KW |
2602 | /* Here, converted the whole of the input */ |
2603 | *is_utf8p = FALSE; | |
2604 | if (first_unconverted) { | |
2605 | *first_unconverted = NULL; | |
170a1c22 | 2606 | } |
976c1b08 KW |
2607 | |
2608 | finish_and_return: | |
46a08a6f KW |
2609 | *d = '\0'; |
2610 | *lenp = d - converted_start; | |
976c1b08 KW |
2611 | |
2612 | /* Trim unused space */ | |
2613 | Renew(converted_start, *lenp + 1, U8); | |
2614 | ||
2615 | return converted_start; | |
f9a63242 JH |
2616 | } |
2617 | ||
2618 | /* | |
87cea99e | 2619 | =for apidoc bytes_to_utf8 |
6940069f | 2620 | |
09af0336 | 2621 | Converts a string C<s> of length C<*lenp> bytes from the native encoding into |
ff97e5cf | 2622 | UTF-8. |
09af0336 | 2623 | Returns a pointer to the newly-created string, and sets C<*lenp> to |
9ff99fb3 KW |
2624 | reflect the new length in bytes. The caller is responsible for arranging for |
2625 | the memory used by this string to get freed. | |
6940069f | 2626 | |
3bc0c78c | 2627 | Upon successful return, the number of variants in the string can be computed by |
23b37b12 | 2628 | having saved the value of C<*lenp> before the call, and subtracting it from the |
3bc0c78c KW |
2629 | after-call value of C<*lenp>. |
2630 | ||
75200dff | 2631 | A C<NUL> character will be written after the end of the string. |
2bbc8d55 SP |
2632 | |
2633 | If you want to convert to UTF-8 from encodings other than | |
2634 | the native (Latin1 or EBCDIC), | |
a1433954 | 2635 | see L</sv_recode_to_utf8>(). |
c9ada85f | 2636 | |
497711e7 | 2637 | =cut |
6940069f GS |
2638 | */ |
2639 | ||
2640 | U8* | |
09af0336 | 2641 | Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp) |
6940069f | 2642 | { |
09af0336 | 2643 | const U8 * const send = s + (*lenp); |
6940069f GS |
2644 | U8 *d; |
2645 | U8 *dst; | |
7918f24d NC |
2646 | |
2647 | PERL_ARGS_ASSERT_BYTES_TO_UTF8; | |
96a5add6 | 2648 | PERL_UNUSED_CONTEXT; |
6940069f | 2649 | |
d4662719 KW |
2650 | /* 1 for each byte + 1 for each byte that expands to two, + trailing NUL */ |
2651 | Newx(d, (*lenp) + variant_under_utf8_count(s, send) + 1, U8); | |
6940069f GS |
2652 | dst = d; |
2653 | ||
2654 | while (s < send) { | |
55d09dc8 KW |
2655 | append_utf8_from_native_byte(*s, &d); |
2656 | s++; | |
6940069f | 2657 | } |
2e11cf67 | 2658 | |
6940069f | 2659 | *d = '\0'; |
09af0336 | 2660 | *lenp = d-dst; |
2e11cf67 | 2661 | |
6940069f GS |
2662 | return dst; |
2663 | } | |
2664 | ||
a0ed51b3 | 2665 | /* |
624504c5 KW |
2666 | * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian), |
2667 | * use utf16_to_utf8_reversed(). | |
a0ed51b3 | 2668 | * |
624504c5 KW |
2669 | * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes. |
2670 | * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes. | |
2671 | * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes. | |
2672 | * | |
2673 | * These functions don't check for overflow. The worst case is every code | |
2674 | * point in the input is 2 bytes, and requires 4 bytes on output. (If the code | |
2675 | * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the | |
2676 | * destination must be pre-extended to 2 times the source length. | |
2677 | * | |
2678 | * Do not use in-place. We optimize for native, for obvious reasons. */ | |
a0ed51b3 LW |
2679 | |
2680 | U8* | |
dea0fc0b | 2681 | Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen) |
a0ed51b3 | 2682 | { |
dea0fc0b JH |
2683 | U8* pend; |
2684 | U8* dstart = d; | |
2685 | ||
7918f24d NC |
2686 | PERL_ARGS_ASSERT_UTF16_TO_UTF8; |
2687 | ||
dea0fc0b | 2688 | if (bytelen & 1) |
56576a04 KW |
2689 | Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf, |
2690 | (UV)bytelen); | |
dea0fc0b JH |
2691 | |
2692 | pend = p + bytelen; | |
2693 | ||
a0ed51b3 | 2694 | while (p < pend) { |
dea0fc0b JH |
2695 | UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */ |
2696 | p += 2; | |
2d1545e5 | 2697 | if (OFFUNI_IS_INVARIANT(uv)) { |
56d37426 | 2698 | *d++ = LATIN1_TO_NATIVE((U8) uv); |
a0ed51b3 LW |
2699 | continue; |
2700 | } | |
56d37426 KW |
2701 | if (uv <= MAX_UTF8_TWO_BYTE) { |
2702 | *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv)); | |
2703 | *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv)); | |
a0ed51b3 LW |
2704 | continue; |
2705 | } | |
ffd0a9d3 | 2706 | |
46956fad KW |
2707 | #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST |
2708 | #define LAST_HIGH_SURROGATE 0xDBFF | |
2709 | #define FIRST_LOW_SURROGATE 0xDC00 | |
2710 | #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST | |
ffd0a9d3 | 2711 | #define FIRST_IN_PLANE1 0x10000 |
e23c50db KW |
2712 | |
2713 | /* This assumes that most uses will be in the first Unicode plane, not | |
2714 | * needing surrogates */ | |
2715 | if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST | |
2716 | && uv <= UNICODE_SURROGATE_LAST)) | |
2717 | { | |
2718 | if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) { | |
2719 | Perl_croak(aTHX_ "Malformed UTF-16 surrogate"); | |
2720 | } | |
2721 | else { | |
01ea242b | 2722 | UV low = (p[0] << 8) + p[1]; |
e23c50db KW |
2723 | if ( UNLIKELY(low < FIRST_LOW_SURROGATE) |
2724 | || UNLIKELY(low > LAST_LOW_SURROGATE)) | |
2725 | { | |
01ea242b | 2726 | Perl_croak(aTHX_ "Malformed UTF-16 surrogate"); |
e23c50db KW |
2727 | } |
2728 | p += 2; | |
46956fad | 2729 | uv = ((uv - FIRST_HIGH_SURROGATE) << 10) |
ffd0a9d3 | 2730 | + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1; |
01ea242b | 2731 | } |
a0ed51b3 | 2732 | } |
56d37426 KW |
2733 | #ifdef EBCDIC |
2734 | d = uvoffuni_to_utf8_flags(d, uv, 0); | |
2735 | #else | |
ffd0a9d3 | 2736 | if (uv < FIRST_IN_PLANE1) { |
eb160463 GS |
2737 | *d++ = (U8)(( uv >> 12) | 0xe0); |
2738 | *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80); | |
2739 | *d++ = (U8)(( uv & 0x3f) | 0x80); | |
a0ed51b3 LW |
2740 | continue; |
2741 | } | |
2742 | else { | |
eb160463 GS |
2743 | *d++ = (U8)(( uv >> 18) | 0xf0); |
2744 | *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80); | |
2745 | *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80); | |
2746 | *d++ = (U8)(( uv & 0x3f) | 0x80); | |
a0ed51b3 LW |
2747 | continue; |
2748 | } | |
56d37426 | 2749 | #endif |
a0ed51b3 | 2750 | } |
dea0fc0b | 2751 | *newlen = d - dstart; |
a0ed51b3 LW |
2752 | return d; |
2753 | } | |
2754 | ||
2755 | /* Note: this one is slightly destructive of the source. */ | |
2756 | ||
2757 | U8* | |
dea0fc0b | 2758 | Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen) |
a0ed51b3 LW |
2759 | { |
2760 | U8* s = (U8*)p; | |
d4c19fe8 | 2761 | U8* const send = s + bytelen; |
7918f24d NC |
2762 | |
2763 | PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED; | |
2764 | ||
e0ea5e2d | 2765 | if (bytelen & 1) |
147e3846 | 2766 | Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf, |
e0ea5e2d NC |
2767 | (UV)bytelen); |
2768 | ||
a0ed51b3 | 2769 | while (s < send) { |
d4c19fe8 | 2770 | const U8 tmp = s[0]; |
a0ed51b3 LW |
2771 | s[0] = s[1]; |
2772 | s[1] = tmp; | |
2773 | s += 2; | |
2774 | } | |
dea0fc0b | 2775 | return utf16_to_utf8(p, d, bytelen, newlen); |
a0ed51b3 LW |
2776 | } |
2777 | ||
922e8cb4 KW |
2778 | bool |
2779 | Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c) | |
2780 | { | |
dc31b55c | 2781 | return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c); |
922e8cb4 KW |
2782 | } |
2783 | ||
f9ae8fb6 JD |
2784 | /* Internal function so we can deprecate the external one, and call |
2785 | this one from other deprecated functions in this file */ | |
2786 | ||
f2645549 KW |
2787 | bool |
2788 | Perl__is_utf8_idstart(pTHX_ const U8 *p) | |
61b19385 | 2789 | { |
f2645549 | 2790 | PERL_ARGS_ASSERT__IS_UTF8_IDSTART; |
61b19385 KW |
2791 | |
2792 | if (*p == '_') | |
2793 | return TRUE; | |
eb1f4bb4 | 2794 | return is_utf8_common(p, PL_utf8_idstart); |
61b19385 KW |
2795 | } |
2796 | ||
5092f92a | 2797 | bool |
eba68aa0 KW |
2798 | Perl__is_uni_perl_idcont(pTHX_ UV c) |
2799 | { | |
c12658c9 | 2800 | return _invlist_contains_cp(PL_utf8_perl_idcont, c); |
eba68aa0 KW |
2801 | } |
2802 | ||
2803 | bool | |
f91dcd13 KW |
2804 | Perl__is_uni_perl_idstart(pTHX_ UV c) |
2805 | { | |
c12658c9 | 2806 | return _invlist_contains_cp(PL_utf8_perl_idstart, c); |
f91dcd13 KW |
2807 | } |
2808 | ||
3a4c58c9 | 2809 | UV |
56576a04 KW |
2810 | Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, |
2811 | const char S_or_s) | |
3a4c58c9 KW |
2812 | { |
2813 | /* We have the latin1-range values compiled into the core, so just use | |
4a4088c4 | 2814 | * those, converting the result to UTF-8. The only difference between upper |
3a4c58c9 KW |
2815 | * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is |
2816 | * either "SS" or "Ss". Which one to use is passed into the routine in | |
2817 | * 'S_or_s' to avoid a test */ | |
2818 | ||
2819 | UV converted = toUPPER_LATIN1_MOD(c); | |
2820 | ||
2821 | PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1; | |
2822 | ||
2823 | assert(S_or_s == 'S' || S_or_s == 's'); | |
2824 | ||
6f2d5cbc | 2825 | if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for |
f4cd282c | 2826 | characters in this range */ |
3a4c58c9 KW |
2827 | *p = (U8) converted; |
2828 | *lenp = 1; | |
2829 | return converted; | |
2830 | } | |
2831 | ||
2832 | /* toUPPER_LATIN1_MOD gives the correct results except for three outliers, | |
2833 | * which it maps to one of them, so as to only have to have one check for | |
2834 | * it in the main case */ | |
2835 | if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) { | |
2836 | switch (c) { | |
2837 | case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS: | |
2838 | converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS; | |
2839 | break; | |
2840 | case MICRO_SIGN: | |
2841 | converted = GREEK_CAPITAL_LETTER_MU; | |
2842 | break; | |
79e064b9 KW |
2843 | #if UNICODE_MAJOR_VERSION > 2 \ |
2844 | || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \ | |
2845 | && UNICODE_DOT_DOT_VERSION >= 8) | |
3a4c58c9 KW |
2846 | case LATIN_SMALL_LETTER_SHARP_S: |
2847 | *(p)++ = 'S'; | |
2848 | *p = S_or_s; | |
2849 | *lenp = 2; | |
2850 | return 'S'; | |
79e064b9 | 2851 | #endif |
3a4c58c9 | 2852 | default: |
56576a04 KW |
2853 | Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect" |
2854 | " '%c' to map to '%c'", | |
2855 | c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS); | |
e5964223 | 2856 | NOT_REACHED; /* NOTREACHED */ |
3a4c58c9 KW |
2857 | } |
2858 | } | |
2859 | ||
2860 | *(p)++ = UTF8_TWO_BYTE_HI(converted); | |
2861 | *p = UTF8_TWO_BYTE_LO(converted); | |
2862 | *lenp = 2; | |
2863 | ||
2864 | return converted; | |
2865 | } | |
2866 | ||
fe63c520 KW |
2867 | /* If compiled on an early Unicode version, there may not be auxiliary tables |
2868 | * */ | |
2869 | #ifndef HAS_UC_AUX_TABLES | |
2870 | # define UC_AUX_TABLE_ptrs NULL | |
2871 | # define UC_AUX_TABLE_lengths NULL | |
2872 | #endif | |
2873 | #ifndef HAS_TC_AUX_TABLES | |
2874 | # define TC_AUX_TABLE_ptrs NULL | |
2875 | # define TC_AUX_TABLE_lengths NULL | |
2876 | #endif | |
2877 | #ifndef HAS_LC_AUX_TABLES | |
2878 | # define LC_AUX_TABLE_ptrs NULL | |
2879 | # define LC_AUX_TABLE_lengths NULL | |
2880 | #endif | |
2881 | #ifndef HAS_CF_AUX_TABLES | |
2882 | # define CF_AUX_TABLE_ptrs NULL | |
2883 | # define CF_AUX_TABLE_lengths NULL | |
2884 | #endif | |
2885 | #ifndef HAS_UC_AUX_TABLES | |
2886 | # define UC_AUX_TABLE_ptrs NULL | |
2887 | # define UC_AUX_TABLE_lengths NULL | |
2888 | #endif | |
2889 | ||
50bda2c3 KW |
2890 | /* Call the function to convert a UTF-8 encoded character to the specified case. |
2891 | * Note that there may be more than one character in the result. | |
6fa2f9bc KW |
2892 | * 's' is a pointer to the first byte of the input character |
2893 | * 'd' will be set to the first byte of the string of changed characters. It | |
50bda2c3 | 2894 | * needs to have space for UTF8_MAXBYTES_CASE+1 bytes |
6fa2f9bc | 2895 | * 'lenp' will be set to the length in bytes of the string of changed characters |
50bda2c3 | 2896 | * |
56576a04 | 2897 | * The functions return the ordinal of the first character in the string of |
6fa2f9bc | 2898 | * 'd' */ |
56576a04 | 2899 | #define CALL_UPPER_CASE(uv, s, d, lenp) \ |
8946fcd9 KW |
2900 | _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \ |
2901 | Uppercase_Mapping_invmap, \ | |
2902 | UC_AUX_TABLE_ptrs, \ | |
2903 | UC_AUX_TABLE_lengths, \ | |
2904 | "uppercase") | |
56576a04 | 2905 | #define CALL_TITLE_CASE(uv, s, d, lenp) \ |
8946fcd9 KW |
2906 | _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \ |
2907 | Titlecase_Mapping_invmap, \ | |
2908 | TC_AUX_TABLE_ptrs, \ | |
2909 | TC_AUX_TABLE_lengths, \ | |
2910 | "titlecase") | |
56576a04 | 2911 | #define CALL_LOWER_CASE(uv, s, d, lenp) \ |
8946fcd9 KW |
2912 | _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \ |
2913 | Lowercase_Mapping_invmap, \ | |
2914 | LC_AUX_TABLE_ptrs, \ | |
2915 | LC_AUX_TABLE_lengths, \ | |
2916 | "lowercase") | |
2917 | ||
50bda2c3 | 2918 | |
b9992569 KW |
2919 | /* This additionally has the input parameter 'specials', which if non-zero will |
2920 | * cause this to use the specials hash for folding (meaning get full case | |
50bda2c3 | 2921 | * folding); otherwise, when zero, this implies a simple case fold */ |
56576a04 | 2922 | #define CALL_FOLD_CASE(uv, s, d, lenp, specials) \ |
8946fcd9 KW |
2923 | (specials) \ |
2924 | ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \ | |
2925 | Case_Folding_invmap, \ | |
2926 | CF_AUX_TABLE_ptrs, \ | |
2927 | CF_AUX_TABLE_lengths, \ | |
2928 | "foldcase") \ | |
2929 | : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \ | |
2930 | Simple_Case_Folding_invmap, \ | |
2931 | NULL, NULL, \ | |
2932 | "foldcase") | |
c3fd2246 | 2933 | |
84afefe6 JH |
2934 | UV |
2935 | Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp) | |
a0ed51b3 | 2936 | { |
a1433954 KW |
2937 | /* Convert the Unicode character whose ordinal is <c> to its uppercase |
2938 | * version and store that in UTF-8 in <p> and its length in bytes in <lenp>. | |
2939 | * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since | |
c3fd2246 KW |
2940 | * the changed version may be longer than the original character. |
2941 | * | |
2942 | * The ordinal of the first character of the changed version is returned | |
2943 | * (but note, as explained above, that there may be more.) */ | |
2944 | ||
7918f24d NC |
2945 | PERL_ARGS_ASSERT_TO_UNI_UPPER; |
2946 | ||
3a4c58c9 KW |
2947 | if (c < 256) { |
2948 | return _to_upper_title_latin1((U8) c, p, lenp, 'S'); | |
2949 | } | |
2950 | ||
a13f1de4 | 2951 | return CALL_UPPER_CASE(c, NULL, p, lenp); |
a0ed51b3 LW |
2952 | } |
2953 | ||
84afefe6 JH |
2954 | UV |
2955 | Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp) | |
a0ed51b3 | 2956 | { |
7918f24d NC |
2957 | PERL_ARGS_ASSERT_TO_UNI_TITLE; |
2958 | ||
3a4c58c9 KW |
2959 | if (c < 256) { |
2960 | return _to_upper_title_latin1((U8) c, p, lenp, 's'); | |
2961 | } | |
2962 | ||
a13f1de4 | 2963 | return CALL_TITLE_CASE(c, NULL, p, lenp); |
a0ed51b3 LW |
2964 | } |
2965 | ||
afc16117 | 2966 | STATIC U8 |
eaf412bf | 2967 | S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy) |
afc16117 KW |
2968 | { |
2969 | /* We have the latin1-range values compiled into the core, so just use | |
4a4088c4 | 2970 | * those, converting the result to UTF-8. Since the result is always just |
a1433954 | 2971 | * one character, we allow <p> to be NULL */ |
afc16117 KW |
2972 | |
2973 | U8 converted = toLOWER_LATIN1(c); | |
2974 | ||
eaf412bf KW |
2975 | PERL_UNUSED_ARG(dummy); |
2976 | ||
afc16117 | 2977 | if (p != NULL) { |
6f2d5cbc | 2978 | if (NATIVE_BYTE_IS_INVARIANT(converted)) { |
afc16117 KW |
2979 | *p = converted; |
2980 | *lenp = 1; | |
2981 | } | |
2982 | else { | |
430c9760 KW |
2983 | /* Result is known to always be < 256, so can use the EIGHT_BIT |
2984 | * macros */ | |
2985 | *p = UTF8_EIGHT_BIT_HI(converted); | |
2986 | *(p+1) = UTF8_EIGHT_BIT_LO(converted); | |
afc16117 KW |
2987 | *lenp = 2; |
2988 | } | |
2989 | } | |
2990 | return converted; | |
2991 | } | |
2992 | ||
84afefe6 JH |
2993 | UV |
2994 | Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp) | |
a0ed51b3 | 2995 | { |
7918f24d NC |
2996 | PERL_ARGS_ASSERT_TO_UNI_LOWER; |
2997 | ||
afc16117 | 2998 | if (c < 256) { |
eaf412bf | 2999 | return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ ); |
bca00c02 KW |
3000 | } |
3001 | ||
a13f1de4 | 3002 | return CALL_LOWER_CASE(c, NULL, p, lenp); |
a0ed51b3 LW |
3003 | } |
3004 | ||
84afefe6 | 3005 | UV |
7c0ab950 | 3006 | Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags) |
a1dde8de | 3007 | { |
51910141 | 3008 | /* Corresponds to to_lower_latin1(); <flags> bits meanings: |
1ca267a5 | 3009 | * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited |
51910141 | 3010 | * FOLD_FLAGS_FULL iff full folding is to be used; |
1ca267a5 KW |
3011 | * |
3012 | * Not to be used for locale folds | |
51910141 | 3013 | */ |
f673fad4 | 3014 | |
a1dde8de KW |
3015 | UV converted; |
3016 | ||
3017 | PERL_ARGS_ASSERT__TO_FOLD_LATIN1; | |
3018 | ||
1ca267a5 KW |
3019 | assert (! (flags & FOLD_FLAGS_LOCALE)); |
3020 | ||
659a7c2d | 3021 | if (UNLIKELY(c == MICRO_SIGN)) { |
a1dde8de KW |
3022 | converted = GREEK_SMALL_LETTER_MU; |
3023 | } | |
9b63e895 KW |
3024 | #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \ |
3025 | || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \ | |
3026 | || UNICODE_DOT_DOT_VERSION > 0) | |
659a7c2d KW |
3027 | else if ( (flags & FOLD_FLAGS_FULL) |
3028 | && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S)) | |
3029 | { | |
1ca267a5 KW |
3030 | /* If can't cross 127/128 boundary, can't return "ss"; instead return |
3031 | * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}") | |
3032 | * under those circumstances. */ | |
3033 | if (flags & FOLD_FLAGS_NOMIX_ASCII) { | |
3034 | *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2; | |
3035 | Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8, | |
3036 | p, *lenp, U8); | |
3037 | return LATIN_SMALL_LETTER_LONG_S; | |
3038 | } | |
3039 | else { | |
4f489194 KW |
3040 | *(p)++ = 's'; |
3041 | *p = 's'; | |
3042 | *lenp = 2; | |
3043 | return 's'; | |
1ca267a5 | 3044 | } |
a1dde8de | 3045 | } |
9b63e895 | 3046 | #endif |
a1dde8de KW |
3047 | else { /* In this range the fold of all other characters is their lower |
3048 | case */ | |
3049 | converted = toLOWER_LATIN1(c); | |
3050 | } | |
3051 | ||
6f2d5cbc | 3052 | if (UVCHR_IS_INVARIANT(converted)) { |
a1dde8de KW |
3053 | *p = (U8) converted; |
3054 | *lenp = 1; | |
3055 | } | |
3056 | else { | |
3057 | *(p)++ = UTF8_TWO_BYTE_HI(converted); | |
3058 | *p = UTF8_TWO_BYTE_LO(converted); | |
3059 | *lenp = 2; | |
3060 | } | |
3061 | ||
3062 | return converted; | |
3063 | } | |
3064 | ||
3065 | UV | |
31f05a37 | 3066 | Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags) |
84afefe6 | 3067 | { |
4b593389 | 3068 | |
a0270393 KW |
3069 | /* Not currently externally documented, and subject to change |
3070 | * <flags> bits meanings: | |
3071 | * FOLD_FLAGS_FULL iff full folding is to be used; | |
31f05a37 KW |
3072 | * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying |
3073 | * locale are to be used. | |
a0270393 KW |
3074 | * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited |
3075 | */ | |
4b593389 | 3076 | |
36bb2ab6 | 3077 | PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS; |
7918f24d | 3078 | |
780fcc9f | 3079 | if (flags & FOLD_FLAGS_LOCALE) { |
8b7358b9 KW |
3080 | /* Treat a UTF-8 locale as not being in locale at all, except for |
3081 | * potentially warning */ | |
3082 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; | |
780fcc9f KW |
3083 | if (IN_UTF8_CTYPE_LOCALE) { |
3084 | flags &= ~FOLD_FLAGS_LOCALE; | |
3085 | } | |
3086 | else { | |
e7b7ac46 | 3087 | goto needs_full_generality; |
780fcc9f | 3088 | } |
31f05a37 KW |
3089 | } |
3090 | ||
a1dde8de | 3091 | if (c < 256) { |
e7b7ac46 | 3092 | return _to_fold_latin1((U8) c, p, lenp, |
31f05a37 | 3093 | flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)); |
a1dde8de KW |
3094 | } |
3095 | ||
2f306ab9 | 3096 | /* Here, above 255. If no special needs, just use the macro */ |
a0270393 | 3097 | if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) { |
a13f1de4 | 3098 | return CALL_FOLD_CASE(c, NULL, p, lenp, flags & FOLD_FLAGS_FULL); |
a0270393 | 3099 | } |
567b353c | 3100 | else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with |
a0270393 KW |
3101 | the special flags. */ |
3102 | U8 utf8_c[UTF8_MAXBYTES + 1]; | |
e7b7ac46 KW |
3103 | |
3104 | needs_full_generality: | |
a0270393 | 3105 | uvchr_to_utf8(utf8_c, c); |
56576a04 KW |
3106 | return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c), |
3107 | p, lenp, flags); | |
a0270393 | 3108 | } |
84afefe6 JH |
3109 | } |
3110 | ||
26483009 | 3111 | PERL_STATIC_INLINE bool |
eb1f4bb4 | 3112 | S_is_utf8_common(pTHX_ const U8 *const p, SV* const invlist) |
bde6a22d | 3113 | { |
ea317ccb | 3114 | /* returns a boolean giving whether or not the UTF8-encoded character that |
eb1f4bb4 | 3115 | * starts at <p> is in the inversion list indicated by <invlist>. |
ea317ccb KW |
3116 | * |
3117 | * Note that it is assumed that the buffer length of <p> is enough to | |
3118 | * contain all the bytes that comprise the character. Thus, <*p> should | |
3119 | * have been checked before this call for mal-formedness enough to assure | |
016c8ffc KW |
3120 | * that. This function, does make sure to not look past any NUL, so it is |
3121 | * safe to use on C, NUL-terminated, strings */ | |
3122 | STRLEN len = my_strnlen((char *) p, UTF8SKIP(p)); | |
ea317ccb | 3123 | |
7918f24d NC |
3124 | PERL_ARGS_ASSERT_IS_UTF8_COMMON; |
3125 | ||
492a624f | 3126 | /* The API should have included a length for the UTF-8 character in <p>, |
28123549 | 3127 | * but it doesn't. We therefore assume that p has been validated at least |
492a624f KW |
3128 | * as far as there being enough bytes available in it to accommodate the |
3129 | * character without reading beyond the end, and pass that number on to the | |
3130 | * validating routine */ | |
016c8ffc KW |
3131 | if (! isUTF8_CHAR(p, p + len)) { |
3132 | _force_out_malformed_utf8_message(p, p + len, _UTF8_NO_CONFIDENCE_IN_CURLEN, | |
86ae6e94 KW |
3133 | 1 /* Die */ ); |
3134 | NOT_REACHED; /* NOTREACHED */ | |
28123549 | 3135 | } |
86ae6e94 | 3136 | |
eb1f4bb4 | 3137 | return is_utf8_common_with_len(p, p + len, invlist); |
bde6a22d NC |
3138 | } |
3139 | ||
da8c1a98 | 3140 | PERL_STATIC_INLINE bool |
56576a04 | 3141 | S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e, |
56576a04 | 3142 | SV* const invlist) |
da8c1a98 KW |
3143 | { |
3144 | /* returns a boolean giving whether or not the UTF8-encoded character that | |
eb1f4bb4 KW |
3145 | * starts at <p>, and extending no further than <e - 1> is in the inversion |
3146 | * list <invlist>. */ | |
da8c1a98 | 3147 | |
b68ffe0c KW |
3148 | UV cp = utf8n_to_uvchr(p, e - p, NULL, 0); |
3149 | ||
da8c1a98 KW |
3150 | PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN; |
3151 | ||
b68ffe0c | 3152 | if (cp == 0 && (p >= e || *p != '\0')) { |
da8c1a98 KW |
3153 | _force_out_malformed_utf8_message(p, e, 0, 1); |
3154 | NOT_REACHED; /* NOTREACHED */ | |
3155 | } | |
3156 | ||
eb1f4bb4 | 3157 | assert(invlist); |
b68ffe0c | 3158 | return _invlist_contains_cp(invlist, cp); |
da8c1a98 KW |
3159 | } |
3160 | ||
34aeb2e9 KW |
3161 | STATIC void |
3162 | S_warn_on_first_deprecated_use(pTHX_ const char * const name, | |
3163 | const char * const alternative, | |
3164 | const bool use_locale, | |
3165 | const char * const file, | |
3166 | const unsigned line) | |
3167 | { | |
3168 | const char * key; | |
3169 | ||
3170 | PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE; | |
3171 | ||
3172 | if (ckWARN_d(WARN_DEPRECATED)) { | |
3173 | ||
3174 | key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line); | |
3175 | if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) { | |
3176 | if (! PL_seen_deprecated_macro) { | |
3177 | PL_seen_deprecated_macro = newHV(); | |
3178 | } | |
3179 | if (! hv_store(PL_seen_deprecated_macro, key, | |
3180 | strlen(key), &PL_sv_undef, 0)) | |
3181 | { | |
3182 | Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed"); | |
3183 | } | |
3184 | ||
c44e9413 | 3185 | if (instr(file, "mathoms.c")) { |
607313a1 KW |
3186 | Perl_warner(aTHX_ WARN_DEPRECATED, |
3187 | "In %s, line %d, starting in Perl v5.30, %s()" | |
3188 | " will be removed. Avoid this message by" | |
3189 | " converting to use %s().\n", | |
3190 | file, line, name, alternative); | |
3191 | } | |
3192 | else { | |
34aeb2e9 KW |
3193 | Perl_warner(aTHX_ WARN_DEPRECATED, |
3194 | "In %s, line %d, starting in Perl v5.30, %s() will" | |
3195 | " require an additional parameter. Avoid this" | |
3196 | " message by converting to use %s().\n", | |
3197 | file, line, name, alternative); | |
607313a1 | 3198 | } |
34aeb2e9 KW |
3199 | } |
3200 | } | |
3201 | } | |
3202 | ||
bde6a22d | 3203 | bool |
34aeb2e9 | 3204 | Perl__is_utf8_FOO(pTHX_ U8 classnum, |
be99e2c2 | 3205 | const U8 * const p, |
34aeb2e9 KW |
3206 | const char * const name, |
3207 | const char * const alternative, | |
3208 | const bool use_utf8, | |
3209 | const bool use_locale, | |
3210 | const char * const file, | |
3211 | const unsigned line) | |
922e8cb4 | 3212 | { |
922e8cb4 KW |
3213 | PERL_ARGS_ASSERT__IS_UTF8_FOO; |
3214 | ||
34aeb2e9 KW |
3215 | warn_on_first_deprecated_use(name, alternative, use_locale, file, line); |
3216 | ||
3217 | if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) { | |
34aeb2e9 KW |
3218 | |
3219 | switch (classnum) { | |
3220 | case _CC_WORDCHAR: | |
3221 | case _CC_DIGIT: | |
3222 | case _CC_ALPHA: | |
3223 | case _CC_LOWER: | |
3224 | case _CC_UPPER: | |
3225 | case _CC_PUNCT: | |
3226 | case _CC_PRINT: | |
3227 | case _CC_ALPHANUMERIC: | |
3228 | case _CC_GRAPH: | |
3229 | case _CC_CASED: | |
3230 | ||
eb1f4bb4 | 3231 | return is_utf8_common(p, PL_XPosix_ptrs[classnum]); |
34aeb2e9 KW |
3232 | |
3233 | case _CC_SPACE: | |
3234 | return is_XPERLSPACE_high(p); | |
3235 | case _CC_BLANK: | |
3236 | return is_HORIZWS_high(p); | |
3237 | case _CC_XDIGIT: | |
3238 | return is_XDIGIT_high(p); | |
3239 | case _CC_CNTRL: | |
3240 | return 0; | |
3241 | case _CC_ASCII: | |
3242 | return 0; | |
3243 | case _CC_VERTSPACE: | |
3244 | return is_VERTWS_high(p); | |
3245 | case _CC_IDFIRST: | |
eb1f4bb4 | 3246 | return is_utf8_common(p, PL_utf8_perl_idstart); |
34aeb2e9 | 3247 | case _CC_IDCONT: |
eb1f4bb4 | 3248 | return is_utf8_common(p, PL_utf8_perl_idcont); |
34aeb2e9 KW |
3249 | } |
3250 | } | |
3251 | ||
3252 | /* idcont is the same as wordchar below 256 */ | |
3253 | if (classnum == _CC_IDCONT) { | |
3254 | classnum = _CC_WORDCHAR; | |
3255 | } | |
3256 | else if (classnum == _CC_IDFIRST) { | |
3257 | if (*p == '_') { | |
3258 | return TRUE; | |
3259 | } | |
3260 | classnum = _CC_ALPHA; | |
3261 | } | |
3262 | ||
3263 | if (! use_locale) { | |
3264 | if (! use_utf8 || UTF8_IS_INVARIANT(*p)) { | |
3265 | return _generic_isCC(*p, classnum); | |
3266 | } | |
922e8cb4 | 3267 | |
34aeb2e9 KW |
3268 | return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum); |
3269 | } | |
3270 | else { | |
3271 | if (! use_utf8 || UTF8_IS_INVARIANT(*p)) { | |
3272 | return isFOO_lc(classnum, *p); | |
3273 | } | |
3274 | ||
3275 | return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 ))); | |
3276 | } | |
3277 | ||
3278 | NOT_REACHED; /* NOTREACHED */ | |
922e8cb4 KW |
3279 | } |
3280 | ||
3281 | bool | |
da8c1a98 KW |
3282 | Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p, |
3283 | const U8 * const e) | |
3284 | { | |
3285 | PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN; | |
3286 | ||
eb1f4bb4 | 3287 | return is_utf8_common_with_len(p, e, PL_XPosix_ptrs[classnum]); |
da8c1a98 KW |
3288 | } |
3289 | ||
3290 | bool | |
da8c1a98 KW |
3291 | Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e) |
3292 | { | |
da8c1a98 KW |
3293 | PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN; |
3294 | ||
eb1f4bb4 | 3295 | return is_utf8_common_with_len(p, e, PL_utf8_perl_idstart); |
da8c1a98 KW |
3296 | } |
3297 | ||
3298 | bool | |
f2645549 | 3299 | Perl__is_utf8_xidstart(pTHX_ const U8 *p) |
c11ff943 | 3300 | { |
f2645549 | 3301 | PERL_ARGS_ASSERT__IS_UTF8_XIDSTART; |
c11ff943 KW |
3302 | |
3303 | if (*p == '_') | |
3304 | return TRUE; | |
eb1f4bb4 | 3305 | return is_utf8_common(p, PL_utf8_xidstart); |
c11ff943 KW |
3306 | } |
3307 | ||
3308 | bool | |
da8c1a98 KW |
3309 | Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e) |
3310 | { | |
da8c1a98 KW |
3311 | PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN; |
3312 | ||
eb1f4bb4 | 3313 | return is_utf8_common_with_len(p, e, PL_utf8_perl_idcont); |
da8c1a98 KW |
3314 | } |
3315 | ||
3316 | bool | |
f2645549 | 3317 | Perl__is_utf8_idcont(pTHX_ const U8 *p) |
82686b01 | 3318 | { |
f2645549 | 3319 | PERL_ARGS_ASSERT__IS_UTF8_IDCONT; |
7918f24d | 3320 | |
eb1f4bb4 | 3321 | return is_utf8_common(p, PL_utf8_idcont); |
a0ed51b3 LW |
3322 | } |
3323 | ||
3324 | bool | |
f2645549 | 3325 | Perl__is_utf8_xidcont(pTHX_ const U8 *p) |
c11ff943 | 3326 | { |
f2645549 | 3327 | PERL_ARGS_ASSERT__IS_UTF8_XIDCONT; |
c11ff943 | 3328 | |
eb1f4bb4 | 3329 | return is_utf8_common(p, PL_utf8_xidcont); |
c11ff943 KW |
3330 | } |
3331 | ||
3332 | bool | |
7dbf68d2 KW |
3333 | Perl__is_utf8_mark(pTHX_ const U8 *p) |
3334 | { | |
7dbf68d2 KW |
3335 | PERL_ARGS_ASSERT__IS_UTF8_MARK; |
3336 | ||
eb1f4bb4 | 3337 | return is_utf8_common(p, PL_utf8_mark); |
7dbf68d2 KW |
3338 | } |
3339 | ||
6a4a25f4 | 3340 | STATIC UV |
30613bdc KW |
3341 | S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, |
3342 | U8* ustrp, STRLEN *lenp, | |
341bb5b7 | 3343 | SV *invlist, const int * const invmap, |
e39a4130 | 3344 | const unsigned int * const * const aux_tables, |
30613bdc KW |
3345 | const U8 * const aux_table_lengths, |
3346 | const char * const normal) | |
b9992569 | 3347 | { |
0134edef | 3348 | STRLEN len = 0; |
7918f24d | 3349 | |
30613bdc KW |
3350 | /* Change the case of code point 'uv1' whose UTF-8 representation (assumed |
3351 | * by this routine to be valid) begins at 'p'. 'normal' is a string to use | |
3352 | * to name the new case in any generated messages, as a fallback if the | |
3353 | * operation being used is not available. The new case is given by the | |
3354 | * data structures in the remaining arguments. | |
3355 | * | |
3356 | * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the | |
3357 | * entire changed case string, and the return value is the first code point | |
3358 | * in that string */ | |
3359 | ||
b9992569 | 3360 | PERL_ARGS_ASSERT__TO_UTF8_CASE; |
7918f24d | 3361 | |
36eaa811 KW |
3362 | /* For code points that don't change case, we already know that the output |
3363 | * of this function is the unchanged input, so we can skip doing look-ups | |
3364 | * for them. Unfortunately the case-changing code points are scattered | |
3365 | * around. But there are some long consecutive ranges where there are no | |
3366 | * case changing code points. By adding tests, we can eliminate the lookup | |
3367 | * for all the ones in such ranges. This is currently done here only for | |
3368 | * just a few cases where the scripts are in common use in modern commerce | |
3369 | * (and scripts adjacent to those which can be included without additional | |
3370 | * tests). */ | |
3371 | ||
3372 | if (uv1 >= 0x0590) { | |
3373 | /* This keeps from needing further processing the code points most | |
3374 | * likely to be used in the following non-cased scripts: Hebrew, | |
3375 | * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari, | |
3376 | * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada, | |
3377 | * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */ | |
3378 | if (uv1 < 0x10A0) { | |
3379 | goto cases_to_self; | |
3380 | } | |
3381 | ||
3382 | /* The following largish code point ranges also don't have case | |
3383 | * changes, but khw didn't think they warranted extra tests to speed | |
3384 | * them up (which would slightly slow down everything else above them): | |
3385 | * 1100..139F Hangul Jamo, Ethiopic | |
3386 | * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic, | |
3387 | * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian, | |
3388 | * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham, | |
3389 | * Combining Diacritical Marks Extended, Balinese, | |
3390 | * Sundanese, Batak, Lepcha, Ol Chiki | |
3391 | * 2000..206F General Punctuation | |
3392 | */ | |
3393 | ||
3394 | if (uv1 >= 0x2D30) { | |
3395 | ||
3396 | /* This keeps the from needing further processing the code points | |
3397 | * most likely to be used in the following non-cased major scripts: | |
3398 | * CJK, Katakana, Hiragana, plus some less-likely scripts. | |
3399 | * | |
3400 | * (0x2D30 above might have to be changed to 2F00 in the unlikely | |
3401 | * event that Unicode eventually allocates the unused block as of | |
3402 | * v8.0 2FE0..2FEF to code points that are cased. khw has verified | |
3403 | * that the test suite will start having failures to alert you | |
3404 | * should that happen) */ | |
3405 | if (uv1 < 0xA640) { | |
3406 | goto cases_to_self; | |
3407 | } | |
3408 | ||
3409 | if (uv1 >= 0xAC00) { | |
3410 | if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) { | |
5af9bc97 KW |
3411 | if (ckWARN_d(WARN_SURROGATE)) { |
3412 | const char* desc = (PL_op) ? OP_DESC(PL_op) : normal; | |
3413 | Perl_warner(aTHX_ packWARN(WARN_SURROGATE), | |
56576a04 KW |
3414 | "Operation \"%s\" returns its argument for" |
3415 | " UTF-16 surrogate U+%04" UVXf, desc, uv1); | |
5af9bc97 KW |
3416 | } |
3417 | goto cases_to_self; | |
3418 | } | |
36eaa811 KW |
3419 | |
3420 | /* AC00..FAFF Catches Hangul syllables and private use, plus | |
3421 | * some others */ | |
3422 | if (uv1 < 0xFB00) { | |
3423 | goto cases_to_self; | |
36eaa811 KW |
3424 | } |
3425 | ||
5af9bc97 | 3426 | if (UNLIKELY(UNICODE_IS_SUPER(uv1))) { |
40606899 | 3427 | if (UNLIKELY(uv1 > MAX_LEGAL_CP)) { |
d22ec717 | 3428 | Perl_croak(aTHX_ cp_above_legal_max, uv1, |
40606899 | 3429 | MAX_LEGAL_CP); |
5af9bc97 KW |
3430 | } |
3431 | if (ckWARN_d(WARN_NON_UNICODE)) { | |
3432 | const char* desc = (PL_op) ? OP_DESC(PL_op) : normal; | |
3433 | Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE), | |
56576a04 KW |
3434 | "Operation \"%s\" returns its argument for" |
3435 | " non-Unicode code point 0x%04" UVXf, desc, uv1); | |
5af9bc97 KW |
3436 | } |
3437 | goto cases_to_self; | |
3438 | } | |
3bfc1e70 KW |
3439 | #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C |
3440 | if (UNLIKELY(uv1 | |
3441 | > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C)) | |
3442 | { | |
3443 | ||
56576a04 KW |
3444 | /* As of Unicode 10.0, this means we avoid swash creation |
3445 | * for anything beyond high Plane 1 (below emojis) */ | |
3bfc1e70 KW |
3446 | goto cases_to_self; |
3447 | } | |
3448 | #endif | |
36eaa811 KW |
3449 | } |
3450 | } | |
9ae3ac1a | 3451 | |
36eaa811 | 3452 | /* Note that non-characters are perfectly legal, so no warning should |
8946fcd9 | 3453 | * be given. */ |
9ae3ac1a KW |
3454 | } |
3455 | ||
8946fcd9 KW |
3456 | { |
3457 | unsigned int i; | |
e39a4130 | 3458 | const unsigned int * cp_list; |
8946fcd9 | 3459 | U8 * d; |
69352d88 KW |
3460 | |
3461 | /* 'index' is guaranteed to be non-negative, as this is an inversion | |
3462 | * map that covers all possible inputs. See [perl #133365] */ | |
8946fcd9 KW |
3463 | SSize_t index = _invlist_search(invlist, uv1); |
3464 | IV base = invmap[index]; | |
0134edef | 3465 | |
30613bdc KW |
3466 | /* The data structures are set up so that if 'base' is non-negative, |
3467 | * the case change is 1-to-1; and if 0, the change is to itself */ | |
8946fcd9 KW |
3468 | if (base >= 0) { |
3469 | IV lc; | |
b08cf34e | 3470 | |
8946fcd9 KW |
3471 | if (base == 0) { |
3472 | goto cases_to_self; | |
4a8240a3 | 3473 | } |
4a8240a3 | 3474 | |
30613bdc | 3475 | /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */ |
8946fcd9 KW |
3476 | lc = base + uv1 - invlist_array(invlist)[index]; |
3477 | *lenp = uvchr_to_utf8(ustrp, lc) - ustrp; | |
3478 | return lc; | |
3479 | } | |
1feea2c7 | 3480 | |
30613bdc KW |
3481 | /* Here 'base' is negative. That means the mapping is 1-to-many, and |
3482 | * requires an auxiliary table look up. abs(base) gives the index into | |
3483 | * a list of such tables which points to the proper aux table. And a | |
3484 | * parallel list gives the length of each corresponding aux table. */ | |
8946fcd9 | 3485 | cp_list = aux_tables[-base]; |
30613bdc KW |
3486 | |
3487 | /* Create the string of UTF-8 from the mapped-to code points */ | |
8946fcd9 KW |
3488 | d = ustrp; |
3489 | for (i = 0; i < aux_table_lengths[-base]; i++) { | |
3490 | d = uvchr_to_utf8(d, cp_list[i]); | |
cbe07460 | 3491 | } |
8946fcd9 KW |
3492 | *d = '\0'; |
3493 | *lenp = d - ustrp; | |
3494 | ||
3495 | return cp_list[0]; | |
cbe07460 KW |
3496 | } |
3497 | ||
3498 | /* Here, there was no mapping defined, which means that the code point maps | |
3499 | * to itself. Return the inputs */ | |
e24dfe9c | 3500 | cases_to_self: |
a13f1de4 KW |
3501 | if (p) { |
3502 | len = UTF8SKIP(p); | |
3503 | if (p != ustrp) { /* Don't copy onto itself */ | |
3504 | Copy(p, ustrp, len, U8); | |
3505 | } | |
3506 | *lenp = len; | |
3507 | } | |
3508 | else { | |
3509 | *lenp = uvchr_to_utf8(ustrp, uv1) - ustrp; | |
ca9fab46 | 3510 | } |
2a37f04d | 3511 | |
f4cd282c | 3512 | return uv1; |
cbe07460 | 3513 | |
a0ed51b3 LW |
3514 | } |
3515 | ||
b74fe592 | 3516 | Size_t |
e39a4130 KW |
3517 | Perl__inverse_folds(pTHX_ const UV cp, unsigned int * first_folds_to, |
3518 | const unsigned int ** remaining_folds_to) | |
b74fe592 KW |
3519 | { |
3520 | /* Returns the count of the number of code points that fold to the input | |
3521 | * 'cp' (besides itself). | |
3522 | * | |
3523 | * If the return is 0, there is nothing else that folds to it, and | |
3524 | * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL. | |
3525 | * | |
3526 | * If the return is 1, '*first_folds_to' is set to the single code point, | |
3527 | * and '*remaining_folds_to' is set to NULL. | |
3528 | * | |
3529 | * Otherwise, '*first_folds_to' is set to a code point, and | |
3530 | * '*remaining_fold_to' is set to an array that contains the others. The | |
3531 | * length of this array is the returned count minus 1. | |
3532 | * | |
3533 | * The reason for this convolution is to avoid having to deal with | |
3534 | * allocating and freeing memory. The lists are already constructed, so | |
3535 | * the return can point to them, but single code points aren't, so would | |
3536 | * need to be constructed if we didn't employ something like this API */ | |
3537 | ||
69352d88 KW |
3538 | /* 'index' is guaranteed to be non-negative, as this is an inversion map |
3539 | * that covers all possible inputs. See [perl #133365] */ | |
b74fe592 KW |
3540 | SSize_t index = _invlist_search(PL_utf8_foldclosures, cp); |
3541 | int base = _Perl_IVCF_invmap[index]; | |
3542 | ||
3543 | PERL_ARGS_ASSERT__INVERSE_FOLDS; | |
3544 | ||
3545 | if (base == 0) { /* No fold */ | |
3546 | *first_folds_to = 0; | |
3547 | *remaining_folds_to = NULL; | |
3548 | return 0; | |
3549 | } | |
3550 | ||
3551 | #ifndef HAS_IVCF_AUX_TABLES /* This Unicode version only has 1-1 folds */ | |
3552 | ||
3553 | assert(base > 0); | |
3554 | ||
3555 | #else | |
3556 | ||
3557 | if (UNLIKELY(base < 0)) { /* Folds to more than one character */ | |
3558 | ||
3559 | /* The data structure is set up so that the absolute value of 'base' is | |
3560 | * an index into a table of pointers to arrays, with the array | |
3561 | * corresponding to the index being the list of code points that fold | |
3562 | * to 'cp', and the parallel array containing the length of the list | |
3563 | * array */ | |
3564 | *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0]; | |
3565 | *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1; /* +1 excludes | |
3566 | *first_folds_to | |
3567 | */ | |
3568 | return IVCF_AUX_TABLE_lengths[-base]; | |
3569 | } | |
3570 | ||
3571 | #endif | |
3572 | ||
3573 | /* Only the single code point. This works like 'fc(G) = G - A + a' */ | |
3574 | *first_folds_to = base + cp - invlist_array(PL_utf8_foldclosures)[index]; | |
3575 | *remaining_folds_to = NULL; | |
3576 | return 1; | |
3577 | } | |
3578 | ||
051a06d4 | 3579 | STATIC UV |
56576a04 KW |
3580 | S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, |
3581 | U8* const ustrp, STRLEN *lenp) | |
051a06d4 | 3582 | { |
4a4088c4 | 3583 | /* This is called when changing the case of a UTF-8-encoded character above |
31f05a37 KW |
3584 | * the Latin1 range, and the operation is in a non-UTF-8 locale. If the |
3585 | * result contains a character that crosses the 255/256 boundary, disallow | |
3586 | * the change, and return the original code point. See L<perlfunc/lc> for | |
3587 | * why; | |
051a06d4 | 3588 | * |
a1433954 KW |
3589 | * p points to the original string whose case was changed; assumed |
3590 | * by this routine to be well-formed | |
051a06d4 | 3591 | * result the code point of the first character in the changed-case string |
56576a04 KW |
3592 | * ustrp points to the changed-case string (<result> represents its |
3593 | * first char) | |
051a06d4 KW |
3594 | * lenp points to the length of <ustrp> */ |
3595 | ||
3596 | UV original; /* To store the first code point of <p> */ | |
3597 | ||
3598 | PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING; | |
3599 | ||
a4f12ed7 | 3600 | assert(UTF8_IS_ABOVE_LATIN1(*p)); |
051a06d4 KW |
3601 | |
3602 | /* We know immediately if the first character in the string crosses the | |
5e45c680 | 3603 | * boundary, so can skip testing */ |
051a06d4 KW |
3604 | if (result > 255) { |
3605 | ||
3606 | /* Look at every character in the result; if any cross the | |
3607 | * boundary, the whole thing is disallowed */ | |
3608 | U8* s = ustrp + UTF8SKIP(ustrp); | |
3609 | U8* e = ustrp + *lenp; | |
3610 | while (s < e) { | |
a4f12ed7 | 3611 | if (! UTF8_IS_ABOVE_LATIN1(*s)) { |
051a06d4 KW |
3612 | goto bad_crossing; |
3613 | } | |
3614 | s += UTF8SKIP(s); | |
3615 | } | |
3616 | ||
613abc6d KW |
3617 | /* Here, no characters crossed, result is ok as-is, but we warn. */ |
3618 | _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p)); | |
051a06d4 KW |
3619 | return result; |
3620 | } | |
3621 | ||
7b52d656 | 3622 | bad_crossing: |
051a06d4 KW |
3623 | |
3624 | /* Failed, have to return the original */ | |
4b88fb76 | 3625 | original = valid_utf8_to_uvchr(p, lenp); |
ab0b796c KW |
3626 | |
3627 | /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ | |
3628 | Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), | |
56576a04 KW |
3629 | "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8" |
3630 | " locale; resolved to \"\\x{%" UVXf "}\".", | |
357aadde | 3631 | OP_DESC(PL_op), |
ab0b796c KW |
3632 | original, |
3633 | original); | |
051a06d4 KW |
3634 | Copy(p, ustrp, *lenp, char); |
3635 | return original; | |
3636 | } | |
3637 | ||
607313a1 KW |
3638 | STATIC U32 |
3639 | S_check_and_deprecate(pTHX_ const U8 *p, | |
3640 | const U8 **e, | |
3641 | const unsigned int type, /* See below */ | |
3642 | const bool use_locale, /* Is this a 'LC_' | |
3643 | macro call? */ | |
3644 | const char * const file, | |
3645 | const unsigned line) | |
3646 | { | |
3647 | /* This is a temporary function to deprecate the unsafe calls to the case | |
3648 | * changing macros and functions. It keeps all the special stuff in just | |
3649 | * one place. | |
3650 | * | |
3651 | * It updates *e with the pointer to the end of the input string. If using | |
3652 | * the old-style macros, *e is NULL on input, and so this function assumes | |
3653 | * the input string is long enough to hold the entire UTF-8 sequence, and | |
3654 | * sets *e accordingly, but it then returns a flag to pass the | |
3655 | * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid | |
3656 | * using the full length if possible. | |
3657 | * | |
3658 | * It also does the assert that *e > p when *e is not NULL. This should be | |
3659 | * migrated to the callers when this function gets deleted. | |
3660 | * | |
3661 | * The 'type' parameter is used for the caller to specify which case | |
3662 | * changing function this is called from: */ | |
3663 | ||
3664 | # define DEPRECATE_TO_UPPER 0 | |
3665 | # define DEPRECATE_TO_TITLE 1 | |
3666 | # define DEPRECATE_TO_LOWER 2 | |
3667 | # define DEPRECATE_TO_FOLD 3 | |
3668 | ||
3669 | U32 utf8n_flags = 0; | |
3670 | const char * name; | |
3671 | const char * alternative; | |
3672 | ||
3673 | PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE; | |
3674 | ||
3675 | if (*e == NULL) { | |
3676 | utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN; | |
6f7ce6e8 KW |
3677 | |
3678 | /* strnlen() makes this function safe for the common case of | |
3679 | * NUL-terminated strings */ | |
3680 | *e = p + my_strnlen((char *) p, UTF8SKIP(p)); | |
607313a1 KW |
3681 | |
3682 | /* For mathoms.c calls, we use the function name we know is stored | |
c44e9413 | 3683 | * there. It could be part of a larger path */ |
607313a1 | 3684 | if (type == DEPRECATE_TO_UPPER) { |
c44e9413 | 3685 | name = instr(file, "mathoms.c") |
607313a1 KW |
3686 | ? "to_utf8_upper" |
3687 | : "toUPPER_utf8"; | |
3688 | alternative = "toUPPER_utf8_safe"; | |
3689 | } | |
3690 | else if (type == DEPRECATE_TO_TITLE) { | |
c44e9413 | 3691 | name = instr(file, "mathoms.c") |
607313a1 KW |
3692 | ? "to_utf8_title" |
3693 | : "toTITLE_utf8"; | |
3694 | alternative = "toTITLE_utf8_safe"; | |
3695 | } | |
3696 | else if (type == DEPRECATE_TO_LOWER) { | |
c44e9413 | 3697 | name = instr(file, "mathoms.c") |
607313a1 KW |
3698 | ? "to_utf8_lower" |
3699 | : "toLOWER_utf8"; | |
3700 | alternative = "toLOWER_utf8_safe"; | |
3701 | } | |
3702 | else if (type == DEPRECATE_TO_FOLD) { | |
c44e9413 | 3703 | name = instr(file, "mathoms.c") |
607313a1 KW |
3704 | ? "to_utf8_fold" |
3705 | : "toFOLD_utf8"; | |
3706 | alternative = "toFOLD_utf8_safe"; | |
3707 | } | |
3708 | else Perl_croak(aTHX_ "panic: Unexpected case change type"); | |
3709 | ||
3710 | warn_on_first_deprecated_use(name, alternative, use_locale, file, line); | |
3711 | } | |
3712 | else { | |
3713 | assert (p < *e); | |
3714 | } | |
3715 | ||
3716 | return utf8n_flags; | |
3717 | } | |
3718 | ||
eaf412bf KW |
3719 | /* The process for changing the case is essentially the same for the four case |
3720 | * change types, except there are complications for folding. Otherwise the | |
3721 | * difference is only which case to change to. To make sure that they all do | |
3722 | * the same thing, the bodies of the functions are extracted out into the | |
3723 | * following two macros. The functions are written with the same variable | |
3724 | * names, and these are known and used inside these macros. It would be | |
3725 | * better, of course, to have inline functions to do it, but since different | |
3726 | * macros are called, depending on which case is being changed to, this is not | |
3727 | * feasible in C (to khw's knowledge). Two macros are created so that the fold | |
3728 | * function can start with the common start macro, then finish with its special | |
3729 | * handling; while the other three cases can just use the common end macro. | |
3730 | * | |
3731 | * The algorithm is to use the proper (passed in) macro or function to change | |
3732 | * the case for code points that are below 256. The macro is used if using | |
3733 | * locale rules for the case change; the function if not. If the code point is | |
3734 | * above 255, it is computed from the input UTF-8, and another macro is called | |
3735 | * to do the conversion. If necessary, the output is converted to UTF-8. If | |
3736 | * using a locale, we have to check that the change did not cross the 255/256 | |
3737 | * boundary, see check_locale_boundary_crossing() for further details. | |
3738 | * | |
3739 | * The macros are split with the correct case change for the below-256 case | |
3740 | * stored into 'result', and in the middle of an else clause for the above-255 | |
3741 | * case. At that point in the 'else', 'result' is not the final result, but is | |
3742 | * the input code point calculated from the UTF-8. The fold code needs to | |
3743 | * realize all this and take it from there. | |
3744 | * | |
3745 | * If you read the two macros as sequential, it's easier to understand what's | |
3746 | * going on. */ | |
3747 | #define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \ | |
3748 | L1_func_extra_param) \ | |
a239b1e2 | 3749 | \ |
eaf412bf | 3750 | if (flags & (locale_flags)) { \ |
8b7358b9 | 3751 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \ |
eaf412bf KW |
3752 | /* Treat a UTF-8 locale as not being in locale at all */ \ |
3753 | if (IN_UTF8_CTYPE_LOCALE) { \ | |
3754 | flags &= ~(locale_flags); \ | |
3755 | } \ | |
eaf412bf KW |
3756 | } \ |
3757 | \ | |
3758 | if (UTF8_IS_INVARIANT(*p)) { \ | |
3759 | if (flags & (locale_flags)) { \ | |
3760 | result = LC_L1_change_macro(*p); \ | |
3761 | } \ | |
3762 | else { \ | |
3763 | return L1_func(*p, ustrp, lenp, L1_func_extra_param); \ | |
3764 | } \ | |
3765 | } \ | |
a239b1e2 | 3766 | else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \ |
1a751160 | 3767 | U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); \ |
eaf412bf | 3768 | if (flags & (locale_flags)) { \ |
1a751160 | 3769 | result = LC_L1_change_macro(c); \ |
eaf412bf KW |
3770 | } \ |
3771 | else { \ | |
1a751160 | 3772 | return L1_func(c, ustrp, lenp, L1_func_extra_param); \ |
eaf412bf KW |
3773 | } \ |
3774 | } \ | |
fa8ab374 KW |
3775 | else { /* malformed UTF-8 or ord above 255 */ \ |
3776 | STRLEN len_result; \ | |
fa8ab374 KW |
3777 | result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \ |
3778 | if (len_result == (STRLEN) -1) { \ | |
607313a1 KW |
3779 | _force_out_malformed_utf8_message(p, e, utf8n_flags, \ |
3780 | 1 /* Die */ ); \ | |
fa8ab374 | 3781 | } |
eaf412bf KW |
3782 | |
3783 | #define CASE_CHANGE_BODY_END(locale_flags, change_macro) \ | |
3784 | result = change_macro(result, p, ustrp, lenp); \ | |
3785 | \ | |
3786 | if (flags & (locale_flags)) { \ | |
3787 | result = check_locale_boundary_crossing(p, result, ustrp, lenp); \ | |
3788 | } \ | |
3789 | return result; \ | |
3790 | } \ | |
3791 | \ | |
3792 | /* Here, used locale rules. Convert back to UTF-8 */ \ | |
3793 | if (UTF8_IS_INVARIANT(result)) { \ | |
3794 | *ustrp = (U8) result; \ | |
3795 | *lenp = 1; \ | |
3796 | } \ | |
3797 | else { \ | |
3798 | *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \ | |
3799 | *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \ | |
3800 | *lenp = 2; \ | |
3801 | } \ | |
3802 | \ | |
3803 | return result; | |
3804 | ||
d3e79532 | 3805 | /* |
87cea99e | 3806 | =for apidoc to_utf8_upper |
d3e79532 | 3807 | |
a239b1e2 | 3808 | Instead use L</toUPPER_utf8_safe>. |
a1433954 | 3809 | |
d3e79532 JH |
3810 | =cut */ |
3811 | ||
051a06d4 | 3812 | /* Not currently externally documented, and subject to change: |
31f05a37 KW |
3813 | * <flags> is set iff iff the rules from the current underlying locale are to |
3814 | * be used. */ | |
051a06d4 | 3815 | |
2104c8d9 | 3816 | UV |
607313a1 KW |
3817 | Perl__to_utf8_upper_flags(pTHX_ const U8 *p, |
3818 | const U8 *e, | |
3819 | U8* ustrp, | |
3820 | STRLEN *lenp, | |
3821 | bool flags, | |
3822 | const char * const file, | |
3823 | const int line) | |
a0ed51b3 | 3824 | { |
051a06d4 | 3825 | UV result; |
607313a1 KW |
3826 | const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER, |
3827 | cBOOL(flags), file, line); | |
051a06d4 KW |
3828 | |
3829 | PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS; | |
7918f24d | 3830 | |
eaf412bf KW |
3831 | /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */ |
3832 | /* 2nd char of uc(U+DF) is 'S' */ | |
3833 | CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S'); | |
3834 | CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE); | |
983ffd37 | 3835 | } |
a0ed51b3 | 3836 | |
d3e79532 | 3837 | /* |
87cea99e | 3838 | =for apidoc to_utf8_title |
d3e79532 | 3839 | |
a239b1e2 | 3840 | Instead use L</toTITLE_utf8_safe>. |
a1433954 | 3841 | |
d3e79532 JH |
3842 | =cut */ |
3843 | ||
051a06d4 | 3844 | /* Not currently externally documented, and subject to change: |
31f05a37 KW |
3845 | * <flags> is set iff the rules from the current underlying locale are to be |
3846 | * used. Since titlecase is not defined in POSIX, for other than a | |
3847 | * UTF-8 locale, uppercase is used instead for code points < 256. | |
445bf929 | 3848 | */ |
051a06d4 | 3849 | |
983ffd37 | 3850 | UV |
607313a1 KW |
3851 | Perl__to_utf8_title_flags(pTHX_ const U8 *p, |
3852 | const U8 *e, | |
3853 | U8* ustrp, | |
3854 | STRLEN *lenp, | |
3855 | bool flags, | |
3856 | const char * const file, | |
3857 | const int line) | |
983ffd37 | 3858 | { |
051a06d4 | 3859 | UV result; |
607313a1 KW |
3860 | const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE, |
3861 | cBOOL(flags), file, line); | |
051a06d4 KW |
3862 | |
3863 | PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS; | |
7918f24d | 3864 | |
eaf412bf KW |
3865 | /* 2nd char of ucfirst(U+DF) is 's' */ |
3866 | CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's'); | |
3867 | CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE); | |
a0ed51b3 LW |
3868 | } |
3869 | ||
d3e79532 | 3870 | /* |
87cea99e | 3871 | =for apidoc to_utf8_lower |
d3e79532 | 3872 | |
a239b1e2 | 3873 | Instead use L</toLOWER_utf8_safe>. |
a1433954 | 3874 | |
d3e79532 JH |
3875 | =cut */ |
3876 | ||
051a06d4 | 3877 | /* Not currently externally documented, and subject to change: |
31f05a37 KW |
3878 | * <flags> is set iff iff the rules from the current underlying locale are to |
3879 | * be used. | |
3880 | */ | |
051a06d4 | 3881 | |
2104c8d9 | 3882 | UV |
607313a1 KW |
3883 | Perl__to_utf8_lower_flags(pTHX_ const U8 *p, |
3884 | const U8 *e, | |
3885 | U8* ustrp, | |
3886 | STRLEN *lenp, | |
3887 | bool flags, | |
3888 | const char * const file, | |
3889 | const int line) | |
a0ed51b3 | 3890 | { |
051a06d4 | 3891 | UV result; |
607313a1 KW |
3892 | const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER, |
3893 | cBOOL(flags), file, line); | |
051a06d4 | 3894 | |
051a06d4 | 3895 | PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS; |
7918f24d | 3896 | |
eaf412bf KW |
3897 | CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */) |
3898 | CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE) | |
b4e400f9 JH |
3899 | } |
3900 | ||
d3e79532 | 3901 | /* |
87cea99e | 3902 | =for apidoc to_utf8_fold |
d3e79532 | 3903 | |
a239b1e2 | 3904 | Instead use L</toFOLD_utf8_safe>. |
a1433954 | 3905 | |
d3e79532 JH |
3906 | =cut */ |
3907 | ||
051a06d4 KW |
3908 | /* Not currently externally documented, and subject to change, |
3909 | * in <flags> | |
31f05a37 KW |
3910 | * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying |
3911 | * locale are to be used. | |
051a06d4 KW |
3912 | * bit FOLD_FLAGS_FULL is set iff full case folds are to be used; |
3913 | * otherwise simple folds | |
a0270393 KW |
3914 | * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are |
3915 | * prohibited | |
445bf929 | 3916 | */ |
36bb2ab6 | 3917 | |
b4e400f9 | 3918 | UV |
607313a1 KW |
3919 | Perl__to_utf8_fold_flags(pTHX_ const U8 *p, |
3920 | const U8 *e, | |
3921 | U8* ustrp, | |
3922 | STRLEN *lenp, | |
3923 | U8 flags, | |
3924 | const char * const file, | |
3925 | const int line) | |
b4e400f9 | 3926 | { |
051a06d4 | 3927 | UV result; |
607313a1 KW |
3928 | const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD, |
3929 | cBOOL(flags), file, line); | |
051a06d4 | 3930 | |
36bb2ab6 | 3931 | PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS; |
7918f24d | 3932 | |
a0270393 KW |
3933 | /* These are mutually exclusive */ |
3934 | assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII))); | |
3935 | ||
50ba90ff KW |
3936 | assert(p != ustrp); /* Otherwise overwrites */ |
3937 | ||
eaf412bf KW |
3938 | CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1, |
3939 | ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII))); | |
31f05a37 | 3940 | |
eaf412bf | 3941 | result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL); |
a1dde8de | 3942 | |
1ca267a5 KW |
3943 | if (flags & FOLD_FLAGS_LOCALE) { |
3944 | ||
76f2ffcd | 3945 | # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8 |
0766489e KW |
3946 | # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8 |
3947 | # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8 | |
76f2ffcd | 3948 | |
538e84ed KW |
3949 | /* Special case these two characters, as what normally gets |
3950 | * returned under locale doesn't work */ | |
db540106 | 3951 | if (memBEGINs((char *) p, e - p, CAP_SHARP_S)) |
1ca267a5 | 3952 | { |
ab0b796c KW |
3953 | /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ |
3954 | Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), | |
3955 | "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; " | |
3956 | "resolved to \"\\x{17F}\\x{17F}\"."); | |
1ca267a5 KW |
3957 | goto return_long_s; |
3958 | } | |
0766489e KW |
3959 | else |
3960 | #endif | |
db540106 | 3961 | if (memBEGINs((char *) p, e - p, LONG_S_T)) |
9fc2026f | 3962 | { |
ab0b796c KW |
3963 | /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ |
3964 | Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), | |
3965 | "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; " | |
3966 | "resolved to \"\\x{FB06}\"."); | |
9fc2026f KW |
3967 | goto return_ligature_st; |
3968 | } | |
74894415 KW |
3969 | |
3970 | #if UNICODE_MAJOR_VERSION == 3 \ | |
3971 | && UNICODE_DOT_VERSION == 0 \ | |
3972 | && UNICODE_DOT_DOT_VERSION == 1 | |
3973 | # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8 | |
3974 | ||
3975 | /* And special case this on this Unicode version only, for the same | |
3976 | * reaons the other two are special cased. They would cross the | |
3977 | * 255/256 boundary which is forbidden under /l, and so the code | |
3978 | * wouldn't catch that they are equivalent (which they are only in | |
3979 | * this release) */ | |
db540106 | 3980 | else if (memBEGINs((char *) p, e - p, DOTTED_I)) { |
74894415 KW |
3981 | /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ |
3982 | Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), | |
3983 | "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; " | |
3984 | "resolved to \"\\x{0131}\"."); | |
3985 | goto return_dotless_i; | |
3986 | } | |
3987 | #endif | |
3988 | ||
357aadde | 3989 | return check_locale_boundary_crossing(p, result, ustrp, lenp); |
051a06d4 | 3990 | } |
a0270393 KW |
3991 | else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) { |
3992 | return result; | |
3993 | } | |
3994 | else { | |
4a4088c4 | 3995 | /* This is called when changing the case of a UTF-8-encoded |
9fc2026f KW |
3996 | * character above the ASCII range, and the result should not |
3997 | * contain an ASCII character. */ | |
a0270393 KW |
3998 | |
3999 | UV original; /* To store the first code point of <p> */ | |
4000 | ||
4001 | /* Look at every character in the result; if any cross the | |
4002 | * boundary, the whole thing is disallowed */ | |
4003 | U8* s = ustrp; | |
4004 | U8* e = ustrp + *lenp; | |
4005 | while (s < e) { | |
4006 | if (isASCII(*s)) { | |
4007 | /* Crossed, have to return the original */ | |
4008 | original = valid_utf8_to_uvchr(p, lenp); | |
1ca267a5 | 4009 | |
9fc2026f | 4010 | /* But in these instances, there is an alternative we can |
1ca267a5 | 4011 | * return that is valid */ |
0766489e KW |
4012 | if (original == LATIN_SMALL_LETTER_SHARP_S |
4013 | #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */ | |
4014 | || original == LATIN_CAPITAL_LETTER_SHARP_S | |
4015 | #endif | |
4016 | ) { | |
1ca267a5 KW |
4017 | goto return_long_s; |
4018 | } | |
9fc2026f KW |
4019 | else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) { |
4020 | goto return_ligature_st; | |
4021 | } | |
74894415 KW |
4022 | #if UNICODE_MAJOR_VERSION == 3 \ |
4023 | && UNICODE_DOT_VERSION == 0 \ | |
4024 | && UNICODE_DOT_DOT_VERSION == 1 | |
4025 | ||
4026 | else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) { | |
4027 | goto return_dotless_i; | |
4028 | } | |
4029 | #endif | |
a0270393 KW |
4030 | Copy(p, ustrp, *lenp, char); |
4031 | return original; | |
4032 | } | |
4033 | s += UTF8SKIP(s); | |
4034 | } | |
051a06d4 | 4035 | |
a0270393 KW |
4036 | /* Here, no characters crossed, result is ok as-is */ |
4037 | return result; | |
4038 | } | |
051a06d4 KW |
4039 | } |
4040 | ||
4a4088c4 | 4041 | /* Here, used locale rules. Convert back to UTF-8 */ |
051a06d4 KW |
4042 | if (UTF8_IS_INVARIANT(result)) { |
4043 | *ustrp = (U8) result; | |
4044 | *lenp = 1; | |
4045 | } | |
4046 | else { | |
62cb07ea KW |
4047 | *ustrp = UTF8_EIGHT_BIT_HI((U8) result); |
4048 | *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); | |
051a06d4 KW |
4049 | *lenp = 2; |
4050 | } | |
4051 | ||
051a06d4 | 4052 | return result; |
1ca267a5 KW |
4053 | |
4054 | return_long_s: | |
4055 | /* Certain folds to 'ss' are prohibited by the options, but they do allow | |
4056 | * folds to a string of two of these characters. By returning this | |
4057 | * instead, then, e.g., | |
4058 | * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}") | |
4059 | * works. */ | |
4060 | ||
4061 | *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2; | |
4062 | Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8, | |
4063 | ustrp, *lenp, U8); | |
4064 | return LATIN_SMALL_LETTER_LONG_S; | |
9fc2026f KW |
4065 | |
4066 | return_ligature_st: | |
4067 | /* Two folds to 'st' are prohibited by the options; instead we pick one and | |
4068 | * have the other one fold to it */ | |
4069 | ||
4070 | *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1; | |
4071 | Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8); | |
4072 | return LATIN_SMALL_LIGATURE_ST; | |
74894415 KW |
4073 | |
4074 | #if UNICODE_MAJOR_VERSION == 3 \ | |
4075 | && UNICODE_DOT_VERSION == 0 \ | |
4076 | && UNICODE_DOT_DOT_VERSION == 1 | |
4077 | ||
4078 | return_dotless_i: | |
4079 | *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1; | |
4080 | Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8); | |
4081 | return LATIN_SMALL_LETTER_DOTLESS_I; | |
4082 | ||
4083 | #endif | |
4084 | ||
a0ed51b3 LW |
4085 | } |
4086 | ||
711a919c | 4087 | /* Note: |
f90a9a02 | 4088 | * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch(). |
711a919c TS |
4089 | * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8". |
4090 | * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl. | |
4091 | */ | |
c4a5db0c | 4092 | |
a0ed51b3 | 4093 | SV* |
56576a04 KW |
4094 | Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, |
4095 | I32 minbits, I32 none) | |
a0ed51b3 | 4096 | { |
c4a5db0c KW |
4097 | PERL_ARGS_ASSERT_SWASH_INIT; |
4098 | ||
4099 | /* Returns a copy of a swash initiated by the called function. This is the | |
4100 | * public interface, and returning a copy prevents others from doing | |
4101 | * mischief on the original */ | |
4102 | ||
56576a04 KW |
4103 | return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, |
4104 | NULL, NULL)); | |
c4a5db0c KW |
4105 | } |
4106 | ||
4107 | SV* | |
56576a04 KW |
4108 | Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, |
4109 | I32 minbits, I32 none, SV* invlist, | |
4110 | U8* const flags_p) | |
c4a5db0c | 4111 | { |
2c1f00b9 YO |
4112 | |
4113 | /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST | |
4114 | * use the following define */ | |
4115 | ||
4116 | #define CORE_SWASH_INIT_RETURN(x) \ | |
4117 | PL_curpm= old_PL_curpm; \ | |
4118 | return x | |
4119 | ||
c4a5db0c | 4120 | /* Initialize and return a swash, creating it if necessary. It does this |
87367d5f KW |
4121 | * by calling utf8_heavy.pl in the general case. The returned value may be |
4122 | * the swash's inversion list instead if the input parameters allow it. | |
4123 | * Which is returned should be immaterial to callers, as the only | |
923b6d4e KW |
4124 | * operations permitted on a swash, swash_fetch(), _get_swash_invlist(), |
4125 | * and swash_to_invlist() handle both these transparently. | |
c4a5db0c KW |
4126 | * |
4127 | * This interface should only be used by functions that won't destroy or | |
4128 | * adversely change the swash, as doing so affects all other uses of the | |
4129 | * swash in the program; the general public should use 'Perl_swash_init' | |
4130 | * instead. | |
4131 | * | |
4132 | * pkg is the name of the package that <name> should be in. | |
4133 | * name is the name of the swash to find. Typically it is a Unicode | |
4134 | * property name, including user-defined ones | |
4135 | * listsv is a string to initialize the swash with. It must be of the form | |
4136 | * documented as the subroutine return value in | |
4137 | * L<perlunicode/User-Defined Character Properties> | |
4138 | * minbits is the number of bits required to represent each data element. | |
4139 | * It is '1' for binary properties. | |
4140 | * none I (khw) do not understand this one, but it is used only in tr///. | |
9a53f6cf | 4141 | * invlist is an inversion list to initialize the swash with (or NULL) |
83199d38 KW |
4142 | * flags_p if non-NULL is the address of various input and output flag bits |
4143 | * to the routine, as follows: ('I' means is input to the routine; | |
4144 | * 'O' means output from the routine. Only flags marked O are | |
4145 | * meaningful on return.) | |
4146 | * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash | |
4147 | * came from a user-defined property. (I O) | |
5d3d13d1 KW |
4148 | * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking |
4149 | * when the swash cannot be located, to simply return NULL. (I) | |
87367d5f KW |
4150 | * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a |
4151 | * return of an inversion list instead of a swash hash if this routine | |
4152 | * thinks that would result in faster execution of swash_fetch() later | |
4153 | * on. (I) | |
9a53f6cf KW |
4154 | * |
4155 | * Thus there are three possible inputs to find the swash: <name>, | |
4156 | * <listsv>, and <invlist>. At least one must be specified. The result | |
4157 | * will be the union of the specified ones, although <listsv>'s various | |
aabbdbda KW |
4158 | * actions can intersect, etc. what <name> gives. To avoid going out to |
4159 | * disk at all, <invlist> should specify completely what the swash should | |
4160 | * have, and <listsv> should be &PL_sv_undef and <name> should be "". | |
9a53f6cf KW |
4161 | * |
4162 | * <invlist> is only valid for binary properties */ | |
c4a5db0c | 4163 | |
2c1f00b9 YO |
4164 | PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */ |
4165 | ||
c4a5db0c | 4166 | SV* retval = &PL_sv_undef; |
83199d38 | 4167 | HV* swash_hv = NULL; |
26141ee3 | 4168 | const bool use_invlist= (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST); |
9a53f6cf KW |
4169 | |
4170 | assert(listsv != &PL_sv_undef || strNE(name, "") || invlist); | |
4171 | assert(! invlist || minbits == 1); | |
4172 | ||
56576a04 KW |
4173 | PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the |
4174 | regex that triggered the swash init and the swash init | |
4175 | perl logic itself. See perl #122747 */ | |
2c1f00b9 | 4176 | |
9a53f6cf KW |
4177 | /* If data was passed in to go out to utf8_heavy to find the swash of, do |
4178 | * so */ | |
4179 | if (listsv != &PL_sv_undef || strNE(name, "")) { | |
69794297 KW |
4180 | dSP; |
4181 | const size_t pkg_len = strlen(pkg); | |
4182 | const size_t name_len = strlen(name); | |
4183 | HV * const stash = gv_stashpvn(pkg, pkg_len, 0); | |
4184 | SV* errsv_save; | |
4185 | GV *method; | |
4186 | ||
4187 | PERL_ARGS_ASSERT__CORE_SWASH_INIT; | |
4188 | ||
4189 | PUSHSTACKi(PERLSI_MAGIC); | |
ce3b816e | 4190 | ENTER; |
69794297 | 4191 | SAVEHINTS(); |
2782061f | 4192 | save_re_context(); |
650f067c JL |
4193 | /* We might get here via a subroutine signature which uses a utf8 |
4194 | * parameter name, at which point PL_subname will have been set | |
4195 | * but not yet used. */ | |
4196 | save_item(PL_subname); | |
69794297 KW |
4197 | if (PL_parser && PL_parser->error_count) |
4198 | SAVEI8(PL_parser->error_count), PL_parser->error_count = 0; | |
4199 | method = gv_fetchmeth(stash, "SWASHNEW", 8, -1); | |
4a4088c4 | 4200 | if (!method) { /* demand load UTF-8 */ |
69794297 | 4201 | ENTER; |
db2c6cb3 FC |
4202 | if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save); |
4203 | GvSV(PL_errgv) = NULL; | |
1a419e6b | 4204 | #ifndef NO_TAINT_SUPPORT |
69794297 KW |
4205 | /* It is assumed that callers of this routine are not passing in |
4206 | * any user derived data. */ | |
2782061f DM |
4207 | /* Need to do this after save_re_context() as it will set |
4208 | * PL_tainted to 1 while saving $1 etc (see the code after getrx: | |
4209 | * in Perl_magic_get). Even line to create errsv_save can turn on | |
4210 | * PL_tainted. */ | |
284167a5 S |
4211 | SAVEBOOL(TAINT_get); |
4212 | TAINT_NOT; | |
4213 | #endif | |
69794297 KW |
4214 | Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len), |
4215 | NULL); | |
eed484f9 | 4216 | { |
db2c6cb3 FC |
4217 | /* Not ERRSV, as there is no need to vivify a scalar we are |
4218 | about to discard. */ | |
4219 | SV * const errsv = GvSV(PL_errgv); | |
4220 | if (!SvTRUE(errsv)) { | |
4221 | GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save); | |
4222 | SvREFCNT_dec(errsv); | |
4223 | } | |
eed484f9 | 4224 | } |
69794297 KW |
4225 | LEAVE; |
4226 | } | |
4227 | SPAGAIN; | |
4228 | PUSHMARK(SP); | |
4229 | EXTEND(SP,5); | |
4230 | mPUSHp(pkg, pkg_len); | |
4231 | mPUSHp(name, name_len); | |
4232 | PUSHs(listsv); | |
4233 | mPUSHi(minbits); | |
4234 | mPUSHi(none); | |
4235 | PUTBACK; | |
db2c6cb3 FC |
4236 | if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save); |
4237 | GvSV(PL_errgv) = NULL; | |
69794297 KW |
4238 | /* If we already have a pointer to the method, no need to use |
4239 | * call_method() to repeat the lookup. */ | |
c41800a8 KW |
4240 | if (method |
4241 | ? call_sv(MUTABLE_SV(method), G_SCALAR) | |
69794297 KW |
4242 | : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD)) |
4243 | { | |
4244 | retval = *PL_stack_sp--; | |
4245 | SvREFCNT_inc(retval); | |
4246 | } | |
eed484f9 | 4247 | { |
db2c6cb3 FC |
4248 | /* Not ERRSV. See above. */ |
4249 | SV * const errsv = GvSV(PL_errgv); | |
4250 | if (!SvTRUE(errsv)) { | |
4251 | GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save); | |
4252 | SvREFCNT_dec(errsv); | |
4253 | } | |
eed484f9 | 4254 | } |
ce3b816e | 4255 | LEAVE; |
69794297 KW |
4256 | POPSTACK; |
4257 | if (IN_PERL_COMPILETIME) { | |
4258 | CopHINTS_set(PL_curcop, PL_hints); | |
4259 | } | |
4260 | if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) { | |
d95e4a00 | 4261 | if (SvPOK(retval)) { |
69794297 KW |
4262 | |
4263 | /* If caller wants to handle missing properties, let them */ | |
5d3d13d1 | 4264 | if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) { |
2c1f00b9 | 4265 | CORE_SWASH_INIT_RETURN(NULL); |
69794297 KW |
4266 | } |
4267 | Perl_croak(aTHX_ | |
147e3846 | 4268 | "Can't find Unicode property definition \"%" SVf "\"", |
69794297 | 4269 | SVfARG(retval)); |
a25b5927 | 4270 | NOT_REACHED; /* NOTREACHED */ |
d95e4a00 | 4271 | } |
69794297 | 4272 | } |
9a53f6cf | 4273 | } /* End of calling the module to find the swash */ |
36eb48b4 | 4274 | |
83199d38 KW |
4275 | /* If this operation fetched a swash, and we will need it later, get it */ |
4276 | if (retval != &PL_sv_undef | |
4277 | && (minbits == 1 || (flags_p | |
4278 | && ! (*flags_p | |
4279 | & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)))) | |
4280 | { | |
4281 | swash_hv = MUTABLE_HV(SvRV(retval)); | |
4282 | ||
4283 | /* If we don't already know that there is a user-defined component to | |
4284 | * this swash, and the user has indicated they wish to know if there is | |
4285 | * one (by passing <flags_p>), find out */ | |
4286 | if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) { | |
4287 | SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE); | |
4288 | if (user_defined && SvUV(*user_defined)) { | |
4289 | *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY; | |
4290 | } | |
4291 | } | |
4292 | } | |
4293 | ||
36eb48b4 KW |
4294 | /* Make sure there is an inversion list for binary properties */ |
4295 | if (minbits == 1) { | |
4296 | SV** swash_invlistsvp = NULL; | |
4297 | SV* swash_invlist = NULL; | |
9a53f6cf | 4298 | bool invlist_in_swash_is_valid = FALSE; |
02c85471 FC |
4299 | bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has |
4300 | an unclaimed reference count */ | |
36eb48b4 | 4301 | |
9a53f6cf | 4302 | /* If this operation fetched a swash, get its already existing |
83199d38 | 4303 | * inversion list, or create one for it */ |
36eb48b4 | 4304 | |
83199d38 | 4305 | if (swash_hv) { |
5c9f4bd2 | 4306 | swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE); |
9a53f6cf KW |
4307 | if (swash_invlistsvp) { |
4308 | swash_invlist = *swash_invlistsvp; | |
4309 | invlist_in_swash_is_valid = TRUE; | |
4310 | } | |
4311 | else { | |
36eb48b4 | 4312 | swash_invlist = _swash_to_invlist(retval); |
02c85471 | 4313 | swash_invlist_unclaimed = TRUE; |
9a53f6cf KW |
4314 | } |
4315 | } | |
4316 | ||
4317 | /* If an inversion list was passed in, have to include it */ | |
4318 | if (invlist) { | |
4319 | ||
4320 | /* Any fetched swash will by now have an inversion list in it; | |
4321 | * otherwise <swash_invlist> will be NULL, indicating that we | |
4322 | * didn't fetch a swash */ | |
4323 | if (swash_invlist) { | |
4324 | ||
4325 | /* Add the passed-in inversion list, which invalidates the one | |
4326 | * already stored in the swash */ | |
4327 | invlist_in_swash_is_valid = FALSE; | |
eee4c920 | 4328 | SvREADONLY_off(swash_invlist); /* Turned on again below */ |
9a53f6cf KW |
4329 | _invlist_union(invlist, swash_invlist, &swash_invlist); |
4330 | } | |
4331 | else { | |
4332 | ||
87367d5f KW |
4333 | /* Here, there is no swash already. Set up a minimal one, if |
4334 | * we are going to return a swash */ | |
26141ee3 | 4335 | if (! use_invlist) { |
971d486f | 4336 | swash_hv = newHV(); |
4aca0fe6 | 4337 | retval = newRV_noinc(MUTABLE_SV(swash_hv)); |
87367d5f | 4338 | } |
9a53f6cf KW |
4339 | swash_invlist = invlist; |
4340 | } | |
9a53f6cf KW |
4341 | } |
4342 | ||
4343 | /* Here, we have computed the union of all the passed-in data. It may | |
4344 | * be that there was an inversion list in the swash which didn't get | |
538e84ed | 4345 | * touched; otherwise save the computed one */ |
26141ee3 | 4346 | if (! invlist_in_swash_is_valid && ! use_invlist) { |
5c9f4bd2 | 4347 | if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist)) |
69794297 KW |
4348 | { |
4349 | Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed"); | |
4350 | } | |
cc34d8c5 FC |
4351 | /* We just stole a reference count. */ |
4352 | if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE; | |
4353 | else SvREFCNT_inc_simple_void_NN(swash_invlist); | |
9a53f6cf | 4354 | } |
87367d5f | 4355 | |
ee3222e3 | 4356 | /* The result is immutable. Forbid attempts to change it. */ |
dbfdbd26 KW |
4357 | SvREADONLY_on(swash_invlist); |
4358 | ||
26141ee3 | 4359 | if (use_invlist) { |
87367d5f | 4360 | SvREFCNT_dec(retval); |
02c85471 FC |
4361 | if (!swash_invlist_unclaimed) |
4362 | SvREFCNT_inc_simple_void_NN(swash_invlist); | |
4363 | retval = newRV_noinc(swash_invlist); | |
87367d5f | 4364 | } |
36eb48b4 KW |
4365 | } |
4366 | ||
2c1f00b9 YO |
4367 | CORE_SWASH_INIT_RETURN(retval); |
4368 | #undef CORE_SWASH_INIT_RETURN | |
a0ed51b3 LW |
4369 | } |
4370 | ||
035d37be JH |
4371 | |
4372 | /* This API is wrong for special case conversions since we may need to | |
4373 | * return several Unicode characters for a single Unicode character | |
4374 | * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is | |
4375 | * the lower-level routine, and it is similarly broken for returning | |
38684baa | 4376 | * multiple values. --jhi |
b9992569 | 4377 | * For those, you should use S__to_utf8_case() instead */ |
b0e3252e | 4378 | /* Now SWASHGET is recasted into S_swatch_get in this file. */ |
680c470c TS |
4379 | |
4380 | /* Note: | |
4381 | * Returns the value of property/mapping C<swash> for the first character | |
4382 | * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is | |
4a4088c4 | 4383 | * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr> |
3d0f8846 | 4384 | * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>. |
af2af982 KW |
4385 | * |
4386 | * A "swash" is a hash which contains initially the keys/values set up by | |
4387 | * SWASHNEW. The purpose is to be able to completely represent a Unicode | |
4388 | * property for all possible code points. Things are stored in a compact form | |
4389 | * (see utf8_heavy.pl) so that calculation is required to find the actual | |
4390 | * property value for a given code point. As code points are looked up, new | |
4391 | * key/value pairs are added to the hash, so that the calculation doesn't have | |
4392 | * to ever be re-done. Further, each calculation is done, not just for the | |
4393 | * desired one, but for a whole block of code points adjacent to that one. | |
4394 | * For binary properties on ASCII machines, the block is usually for 64 code | |
4395 | * points, starting with a code point evenly divisible by 64. Thus if the | |
4396 | * property value for code point 257 is requested, the code goes out and | |
4397 | * calculates the property values for all 64 code points between 256 and 319, | |
4398 | * and stores these as a single 64-bit long bit vector, called a "swatch", | |
4399 | * under the key for code point 256. The key is the UTF-8 encoding for code | |
4400 | * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding | |
4401 | * for a code point is 13 bytes, the key will be 12 bytes long. If the value | |
4402 | * for code point 258 is then requested, this code realizes that it would be | |
4403 | * stored under the key for 256, and would find that value and extract the | |
4404 | * relevant bit, offset from 256. | |
4405 | * | |
4406 | * Non-binary properties are stored in as many bits as necessary to represent | |
4407 | * their values (32 currently, though the code is more general than that), not | |
fc273927 | 4408 | * as single bits, but the principle is the same: the value for each key is a |
af2af982 KW |
4409 | * vector that encompasses the property values for all code points whose UTF-8 |
4410 | * representations are represented by the key. That is, for all code points | |
4411 | * whose UTF-8 representations are length N bytes, and the key is the first N-1 | |
4412 | * bytes of that. | |
680c470c | 4413 | */ |
a0ed51b3 | 4414 | UV |
680c470c | 4415 | Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8) |
a0ed51b3 | 4416 | { |
ef8f7699 | 4417 | HV *const hv = MUTABLE_HV(SvRV(swash)); |
3568d838 JH |
4418 | U32 klen; |
4419 | U32 off; | |
9b56a019 | 4420 | STRLEN slen = 0; |
7d85a32c | 4421 | STRLEN needents; |
cfd0369c | 4422 | const U8 *tmps = NULL; |
979f2922 | 4423 | SV *swatch; |
08fb1ac5 | 4424 | const U8 c = *ptr; |
3568d838 | 4425 | |
7918f24d NC |
4426 | PERL_ARGS_ASSERT_SWASH_FETCH; |
4427 | ||
87367d5f KW |
4428 | /* If it really isn't a hash, it isn't really swash; must be an inversion |
4429 | * list */ | |
4430 | if (SvTYPE(hv) != SVt_PVHV) { | |
4431 | return _invlist_contains_cp((SV*)hv, | |
4432 | (do_utf8) | |
4433 | ? valid_utf8_to_uvchr(ptr, NULL) | |
4434 | : c); | |
4435 | } | |
4436 | ||
08fb1ac5 KW |
4437 | /* We store the values in a "swatch" which is a vec() value in a swash |
4438 | * hash. Code points 0-255 are a single vec() stored with key length | |
4439 | * (klen) 0. All other code points have a UTF-8 representation | |
4440 | * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which | |
4441 | * share 0xAA..0xYY, which is the key in the hash to that vec. So the key | |
4442 | * length for them is the length of the encoded char - 1. ptr[klen] is the | |
4443 | * final byte in the sequence representing the character */ | |
4444 | if (!do_utf8 || UTF8_IS_INVARIANT(c)) { | |
4445 | klen = 0; | |
4446 | needents = 256; | |
4447 | off = c; | |
3568d838 | 4448 | } |
08fb1ac5 KW |
4449 | else if (UTF8_IS_DOWNGRADEABLE_START(c)) { |
4450 | klen = 0; | |
4451 | needents = 256; | |
a62b247b | 4452 | off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1)); |
979f2922 TS |
4453 | } |
4454 | else { | |
08fb1ac5 KW |
4455 | klen = UTF8SKIP(ptr) - 1; |
4456 | ||
4457 | /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into | |
4458 | * the vec is the final byte in the sequence. (In EBCDIC this is | |
4459 | * converted to I8 to get consecutive values.) To help you visualize | |
4460 | * all this: | |
4461 | * Straight 1047 After final byte | |
4462 | * UTF-8 UTF-EBCDIC I8 transform | |
4463 | * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0 | |
4464 | * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1 | |
4465 | * ... | |
4466 | * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9 | |
4467 | * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA | |
4468 | * ... | |
4469 | * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2 | |
4470 | * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3 | |
4471 | * ... | |
4472 | * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB | |
4473 | * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC | |
4474 | * ... | |
4475 | * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF | |
4476 | * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41 | |
4477 | * | |
4478 | * (There are no discontinuities in the elided (...) entries.) | |
4479 | * The UTF-8 key for these 33 code points is '\xD0' (which also is the | |
4480 | * key for the next 31, up through U+043F, whose UTF-8 final byte is | |
4481 | * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points. | |
4482 | * The final UTF-8 byte, which ranges between \x80 and \xBF, is an | |
4483 | * index into the vec() swatch (after subtracting 0x80, which we | |
4484 | * actually do with an '&'). | |
4485 | * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32 | |
4486 | * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has | |
4487 | * dicontinuities which go away by transforming it into I8, and we | |
4488 | * effectively subtract 0xA0 to get the index. */ | |
979f2922 | 4489 | needents = (1 << UTF_ACCUMULATION_SHIFT); |
bc3632a8 | 4490 | off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK; |
979f2922 | 4491 | } |
7d85a32c | 4492 | |
a0ed51b3 | 4493 | /* |
4a4088c4 | 4494 | * This single-entry cache saves about 1/3 of the UTF-8 overhead in test |
a0ed51b3 LW |
4495 | * suite. (That is, only 7-8% overall over just a hash cache. Still, |
4496 | * it's nothing to sniff at.) Pity we usually come through at least | |
4497 | * two function calls to get here... | |
4498 | * | |
4499 | * NB: this code assumes that swatches are never modified, once generated! | |
4500 | */ | |
4501 | ||
3568d838 | 4502 | if (hv == PL_last_swash_hv && |
a0ed51b3 | 4503 | klen == PL_last_swash_klen && |
27da23d5 | 4504 | (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) ) |
a0ed51b3 LW |
4505 | { |
4506 | tmps = PL_last_swash_tmps; | |
4507 | slen = PL_last_swash_slen; | |
4508 | } | |
4509 | else { | |
4510 | /* Try our second-level swatch cache, kept in a hash. */ | |
e1ec3a88 | 4511 | SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE); |
a0ed51b3 | 4512 | |
b0e3252e | 4513 | /* If not cached, generate it via swatch_get */ |
979f2922 | 4514 | if (!svp || !SvPOK(*svp) |
08fb1ac5 KW |
4515 | || !(tmps = (const U8*)SvPV_const(*svp, slen))) |
4516 | { | |
4517 | if (klen) { | |
4518 | const UV code_point = valid_utf8_to_uvchr(ptr, NULL); | |
4519 | swatch = swatch_get(swash, | |
4520 | code_point & ~((UV)needents - 1), | |
4521 | needents); | |
4522 | } | |
4523 | else { /* For the first 256 code points, the swatch has a key of | |
4524 | length 0 */ | |
4525 | swatch = swatch_get(swash, 0, needents); | |
4526 | } | |
979f2922 | 4527 | |
923e4eb5 | 4528 | if (IN_PERL_COMPILETIME) |
623e6609 | 4529 | CopHINTS_set(PL_curcop, PL_hints); |
a0ed51b3 | 4530 | |
979f2922 | 4531 | svp = hv_store(hv, (const char *)ptr, klen, swatch, 0); |
a0ed51b3 | 4532 | |
979f2922 TS |
4533 | if (!svp || !(tmps = (U8*)SvPV(*svp, slen)) |
4534 | || (slen << 3) < needents) | |
5637ef5b | 4535 | Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, " |
147e3846 | 4536 | "svp=%p, tmps=%p, slen=%" UVuf ", needents=%" UVuf, |
5637ef5b | 4537 | svp, tmps, (UV)slen, (UV)needents); |
a0ed51b3 LW |
4538 | } |
4539 | ||
4540 | PL_last_swash_hv = hv; | |
16d8f38a | 4541 | assert(klen <= sizeof(PL_last_swash_key)); |
eac04b2e | 4542 | PL_last_swash_klen = (U8)klen; |
cfd0369c NC |
4543 | /* FIXME change interpvar.h? */ |
4544 | PL_last_swash_tmps = (U8 *) tmps; | |
a0ed51b3 LW |
4545 | PL_last_swash_slen = slen; |
4546 | if (klen) | |
4547 | Copy(ptr, PL_last_swash_key, klen, U8); | |
4548 | } | |
4549 | ||
9faf8d75 | 4550 | switch ((int)((slen << 3) / needents)) { |
a0ed51b3 | 4551 | case 1: |
e7aca353 | 4552 | return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0; |
a0ed51b3 | 4553 | case 8: |
e7aca353 | 4554 | return ((UV) tmps[off]); |
a0ed51b3 LW |
4555 | case 16: |
4556 | off <<= 1; | |
e7aca353 JH |
4557 | return |
4558 | ((UV) tmps[off ] << 8) + | |
4559 | ((UV) tmps[off + 1]); | |
a0ed51b3 LW |
4560 | case 32: |
4561 | off <<= 2; | |
e7aca353 JH |
4562 | return |
4563 | ((UV) tmps[off ] << 24) + | |
4564 | ((UV) tmps[off + 1] << 16) + | |
4565 | ((UV) tmps[off + 2] << 8) + | |
4566 | ((UV) tmps[off + 3]); | |
a0ed51b3 | 4567 | } |
5637ef5b | 4568 | Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, " |
147e3846 | 4569 | "slen=%" UVuf ", needents=%" UVuf, (UV)slen, (UV)needents); |
670f1322 | 4570 | NORETURN_FUNCTION_END; |
a0ed51b3 | 4571 | } |
2b9d42f0 | 4572 | |
319009ee KW |
4573 | /* Read a single line of the main body of the swash input text. These are of |
4574 | * the form: | |
4575 | * 0053 0056 0073 | |
4576 | * where each number is hex. The first two numbers form the minimum and | |
4577 | * maximum of a range, and the third is the value associated with the range. | |
4578 | * Not all swashes should have a third number | |
4579 | * | |
4580 | * On input: l points to the beginning of the line to be examined; it points | |
4581 | * to somewhere in the string of the whole input text, and is | |
4582 | * terminated by a \n or the null string terminator. | |
4583 | * lend points to the null terminator of that string | |
4584 | * wants_value is non-zero if the swash expects a third number | |
4585 | * typestr is the name of the swash's mapping, like 'ToLower' | |
4586 | * On output: *min, *max, and *val are set to the values read from the line. | |
4587 | * returns a pointer just beyond the line examined. If there was no | |
4588 | * valid min number on the line, returns lend+1 | |
4589 | */ | |
4590 | ||
4591 | STATIC U8* | |
4592 | S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val, | |
4593 | const bool wants_value, const U8* const typestr) | |
4594 | { | |
4595 | const int typeto = typestr[0] == 'T' && typestr[1] == 'o'; | |
4596 | STRLEN numlen; /* Length of the number */ | |
02470786 KW |
4597 | I32 flags = PERL_SCAN_SILENT_ILLDIGIT |
4598 | | PERL_SCAN_DISALLOW_PREFIX | |
4599 | | PERL_SCAN_SILENT_NON_PORTABLE; | |
319009ee KW |
4600 | |
4601 | /* nl points to the next \n in the scan */ | |
4602 | U8* const nl = (U8*)memchr(l, '\n', lend - l); | |
4603 | ||
95543e92 KW |
4604 | PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE; |
4605 | ||
319009ee KW |
4606 | /* Get the first number on the line: the range minimum */ |
4607 | numlen = lend - l; | |
4608 | *min = grok_hex((char *)l, &numlen, &flags, NULL); | |
c88850db | 4609 | *max = *min; /* So can never return without setting max */ |
319009ee KW |
4610 | if (numlen) /* If found a hex number, position past it */ |
4611 | l += numlen; | |
4612 | else if (nl) { /* Else, go handle next line, if any */ | |
4613 | return nl + 1; /* 1 is length of "\n" */ | |
4614 | } | |
4615 | else { /* Else, no next line */ | |
4616 | return lend + 1; /* to LIST's end at which \n is not found */ | |
4617 | } | |
4618 | ||
4619 | /* The max range value follows, separated by a BLANK */ | |
4620 | if (isBLANK(*l)) { | |
4621 | ++l; | |
02470786 KW |
4622 | flags = PERL_SCAN_SILENT_ILLDIGIT |
4623 | | PERL_SCAN_DISALLOW_PREFIX | |
4624 | | PERL_SCAN_SILENT_NON_PORTABLE; | |
319009ee KW |
4625 | numlen = lend - l; |
4626 | *max = grok_hex((char *)l, &numlen, &flags, NULL); | |
4627 | if (numlen) | |
4628 | l += numlen; | |
4629 | else /* If no value here, it is a single element range */ | |
4630 | *max = *min; | |
4631 | ||
4632 | /* Non-binary tables have a third entry: what the first element of the | |
24303724 | 4633 | * range maps to. The map for those currently read here is in hex */ |
319009ee KW |
4634 | if (wants_value) { |
4635 | if (isBLANK(*l)) { | |
4636 | ++l; | |
f2a7d0fc KW |
4637 | flags = PERL_SCAN_SILENT_ILLDIGIT |
4638 | | PERL_SCAN_DISALLOW_PREFIX | |
4639 | | PERL_SCAN_SILENT_NON_PORTABLE; | |
4640 | numlen = lend - l; | |
4641 | *val = grok_hex((char *)l, &numlen, &flags, NULL); | |
4642 | if (numlen) | |
4643 | l += numlen; | |
4644 | else | |
4645 | *val = 0; | |
319009ee KW |
4646 | } |
4647 | else { | |
4648 | *val = 0; | |
4649 | if (typeto) { | |
dcbac5bb | 4650 | /* diag_listed_as: To%s: illegal mapping '%s' */ |
319009ee KW |
4651 | Perl_croak(aTHX_ "%s: illegal mapping '%s'", |
4652 | typestr, l); | |
4653 | } | |
4654 | } | |
4655 | } | |
4656 | else | |
4657 | *val = 0; /* bits == 1, then any val should be ignored */ | |
4658 | } | |
4659 | else { /* Nothing following range min, should be single element with no | |
4660 | mapping expected */ | |
319009ee KW |
4661 | if (wants_value) { |
4662 | *val = 0; | |
4663 | if (typeto) { | |
dcbac5bb | 4664 | /* diag_listed_as: To%s: illegal mapping '%s' */ |
319009ee KW |
4665 | Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l); |
4666 | } | |
4667 | } | |
4668 | else | |
4669 | *val = 0; /* bits == 1, then val should be ignored */ | |
4670 | } | |
4671 | ||
4672 | /* Position to next line if any, or EOF */ | |
4673 | if (nl) | |
4674 | l = nl + 1; | |
4675 | else | |
4676 | l = lend; | |
4677 | ||
4678 | return l; | |
4679 | } | |
4680 | ||
979f2922 TS |
4681 | /* Note: |
4682 | * Returns a swatch (a bit vector string) for a code point sequence | |
4683 | * that starts from the value C<start> and comprises the number C<span>. | |
4684 | * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl). | |
4685 | * Should be used via swash_fetch, which will cache the swatch in C<swash>. | |
4686 | */ | |
4687 | STATIC SV* | |
b0e3252e | 4688 | S_swatch_get(pTHX_ SV* swash, UV start, UV span) |
979f2922 TS |
4689 | { |
4690 | SV *swatch; | |
77f9f126 | 4691 | U8 *l, *lend, *x, *xend, *s, *send; |
979f2922 | 4692 | STRLEN lcur, xcur, scur; |
ef8f7699 | 4693 | HV *const hv = MUTABLE_HV(SvRV(swash)); |
5c9f4bd2 | 4694 | SV** const invlistsvp = hv_fetchs(hv, "V", FALSE); |
36eb48b4 | 4695 | |
88d45d28 KW |
4696 | SV** listsvp = NULL; /* The string containing the main body of the table */ |
4697 | SV** extssvp = NULL; | |
4698 | SV** invert_it_svp = NULL; | |
4699 | U8* typestr = NULL; | |
786861f5 KW |
4700 | STRLEN bits; |
4701 | STRLEN octets; /* if bits == 1, then octets == 0 */ | |
4702 | UV none; | |
4703 | UV end = start + span; | |
972dd592 | 4704 | |
36eb48b4 | 4705 | if (invlistsvp == NULL) { |
786861f5 KW |
4706 | SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE); |
4707 | SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE); | |
4708 | SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE); | |
4709 | extssvp = hv_fetchs(hv, "EXTRAS", FALSE); | |
4710 | listsvp = hv_fetchs(hv, "LIST", FALSE); | |
4711 | invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE); | |
4712 | ||
4713 | bits = SvUV(*bitssvp); | |
4714 | none = SvUV(*nonesvp); | |
4715 | typestr = (U8*)SvPV_nolen(*typesvp); | |
4716 | } | |
36eb48b4 KW |
4717 | else { |
4718 | bits = 1; | |
4719 | none = 0; | |
4720 | } | |
786861f5 | 4721 | octets = bits >> 3; /* if bits == 1, then octets == 0 */ |
979f2922 | 4722 | |
b0e3252e | 4723 | PERL_ARGS_ASSERT_SWATCH_GET; |
7918f24d | 4724 | |
979f2922 | 4725 | if (bits != 1 && bits != 8 && bits != 16 && bits != 32) { |
147e3846 | 4726 | Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %" UVuf, |
660a4616 | 4727 | (UV)bits); |
979f2922 TS |
4728 | } |
4729 | ||
84ea5ef6 KW |
4730 | /* If overflowed, use the max possible */ |
4731 | if (end < start) { | |
4732 | end = UV_MAX; | |
4733 | span = end - start; | |
4734 | } | |
4735 | ||
979f2922 | 4736 | /* create and initialize $swatch */ |
979f2922 | 4737 | scur = octets ? (span * octets) : (span + 7) / 8; |
e524fe40 NC |
4738 | swatch = newSV(scur); |
4739 | SvPOK_on(swatch); | |
979f2922 TS |
4740 | s = (U8*)SvPVX(swatch); |
4741 | if (octets && none) { | |
0bd48802 | 4742 | const U8* const e = s + scur; |
979f2922 TS |
4743 | while (s < e) { |
4744 | if (bits == 8) | |
4745 | *s++ = (U8)(none & 0xff); | |
4746 | else if (bits == 16) { | |
4747 | *s++ = (U8)((none >> 8) & 0xff); | |
4748 | *s++ = (U8)( none & 0xff); | |
4749 | } | |
4750 | else if (bits == 32) { | |
4751 | *s++ = (U8)((none >> 24) & 0xff); | |
4752 | *s++ = (U8)((none >> 16) & 0xff); | |
4753 | *s++ = (U8)((none >> 8) & 0xff); | |
4754 | *s++ = (U8)( none & 0xff); | |
4755 | } | |
4756 | } | |
4757 | *s = '\0'; | |
4758 | } | |
4759 | else { | |
4760 | (void)memzero((U8*)s, scur + 1); | |
4761 | } | |
4762 | SvCUR_set(swatch, scur); | |
4763 | s = (U8*)SvPVX(swatch); | |
4764 | ||
36eb48b4 KW |
4765 | if (invlistsvp) { /* If has an inversion list set up use that */ |
4766 | _invlist_populate_swatch(*invlistsvp, start, end, s); | |
4767 | return swatch; | |
4768 | } | |
4769 | ||
4770 | /* read $swash->{LIST} */ | |
979f2922 TS |
4771 | l = (U8*)SvPV(*listsvp, lcur); |
4772 | lend = l + lcur; | |
4773 | while (l < lend) { | |
8ed25d53 | 4774 | UV min, max, val, upper; |
95543e92 KW |
4775 | l = swash_scan_list_line(l, lend, &min, &max, &val, |
4776 | cBOOL(octets), typestr); | |
319009ee | 4777 | if (l > lend) { |
979f2922 TS |
4778 | break; |
4779 | } | |
4780 | ||
972dd592 | 4781 | /* If looking for something beyond this range, go try the next one */ |
979f2922 TS |
4782 | if (max < start) |
4783 | continue; | |
4784 | ||
8ed25d53 KW |
4785 | /* <end> is generally 1 beyond where we want to set things, but at the |
4786 | * platform's infinity, where we can't go any higher, we want to | |
4787 | * include the code point at <end> */ | |
4788 | upper = (max < end) | |
4789 | ? max | |
4790 | : (max != UV_MAX || end != UV_MAX) | |
4791 | ? end - 1 | |
4792 | : end; | |
4793 | ||
979f2922 | 4794 | if (octets) { |
35da51f7 | 4795 | UV key; |
979f2922 TS |
4796 | if (min < start) { |
4797 | if (!none || val < none) { | |
4798 | val += start - min; | |
4799 | } | |
4800 | min = start; | |
4801 | } | |
8ed25d53 | 4802 | for (key = min; key <= upper; key++) { |
979f2922 | 4803 | STRLEN offset; |
979f2922 TS |
4804 | /* offset must be non-negative (start <= min <= key < end) */ |
4805 | offset = octets * (key - start); | |
4806 | if (bits == 8) | |
4807 | s[offset] = (U8)(val & 0xff); | |
4808 | else if (bits == 16) { | |
4809 | s[offset ] = (U8)((val >> 8) & 0xff); | |
4810 | s[offset + 1] = (U8)( val & 0xff); | |
4811 | } | |
4812 | else if (bits == 32) { | |
4813 | s[offset ] = (U8)((val >> 24) & 0xff); | |
4814 | s[offset + 1] = (U8)((val >> 16) & 0xff); | |
4815 | s[offset + 2] = (U8)((val >> 8) & 0xff); | |
4816 | s[offset + 3] = (U8)( val & 0xff); | |
4817 | } | |
4818 | ||
4819 | if (!none || val < none) | |
4820 | ++val; | |
4821 | } | |
4822 | } | |
711a919c | 4823 | else { /* bits == 1, then val should be ignored */ |
35da51f7 | 4824 | UV key; |
979f2922 TS |
4825 | if (min < start) |
4826 | min = start; | |
6cb05c12 | 4827 | |
8ed25d53 | 4828 | for (key = min; key <= upper; key++) { |
0bd48802 | 4829 | const STRLEN offset = (STRLEN)(key - start); |
979f2922 TS |
4830 | s[offset >> 3] |= 1 << (offset & 7); |
4831 | } | |
4832 | } | |
4833 | } /* while */ | |
979f2922 | 4834 | |
9479a769 | 4835 | /* Invert if the data says it should be. Assumes that bits == 1 */ |
77f9f126 | 4836 | if (invert_it_svp && SvUV(*invert_it_svp)) { |
0bda3001 KW |
4837 | |
4838 | /* Unicode properties should come with all bits above PERL_UNICODE_MAX | |
4839 | * be 0, and their inversion should also be 0, as we don't succeed any | |
4840 | * Unicode property matches for non-Unicode code points */ | |
4841 | if (start <= PERL_UNICODE_MAX) { | |
4842 | ||
4843 | /* The code below assumes that we never cross the | |
4844 | * Unicode/above-Unicode boundary in a range, as otherwise we would | |
4845 | * have to figure out where to stop flipping the bits. Since this | |
4846 | * boundary is divisible by a large power of 2, and swatches comes | |
4847 | * in small powers of 2, this should be a valid assumption */ | |
4848 | assert(start + span - 1 <= PERL_UNICODE_MAX); | |
4849 | ||
507a8485 KW |
4850 | send = s + scur; |
4851 | while (s < send) { | |
4852 | *s = ~(*s); | |
4853 | s++; | |
4854 | } | |
0bda3001 | 4855 | } |
77f9f126 KW |
4856 | } |
4857 | ||
d73c39c5 KW |
4858 | /* read $swash->{EXTRAS} |
4859 | * This code also copied to swash_to_invlist() below */ | |
979f2922 TS |
4860 | x = (U8*)SvPV(*extssvp, xcur); |
4861 | xend = x + xcur; | |
4862 | while (x < xend) { | |
4863 | STRLEN namelen; | |
4864 | U8 *namestr; | |
4865 | SV** othersvp; | |
4866 | HV* otherhv; | |
4867 | STRLEN otherbits; | |
4868 | SV **otherbitssvp, *other; | |
711a919c | 4869 | U8 *s, *o, *nl; |
979f2922 TS |
4870 | STRLEN slen, olen; |
4871 | ||
35da51f7 | 4872 | const U8 opc = *x++; |
979f2922 TS |
4873 | if (opc == '\n') |
4874 | continue; | |
4875 | ||
4876 | nl = (U8*)memchr(x, '\n', xend - x); | |
4877 | ||
4878 | if (opc != '-' && opc != '+' && opc != '!' && opc != '&') { | |
4879 | if (nl) { | |
4880 | x = nl + 1; /* 1 is length of "\n" */ | |
4881 | continue; | |
4882 | } | |
4883 | else { | |
4884 | x = xend; /* to EXTRAS' end at which \n is not found */ | |
4885 | break; | |
4886 | } | |
4887 | } | |
4888 | ||
4889 | namestr = x; | |
4890 | if (nl) { | |
4891 | namelen = nl - namestr; | |
4892 | x = nl + 1; | |
4893 | } | |
4894 | else { | |
4895 | namelen = xend - namestr; | |
4896 | x = xend; | |
4897 | } | |
4898 | ||
4899 | othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE); | |
ef8f7699 | 4900 | otherhv = MUTABLE_HV(SvRV(*othersvp)); |
017a3ce5 | 4901 | otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE); |
979f2922 TS |
4902 | otherbits = (STRLEN)SvUV(*otherbitssvp); |
4903 | if (bits < otherbits) | |
5637ef5b | 4904 | Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, " |
147e3846 | 4905 | "bits=%" UVuf ", otherbits=%" UVuf, (UV)bits, (UV)otherbits); |
979f2922 TS |
4906 | |
4907 | /* The "other" swatch must be destroyed after. */ | |
b0e3252e | 4908 | other = swatch_get(*othersvp, start, span); |
979f2922 TS |
4909 | o = (U8*)SvPV(other, olen); |
4910 | ||
4911 | if (!olen) | |
b0e3252e | 4912 | Perl_croak(aTHX_ "panic: swatch_get got improper swatch"); |
979f2922 TS |
4913 | |
4914 | s = (U8*)SvPV(swatch, slen); | |
4915 | if (bits == 1 && otherbits == 1) { | |
4916 | if (slen != olen) | |
5637ef5b | 4917 | Perl_croak(aTHX_ "panic: swatch_get found swatch length " |
147e3846 | 4918 | "mismatch, slen=%" UVuf ", olen=%" UVuf, |
5637ef5b | 4919 | (UV)slen, (UV)olen); |
979f2922 TS |
4920 | |
4921 | switch (opc) { | |
4922 | case '+': | |
4923 | while (slen--) | |
4924 | *s++ |= *o++; | |
4925 | break; | |
4926 | case '!': | |
4927 | while (slen--) | |
4928 | *s++ |= ~*o++; | |
4929 | break; | |
4930 | case '-': | |
4931 | while (slen--) | |
4932 | *s++ &= ~*o++; | |
4933 | break; | |
4934 | case '&': | |
4935 | while (slen--) | |
4936 | *s++ &= *o++; | |
4937 | break; | |
4938 | default: | |
4939 | break; | |
4940 | } | |
4941 | } | |
711a919c | 4942 | else { |
979f2922 TS |
4943 | STRLEN otheroctets = otherbits >> 3; |
4944 | STRLEN offset = 0; | |
35da51f7 | 4945 | U8* const send = s + slen; |
979f2922 TS |
4946 | |
4947 | while (s < send) { | |
4948 | UV otherval = 0; | |
4949 | ||
4950 | if (otherbits == 1) { | |
4951 | otherval = (o[offset >> 3] >> (offset & 7)) & 1; | |
4952 | ++offset; | |
4953 | } | |
4954 | else { | |
4955 | STRLEN vlen = otheroctets; | |
4956 | otherval = *o++; | |
4957 | while (--vlen) { | |
4958 | otherval <<= 8; | |
4959 | otherval |= *o++; | |
4960 | } | |
4961 | } | |
4962 | ||
711a919c | 4963 | if (opc == '+' && otherval) |
6f207bd3 | 4964 | NOOP; /* replace with otherval */ |
979f2922 TS |
4965 | else if (opc == '!' && !otherval) |
4966 | otherval = 1; | |
4967 | else if (opc == '-' && otherval) | |
4968 | otherval = 0; | |
4969 | else if (opc == '&' && !otherval) | |
4970 | otherval = 0; | |
4971 | else { | |
711a919c | 4972 | s += octets; /* no replacement */ |
979f2922 TS |
4973 | continue; |
4974 | } | |
4975 | ||
4976 | if (bits == 8) | |
4977 | *s++ = (U8)( otherval & 0xff); | |
4978 | else if (bits == 16) { | |
4979 | *s++ = (U8)((otherval >> 8) & 0xff); | |
4980 | *s++ = (U8)( otherval & 0xff); | |
4981 | } | |
4982 | else if (bits == 32) { | |
4983 | *s++ = (U8)((otherval >> 24) & 0xff); | |
4984 | *s++ = (U8)((otherval >> 16) & 0xff); | |
4985 | *s++ = (U8)((otherval >> 8) & 0xff); | |
4986 | *s++ = (U8)( otherval & 0xff); | |
4987 | } | |
4988 | } | |
4989 | } | |
4990 | sv_free(other); /* through with it! */ | |
4991 | } /* while */ | |
4992 | return swatch; | |
4993 | } | |
4994 | ||
a25abddc | 4995 | SV* |
d764b54e KW |
4996 | Perl__swash_to_invlist(pTHX_ SV* const swash) |
4997 | { | |
4998 | ||
ed92f1b3 KW |
4999 | /* Subject to change or removal. For use only in one place in regcomp.c. |
5000 | * Ownership is given to one reference count in the returned SV* */ | |
d764b54e KW |
5001 | |
5002 | U8 *l, *lend; | |
5003 | char *loc; | |
5004 | STRLEN lcur; | |
5005 | HV *const hv = MUTABLE_HV(SvRV(swash)); | |
5006 | UV elements = 0; /* Number of elements in the inversion list */ | |
b443038a | 5007 | U8 empty[] = ""; |
923b6d4e KW |
5008 | SV** listsvp; |
5009 | SV** typesvp; | |
5010 | SV** bitssvp; | |
5011 | SV** extssvp; | |
5012 | SV** invert_it_svp; | |
d764b54e | 5013 | |
923b6d4e KW |
5014 | U8* typestr; |
5015 | STRLEN bits; | |
5016 | STRLEN octets; /* if bits == 1, then octets == 0 */ | |
d73c39c5 KW |
5017 | U8 *x, *xend; |
5018 | STRLEN xcur; | |
d764b54e | 5019 | |
a25abddc | 5020 | SV* invlist; |
d764b54e | 5021 | |
b81740c0 KW |
5022 | PERL_ARGS_ASSERT__SWASH_TO_INVLIST; |
5023 | ||
923b6d4e KW |
5024 | /* If not a hash, it must be the swash's inversion list instead */ |
5025 | if (SvTYPE(hv) != SVt_PVHV) { | |
ed92f1b3 | 5026 | return SvREFCNT_inc_simple_NN((SV*) hv); |
923b6d4e KW |
5027 | } |
5028 | ||
5029 | /* The string containing the main body of the table */ | |
5030 | listsvp = hv_fetchs(hv, "LIST", FALSE); | |
5031 | typesvp = hv_fetchs(hv, "TYPE", FALSE); | |
5032 | bitssvp = hv_fetchs(hv, "BITS", FALSE); | |
5033 | extssvp = hv_fetchs(hv, "EXTRAS", FALSE); | |
5034 | invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE); | |
5035 | ||
5036 | typestr = (U8*)SvPV_nolen(*typesvp); | |
5037 | bits = SvUV(*bitssvp); | |
5038 | octets = bits >> 3; /* if bits == 1, then octets == 0 */ | |
5039 | ||
d764b54e | 5040 | /* read $swash->{LIST} */ |
b443038a KW |
5041 | if (SvPOK(*listsvp)) { |
5042 | l = (U8*)SvPV(*listsvp, lcur); | |
5043 | } | |
5044 | else { | |
5045 | /* LIST legitimately doesn't contain a string during compilation phases | |
5046 | * of Perl itself, before the Unicode tables are generated. In this | |
5047 | * case, just fake things up by creating an empty list */ | |
5048 | l = empty; | |
5049 | lcur = 0; | |
5050 | } | |
d764b54e KW |
5051 | loc = (char *) l; |
5052 | lend = l + lcur; | |
5053 | ||
31aa6e0b | 5054 | if (*l == 'V') { /* Inversion list format */ |
99944878 | 5055 | const char *after_atou = (char *) lend; |
31aa6e0b KW |
5056 | UV element0; |
5057 | UV* other_elements_ptr; | |
5058 | ||
5059 | /* The first number is a count of the rest */ | |
5060 | l++; | |
22ff3130 | 5061 | if (!grok_atoUV((const char *)l, &elements, &after_atou)) { |
56576a04 KW |
5062 | Perl_croak(aTHX_ "panic: Expecting a valid count of elements" |
5063 | " at start of inversion list"); | |
22ff3130 | 5064 | } |
eb092534 KW |
5065 | if (elements == 0) { |
5066 | invlist = _new_invlist(0); | |
5067 | } | |
5068 | else { | |
99944878 | 5069 | l = (U8 *) after_atou; |
1f9f7d4c | 5070 | |
56576a04 KW |
5071 | /* Get the 0th element, which is needed to setup the inversion list |
5072 | * */ | |
99944878 | 5073 | while (isSPACE(*l)) l++; |
5d4a52b5 | 5074 | after_atou = (char *) lend; |
22ff3130 | 5075 | if (!grok_atoUV((const char *)l, &element0, &after_atou)) { |
56576a04 KW |
5076 | Perl_croak(aTHX_ "panic: Expecting a valid 0th element for" |
5077 | " inversion list"); | |
22ff3130 | 5078 | } |
99944878 | 5079 | l = (U8 *) after_atou; |
56576a04 KW |
5080 | invlist = _setup_canned_invlist(elements, element0, |
5081 | &other_elements_ptr); | |
1f9f7d4c KW |
5082 | elements--; |
5083 | ||
5084 | /* Then just populate the rest of the input */ | |
5085 | while (elements-- > 0) { | |
5086 | if (l > lend) { | |
56576a04 KW |
5087 | Perl_croak(aTHX_ "panic: Expecting %" UVuf " more" |
5088 | " elements than available", elements); | |
1f9f7d4c | 5089 | } |
99944878 | 5090 | while (isSPACE(*l)) l++; |
5d4a52b5 | 5091 | after_atou = (char *) lend; |
56576a04 KW |
5092 | if (!grok_atoUV((const char *)l, other_elements_ptr++, |
5093 | &after_atou)) | |
5094 | { | |
5095 | Perl_croak(aTHX_ "panic: Expecting a valid element" | |
5096 | " in inversion list"); | |
22ff3130 | 5097 | } |
99944878 | 5098 | l = (U8 *) after_atou; |
1f9f7d4c | 5099 | } |
eb092534 | 5100 | } |
31aa6e0b KW |
5101 | } |
5102 | else { | |
5103 | ||
1784d2f9 KW |
5104 | /* Scan the input to count the number of lines to preallocate array |
5105 | * size based on worst possible case, which is each line in the input | |
5106 | * creates 2 elements in the inversion list: 1) the beginning of a | |
5107 | * range in the list; 2) the beginning of a range not in the list. */ | |
320d4f1b | 5108 | while ((loc = (char *) memchr(loc, '\n', lend - (U8 *) loc)) != NULL) { |
1784d2f9 KW |
5109 | elements += 2; |
5110 | loc++; | |
5111 | } | |
d764b54e | 5112 | |
1784d2f9 KW |
5113 | /* If the ending is somehow corrupt and isn't a new line, add another |
5114 | * element for the final range that isn't in the inversion list */ | |
5115 | if (! (*lend == '\n' | |
5116 | || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n')))) | |
5117 | { | |
5118 | elements++; | |
5119 | } | |
d764b54e | 5120 | |
1784d2f9 | 5121 | invlist = _new_invlist(elements); |
d764b54e | 5122 | |
1784d2f9 KW |
5123 | /* Now go through the input again, adding each range to the list */ |
5124 | while (l < lend) { | |
5125 | UV start, end; | |
5126 | UV val; /* Not used by this function */ | |
d764b54e | 5127 | |
95543e92 KW |
5128 | l = swash_scan_list_line(l, lend, &start, &end, &val, |
5129 | cBOOL(octets), typestr); | |
d764b54e | 5130 | |
1784d2f9 KW |
5131 | if (l > lend) { |
5132 | break; | |
5133 | } | |
5134 | ||
5135 | invlist = _add_range_to_invlist(invlist, start, end); | |
5136 | } | |
31aa6e0b | 5137 | } |
d764b54e | 5138 | |
77f9f126 KW |
5139 | /* Invert if the data says it should be */ |
5140 | if (invert_it_svp && SvUV(*invert_it_svp)) { | |
25151030 | 5141 | _invlist_invert(invlist); |
77f9f126 KW |
5142 | } |
5143 | ||
b0e3252e | 5144 | /* This code is copied from swatch_get() |
d73c39c5 KW |
5145 | * read $swash->{EXTRAS} */ |
5146 | x = (U8*)SvPV(*extssvp, xcur); | |
5147 | xend = x + xcur; | |
5148 | while (x < xend) { | |
5149 | STRLEN namelen; | |
5150 | U8 *namestr; | |
5151 | SV** othersvp; | |
5152 | HV* otherhv; | |
5153 | STRLEN otherbits; | |
5154 | SV **otherbitssvp, *other; | |
5155 | U8 *nl; | |
5156 | ||
5157 | const U8 opc = *x++; | |
5158 | if (opc == '\n') | |
5159 | continue; | |
5160 | ||
5161 | nl = (U8*)memchr(x, '\n', xend - x); | |
5162 | ||
5163 | if (opc != '-' && opc != '+' && opc != '!' && opc != '&') { | |
5164 | if (nl) { | |
5165 | x = nl + 1; /* 1 is length of "\n" */ | |
5166 | continue; | |
5167 | } | |
5168 | else { | |
5169 | x = xend; /* to EXTRAS' end at which \n is not found */ | |
5170 | break; | |
5171 | } | |
5172 | } | |
5173 | ||
5174 | namestr = x; | |
5175 | if (nl) { | |
5176 | namelen = nl - namestr; | |
5177 | x = nl + 1; | |
5178 | } | |
5179 | else { | |
5180 | namelen = xend - namestr; | |
5181 | x = xend; | |
5182 | } | |
5183 | ||
5184 | othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE); | |
5185 | otherhv = MUTABLE_HV(SvRV(*othersvp)); | |
5186 | otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE); | |
5187 | otherbits = (STRLEN)SvUV(*otherbitssvp); | |
5188 | ||
5189 | if (bits != otherbits || bits != 1) { | |
5637ef5b | 5190 | Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean " |
147e3846 | 5191 | "properties, bits=%" UVuf ", otherbits=%" UVuf, |
5637ef5b | 5192 | (UV)bits, (UV)otherbits); |
d73c39c5 KW |
5193 | } |
5194 | ||
5195 | /* The "other" swatch must be destroyed after. */ | |
5196 | other = _swash_to_invlist((SV *)*othersvp); | |
5197 | ||
b0e3252e | 5198 | /* End of code copied from swatch_get() */ |
d73c39c5 KW |
5199 | switch (opc) { |
5200 | case '+': | |
5201 | _invlist_union(invlist, other, &invlist); | |
5202 | break; | |
5203 | case '!': | |
6c46377d | 5204 | _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist); |
d73c39c5 KW |
5205 | break; |
5206 | case '-': | |
5207 | _invlist_subtract(invlist, other, &invlist); | |
5208 | break; | |
5209 | case '&': | |
5210 | _invlist_intersection(invlist, other, &invlist); | |
5211 | break; | |
5212 | default: | |
5213 | break; | |
5214 | } | |
5215 | sv_free(other); /* through with it! */ | |
5216 | } | |
5217 | ||
dbfdbd26 | 5218 | SvREADONLY_on(invlist); |
d764b54e KW |
5219 | return invlist; |
5220 | } | |
5221 | ||
3fdfee00 KW |
5222 | SV* |
5223 | Perl__get_swash_invlist(pTHX_ SV* const swash) | |
5224 | { | |
872dd7e0 | 5225 | SV** ptr; |
3fdfee00 KW |
5226 | |
5227 | PERL_ARGS_ASSERT__GET_SWASH_INVLIST; | |
5228 | ||
87367d5f | 5229 | if (! SvROK(swash)) { |
872dd7e0 KW |
5230 | return NULL; |
5231 | } | |
5232 | ||
87367d5f KW |
5233 | /* If it really isn't a hash, it isn't really swash; must be an inversion |
5234 | * list */ | |
5235 | if (SvTYPE(SvRV(swash)) != SVt_PVHV) { | |
5236 | return SvRV(swash); | |
5237 | } | |
872dd7e0 | 5238 | |
87367d5f | 5239 | ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE); |
3fdfee00 KW |
5240 | if (! ptr) { |
5241 | return NULL; | |
5242 | } | |
5243 | ||
5244 | return *ptr; | |
5245 | } | |
5246 | ||
0876b9a0 | 5247 | bool |
5aaab254 | 5248 | Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len) |
0876b9a0 KW |
5249 | { |
5250 | /* May change: warns if surrogates, non-character code points, or | |
56576a04 KW |
5251 | * non-Unicode code points are in 's' which has length 'len' bytes. |
5252 | * Returns TRUE if none found; FALSE otherwise. The only other validity | |
d22ec717 KW |
5253 | * check is to make sure that this won't exceed the string's length nor |
5254 | * overflow */ | |
0876b9a0 KW |
5255 | |
5256 | const U8* const e = s + len; | |
5257 | bool ok = TRUE; | |
5258 | ||
5259 | PERL_ARGS_ASSERT_CHECK_UTF8_PRINT; | |
5260 | ||
5261 | while (s < e) { | |
5262 | if (UTF8SKIP(s) > len) { | |
5263 | Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), | |
5264 | "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print"); | |
5265 | return FALSE; | |
5266 | } | |
ac6f1fbe | 5267 | if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) { |
f2bf18cc | 5268 | if (UNLIKELY(UTF8_IS_SUPER(s, e))) { |
760c7c2f | 5269 | if ( ckWARN_d(WARN_NON_UNICODE) |
e050c007 KW |
5270 | || UNLIKELY(0 < does_utf8_overflow(s, s + len, |
5271 | 0 /* Don't consider overlongs */ | |
5272 | ))) | |
5273 | { | |
15ca5930 | 5274 | /* A side effect of this function will be to warn */ |
2db24202 | 5275 | (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_SUPER); |
7ee537e6 KW |
5276 | ok = FALSE; |
5277 | } | |
0876b9a0 | 5278 | } |
f2bf18cc | 5279 | else if (UNLIKELY(UTF8_IS_SURROGATE(s, e))) { |
8457b38f | 5280 | if (ckWARN_d(WARN_SURROGATE)) { |
15ca5930 KW |
5281 | /* This has a different warning than the one the called |
5282 | * function would output, so can't just call it, unlike we | |
5283 | * do for the non-chars and above-unicodes */ | |
2db24202 | 5284 | UV uv = utf8_to_uvchr_buf(s, e, NULL); |
8457b38f | 5285 | Perl_warner(aTHX_ packWARN(WARN_SURROGATE), |
56576a04 KW |
5286 | "Unicode surrogate U+%04" UVXf " is illegal in UTF-8", |
5287 | uv); | |
8457b38f KW |
5288 | ok = FALSE; |
5289 | } | |
0876b9a0 | 5290 | } |
56576a04 KW |
5291 | else if ( UNLIKELY(UTF8_IS_NONCHAR(s, e)) |
5292 | && (ckWARN_d(WARN_NONCHAR))) | |
5293 | { | |
15ca5930 | 5294 | /* A side effect of this function will be to warn */ |
2db24202 | 5295 | (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_NONCHAR); |
0876b9a0 KW |
5296 | ok = FALSE; |
5297 | } | |
5298 | } | |
5299 | s += UTF8SKIP(s); | |
5300 | } | |
5301 | ||
5302 | return ok; | |
5303 | } | |
5304 | ||
0f830e0b | 5305 | /* |
87cea99e | 5306 | =for apidoc pv_uni_display |
d2cc3551 | 5307 | |
a1433954 KW |
5308 | Build to the scalar C<dsv> a displayable version of the string C<spv>, |
5309 | length C<len>, the displayable version being at most C<pvlim> bytes long | |
796b6530 | 5310 | (if longer, the rest is truncated and C<"..."> will be appended). |
0a2ef054 | 5311 | |
796b6530 KW |
5312 | The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display |
5313 | C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH> | |
5314 | to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">) | |
5315 | (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">). | |
5316 | C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both | |
5317 | C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on. | |
0a2ef054 | 5318 | |
a1433954 | 5319 | The pointer to the PV of the C<dsv> is returned. |
d2cc3551 | 5320 | |
119bc988 KW |
5321 | See also L</sv_uni_display>. |
5322 | ||
d2cc3551 | 5323 | =cut */ |
e6b2e755 | 5324 | char * |
56576a04 KW |
5325 | Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, |
5326 | UV flags) | |
e6b2e755 JH |
5327 | { |
5328 | int truncated = 0; | |
e1ec3a88 | 5329 | const char *s, *e; |
e6b2e755 | 5330 | |
7918f24d NC |
5331 | PERL_ARGS_ASSERT_PV_UNI_DISPLAY; |
5332 | ||
9e2aa2e7 | 5333 | SvPVCLEAR(dsv); |
7fddd944 | 5334 | SvUTF8_off(dsv); |
e1ec3a88 | 5335 | for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) { |
e6b2e755 | 5336 | UV u; |
a49f32c6 NC |
5337 | /* This serves double duty as a flag and a character to print after |
5338 | a \ when flags & UNI_DISPLAY_BACKSLASH is true. | |
5339 | */ | |
5340 | char ok = 0; | |
c728cb41 | 5341 | |
e6b2e755 JH |
5342 | if (pvlim && SvCUR(dsv) >= pvlim) { |
5343 | truncated++; | |
5344 | break; | |
5345 | } | |
4b88fb76 | 5346 | u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0); |
c728cb41 | 5347 | if (u < 256) { |
a3b680e6 | 5348 | const unsigned char c = (unsigned char)u & 0xFF; |
0bd48802 | 5349 | if (flags & UNI_DISPLAY_BACKSLASH) { |
a49f32c6 | 5350 | switch (c) { |
c728cb41 | 5351 | case '\n': |
a49f32c6 | 5352 | ok = 'n'; break; |
c728cb41 | 5353 | case '\r': |
a49f32c6 | 5354 | ok = 'r'; break; |
c728cb41 | 5355 | case '\t': |
a49f32c6 | 5356 | ok = 't'; break; |
c728cb41 | 5357 | case '\f': |
a49f32c6 | 5358 | ok = 'f'; break; |
c728cb41 | 5359 | case '\a': |
a49f32c6 | 5360 | ok = 'a'; break; |
c728cb41 | 5361 | case '\\': |
a49f32c6 | 5362 | ok = '\\'; break; |
c728cb41 JH |
5363 | default: break; |
5364 | } | |
a49f32c6 | 5365 | if (ok) { |
88c9ea1e | 5366 | const char string = ok; |
76f68e9b | 5367 | sv_catpvs(dsv, "\\"); |
5e7aa789 | 5368 | sv_catpvn(dsv, &string, 1); |
a49f32c6 | 5369 | } |
c728cb41 | 5370 | } |
00e86452 | 5371 | /* isPRINT() is the locale-blind version. */ |
a49f32c6 | 5372 | if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) { |
88c9ea1e | 5373 | const char string = c; |
5e7aa789 | 5374 | sv_catpvn(dsv, &string, 1); |
a49f32c6 | 5375 | ok = 1; |
0a2ef054 | 5376 | } |
c728cb41 JH |
5377 | } |
5378 | if (!ok) | |
147e3846 | 5379 | Perl_sv_catpvf(aTHX_ dsv, "\\x{%" UVxf "}", u); |
e6b2e755 JH |
5380 | } |
5381 | if (truncated) | |
396482e1 | 5382 | sv_catpvs(dsv, "..."); |
48ef279e | 5383 | |
e6b2e755 JH |
5384 | return SvPVX(dsv); |
5385 | } | |
2b9d42f0 | 5386 | |
d2cc3551 | 5387 | /* |
87cea99e | 5388 | =for apidoc sv_uni_display |
d2cc3551 | 5389 | |
a1433954 KW |
5390 | Build to the scalar C<dsv> a displayable version of the scalar C<sv>, |
5391 | the displayable version being at most C<pvlim> bytes long | |
d2cc3551 | 5392 | (if longer, the rest is truncated and "..." will be appended). |
0a2ef054 | 5393 | |
a1433954 | 5394 | The C<flags> argument is as in L</pv_uni_display>(). |
0a2ef054 | 5395 | |
a1433954 | 5396 | The pointer to the PV of the C<dsv> is returned. |
d2cc3551 | 5397 | |
d4c19fe8 AL |
5398 | =cut |
5399 | */ | |
e6b2e755 JH |
5400 | char * |
5401 | Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags) | |
5402 | { | |
8cdde9f8 NC |
5403 | const char * const ptr = |
5404 | isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv); | |
5405 | ||
7918f24d NC |
5406 | PERL_ARGS_ASSERT_SV_UNI_DISPLAY; |
5407 | ||
8cdde9f8 | 5408 | return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr, |
cfd0369c | 5409 | SvCUR(ssv), pvlim, flags); |
701a277b JH |
5410 | } |
5411 | ||
d2cc3551 | 5412 | /* |
e6226b18 | 5413 | =for apidoc foldEQ_utf8 |
d2cc3551 | 5414 | |
56576a04 KW |
5415 | Returns true if the leading portions of the strings C<s1> and C<s2> (either or |
5416 | both of which may be in UTF-8) are the same case-insensitively; false | |
5417 | otherwise. How far into the strings to compare is determined by other input | |
5418 | parameters. | |
8b35872c | 5419 | |
a1433954 | 5420 | If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode; |
56576a04 KW |
5421 | otherwise it is assumed to be in native 8-bit encoding. Correspondingly for |
5422 | C<u2> with respect to C<s2>. | |
5423 | ||
5424 | If the byte length C<l1> is non-zero, it says how far into C<s1> to check for | |
5425 | fold equality. In other words, C<s1>+C<l1> will be used as a goal to reach. | |
5426 | The scan will not be considered to be a match unless the goal is reached, and | |
5427 | scanning won't continue past that goal. Correspondingly for C<l2> with respect | |
5428 | to C<s2>. | |
5429 | ||
5430 | If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that | |
5431 | pointer is considered an end pointer to the position 1 byte past the maximum | |
5432 | point in C<s1> beyond which scanning will not continue under any circumstances. | |
03bb5c85 | 5433 | (This routine assumes that UTF-8 encoded input strings are not malformed; |
56576a04 KW |
5434 | malformed input can cause it to read past C<pe1>). This means that if both |
5435 | C<l1> and C<pe1> are specified, and C<pe1> is less than C<s1>+C<l1>, the match | |
5436 | will never be successful because it can never | |
d51c1b21 | 5437 | get as far as its goal (and in fact is asserted against). Correspondingly for |
a1433954 | 5438 | C<pe2> with respect to C<s2>. |
8b35872c | 5439 | |
a1433954 KW |
5440 | At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and |
5441 | C<l2> must be non-zero), and if both do, both have to be | |
8b35872c KW |
5442 | reached for a successful match. Also, if the fold of a character is multiple |
5443 | characters, all of them must be matched (see tr21 reference below for | |
5444 | 'folding'). | |
5445 | ||
796b6530 | 5446 | Upon a successful match, if C<pe1> is non-C<NULL>, |
a1433954 KW |
5447 | it will be set to point to the beginning of the I<next> character of C<s1> |
5448 | beyond what was matched. Correspondingly for C<pe2> and C<s2>. | |
d2cc3551 JH |
5449 | |
5450 | For case-insensitiveness, the "casefolding" of Unicode is used | |
5451 | instead of upper/lowercasing both the characters, see | |
a1433954 | 5452 | L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings). |
d2cc3551 JH |
5453 | |
5454 | =cut */ | |
a33c29bc KW |
5455 | |
5456 | /* A flags parameter has been added which may change, and hence isn't | |
5457 | * externally documented. Currently it is: | |
5458 | * 0 for as-documented above | |
5459 | * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an | |
5460 | ASCII one, to not match | |
31f05a37 KW |
5461 | * FOLDEQ_LOCALE is set iff the rules from the current underlying |
5462 | * locale are to be used. | |
5463 | * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this | |
aa8ebe62 KW |
5464 | * routine. This allows that step to be skipped. |
5465 | * Currently, this requires s1 to be encoded as UTF-8 | |
5466 | * (u1 must be true), which is asserted for. | |
d635b710 KW |
5467 | * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may |
5468 | * cross certain boundaries. Hence, the caller should | |
5469 | * let this function do the folding instead of | |
5470 | * pre-folding. This code contains an assertion to | |
5471 | * that effect. However, if the caller knows what | |
5472 | * it's doing, it can pass this flag to indicate that, | |
5473 | * and the assertion is skipped. | |
31f05a37 | 5474 | * FOLDEQ_S2_ALREADY_FOLDED Similarly. |
d635b710 | 5475 | * FOLDEQ_S2_FOLDS_SANE |
a33c29bc | 5476 | */ |
701a277b | 5477 | I32 |
56576a04 KW |
5478 | Perl_foldEQ_utf8_flags(pTHX_ const char *s1, char **pe1, UV l1, bool u1, |
5479 | const char *s2, char **pe2, UV l2, bool u2, | |
5480 | U32 flags) | |
332ddc25 | 5481 | { |
eb578fdb KW |
5482 | const U8 *p1 = (const U8*)s1; /* Point to current char */ |
5483 | const U8 *p2 = (const U8*)s2; | |
5484 | const U8 *g1 = NULL; /* goal for s1 */ | |
5485 | const U8 *g2 = NULL; | |
5486 | const U8 *e1 = NULL; /* Don't scan s1 past this */ | |
5487 | U8 *f1 = NULL; /* Point to current folded */ | |
5488 | const U8 *e2 = NULL; | |
5489 | U8 *f2 = NULL; | |
48ef279e | 5490 | STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */ |
8b35872c KW |
5491 | U8 foldbuf1[UTF8_MAXBYTES_CASE+1]; |
5492 | U8 foldbuf2[UTF8_MAXBYTES_CASE+1]; | |
1d39b2cd | 5493 | U8 flags_for_folder = FOLD_FLAGS_FULL; |
8b35872c | 5494 | |
eda9cac1 | 5495 | PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS; |
8b35872c | 5496 | |
cea315b6 | 5497 | assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE)) |
d635b710 KW |
5498 | && (((flags & FOLDEQ_S1_ALREADY_FOLDED) |
5499 | && !(flags & FOLDEQ_S1_FOLDS_SANE)) | |
5500 | || ((flags & FOLDEQ_S2_ALREADY_FOLDED) | |
5501 | && !(flags & FOLDEQ_S2_FOLDS_SANE))))); | |
b08f1bd5 KW |
5502 | /* The algorithm is to trial the folds without regard to the flags on |
5503 | * the first line of the above assert(), and then see if the result | |
5504 | * violates them. This means that the inputs can't be pre-folded to a | |
5505 | * violating result, hence the assert. This could be changed, with the | |
5506 | * addition of extra tests here for the already-folded case, which would | |
5507 | * slow it down. That cost is more than any possible gain for when these | |
5508 | * flags are specified, as the flags indicate /il or /iaa matching which | |
5509 | * is less common than /iu, and I (khw) also believe that real-world /il | |
5510 | * and /iaa matches are most likely to involve code points 0-255, and this | |
5511 | * function only under rare conditions gets called for 0-255. */ | |
18f762c3 | 5512 | |
1d39b2cd KW |
5513 | if (flags & FOLDEQ_LOCALE) { |
5514 | if (IN_UTF8_CTYPE_LOCALE) { | |
5515 | flags &= ~FOLDEQ_LOCALE; | |
5516 | } | |
5517 | else { | |
5518 | flags_for_folder |= FOLD_FLAGS_LOCALE; | |
5519 | } | |
31f05a37 KW |
5520 | } |
5521 | ||
8b35872c | 5522 | if (pe1) { |
48ef279e | 5523 | e1 = *(U8**)pe1; |
8b35872c KW |
5524 | } |
5525 | ||
5526 | if (l1) { | |
48ef279e | 5527 | g1 = (const U8*)s1 + l1; |
8b35872c KW |
5528 | } |
5529 | ||
5530 | if (pe2) { | |
48ef279e | 5531 | e2 = *(U8**)pe2; |
8b35872c KW |
5532 | } |
5533 | ||
5534 | if (l2) { | |
48ef279e | 5535 | g2 = (const U8*)s2 + l2; |
8b35872c KW |
5536 | } |
5537 | ||
5538 | /* Must have at least one goal */ | |
5539 | assert(g1 || g2); | |
5540 | ||
5541 | if (g1) { | |
5542 | ||
48ef279e KW |
5543 | /* Will never match if goal is out-of-bounds */ |
5544 | assert(! e1 || e1 >= g1); | |
8b35872c | 5545 | |
48ef279e KW |
5546 | /* Here, there isn't an end pointer, or it is beyond the goal. We |
5547 | * only go as far as the goal */ | |
5548 | e1 = g1; | |
8b35872c | 5549 | } |
313b38e5 NC |
5550 | else { |
5551 | assert(e1); /* Must have an end for looking at s1 */ | |
5552 | } | |
8b35872c KW |
5553 | |
5554 | /* Same for goal for s2 */ | |
5555 | if (g2) { | |
48ef279e KW |
5556 | assert(! e2 || e2 >= g2); |
5557 | e2 = g2; | |
8b35872c | 5558 | } |
313b38e5 NC |
5559 | else { |
5560 | assert(e2); | |
5561 | } | |
8b35872c | 5562 | |
18f762c3 KW |
5563 | /* If both operands are already folded, we could just do a memEQ on the |
5564 | * whole strings at once, but it would be better if the caller realized | |
5565 | * this and didn't even call us */ | |
5566 | ||
8b35872c KW |
5567 | /* Look through both strings, a character at a time */ |
5568 | while (p1 < e1 && p2 < e2) { | |
5569 | ||
d51c1b21 | 5570 | /* If at the beginning of a new character in s1, get its fold to use |
1d39b2cd | 5571 | * and the length of the fold. */ |
48ef279e | 5572 | if (n1 == 0) { |
18f762c3 KW |
5573 | if (flags & FOLDEQ_S1_ALREADY_FOLDED) { |
5574 | f1 = (U8 *) p1; | |
aa8ebe62 | 5575 | assert(u1); |
18f762c3 | 5576 | n1 = UTF8SKIP(f1); |
18f762c3 KW |
5577 | } |
5578 | else { | |
1d39b2cd KW |
5579 | if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) { |
5580 | ||
5581 | /* We have to forbid mixing ASCII with non-ASCII if the | |
5582 | * flags so indicate. And, we can short circuit having to | |
5583 | * call the general functions for this common ASCII case, | |
5584 | * all of whose non-locale folds are also ASCII, and hence | |
5585 | * UTF-8 invariants, so the UTF8ness of the strings is not | |
5586 | * relevant. */ | |
5587 | if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) { | |
5588 | return 0; | |
5589 | } | |
5590 | n1 = 1; | |
5591 | *foldbuf1 = toFOLD(*p1); | |
5592 | } | |
5593 | else if (u1) { | |
a1a5ec35 | 5594 | _toFOLD_utf8_flags(p1, e1, foldbuf1, &n1, flags_for_folder); |
1d39b2cd | 5595 | } |
4a4088c4 | 5596 | else { /* Not UTF-8, get UTF-8 fold */ |
1d39b2cd KW |
5597 | _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder); |
5598 | } | |
5599 | f1 = foldbuf1; | |
5600 | } | |
48ef279e | 5601 | } |
8b35872c | 5602 | |
48ef279e | 5603 | if (n2 == 0) { /* Same for s2 */ |
18f762c3 KW |
5604 | if (flags & FOLDEQ_S2_ALREADY_FOLDED) { |
5605 | f2 = (U8 *) p2; | |
aa8ebe62 | 5606 | assert(u2); |
18f762c3 KW |
5607 | n2 = UTF8SKIP(f2); |
5608 | } | |
5609 | else { | |
1d39b2cd KW |
5610 | if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) { |
5611 | if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) { | |
5612 | return 0; | |
5613 | } | |
5614 | n2 = 1; | |
5615 | *foldbuf2 = toFOLD(*p2); | |
5616 | } | |
5617 | else if (u2) { | |
a1a5ec35 | 5618 | _toFOLD_utf8_flags(p2, e2, foldbuf2, &n2, flags_for_folder); |
1d39b2cd KW |
5619 | } |
5620 | else { | |
5621 | _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder); | |
5622 | } | |
5623 | f2 = foldbuf2; | |
18f762c3 | 5624 | } |
48ef279e | 5625 | } |
8b35872c | 5626 | |
5001101e | 5627 | /* Here f1 and f2 point to the beginning of the strings to compare. |
227968da | 5628 | * These strings are the folds of the next character from each input |
4a4088c4 | 5629 | * string, stored in UTF-8. */ |
5e64d0fa | 5630 | |
48ef279e KW |
5631 | /* While there is more to look for in both folds, see if they |
5632 | * continue to match */ | |
5633 | while (n1 && n2) { | |
5634 | U8 fold_length = UTF8SKIP(f1); | |
5635 | if (fold_length != UTF8SKIP(f2) | |
5636 | || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE | |
5637 | function call for single | |
a6d5f321 | 5638 | byte */ |
48ef279e KW |
5639 | || memNE((char*)f1, (char*)f2, fold_length)) |
5640 | { | |
e6226b18 | 5641 | return 0; /* mismatch */ |
48ef279e KW |
5642 | } |
5643 | ||
5644 | /* Here, they matched, advance past them */ | |
5645 | n1 -= fold_length; | |
5646 | f1 += fold_length; | |
5647 | n2 -= fold_length; | |
5648 | f2 += fold_length; | |
5649 | } | |
8b35872c | 5650 | |
48ef279e KW |
5651 | /* When reach the end of any fold, advance the input past it */ |
5652 | if (n1 == 0) { | |
5653 | p1 += u1 ? UTF8SKIP(p1) : 1; | |
5654 | } | |
5655 | if (n2 == 0) { | |
5656 | p2 += u2 ? UTF8SKIP(p2) : 1; | |
5657 | } | |
8b35872c KW |
5658 | } /* End of loop through both strings */ |
5659 | ||
5660 | /* A match is defined by each scan that specified an explicit length | |
5661 | * reaching its final goal, and the other not having matched a partial | |
5662 | * character (which can happen when the fold of a character is more than one | |
5663 | * character). */ | |
5664 | if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) { | |
e6226b18 | 5665 | return 0; |
8b35872c KW |
5666 | } |
5667 | ||
5668 | /* Successful match. Set output pointers */ | |
5669 | if (pe1) { | |
48ef279e | 5670 | *pe1 = (char*)p1; |
8b35872c KW |
5671 | } |
5672 | if (pe2) { | |
48ef279e | 5673 | *pe2 = (char*)p2; |
8b35872c | 5674 | } |
e6226b18 | 5675 | return 1; |
e6b2e755 | 5676 | } |
701a277b | 5677 | |
f2645549 | 5678 | /* XXX The next two functions should likely be moved to mathoms.c once all |
37e7596b KW |
5679 | * occurrences of them are removed from the core; some cpan-upstream modules |
5680 | * still use them */ | |
5681 | ||
5682 | U8 * | |
5683 | Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv) | |
5684 | { | |
5685 | PERL_ARGS_ASSERT_UVUNI_TO_UTF8; | |
5686 | ||
33f38593 | 5687 | return uvoffuni_to_utf8_flags(d, uv, 0); |
37e7596b KW |
5688 | } |
5689 | ||
e505af10 KW |
5690 | /* |
5691 | =for apidoc utf8n_to_uvuni | |
5692 | ||
5693 | Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>. | |
5694 | ||
5695 | This function was useful for code that wanted to handle both EBCDIC and | |
5696 | ASCII platforms with Unicode properties, but starting in Perl v5.20, the | |
5697 | distinctions between the platforms have mostly been made invisible to most | |
5698 | code, so this function is quite unlikely to be what you want. If you do need | |
5699 | this precise functionality, use instead | |
5700 | C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> | |
5701 | or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>. | |
5702 | ||
5703 | =cut | |
5704 | */ | |
5705 | ||
37e7596b KW |
5706 | UV |
5707 | Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags) | |
5708 | { | |
5709 | PERL_ARGS_ASSERT_UTF8N_TO_UVUNI; | |
5710 | ||
5711 | return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags)); | |
5712 | } | |
5713 | ||
5714 | /* | |
5715 | =for apidoc uvuni_to_utf8_flags | |
5716 | ||
5717 | Instead you almost certainly want to use L</uvchr_to_utf8> or | |
efa9cd84 | 5718 | L</uvchr_to_utf8_flags>. |
37e7596b KW |
5719 | |
5720 | This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>, | |
5721 | which itself, while not deprecated, should be used only in isolated | |
5722 | circumstances. These functions were useful for code that wanted to handle | |
5723 | both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl | |
5724 | v5.20, the distinctions between the platforms have mostly been made invisible | |
5725 | to most code, so this function is quite unlikely to be what you want. | |
5726 | ||
5727 | =cut | |
5728 | */ | |
5729 | ||
5730 | U8 * | |
5731 | Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags) | |
5732 | { | |
5733 | PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS; | |
5734 | ||
5735 | return uvoffuni_to_utf8_flags(d, uv, flags); | |
5736 | } | |
5737 | ||
5738 | /* | |
7723e007 KW |
5739 | =for apidoc utf8_to_uvchr |
5740 | ||
5741 | Returns the native code point of the first character in the string C<s> | |
5742 | which is assumed to be in UTF-8 encoding; C<retlen> will be set to the | |
5743 | length, in bytes, of that character. | |
5744 | ||
5745 | Some, but not all, UTF-8 malformations are detected, and in fact, some | |
5746 | malformed input could cause reading beyond the end of the input buffer, which | |
5747 | is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead. | |
5748 | ||
5749 | If C<s> points to one of the detected malformations, and UTF8 warnings are | |
5750 | enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't | |
5751 | C<NULL>) to -1. If those warnings are off, the computed value if well-defined (or | |
5752 | the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen> | |
5753 | is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the | |
5754 | next possible position in C<s> that could begin a non-malformed character. | |
5755 | See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned. | |
5756 | ||
5757 | =cut | |
5758 | */ | |
5759 | ||
5760 | UV | |
5761 | Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen) | |
5762 | { | |
5763 | PERL_ARGS_ASSERT_UTF8_TO_UVCHR; | |
5764 | ||
aa3c16bd KW |
5765 | /* This function is unsafe if malformed UTF-8 input is given it, which is |
5766 | * why the function is deprecated. If the first byte of the input | |
5767 | * indicates that there are more bytes remaining in the sequence that forms | |
5768 | * the character than there are in the input buffer, it can read past the | |
5769 | * end. But we can make it safe if the input string happens to be | |
5770 | * NUL-terminated, as many strings in Perl are, by refusing to read past a | |
5771 | * NUL. A NUL indicates the start of the next character anyway. If the | |
5772 | * input isn't NUL-terminated, the function remains unsafe, as it always | |
5773 | * has been. | |
5774 | * | |
5775 | * An initial NUL has to be handled separately, but all ASCIIs can be | |
5776 | * handled the same way, speeding up this common case */ | |
5777 | ||
5778 | if (UTF8_IS_INVARIANT(*s)) { /* Assumes 's' contains at least 1 byte */ | |
5779 | return (UV) *s; | |
5780 | } | |
5781 | ||
5782 | return utf8_to_uvchr_buf(s, | |
2951abb4 KW |
5783 | s + my_strnlen((char *) s, UTF8SKIP(s)), |
5784 | retlen); | |
7723e007 KW |
5785 | } |
5786 | ||
5787 | /* | |
14d04a33 | 5788 | * ex: set ts=8 sts=4 sw=4 et: |
37442d52 | 5789 | */ |