]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Lexer for source files, ToNumber() string conversions, RegExp expressions, | |
3 | * and JSON. | |
4 | * | |
5 | * Provides a stream of Ecmascript tokens from an UTF-8/CESU-8 buffer. The | |
6 | * caller can also rewind the token stream into a certain position which is | |
7 | * needed by the compiler part for multi-pass scanning. Tokens are | |
8 | * represented as duk_token structures, and contain line number information. | |
9 | * Token types are identified with DUK_TOK_* defines. | |
10 | * | |
11 | * Characters are decoded into a fixed size lookup window consisting of | |
12 | * decoded Unicode code points, with window positions past the end of the | |
13 | * input filled with an invalid codepoint (-1). The tokenizer can thus | |
14 | * perform multiple character lookups efficiently and with few sanity | |
15 | * checks (such as access outside the end of the input), which keeps the | |
16 | * tokenization code small at the cost of performance. | |
17 | * | |
18 | * Character data in tokens, such as identifier names and string literals, | |
19 | * is encoded into CESU-8 format on-the-fly while parsing the token in | |
20 | * question. The string data is made reachable to garbage collection by | |
21 | * placing the token-related values in value stack entries allocated for | |
22 | * this purpose by the caller. The characters exist in Unicode code point | |
23 | * form only in the fixed size lookup window, which keeps character data | |
24 | * expansion (of especially ASCII data) low. | |
25 | * | |
26 | * Token parsing supports the full range of Unicode characters as described | |
27 | * in the E5 specification. Parsing has been optimized for ASCII characters | |
28 | * because ordinary Ecmascript code consists almost entirely of ASCII | |
29 | * characters. Matching of complex Unicode codepoint sets (such as in the | |
30 | * IdentifierStart and IdentifierPart productions) is optimized for size, | |
31 | * and is done using a linear scan of a bit-packed list of ranges. This is | |
32 | * very slow, but should never be entered unless the source code actually | |
33 | * contains Unicode characters. | |
34 | * | |
35 | * Ecmascript tokenization is partially context sensitive. First, | |
36 | * additional future reserved words are recognized in strict mode (see E5 | |
37 | * Section 7.6.1.2). Second, a forward slash character ('/') can be | |
38 | * recognized either as starting a RegExp literal or as a division operator, | |
39 | * depending on context. The caller must provide necessary context flags | |
40 | * when requesting a new token. | |
41 | * | |
42 | * Future work: | |
43 | * | |
44 | * * Make line number tracking optional, as it consumes space. | |
45 | * | |
46 | * * Add a feature flag for disabling UTF-8 decoding of input, as most | |
47 | * source code is ASCII. Because of Unicode escapes written in ASCII, | |
48 | * this does not allow Unicode support to be removed from e.g. | |
49 | * duk_unicode_is_identifier_start() nor does it allow removal of CESU-8 | |
50 | * encoding of e.g. string literals. | |
51 | * | |
52 | * * Add a feature flag for disabling Unicode compliance of e.g. identifier | |
53 | * names. This allows for a build more than a kilobyte smaller, because | |
54 | * Unicode ranges needed by duk_unicode_is_identifier_start() and | |
55 | * duk_unicode_is_identifier_part() can be dropped. String literals | |
56 | * should still be allowed to contain escaped Unicode, so this still does | |
57 | * not allow removal of CESU-8 encoding of e.g. string literals. | |
58 | * | |
59 | * * Character lookup tables for codepoints above BMP could be stripped. | |
60 | * | |
61 | * * Strictly speaking, E5 specification requires that source code consists | |
62 | * of 16-bit code units, and if not, must be conceptually converted to | |
63 | * that format first. The current lexer processes Unicode code points | |
64 | * and allows characters outside the BMP. These should be converted to | |
65 | * surrogate pairs while reading the source characters into the window, | |
66 | * not after tokens have been formed (as is done now). However, the fix | |
67 | * is not trivial because two characters are decoded from one codepoint. | |
68 | * | |
69 | * * Optimize for speed as well as size. Large if-else ladders are (at | |
70 | * least potentially) slow. | |
71 | */ | |
72 | ||
73 | #include "duk_internal.h" | |
74 | ||
75 | /* | |
76 | * Various defines and file specific helper macros | |
77 | */ | |
78 | ||
79 | #define DUK__MAX_RE_DECESC_DIGITS 9 | |
80 | #define DUK__MAX_RE_QUANT_DIGITS 9 /* Does not allow e.g. 2**31-1, but one more would allow overflows of u32. */ | |
81 | ||
82 | /* whether to use macros or helper function depends on call count */ | |
83 | #define DUK__ISDIGIT(x) ((x) >= DUK_ASC_0 && (x) <= DUK_ASC_9) | |
84 | #define DUK__ISHEXDIGIT(x) duk__is_hex_digit((x)) | |
85 | #define DUK__ISOCTDIGIT(x) ((x) >= DUK_ASC_0 && (x) <= DUK_ASC_7) | |
86 | #define DUK__ISDIGIT03(x) ((x) >= DUK_ASC_0 && (x) <= DUK_ASC_3) | |
87 | #define DUK__ISDIGIT47(x) ((x) >= DUK_ASC_4 && (x) <= DUK_ASC_7) | |
88 | ||
89 | /* lexer character window helpers */ | |
90 | #define DUK__LOOKUP(lex_ctx,index) ((lex_ctx)->window[(index)].codepoint) | |
91 | #define DUK__ADVANCECHARS(lex_ctx,count) duk__advance_bytes((lex_ctx), (count) * sizeof(duk_lexer_codepoint)) | |
92 | #define DUK__ADVANCEBYTES(lex_ctx,count) duk__advance_bytes((lex_ctx), (count)) | |
93 | #define DUK__INITBUFFER(lex_ctx) duk__initbuffer((lex_ctx)) | |
94 | #define DUK__APPENDBUFFER(lex_ctx,x) duk__appendbuffer((lex_ctx), (duk_codepoint_t) (x)) | |
95 | ||
96 | /* lookup shorthands (note: assume context variable is named 'lex_ctx') */ | |
97 | #define DUK__L0() DUK__LOOKUP(lex_ctx, 0) | |
98 | #define DUK__L1() DUK__LOOKUP(lex_ctx, 1) | |
99 | #define DUK__L2() DUK__LOOKUP(lex_ctx, 2) | |
100 | #define DUK__L3() DUK__LOOKUP(lex_ctx, 3) | |
101 | #define DUK__L4() DUK__LOOKUP(lex_ctx, 4) | |
102 | #define DUK__L5() DUK__LOOKUP(lex_ctx, 5) | |
103 | ||
104 | /* packed advance/token number macro used by multiple functions */ | |
105 | #define DUK__ADVTOK(advbytes,tok) ((((advbytes) * sizeof(duk_lexer_codepoint)) << 8) + (tok)) | |
106 | ||
107 | /* | |
108 | * Advance lookup window by N characters, filling in new characters as | |
109 | * necessary. After returning caller is guaranteed a character window of | |
110 | * at least DUK_LEXER_WINDOW_SIZE characters. | |
111 | * | |
112 | * The main function duk__advance_bytes() is called at least once per every | |
113 | * token so it has a major lexer/compiler performance impact. There are two | |
114 | * variants for the main duk__advance_bytes() algorithm: a sliding window | |
115 | * approach which is slightly faster at the cost of larger code footprint, | |
116 | * and a simple copying one. | |
117 | * | |
118 | * Decoding directly from the source string would be another lexing option. | |
119 | * But the lookup window based approach has the advantage of hiding the | |
120 | * source string and its encoding effectively which gives more flexibility | |
121 | * going forward to e.g. support chunked streaming of source from flash. | |
122 | * | |
123 | * Decodes UTF-8/CESU-8 leniently with support for code points from U+0000 to | |
124 | * U+10FFFF, causing an error if the input is unparseable. Leniency means: | |
125 | * | |
126 | * * Unicode code point validation is intentionally not performed, | |
127 | * except to check that the codepoint does not exceed 0x10ffff. | |
128 | * | |
129 | * * In particular, surrogate pairs are allowed and not combined, which | |
130 | * allows source files to represent all SourceCharacters with CESU-8. | |
131 | * Broken surrogate pairs are allowed, as Ecmascript does not mandate | |
132 | * their validation. | |
133 | * | |
134 | * * Allow non-shortest UTF-8 encodings. | |
135 | * | |
136 | * Leniency here causes few security concerns because all character data is | |
137 | * decoded into Unicode codepoints before lexer processing, and is then | |
138 | * re-encoded into CESU-8. The source can be parsed as strict UTF-8 with | |
139 | * a compiler option. However, Ecmascript source characters include -all- | |
140 | * 16-bit unsigned integer codepoints, so leniency seems to be appropriate. | |
141 | * | |
142 | * Note that codepoints above the BMP are not strictly SourceCharacters, | |
143 | * but the lexer still accepts them as such. Before ending up in a string | |
144 | * or an identifier name, codepoints above BMP are converted into surrogate | |
145 | * pairs and then CESU-8 encoded, resulting in 16-bit Unicode data as | |
146 | * expected by Ecmascript. | |
147 | * | |
148 | * An alternative approach to dealing with invalid or partial sequences | |
149 | * would be to skip them and replace them with e.g. the Unicode replacement | |
150 | * character U+FFFD. This has limited utility because a replacement character | |
151 | * will most likely cause a parse error, unless it occurs inside a string. | |
152 | * Further, Ecmascript source is typically pure ASCII. | |
153 | * | |
154 | * See: | |
155 | * | |
156 | * http://en.wikipedia.org/wiki/UTF-8 | |
157 | * http://en.wikipedia.org/wiki/CESU-8 | |
158 | * http://tools.ietf.org/html/rfc3629 | |
159 | * http://en.wikipedia.org/wiki/UTF-8#Invalid_byte_sequences | |
160 | * | |
161 | * Future work: | |
162 | * | |
163 | * * Reject other invalid Unicode sequences (see Wikipedia entry for examples) | |
164 | * in strict UTF-8 mode. | |
165 | * | |
166 | * * Size optimize. An attempt to use a 16-byte lookup table for the first | |
167 | * byte resulted in a code increase though. | |
168 | * | |
169 | * * Is checking against maximum 0x10ffff really useful? 4-byte encoding | |
170 | * imposes a certain limit anyway. | |
171 | * | |
172 | * * Support chunked streaming of source code. Can be implemented either | |
173 | * by streaming chunks of bytes or chunks of codepoints. | |
174 | */ | |
175 | ||
176 | #if defined(DUK_USE_LEXER_SLIDING_WINDOW) | |
177 | DUK_LOCAL void duk__fill_lexer_buffer(duk_lexer_ctx *lex_ctx, duk_small_uint_t start_offset_bytes) { | |
178 | duk_lexer_codepoint *cp, *cp_end; | |
179 | duk_ucodepoint_t x; | |
180 | duk_small_uint_t contlen; | |
181 | const duk_uint8_t *p, *p_end; | |
182 | #if defined(DUK_USE_STRICT_UTF8_SOURCE) | |
183 | duk_ucodepoint_t mincp; | |
184 | #endif | |
185 | duk_int_t input_line; | |
186 | ||
187 | /* Use temporaries and update lex_ctx only when finished. */ | |
188 | input_line = lex_ctx->input_line; | |
189 | p = lex_ctx->input + lex_ctx->input_offset; | |
190 | p_end = lex_ctx->input + lex_ctx->input_length; | |
191 | ||
192 | cp = (duk_lexer_codepoint *) (void *) ((duk_uint8_t *) lex_ctx->buffer + start_offset_bytes); | |
193 | cp_end = lex_ctx->buffer + DUK_LEXER_BUFFER_SIZE; | |
194 | ||
195 | for (; cp != cp_end; cp++) { | |
196 | cp->offset = (duk_size_t) (p - lex_ctx->input); | |
197 | cp->line = input_line; | |
198 | ||
199 | /* XXX: potential issue with signed pointers, p_end < p. */ | |
200 | if (DUK_UNLIKELY(p >= p_end)) { | |
201 | /* If input_offset were assigned a negative value, it would | |
202 | * result in a large positive value. Most likely it would be | |
203 | * larger than input_length and be caught here. In any case | |
204 | * no memory unsafe behavior would happen. | |
205 | */ | |
206 | cp->codepoint = -1; | |
207 | continue; | |
208 | } | |
209 | ||
210 | x = (duk_ucodepoint_t) (*p++); | |
211 | ||
212 | /* Fast path. */ | |
213 | ||
214 | if (DUK_LIKELY(x < 0x80UL)) { | |
215 | DUK_ASSERT(x != 0x2028UL && x != 0x2029UL); /* not LS/PS */ | |
216 | if (DUK_UNLIKELY(x <= 0x000dUL)) { | |
217 | if ((x == 0x000aUL) || | |
218 | ((x == 0x000dUL) && (p >= p_end || *p != 0x000aUL))) { | |
219 | /* lookup for 0x000a above assumes shortest encoding now */ | |
220 | ||
221 | /* E5 Section 7.3, treat the following as newlines: | |
222 | * LF | |
223 | * CR [not followed by LF] | |
224 | * LS | |
225 | * PS | |
226 | * | |
227 | * For CR LF, CR is ignored if it is followed by LF, and the LF will bump | |
228 | * the line number. | |
229 | */ | |
230 | input_line++; | |
231 | } | |
232 | } | |
233 | ||
234 | cp->codepoint = (duk_codepoint_t) x; | |
235 | continue; | |
236 | } | |
237 | ||
238 | /* Slow path. */ | |
239 | ||
240 | if (x < 0xc0UL) { | |
241 | /* 10xx xxxx -> invalid */ | |
242 | goto error_encoding; | |
243 | } else if (x < 0xe0UL) { | |
244 | /* 110x xxxx 10xx xxxx */ | |
245 | contlen = 1; | |
246 | #if defined(DUK_USE_STRICT_UTF8_SOURCE) | |
247 | mincp = 0x80UL; | |
248 | #endif | |
249 | x = x & 0x1fUL; | |
250 | } else if (x < 0xf0UL) { | |
251 | /* 1110 xxxx 10xx xxxx 10xx xxxx */ | |
252 | contlen = 2; | |
253 | #if defined(DUK_USE_STRICT_UTF8_SOURCE) | |
254 | mincp = 0x800UL; | |
255 | #endif | |
256 | x = x & 0x0fUL; | |
257 | } else if (x < 0xf8UL) { | |
258 | /* 1111 0xxx 10xx xxxx 10xx xxxx 10xx xxxx */ | |
259 | contlen = 3; | |
260 | #if defined(DUK_USE_STRICT_UTF8_SOURCE) | |
261 | mincp = 0x10000UL; | |
262 | #endif | |
263 | x = x & 0x07UL; | |
264 | } else { | |
265 | /* no point in supporting encodings of 5 or more bytes */ | |
266 | goto error_encoding; | |
267 | } | |
268 | ||
269 | DUK_ASSERT(p_end >= p); | |
270 | if ((duk_size_t) contlen > (duk_size_t) (p_end - p)) { | |
271 | goto error_clipped; | |
272 | } | |
273 | ||
274 | while (contlen > 0) { | |
275 | duk_small_uint_t y; | |
276 | y = *p++; | |
277 | if ((y & 0xc0U) != 0x80U) { | |
278 | /* check that byte has the form 10xx xxxx */ | |
279 | goto error_encoding; | |
280 | } | |
281 | x = x << 6; | |
282 | x += y & 0x3fUL; | |
283 | contlen--; | |
284 | } | |
285 | ||
286 | /* check final character validity */ | |
287 | ||
288 | if (x > 0x10ffffUL) { | |
289 | goto error_encoding; | |
290 | } | |
291 | #if defined(DUK_USE_STRICT_UTF8_SOURCE) | |
292 | if (x < mincp || (x >= 0xd800UL && x <= 0xdfffUL) || x == 0xfffeUL) { | |
293 | goto error_encoding; | |
294 | } | |
295 | #endif | |
296 | ||
297 | DUK_ASSERT(x != 0x000aUL && x != 0x000dUL); | |
298 | if ((x == 0x2028UL) || (x == 0x2029UL)) { | |
299 | input_line++; | |
300 | } | |
301 | ||
302 | cp->codepoint = (duk_codepoint_t) x; | |
303 | } | |
304 | ||
305 | lex_ctx->input_offset = (duk_size_t) (p - lex_ctx->input); | |
306 | lex_ctx->input_line = input_line; | |
307 | return; | |
308 | ||
309 | error_clipped: /* clipped codepoint */ | |
310 | error_encoding: /* invalid codepoint encoding or codepoint */ | |
311 | lex_ctx->input_offset = (duk_size_t) (p - lex_ctx->input); | |
312 | lex_ctx->input_line = input_line; | |
313 | ||
314 | DUK_ERROR_SYNTAX(lex_ctx->thr, "utf-8 decode failed"); | |
315 | } | |
316 | ||
317 | DUK_LOCAL void duk__advance_bytes(duk_lexer_ctx *lex_ctx, duk_small_uint_t count_bytes) { | |
318 | duk_small_uint_t used_bytes, avail_bytes; | |
319 | ||
320 | DUK_ASSERT_DISABLE(count_bytes >= 0); /* unsigned */ | |
321 | DUK_ASSERT(count_bytes <= (duk_small_uint_t) (DUK_LEXER_WINDOW_SIZE * sizeof(duk_lexer_codepoint))); | |
322 | DUK_ASSERT(lex_ctx->window >= lex_ctx->buffer); | |
323 | DUK_ASSERT(lex_ctx->window < lex_ctx->buffer + DUK_LEXER_BUFFER_SIZE); | |
324 | DUK_ASSERT((duk_uint8_t *) lex_ctx->window + count_bytes <= (duk_uint8_t *) lex_ctx->buffer + DUK_LEXER_BUFFER_SIZE * sizeof(duk_lexer_codepoint)); | |
325 | ||
326 | /* Zero 'count' is also allowed to make call sites easier. | |
327 | * Arithmetic in bytes generates better code in GCC. | |
328 | */ | |
329 | ||
330 | lex_ctx->window = (duk_lexer_codepoint *) (void *) ((duk_uint8_t *) lex_ctx->window + count_bytes); /* avoid multiply */ | |
331 | used_bytes = (duk_small_uint_t) ((duk_uint8_t *) lex_ctx->window - (duk_uint8_t *) lex_ctx->buffer); | |
332 | avail_bytes = DUK_LEXER_BUFFER_SIZE * sizeof(duk_lexer_codepoint) - used_bytes; | |
333 | if (avail_bytes < (duk_small_uint_t) (DUK_LEXER_WINDOW_SIZE * sizeof(duk_lexer_codepoint))) { | |
334 | /* Not enough data to provide a full window, so "scroll" window to | |
335 | * start of buffer and fill up the rest. | |
336 | */ | |
337 | DUK_MEMMOVE((void *) lex_ctx->buffer, | |
338 | (const void *) lex_ctx->window, | |
339 | (size_t) avail_bytes); | |
340 | lex_ctx->window = lex_ctx->buffer; | |
341 | duk__fill_lexer_buffer(lex_ctx, avail_bytes); | |
342 | } | |
343 | } | |
344 | ||
345 | DUK_LOCAL void duk__init_lexer_window(duk_lexer_ctx *lex_ctx) { | |
346 | lex_ctx->window = lex_ctx->buffer; | |
347 | duk__fill_lexer_buffer(lex_ctx, 0); | |
348 | } | |
349 | #else /* DUK_USE_LEXER_SLIDING_WINDOW */ | |
350 | DUK_LOCAL duk_codepoint_t duk__read_char(duk_lexer_ctx *lex_ctx) { | |
351 | duk_ucodepoint_t x; | |
352 | duk_small_uint_t len; | |
353 | duk_small_uint_t i; | |
354 | const duk_uint8_t *p; | |
355 | #if defined(DUK_USE_STRICT_UTF8_SOURCE) | |
356 | duk_ucodepoint_t mincp; | |
357 | #endif | |
358 | duk_size_t input_offset; | |
359 | ||
360 | input_offset = lex_ctx->input_offset; | |
361 | if (DUK_UNLIKELY(input_offset >= lex_ctx->input_length)) { | |
362 | /* If input_offset were assigned a negative value, it would | |
363 | * result in a large positive value. Most likely it would be | |
364 | * larger than input_length and be caught here. In any case | |
365 | * no memory unsafe behavior would happen. | |
366 | */ | |
367 | return -1; | |
368 | } | |
369 | ||
370 | p = lex_ctx->input + input_offset; | |
371 | x = (duk_ucodepoint_t) (*p); | |
372 | ||
373 | if (DUK_LIKELY(x < 0x80UL)) { | |
374 | /* 0xxx xxxx -> fast path */ | |
375 | ||
376 | /* input offset tracking */ | |
377 | lex_ctx->input_offset++; | |
378 | ||
379 | DUK_ASSERT(x != 0x2028UL && x != 0x2029UL); /* not LS/PS */ | |
380 | if (DUK_UNLIKELY(x <= 0x000dUL)) { | |
381 | if ((x == 0x000aUL) || | |
382 | ((x == 0x000dUL) && (lex_ctx->input_offset >= lex_ctx->input_length || | |
383 | lex_ctx->input[lex_ctx->input_offset] != 0x000aUL))) { | |
384 | /* lookup for 0x000a above assumes shortest encoding now */ | |
385 | ||
386 | /* E5 Section 7.3, treat the following as newlines: | |
387 | * LF | |
388 | * CR [not followed by LF] | |
389 | * LS | |
390 | * PS | |
391 | * | |
392 | * For CR LF, CR is ignored if it is followed by LF, and the LF will bump | |
393 | * the line number. | |
394 | */ | |
395 | lex_ctx->input_line++; | |
396 | } | |
397 | } | |
398 | ||
399 | return (duk_codepoint_t) x; | |
400 | } | |
401 | ||
402 | /* Slow path. */ | |
403 | ||
404 | if (x < 0xc0UL) { | |
405 | /* 10xx xxxx -> invalid */ | |
406 | goto error_encoding; | |
407 | } else if (x < 0xe0UL) { | |
408 | /* 110x xxxx 10xx xxxx */ | |
409 | len = 2; | |
410 | #if defined(DUK_USE_STRICT_UTF8_SOURCE) | |
411 | mincp = 0x80UL; | |
412 | #endif | |
413 | x = x & 0x1fUL; | |
414 | } else if (x < 0xf0UL) { | |
415 | /* 1110 xxxx 10xx xxxx 10xx xxxx */ | |
416 | len = 3; | |
417 | #if defined(DUK_USE_STRICT_UTF8_SOURCE) | |
418 | mincp = 0x800UL; | |
419 | #endif | |
420 | x = x & 0x0fUL; | |
421 | } else if (x < 0xf8UL) { | |
422 | /* 1111 0xxx 10xx xxxx 10xx xxxx 10xx xxxx */ | |
423 | len = 4; | |
424 | #if defined(DUK_USE_STRICT_UTF8_SOURCE) | |
425 | mincp = 0x10000UL; | |
426 | #endif | |
427 | x = x & 0x07UL; | |
428 | } else { | |
429 | /* no point in supporting encodings of 5 or more bytes */ | |
430 | goto error_encoding; | |
431 | } | |
432 | ||
433 | DUK_ASSERT(lex_ctx->input_length >= lex_ctx->input_offset); | |
434 | if ((duk_size_t) len > (duk_size_t) (lex_ctx->input_length - lex_ctx->input_offset)) { | |
435 | goto error_clipped; | |
436 | } | |
437 | ||
438 | p++; | |
439 | for (i = 1; i < len; i++) { | |
440 | duk_small_uint_t y; | |
441 | y = *p++; | |
442 | if ((y & 0xc0U) != 0x80U) { | |
443 | /* check that byte has the form 10xx xxxx */ | |
444 | goto error_encoding; | |
445 | } | |
446 | x = x << 6; | |
447 | x += y & 0x3fUL; | |
448 | } | |
449 | ||
450 | /* check final character validity */ | |
451 | ||
452 | if (x > 0x10ffffUL) { | |
453 | goto error_encoding; | |
454 | } | |
455 | #if defined(DUK_USE_STRICT_UTF8_SOURCE) | |
456 | if (x < mincp || (x >= 0xd800UL && x <= 0xdfffUL) || x == 0xfffeUL) { | |
457 | goto error_encoding; | |
458 | } | |
459 | #endif | |
460 | ||
461 | /* input offset tracking */ | |
462 | lex_ctx->input_offset += len; | |
463 | ||
464 | /* line tracking */ | |
465 | DUK_ASSERT(x != 0x000aUL && x != 0x000dUL); | |
466 | if ((x == 0x2028UL) || (x == 0x2029UL)) { | |
467 | lex_ctx->input_line++; | |
468 | } | |
469 | ||
470 | return (duk_codepoint_t) x; | |
471 | ||
472 | error_clipped: /* clipped codepoint */ | |
473 | error_encoding: /* invalid codepoint encoding or codepoint */ | |
474 | DUK_ERROR_SYNTAX(lex_ctx->thr, "utf-8 decode failed"); | |
475 | return 0; | |
476 | } | |
477 | ||
478 | DUK_LOCAL void duk__advance_bytes(duk_lexer_ctx *lex_ctx, duk_small_uint_t count_bytes) { | |
479 | duk_small_uint_t keep_bytes; | |
480 | duk_lexer_codepoint *cp, *cp_end; | |
481 | ||
482 | DUK_ASSERT_DISABLE(count_bytes >= 0); /* unsigned */ | |
483 | DUK_ASSERT(count_bytes <= (duk_small_uint_t) (DUK_LEXER_WINDOW_SIZE * sizeof(duk_lexer_codepoint))); | |
484 | ||
485 | /* Zero 'count' is also allowed to make call sites easier. */ | |
486 | ||
487 | keep_bytes = DUK_LEXER_WINDOW_SIZE * sizeof(duk_lexer_codepoint) - count_bytes; | |
488 | DUK_MEMMOVE((void *) lex_ctx->window, | |
489 | (const void *) ((duk_uint8_t *) lex_ctx->window + count_bytes), | |
490 | (size_t) keep_bytes); | |
491 | ||
492 | cp = (duk_lexer_codepoint *) ((duk_uint8_t *) lex_ctx->window + keep_bytes); | |
493 | cp_end = lex_ctx->window + DUK_LEXER_WINDOW_SIZE; | |
494 | for (; cp != cp_end; cp++) { | |
495 | cp->offset = lex_ctx->input_offset; | |
496 | cp->line = lex_ctx->input_line; | |
497 | cp->codepoint = duk__read_char(lex_ctx); | |
498 | } | |
499 | } | |
500 | ||
501 | DUK_LOCAL void duk__init_lexer_window(duk_lexer_ctx *lex_ctx) { | |
502 | /* Call with count == DUK_LEXER_WINDOW_SIZE to fill buffer initially. */ | |
503 | duk__advance_bytes(lex_ctx, DUK_LEXER_WINDOW_SIZE * sizeof(duk_lexer_codepoint)); /* fill window */ | |
504 | } | |
505 | #endif /* DUK_USE_LEXER_SLIDING_WINDOW */ | |
506 | ||
507 | /* | |
508 | * (Re)initialize the temporary byte buffer. May be called extra times | |
509 | * with little impact. | |
510 | */ | |
511 | ||
512 | DUK_LOCAL void duk__initbuffer(duk_lexer_ctx *lex_ctx) { | |
513 | /* Reuse buffer as is unless buffer has grown large. */ | |
514 | if (DUK_HBUFFER_DYNAMIC_GET_SIZE(lex_ctx->buf) < DUK_LEXER_TEMP_BUF_LIMIT) { | |
515 | /* Keep current size */ | |
516 | } else { | |
517 | duk_hbuffer_resize(lex_ctx->thr, lex_ctx->buf, DUK_LEXER_TEMP_BUF_LIMIT); | |
518 | } | |
519 | ||
520 | DUK_BW_INIT_WITHBUF(lex_ctx->thr, &lex_ctx->bw, lex_ctx->buf); | |
521 | } | |
522 | ||
523 | /* | |
524 | * Append a Unicode codepoint to the temporary byte buffer. Performs | |
525 | * CESU-8 surrogate pair encoding for codepoints above the BMP. | |
526 | * Existing surrogate pairs are allowed and also encoded into CESU-8. | |
527 | */ | |
528 | ||
529 | DUK_LOCAL void duk__appendbuffer(duk_lexer_ctx *lex_ctx, duk_codepoint_t x) { | |
530 | /* | |
531 | * Since character data is only generated by decoding the source or by | |
532 | * the compiler itself, we rely on the input codepoints being correct | |
533 | * and avoid a check here. | |
534 | * | |
535 | * Character data can also come here through decoding of Unicode | |
536 | * escapes ("\udead\ubeef") so all 16-but unsigned values can be | |
537 | * present, even when the source file itself is strict UTF-8. | |
538 | */ | |
539 | ||
540 | DUK_ASSERT(x >= 0 && x <= 0x10ffff); | |
541 | ||
542 | DUK_BW_WRITE_ENSURE_CESU8(lex_ctx->thr, &lex_ctx->bw, (duk_ucodepoint_t) x); | |
543 | } | |
544 | ||
545 | /* | |
546 | * Intern the temporary byte buffer into a valstack slot | |
547 | * (in practice, slot1 or slot2). | |
548 | */ | |
549 | ||
550 | DUK_LOCAL void duk__internbuffer(duk_lexer_ctx *lex_ctx, duk_idx_t valstack_idx) { | |
551 | duk_context *ctx = (duk_context *) lex_ctx->thr; | |
552 | ||
553 | DUK_ASSERT(valstack_idx == lex_ctx->slot1_idx || valstack_idx == lex_ctx->slot2_idx); | |
554 | ||
555 | DUK_BW_PUSH_AS_STRING(lex_ctx->thr, &lex_ctx->bw); | |
556 | duk_replace(ctx, valstack_idx); | |
557 | } | |
558 | ||
559 | /* | |
560 | * Init lexer context | |
561 | */ | |
562 | ||
563 | DUK_INTERNAL void duk_lexer_initctx(duk_lexer_ctx *lex_ctx) { | |
564 | DUK_ASSERT(lex_ctx != NULL); | |
565 | ||
566 | DUK_MEMZERO(lex_ctx, sizeof(*lex_ctx)); | |
567 | #if defined(DUK_USE_EXPLICIT_NULL_INIT) | |
568 | #if defined(DUK_USE_LEXER_SLIDING_WINDOW) | |
569 | lex_ctx->window = NULL; | |
570 | #endif | |
571 | lex_ctx->thr = NULL; | |
572 | lex_ctx->input = NULL; | |
573 | lex_ctx->buf = NULL; | |
574 | #endif | |
575 | } | |
576 | ||
577 | /* | |
578 | * Set lexer input position and reinitialize lookup window. | |
579 | */ | |
580 | ||
581 | /* NB: duk_lexer_getpoint() is a macro only */ | |
582 | ||
583 | DUK_INTERNAL void duk_lexer_setpoint(duk_lexer_ctx *lex_ctx, duk_lexer_point *pt) { | |
584 | DUK_ASSERT_DISABLE(pt->offset >= 0); /* unsigned */ | |
585 | DUK_ASSERT(pt->line >= 1); | |
586 | lex_ctx->input_offset = pt->offset; | |
587 | lex_ctx->input_line = pt->line; | |
588 | duk__init_lexer_window(lex_ctx); | |
589 | } | |
590 | ||
591 | /* | |
592 | * Lexing helpers | |
593 | */ | |
594 | ||
595 | /* numeric value of a hex digit (also covers octal and decimal digits) */ | |
596 | DUK_LOCAL duk_codepoint_t duk__hexval(duk_lexer_ctx *lex_ctx, duk_codepoint_t x) { | |
597 | duk_small_int_t t; | |
598 | ||
599 | /* Here 'x' is a Unicode codepoint */ | |
600 | if (DUK_LIKELY(x >= 0 && x <= 0xff)) { | |
601 | t = duk_hex_dectab[x]; | |
602 | if (DUK_LIKELY(t >= 0)) { | |
603 | return t; | |
604 | } | |
605 | } | |
606 | ||
607 | /* Throwing an error this deep makes the error rather vague, but | |
608 | * saves hundreds of bytes of code. | |
609 | */ | |
610 | DUK_ERROR_SYNTAX(lex_ctx->thr, "decode error"); | |
611 | return 0; | |
612 | } | |
613 | ||
614 | /* having this as a separate function provided a size benefit */ | |
615 | DUK_LOCAL duk_bool_t duk__is_hex_digit(duk_codepoint_t x) { | |
616 | if (DUK_LIKELY(x >= 0 && x <= 0xff)) { | |
617 | return (duk_hex_dectab[x] >= 0); | |
618 | } | |
619 | return 0; | |
620 | } | |
621 | ||
622 | DUK_LOCAL duk_codepoint_t duk__decode_hexesc_from_window(duk_lexer_ctx *lex_ctx, duk_small_int_t lookup_offset) { | |
623 | /* validation performed by duk__hexval */ | |
624 | return (duk__hexval(lex_ctx, lex_ctx->window[lookup_offset].codepoint) << 4) | | |
625 | (duk__hexval(lex_ctx, lex_ctx->window[lookup_offset + 1].codepoint)); | |
626 | } | |
627 | ||
628 | DUK_LOCAL duk_codepoint_t duk__decode_uniesc_from_window(duk_lexer_ctx *lex_ctx, duk_small_int_t lookup_offset) { | |
629 | /* validation performed by duk__hexval */ | |
630 | return (duk__hexval(lex_ctx, lex_ctx->window[lookup_offset].codepoint) << 12) | | |
631 | (duk__hexval(lex_ctx, lex_ctx->window[lookup_offset + 1].codepoint) << 8) | | |
632 | (duk__hexval(lex_ctx, lex_ctx->window[lookup_offset + 2].codepoint) << 4) | | |
633 | (duk__hexval(lex_ctx, lex_ctx->window[lookup_offset + 3].codepoint)); | |
634 | } | |
635 | ||
636 | /* | |
637 | * Parse Ecmascript source InputElementDiv or InputElementRegExp | |
638 | * (E5 Section 7), skipping whitespace, comments, and line terminators. | |
639 | * | |
640 | * Possible results are: | |
641 | * (1) a token | |
642 | * (2) a line terminator (skipped) | |
643 | * (3) a comment (skipped) | |
644 | * (4) EOF | |
645 | * | |
646 | * White space is automatically skipped from the current position (but | |
647 | * not after the input element). If input has already ended, returns | |
648 | * DUK_TOK_EOF indefinitely. If a parse error occurs, uses an DUK_ERROR() | |
649 | * macro call (and hence a longjmp through current heap longjmp context). | |
650 | * Comments and line terminator tokens are automatically skipped. | |
651 | * | |
652 | * The input element being matched is determined by regexp_mode; if set, | |
653 | * parses a InputElementRegExp, otherwise a InputElementDiv. The | |
654 | * difference between these are handling of productions starting with a | |
655 | * forward slash. | |
656 | * | |
657 | * If strict_mode is set, recognizes additional future reserved words | |
658 | * specific to strict mode, and refuses to parse octal literals. | |
659 | * | |
660 | * The matching strategy below is to (currently) use a six character | |
661 | * lookup window to quickly determine which production is the -longest- | |
662 | * matching one, and then parse that. The top-level if-else clauses | |
663 | * match the first character, and the code blocks for each clause | |
664 | * handle -all- alternatives for that first character. Ecmascript | |
665 | * specification uses the "longest match wins" semantics, so the order | |
666 | * of the if-clauses matters. | |
667 | * | |
668 | * Misc notes: | |
669 | * | |
670 | * * Ecmascript numeric literals do not accept a sign character. | |
671 | * Consequently e.g. "-1.0" is parsed as two tokens: a negative | |
672 | * sign and a positive numeric literal. The compiler performs | |
673 | * the negation during compilation, so this has no adverse impact. | |
674 | * | |
675 | * * There is no token for "undefined": it is just a value available | |
676 | * from the global object (or simply established by doing a reference | |
677 | * to an undefined value). | |
678 | * | |
679 | * * Some contexts want Identifier tokens, which are IdentifierNames | |
680 | * excluding reserved words, while some contexts want IdentifierNames | |
681 | * directly. In the latter case e.g. "while" is interpreted as an | |
682 | * identifier name, not a DUK_TOK_WHILE token. The solution here is | |
683 | * to provide both token types: DUK_TOK_WHILE goes to 't' while | |
684 | * DUK_TOK_IDENTIFIER goes to 't_nores', and 'slot1' always contains | |
685 | * the identifier / keyword name. | |
686 | * | |
687 | * * Directive prologue needs to identify string literals such as | |
688 | * "use strict" and 'use strict', which are sensitive to line | |
689 | * continuations and escape sequences. For instance, "use\u0020strict" | |
690 | * is a valid directive but is distinct from "use strict". The solution | |
691 | * here is to decode escapes while tokenizing, but to keep track of the | |
692 | * number of escapes. Directive detection can then check that the | |
693 | * number of escapes is zero. | |
694 | * | |
695 | * * Multi-line comments with one or more internal LineTerminator are | |
696 | * treated like a line terminator to comply with automatic semicolon | |
697 | * insertion. | |
698 | */ | |
699 | ||
700 | DUK_INTERNAL | |
701 | void duk_lexer_parse_js_input_element(duk_lexer_ctx *lex_ctx, | |
702 | duk_token *out_token, | |
703 | duk_bool_t strict_mode, | |
704 | duk_bool_t regexp_mode) { | |
705 | duk_codepoint_t x; /* temporary, must be signed and 32-bit to hold Unicode code points */ | |
706 | duk_small_uint_t advtok = 0; /* (advance << 8) + token_type, updated at function end, | |
707 | * init is unnecessary but suppresses "may be used uninitialized" warnings. | |
708 | */ | |
709 | duk_bool_t got_lineterm = 0; /* got lineterm preceding non-whitespace, non-lineterm token */ | |
710 | ||
711 | if (++lex_ctx->token_count >= lex_ctx->token_limit) { | |
712 | DUK_ERROR_RANGE(lex_ctx->thr, "token limit"); | |
713 | return; /* unreachable */ | |
714 | } | |
715 | ||
716 | out_token->t = DUK_TOK_EOF; | |
717 | out_token->t_nores = -1; /* marker: copy t if not changed */ | |
718 | #if 0 /* not necessary to init, disabled for faster parsing */ | |
719 | out_token->num = DUK_DOUBLE_NAN; | |
720 | out_token->str1 = NULL; | |
721 | out_token->str2 = NULL; | |
722 | #endif | |
723 | out_token->num_escapes = 0; | |
724 | /* out_token->lineterm set by caller */ | |
725 | ||
726 | /* This would be nice, but parsing is faster without resetting the | |
727 | * value slots. The only side effect is that references to temporary | |
728 | * string values may linger until lexing is finished; they're then | |
729 | * freed normally. | |
730 | */ | |
731 | #if 0 | |
732 | duk_to_undefined((duk_context *) lex_ctx->thr, lex_ctx->slot1_idx); | |
733 | duk_to_undefined((duk_context *) lex_ctx->thr, lex_ctx->slot2_idx); | |
734 | #endif | |
735 | ||
736 | /* 'advtok' indicates how much to advance and which token id to assign | |
737 | * at the end. This shared functionality minimizes code size. All | |
738 | * code paths are required to set 'advtok' to some value, so no default | |
739 | * init value is used. Code paths calling DUK_ERROR() never return so | |
740 | * they don't need to set advtok. | |
741 | */ | |
742 | ||
743 | /* | |
744 | * Matching order: | |
745 | * | |
746 | * Punctuator first chars, also covers comments, regexps | |
747 | * LineTerminator | |
748 | * Identifier or reserved word, also covers null/true/false literals | |
749 | * NumericLiteral | |
750 | * StringLiteral | |
751 | * EOF | |
752 | * | |
753 | * The order does not matter as long as the longest match is | |
754 | * always correctly identified. There are order dependencies | |
755 | * in the clauses, so it's not trivial to convert to a switch. | |
756 | */ | |
757 | ||
758 | restart_lineupdate: | |
759 | out_token->start_line = lex_ctx->window[0].line; | |
760 | ||
761 | restart: | |
762 | out_token->start_offset = lex_ctx->window[0].offset; | |
763 | ||
764 | x = DUK__L0(); | |
765 | ||
766 | switch (x) { | |
767 | case DUK_ASC_SPACE: | |
768 | case DUK_ASC_HT: /* fast paths for space and tab */ | |
769 | DUK__ADVANCECHARS(lex_ctx, 1); | |
770 | goto restart; | |
771 | case DUK_ASC_LF: /* LF line terminator; CR LF and Unicode lineterms are handled in slow path */ | |
772 | DUK__ADVANCECHARS(lex_ctx, 1); | |
773 | got_lineterm = 1; | |
774 | goto restart_lineupdate; | |
775 | case DUK_ASC_SLASH: /* '/' */ | |
776 | if (DUK__L1() == '/') { | |
777 | /* | |
778 | * E5 Section 7.4, allow SourceCharacter (which is any 16-bit | |
779 | * code point). | |
780 | */ | |
781 | ||
782 | /* DUK__ADVANCECHARS(lex_ctx, 2) would be correct here, but it unnecessary */ | |
783 | for (;;) { | |
784 | x = DUK__L0(); | |
785 | if (x < 0 || duk_unicode_is_line_terminator(x)) { | |
786 | break; | |
787 | } | |
788 | DUK__ADVANCECHARS(lex_ctx, 1); | |
789 | } | |
790 | goto restart; /* line terminator will be handled on next round */ | |
791 | } else if (DUK__L1() == '*') { | |
792 | /* | |
793 | * E5 Section 7.4. If the multi-line comment contains a newline, | |
794 | * it is treated like a single line terminator for automatic | |
795 | * semicolon insertion. | |
796 | */ | |
797 | ||
798 | duk_bool_t last_asterisk = 0; | |
799 | DUK__ADVANCECHARS(lex_ctx, 2); | |
800 | for (;;) { | |
801 | x = DUK__L0(); | |
802 | if (x < 0) { | |
803 | DUK_ERROR_SYNTAX(lex_ctx->thr, "eof in multiline comment"); | |
804 | } | |
805 | DUK__ADVANCECHARS(lex_ctx, 1); | |
806 | if (last_asterisk && x == '/') { | |
807 | break; | |
808 | } | |
809 | if (duk_unicode_is_line_terminator(x)) { | |
810 | got_lineterm = 1; | |
811 | } | |
812 | last_asterisk = (x == '*'); | |
813 | } | |
814 | goto restart_lineupdate; | |
815 | } else if (regexp_mode) { | |
816 | #if defined(DUK_USE_REGEXP_SUPPORT) | |
817 | /* | |
818 | * "/" followed by something in regexp mode. See E5 Section 7.8.5. | |
819 | * | |
820 | * RegExp parsing is a bit complex. First, the regexp body is delimited | |
821 | * by forward slashes, but the body may also contain forward slashes as | |
822 | * part of an escape sequence or inside a character class (delimited by | |
823 | * square brackets). A mini state machine is used to implement these. | |
824 | * | |
825 | * Further, an early (parse time) error must be thrown if the regexp | |
826 | * would cause a run-time error when used in the expression new RegExp(...). | |
827 | * Parsing here simply extracts the (candidate) regexp, and also accepts | |
828 | * invalid regular expressions (which are delimited properly). The caller | |
829 | * (compiler) must perform final validation and regexp compilation. | |
830 | * | |
831 | * RegExp first char may not be '/' (single line comment) or '*' (multi- | |
832 | * line comment). These have already been checked above, so there is no | |
833 | * need below for special handling of the first regexp character as in | |
834 | * the E5 productions. | |
835 | * | |
836 | * About unicode escapes within regexp literals: | |
837 | * | |
838 | * E5 Section 7.8.5 grammar does NOT accept \uHHHH escapes. | |
839 | * However, Section 6 states that regexps accept the escapes, | |
840 | * see paragraph starting with "In string literals...". | |
841 | * The regexp grammar, which sees the decoded regexp literal | |
842 | * (after lexical parsing) DOES have a \uHHHH unicode escape. | |
843 | * So, for instance: | |
844 | * | |
845 | * /\u1234/ | |
846 | * | |
847 | * should first be parsed by the lexical grammar as: | |
848 | * | |
849 | * '\' 'u' RegularExpressionBackslashSequence | |
850 | * '1' RegularExpressionNonTerminator | |
851 | * '2' RegularExpressionNonTerminator | |
852 | * '3' RegularExpressionNonTerminator | |
853 | * '4' RegularExpressionNonTerminator | |
854 | * | |
855 | * and the escape itself is then parsed by the regexp engine. | |
856 | * This is the current implementation. | |
857 | * | |
858 | * Minor spec inconsistency: | |
859 | * | |
860 | * E5 Section 7.8.5 RegularExpressionBackslashSequence is: | |
861 | * | |
862 | * \ RegularExpressionNonTerminator | |
863 | * | |
864 | * while Section A.1 RegularExpressionBackslashSequence is: | |
865 | * | |
866 | * \ NonTerminator | |
867 | * | |
868 | * The latter is not normative and a typo. | |
869 | * | |
870 | */ | |
871 | ||
872 | /* first, parse regexp body roughly */ | |
873 | ||
874 | duk_small_int_t state = 0; /* 0=base, 1=esc, 2=class, 3=class+esc */ | |
875 | ||
876 | DUK__INITBUFFER(lex_ctx); | |
877 | for (;;) { | |
878 | DUK__ADVANCECHARS(lex_ctx, 1); /* skip opening slash on first loop */ | |
879 | x = DUK__L0(); | |
880 | if (x < 0 || duk_unicode_is_line_terminator(x)) { | |
881 | DUK_ERROR_SYNTAX(lex_ctx->thr, "eof or line terminator in regexp"); | |
882 | } | |
883 | x = DUK__L0(); /* re-read to avoid spill / fetch */ | |
884 | if (state == 0) { | |
885 | if (x == '/') { | |
886 | DUK__ADVANCECHARS(lex_ctx, 1); /* eat closing slash */ | |
887 | break; | |
888 | } else if (x == '\\') { | |
889 | state = 1; | |
890 | } else if (x == '[') { | |
891 | state = 2; | |
892 | } | |
893 | } else if (state == 1) { | |
894 | state = 0; | |
895 | } else if (state == 2) { | |
896 | if (x == ']') { | |
897 | state = 0; | |
898 | } else if (x == '\\') { | |
899 | state = 3; | |
900 | } | |
901 | } else { /* state == 3 */ | |
902 | state = 2; | |
903 | } | |
904 | DUK__APPENDBUFFER(lex_ctx, x); | |
905 | } | |
906 | duk__internbuffer(lex_ctx, lex_ctx->slot1_idx); | |
907 | out_token->str1 = duk_get_hstring((duk_context *) lex_ctx->thr, lex_ctx->slot1_idx); | |
908 | ||
909 | /* second, parse flags */ | |
910 | ||
911 | DUK__INITBUFFER(lex_ctx); | |
912 | for (;;) { | |
913 | x = DUK__L0(); | |
914 | if (!duk_unicode_is_identifier_part(x)) { | |
915 | break; | |
916 | } | |
917 | x = DUK__L0(); /* re-read to avoid spill / fetch */ | |
918 | DUK__APPENDBUFFER(lex_ctx, x); | |
919 | DUK__ADVANCECHARS(lex_ctx, 1); | |
920 | } | |
921 | duk__internbuffer(lex_ctx, lex_ctx->slot2_idx); | |
922 | out_token->str2 = duk_get_hstring((duk_context *) lex_ctx->thr, lex_ctx->slot2_idx); | |
923 | ||
924 | DUK__INITBUFFER(lex_ctx); /* free some memory */ | |
925 | ||
926 | /* validation of the regexp is caller's responsibility */ | |
927 | ||
928 | advtok = DUK__ADVTOK(0, DUK_TOK_REGEXP); | |
929 | #else | |
930 | DUK_ERROR_SYNTAX(lex_ctx->thr, "regexp support disabled"); | |
931 | #endif | |
932 | } else if (DUK__L1() == '=') { | |
933 | /* "/=" and not in regexp mode */ | |
934 | advtok = DUK__ADVTOK(2, DUK_TOK_DIV_EQ); | |
935 | } else { | |
936 | /* "/" and not in regexp mode */ | |
937 | advtok = DUK__ADVTOK(1, DUK_TOK_DIV); | |
938 | } | |
939 | break; | |
940 | case DUK_ASC_LCURLY: /* '{' */ | |
941 | advtok = DUK__ADVTOK(1, DUK_TOK_LCURLY); | |
942 | break; | |
943 | case DUK_ASC_RCURLY: /* '}' */ | |
944 | advtok = DUK__ADVTOK(1, DUK_TOK_RCURLY); | |
945 | break; | |
946 | case DUK_ASC_LPAREN: /* '(' */ | |
947 | advtok = DUK__ADVTOK(1, DUK_TOK_LPAREN); | |
948 | break; | |
949 | case DUK_ASC_RPAREN: /* ')' */ | |
950 | advtok = DUK__ADVTOK(1, DUK_TOK_RPAREN); | |
951 | break; | |
952 | case DUK_ASC_LBRACKET: /* '[' */ | |
953 | advtok = DUK__ADVTOK(1, DUK_TOK_LBRACKET); | |
954 | break; | |
955 | case DUK_ASC_RBRACKET: /* ']' */ | |
956 | advtok = DUK__ADVTOK(1, DUK_TOK_RBRACKET); | |
957 | break; | |
958 | case DUK_ASC_PERIOD: /* '.' */ | |
959 | if (DUK__ISDIGIT(DUK__L1())) { | |
960 | /* Period followed by a digit can only start DecimalLiteral | |
961 | * (handled in slow path). We could jump straight into the | |
962 | * DecimalLiteral handling but should avoid goto to inside | |
963 | * a block. | |
964 | */ | |
965 | goto slow_path; | |
966 | } | |
967 | advtok = DUK__ADVTOK(1, DUK_TOK_PERIOD); | |
968 | break; | |
969 | case DUK_ASC_SEMICOLON: /* ';' */ | |
970 | advtok = DUK__ADVTOK(1, DUK_TOK_SEMICOLON); | |
971 | break; | |
972 | case DUK_ASC_COMMA: /* ',' */ | |
973 | advtok = DUK__ADVTOK(1, DUK_TOK_COMMA); | |
974 | break; | |
975 | case DUK_ASC_LANGLE: /* '<' */ | |
976 | if (DUK__L1() == '<' && DUK__L2() == '=') { | |
977 | advtok = DUK__ADVTOK(3, DUK_TOK_ALSHIFT_EQ); | |
978 | } else if (DUK__L1() == '=') { | |
979 | advtok = DUK__ADVTOK(2, DUK_TOK_LE); | |
980 | } else if (DUK__L1() == '<') { | |
981 | advtok = DUK__ADVTOK(2, DUK_TOK_ALSHIFT); | |
982 | } else { | |
983 | advtok = DUK__ADVTOK(1, DUK_TOK_LT); | |
984 | } | |
985 | break; | |
986 | case DUK_ASC_RANGLE: /* '>' */ | |
987 | if (DUK__L1() == '>' && DUK__L2() == '>' && DUK__L3() == '=') { | |
988 | advtok = DUK__ADVTOK(4, DUK_TOK_RSHIFT_EQ); | |
989 | } else if (DUK__L1() == '>' && DUK__L2() == '>') { | |
990 | advtok = DUK__ADVTOK(3, DUK_TOK_RSHIFT); | |
991 | } else if (DUK__L1() == '>' && DUK__L2() == '=') { | |
992 | advtok = DUK__ADVTOK(3, DUK_TOK_ARSHIFT_EQ); | |
993 | } else if (DUK__L1() == '=') { | |
994 | advtok = DUK__ADVTOK(2, DUK_TOK_GE); | |
995 | } else if (DUK__L1() == '>') { | |
996 | advtok = DUK__ADVTOK(2, DUK_TOK_ARSHIFT); | |
997 | } else { | |
998 | advtok = DUK__ADVTOK(1, DUK_TOK_GT); | |
999 | } | |
1000 | break; | |
1001 | case DUK_ASC_EQUALS: /* '=' */ | |
1002 | if (DUK__L1() == '=' && DUK__L2() == '=') { | |
1003 | advtok = DUK__ADVTOK(3, DUK_TOK_SEQ); | |
1004 | } else if (DUK__L1() == '=') { | |
1005 | advtok = DUK__ADVTOK(2, DUK_TOK_EQ); | |
1006 | } else { | |
1007 | advtok = DUK__ADVTOK(1, DUK_TOK_EQUALSIGN); | |
1008 | } | |
1009 | break; | |
1010 | case DUK_ASC_EXCLAMATION: /* '!' */ | |
1011 | if (DUK__L1() == '=' && DUK__L2() == '=') { | |
1012 | advtok = DUK__ADVTOK(3, DUK_TOK_SNEQ); | |
1013 | } else if (DUK__L1() == '=') { | |
1014 | advtok = DUK__ADVTOK(2, DUK_TOK_NEQ); | |
1015 | } else { | |
1016 | advtok = DUK__ADVTOK(1, DUK_TOK_LNOT); | |
1017 | } | |
1018 | break; | |
1019 | case DUK_ASC_PLUS: /* '+' */ | |
1020 | if (DUK__L1() == '+') { | |
1021 | advtok = DUK__ADVTOK(2, DUK_TOK_INCREMENT); | |
1022 | } else if (DUK__L1() == '=') { | |
1023 | advtok = DUK__ADVTOK(2, DUK_TOK_ADD_EQ); | |
1024 | } else { | |
1025 | advtok = DUK__ADVTOK(1, DUK_TOK_ADD); | |
1026 | } | |
1027 | break; | |
1028 | case DUK_ASC_MINUS: /* '-' */ | |
1029 | if (DUK__L1() == '-') { | |
1030 | advtok = DUK__ADVTOK(2, DUK_TOK_DECREMENT); | |
1031 | } else if (DUK__L1() == '=') { | |
1032 | advtok = DUK__ADVTOK(2, DUK_TOK_SUB_EQ); | |
1033 | } else { | |
1034 | advtok = DUK__ADVTOK(1, DUK_TOK_SUB); | |
1035 | } | |
1036 | break; | |
1037 | case DUK_ASC_STAR: /* '*' */ | |
1038 | if (DUK__L1() == '=') { | |
1039 | advtok = DUK__ADVTOK(2, DUK_TOK_MUL_EQ); | |
1040 | } else { | |
1041 | advtok = DUK__ADVTOK(1, DUK_TOK_MUL); | |
1042 | } | |
1043 | break; | |
1044 | case DUK_ASC_PERCENT: /* '%' */ | |
1045 | if (DUK__L1() == '=') { | |
1046 | advtok = DUK__ADVTOK(2, DUK_TOK_MOD_EQ); | |
1047 | } else { | |
1048 | advtok = DUK__ADVTOK(1, DUK_TOK_MOD); | |
1049 | } | |
1050 | break; | |
1051 | case DUK_ASC_AMP: /* '&' */ | |
1052 | if (DUK__L1() == '&') { | |
1053 | advtok = DUK__ADVTOK(2, DUK_TOK_LAND); | |
1054 | } else if (DUK__L1() == '=') { | |
1055 | advtok = DUK__ADVTOK(2, DUK_TOK_BAND_EQ); | |
1056 | } else { | |
1057 | advtok = DUK__ADVTOK(1, DUK_TOK_BAND); | |
1058 | } | |
1059 | break; | |
1060 | case DUK_ASC_PIPE: /* '|' */ | |
1061 | if (DUK__L1() == '|') { | |
1062 | advtok = DUK__ADVTOK(2, DUK_TOK_LOR); | |
1063 | } else if (DUK__L1() == '=') { | |
1064 | advtok = DUK__ADVTOK(2, DUK_TOK_BOR_EQ); | |
1065 | } else { | |
1066 | advtok = DUK__ADVTOK(1, DUK_TOK_BOR); | |
1067 | } | |
1068 | break; | |
1069 | case DUK_ASC_CARET: /* '^' */ | |
1070 | if (DUK__L1() == '=') { | |
1071 | advtok = DUK__ADVTOK(2, DUK_TOK_BXOR_EQ); | |
1072 | } else { | |
1073 | advtok = DUK__ADVTOK(1, DUK_TOK_BXOR); | |
1074 | } | |
1075 | break; | |
1076 | case DUK_ASC_TILDE: /* '~' */ | |
1077 | advtok = DUK__ADVTOK(1, DUK_TOK_BNOT); | |
1078 | break; | |
1079 | case DUK_ASC_QUESTION: /* '?' */ | |
1080 | advtok = DUK__ADVTOK(1, DUK_TOK_QUESTION); | |
1081 | break; | |
1082 | case DUK_ASC_COLON: /* ':' */ | |
1083 | advtok = DUK__ADVTOK(1, DUK_TOK_COLON); | |
1084 | break; | |
1085 | case DUK_ASC_DOUBLEQUOTE: /* '"' */ | |
1086 | case DUK_ASC_SINGLEQUOTE: { /* '\'' */ | |
1087 | duk_small_int_t quote = x; /* Note: duk_uint8_t type yields larger code */ | |
1088 | duk_small_int_t adv; | |
1089 | ||
1090 | DUK__INITBUFFER(lex_ctx); | |
1091 | for (;;) { | |
1092 | DUK__ADVANCECHARS(lex_ctx, 1); /* eat opening quote on first loop */ | |
1093 | x = DUK__L0(); | |
1094 | if (x < 0 || duk_unicode_is_line_terminator(x)) { | |
1095 | DUK_ERROR_SYNTAX(lex_ctx->thr, "eof or line terminator in string literal"); | |
1096 | } | |
1097 | if (x == quote) { | |
1098 | DUK__ADVANCECHARS(lex_ctx, 1); /* eat closing quote */ | |
1099 | break; | |
1100 | } | |
1101 | if (x == '\\') { | |
1102 | /* DUK__L0 -> '\' char | |
1103 | * DUK__L1 ... DUK__L5 -> more lookup | |
1104 | */ | |
1105 | ||
1106 | x = DUK__L1(); | |
1107 | ||
1108 | /* How much to advance before next loop; note that next loop | |
1109 | * will advance by 1 anyway, so -1 from the total escape | |
1110 | * length (e.g. len('\uXXXX') - 1 = 6 - 1). As a default, | |
1111 | * 1 is good. | |
1112 | */ | |
1113 | adv = 2 - 1; /* note: long live range */ | |
1114 | ||
1115 | if (x < 0) { | |
1116 | DUK_ERROR_SYNTAX(lex_ctx->thr, "eof or line terminator in string literal"); | |
1117 | } | |
1118 | if (duk_unicode_is_line_terminator(x)) { | |
1119 | /* line continuation */ | |
1120 | if (x == 0x000d && DUK__L2() == 0x000a) { | |
1121 | /* CR LF again a special case */ | |
1122 | adv = 3 - 1; | |
1123 | } | |
1124 | } else if (x == '\'') { | |
1125 | DUK__APPENDBUFFER(lex_ctx, 0x0027); | |
1126 | } else if (x == '"') { | |
1127 | DUK__APPENDBUFFER(lex_ctx, 0x0022); | |
1128 | } else if (x == '\\') { | |
1129 | DUK__APPENDBUFFER(lex_ctx, 0x005c); | |
1130 | } else if (x == 'b') { | |
1131 | DUK__APPENDBUFFER(lex_ctx, 0x0008); | |
1132 | } else if (x == 'f') { | |
1133 | DUK__APPENDBUFFER(lex_ctx, 0x000c); | |
1134 | } else if (x == 'n') { | |
1135 | DUK__APPENDBUFFER(lex_ctx, 0x000a); | |
1136 | } else if (x == 'r') { | |
1137 | DUK__APPENDBUFFER(lex_ctx, 0x000d); | |
1138 | } else if (x == 't') { | |
1139 | DUK__APPENDBUFFER(lex_ctx, 0x0009); | |
1140 | } else if (x == 'v') { | |
1141 | DUK__APPENDBUFFER(lex_ctx, 0x000b); | |
1142 | } else if (x == 'x') { | |
1143 | adv = 4 - 1; | |
1144 | DUK__APPENDBUFFER(lex_ctx, duk__decode_hexesc_from_window(lex_ctx, 2)); | |
1145 | } else if (x == 'u') { | |
1146 | adv = 6 - 1; | |
1147 | DUK__APPENDBUFFER(lex_ctx, duk__decode_uniesc_from_window(lex_ctx, 2)); | |
1148 | } else if (DUK__ISDIGIT(x)) { | |
1149 | duk_codepoint_t ch = 0; /* initialized to avoid warnings of unused var */ | |
1150 | ||
1151 | /* | |
1152 | * Octal escape or zero escape: | |
1153 | * \0 (lookahead not DecimalDigit) | |
1154 | * \1 ... \7 (lookahead not DecimalDigit) | |
1155 | * \ZeroToThree OctalDigit (lookahead not DecimalDigit) | |
1156 | * \FourToSeven OctalDigit (no lookahead restrictions) | |
1157 | * \ZeroToThree OctalDigit OctalDigit (no lookahead restrictions) | |
1158 | * | |
1159 | * Zero escape is part of the standard syntax. Octal escapes are | |
1160 | * defined in E5 Section B.1.2, and are only allowed in non-strict mode. | |
1161 | * Any other productions starting with a decimal digit are invalid. | |
1162 | */ | |
1163 | ||
1164 | if (x == '0' && !DUK__ISDIGIT(DUK__L2())) { | |
1165 | /* Zero escape (also allowed in non-strict mode) */ | |
1166 | ch = 0; | |
1167 | /* adv = 2 - 1 default OK */ | |
1168 | #if defined(DUK_USE_OCTAL_SUPPORT) | |
1169 | } else if (strict_mode) { | |
1170 | /* No other escape beginning with a digit in strict mode */ | |
1171 | DUK_ERROR_SYNTAX(lex_ctx->thr, "invalid escape in string literal"); | |
1172 | } else if (DUK__ISDIGIT03(x) && DUK__ISOCTDIGIT(DUK__L2()) && DUK__ISOCTDIGIT(DUK__L3())) { | |
1173 | /* Three digit octal escape, digits validated. */ | |
1174 | adv = 4 - 1; | |
1175 | ch = (duk__hexval(lex_ctx, x) << 6) + | |
1176 | (duk__hexval(lex_ctx, DUK__L2()) << 3) + | |
1177 | duk__hexval(lex_ctx, DUK__L3()); | |
1178 | } else if (((DUK__ISDIGIT03(x) && !DUK__ISDIGIT(DUK__L3())) || DUK__ISDIGIT47(x)) && | |
1179 | DUK__ISOCTDIGIT(DUK__L2())) { | |
1180 | /* Two digit octal escape, digits validated. | |
1181 | * | |
1182 | * The if-condition is a bit tricky. We could catch e.g. | |
1183 | * '\039' in the three-digit escape and fail it there (by | |
1184 | * validating the digits), but we want to avoid extra | |
1185 | * additional validation code. | |
1186 | */ | |
1187 | adv = 3 - 1; | |
1188 | ch = (duk__hexval(lex_ctx, x) << 3) + | |
1189 | duk__hexval(lex_ctx, DUK__L2()); | |
1190 | } else if (DUK__ISDIGIT(x) && !DUK__ISDIGIT(DUK__L2())) { | |
1191 | /* One digit octal escape, digit validated. */ | |
1192 | /* adv = 2 default OK */ | |
1193 | ch = duk__hexval(lex_ctx, x); | |
1194 | #else | |
1195 | /* fall through to error */ | |
1196 | #endif | |
1197 | } else { | |
1198 | DUK_ERROR_SYNTAX(lex_ctx->thr, "invalid escape in string literal"); | |
1199 | } | |
1200 | ||
1201 | DUK__APPENDBUFFER(lex_ctx, ch); | |
1202 | } else { | |
1203 | /* escaped NonEscapeCharacter */ | |
1204 | DUK__APPENDBUFFER(lex_ctx, x); | |
1205 | } | |
1206 | DUK__ADVANCECHARS(lex_ctx, adv); | |
1207 | ||
1208 | /* Track number of escapes; count not really needed but directive | |
1209 | * prologues need to detect whether there were any escapes or line | |
1210 | * continuations or not. | |
1211 | */ | |
1212 | out_token->num_escapes++; | |
1213 | } else { | |
1214 | /* part of string */ | |
1215 | DUK__APPENDBUFFER(lex_ctx, x); | |
1216 | } | |
1217 | } | |
1218 | ||
1219 | duk__internbuffer(lex_ctx, lex_ctx->slot1_idx); | |
1220 | out_token->str1 = duk_get_hstring((duk_context *) lex_ctx->thr, lex_ctx->slot1_idx); | |
1221 | ||
1222 | DUK__INITBUFFER(lex_ctx); /* free some memory */ | |
1223 | ||
1224 | advtok = DUK__ADVTOK(0, DUK_TOK_STRING); | |
1225 | break; | |
1226 | } | |
1227 | default: | |
1228 | goto slow_path; | |
1229 | } /* switch */ | |
1230 | ||
1231 | goto skip_slow_path; | |
1232 | ||
1233 | slow_path: | |
1234 | if (duk_unicode_is_line_terminator(x)) { | |
1235 | if (x == 0x000d && DUK__L1() == 0x000a) { | |
1236 | /* | |
1237 | * E5 Section 7.3: CR LF is detected as a single line terminator for | |
1238 | * line numbers. Here we also detect it as a single line terminator | |
1239 | * token. | |
1240 | */ | |
1241 | DUK__ADVANCECHARS(lex_ctx, 2); | |
1242 | } else { | |
1243 | DUK__ADVANCECHARS(lex_ctx, 1); | |
1244 | } | |
1245 | got_lineterm = 1; | |
1246 | goto restart_lineupdate; | |
1247 | } else if (duk_unicode_is_identifier_start(x) || x == '\\') { | |
1248 | /* | |
1249 | * Parse an identifier and then check whether it is: | |
1250 | * - reserved word (keyword or other reserved word) | |
1251 | * - "null" (NullLiteral) | |
1252 | * - "true" (BooleanLiteral) | |
1253 | * - "false" (BooleanLiteral) | |
1254 | * - anything else => identifier | |
1255 | * | |
1256 | * This does not follow the E5 productions cleanly, but is | |
1257 | * useful and compact. | |
1258 | * | |
1259 | * Note that identifiers may contain Unicode escapes, | |
1260 | * see E5 Sections 6 and 7.6. They must be decoded first, | |
1261 | * and the result checked against allowed characters. | |
1262 | * The above if-clause accepts an identifier start and an | |
1263 | * '\' character -- no other token can begin with a '\'. | |
1264 | * | |
1265 | * Note that "get" and "set" are not reserved words in E5 | |
1266 | * specification so they are recognized as plain identifiers | |
1267 | * (the tokens DUK_TOK_GET and DUK_TOK_SET are actually not | |
1268 | * used now). The compiler needs to work around this. | |
1269 | * | |
1270 | * Strictly speaking, following Ecmascript longest match | |
1271 | * specification, an invalid escape for the first character | |
1272 | * should cause a syntax error. However, an invalid escape | |
1273 | * for IdentifierParts should just terminate the identifier | |
1274 | * early (longest match), and let the next tokenization | |
1275 | * fail. For instance Rhino croaks with 'foo\z' when | |
1276 | * parsing the identifier. This has little practical impact. | |
1277 | */ | |
1278 | ||
1279 | duk_small_int_t i, i_end; | |
1280 | duk_bool_t first = 1; | |
1281 | duk_hstring *str; | |
1282 | ||
1283 | DUK__INITBUFFER(lex_ctx); | |
1284 | for (;;) { | |
1285 | /* re-lookup first char on first loop */ | |
1286 | if (DUK__L0() == '\\') { | |
1287 | duk_codepoint_t ch; | |
1288 | if (DUK__L1() != 'u') { | |
1289 | DUK_ERROR_SYNTAX(lex_ctx->thr, "invalid unicode escape in identifier"); | |
1290 | } | |
1291 | ||
1292 | ch = duk__decode_uniesc_from_window(lex_ctx, 2); | |
1293 | ||
1294 | /* IdentifierStart is stricter than IdentifierPart, so if the first | |
1295 | * character is escaped, must have a stricter check here. | |
1296 | */ | |
1297 | if (!(first ? duk_unicode_is_identifier_start(ch) : duk_unicode_is_identifier_part(ch))) { | |
1298 | DUK_ERROR_SYNTAX(lex_ctx->thr, "invalid unicode escape in identifier"); | |
1299 | } | |
1300 | DUK__APPENDBUFFER(lex_ctx, ch); | |
1301 | DUK__ADVANCECHARS(lex_ctx, 6); | |
1302 | ||
1303 | /* Track number of escapes: necessary for proper keyword | |
1304 | * detection. | |
1305 | */ | |
1306 | out_token->num_escapes++; | |
1307 | } else { | |
1308 | /* Note: first character is checked against this. But because | |
1309 | * IdentifierPart includes all IdentifierStart characters, and | |
1310 | * the first character (if unescaped) has already been checked | |
1311 | * in the if condition, this is OK. | |
1312 | */ | |
1313 | if (!duk_unicode_is_identifier_part(DUK__L0())) { | |
1314 | break; | |
1315 | } | |
1316 | DUK__APPENDBUFFER(lex_ctx, DUK__L0()); | |
1317 | DUK__ADVANCECHARS(lex_ctx, 1); | |
1318 | } | |
1319 | first = 0; | |
1320 | } | |
1321 | ||
1322 | duk__internbuffer(lex_ctx, lex_ctx->slot1_idx); | |
1323 | out_token->str1 = duk_get_hstring((duk_context *) lex_ctx->thr, lex_ctx->slot1_idx); | |
1324 | str = out_token->str1; | |
1325 | DUK_ASSERT(str != NULL); | |
1326 | out_token->t_nores = DUK_TOK_IDENTIFIER; | |
1327 | ||
1328 | DUK__INITBUFFER(lex_ctx); /* free some memory */ | |
1329 | ||
1330 | /* | |
1331 | * Interned identifier is compared against reserved words, which are | |
1332 | * currently interned into the heap context. See genbuiltins.py. | |
1333 | * | |
1334 | * Note that an escape in the identifier disables recognition of | |
1335 | * keywords; e.g. "\u0069f = 1;" is a valid statement (assigns to | |
1336 | * identifier named "if"). This is not necessarily compliant, | |
1337 | * see test-dec-escaped-char-in-keyword.js. | |
1338 | * | |
1339 | * Note: "get" and "set" are awkward. They are not officially | |
1340 | * ReservedWords (and indeed e.g. "var set = 1;" is valid), and | |
1341 | * must come out as DUK_TOK_IDENTIFIER. The compiler needs to | |
1342 | * work around this a bit. | |
1343 | */ | |
1344 | ||
1345 | /* XXX: optimize by adding the token numbers directly into the | |
1346 | * always interned duk_hstring objects (there should be enough | |
1347 | * flag bits free for that)? | |
1348 | */ | |
1349 | ||
1350 | i_end = (strict_mode ? DUK_STRIDX_END_RESERVED : DUK_STRIDX_START_STRICT_RESERVED); | |
1351 | ||
1352 | advtok = DUK__ADVTOK(0, DUK_TOK_IDENTIFIER); | |
1353 | if (out_token->num_escapes == 0) { | |
1354 | for (i = DUK_STRIDX_START_RESERVED; i < i_end; i++) { | |
1355 | DUK_ASSERT(i >= 0 && i < DUK_HEAP_NUM_STRINGS); | |
1356 | if (DUK_HTHREAD_GET_STRING(lex_ctx->thr, i) == str) { | |
1357 | advtok = DUK__ADVTOK(0, DUK_STRIDX_TO_TOK(i)); | |
1358 | break; | |
1359 | } | |
1360 | } | |
1361 | } | |
1362 | } else if (DUK__ISDIGIT(x) || (x == '.')) { | |
1363 | /* Note: decimal number may start with a period, but must be followed by a digit */ | |
1364 | ||
1365 | /* | |
1366 | * DecimalLiteral, HexIntegerLiteral, OctalIntegerLiteral | |
1367 | * "pre-parsing", followed by an actual, accurate parser step. | |
1368 | * | |
1369 | * Note: the leading sign character ('+' or '-') is -not- part of | |
1370 | * the production in E5 grammar, and that the a DecimalLiteral | |
1371 | * starting with a '0' must be followed by a non-digit. Leading | |
1372 | * zeroes are syntax errors and must be checked for. | |
1373 | * | |
1374 | * XXX: the two step parsing process is quite awkward, it would | |
1375 | * be more straightforward to allow numconv to parse the longest | |
1376 | * valid prefix (it already does that, it only needs to indicate | |
1377 | * where the input ended). However, the lexer decodes characters | |
1378 | * using a lookup window, so this is not a trivial change. | |
1379 | */ | |
1380 | ||
1381 | /* XXX: because of the final check below (that the literal is not | |
1382 | * followed by a digit), this could maybe be simplified, if we bail | |
1383 | * out early from a leading zero (and if there are no periods etc). | |
1384 | * Maybe too complex. | |
1385 | */ | |
1386 | ||
1387 | duk_double_t val; | |
1388 | duk_bool_t int_only = 0; | |
1389 | duk_bool_t allow_hex = 0; | |
1390 | duk_small_int_t state; /* 0=before period/exp, | |
1391 | * 1=after period, before exp | |
1392 | * 2=after exp, allow '+' or '-' | |
1393 | * 3=after exp and exp sign | |
1394 | */ | |
1395 | duk_small_uint_t s2n_flags; | |
1396 | duk_codepoint_t y; | |
1397 | ||
1398 | DUK__INITBUFFER(lex_ctx); | |
1399 | y = DUK__L1(); | |
1400 | if (x == '0' && (y == 'x' || y == 'X')) { | |
1401 | DUK__APPENDBUFFER(lex_ctx, x); | |
1402 | DUK__APPENDBUFFER(lex_ctx, y); | |
1403 | DUK__ADVANCECHARS(lex_ctx, 2); | |
1404 | int_only = 1; | |
1405 | allow_hex = 1; | |
1406 | #if defined(DUK_USE_OCTAL_SUPPORT) | |
1407 | } else if (!strict_mode && x == '0' && DUK__ISDIGIT(y)) { | |
1408 | /* Note: if DecimalLiteral starts with a '0', it can only be | |
1409 | * followed by a period or an exponent indicator which starts | |
1410 | * with 'e' or 'E'. Hence the if-check above ensures that | |
1411 | * OctalIntegerLiteral is the only valid NumericLiteral | |
1412 | * alternative at this point (even if y is, say, '9'). | |
1413 | */ | |
1414 | ||
1415 | DUK__APPENDBUFFER(lex_ctx, x); | |
1416 | DUK__ADVANCECHARS(lex_ctx, 1); | |
1417 | int_only = 1; | |
1418 | #endif | |
1419 | } | |
1420 | ||
1421 | state = 0; | |
1422 | for (;;) { | |
1423 | x = DUK__L0(); /* re-lookup curr char on first round */ | |
1424 | if (DUK__ISDIGIT(x)) { | |
1425 | /* Note: intentionally allow leading zeroes here, as the | |
1426 | * actual parser will check for them. | |
1427 | */ | |
1428 | if (state == 2) { | |
1429 | state = 3; | |
1430 | } | |
1431 | } else if (allow_hex && DUK__ISHEXDIGIT(x)) { | |
1432 | /* Note: 'e' and 'E' are also accepted here. */ | |
1433 | ; | |
1434 | } else if (x == '.') { | |
1435 | if (state >= 1 || int_only) { | |
1436 | break; | |
1437 | } else { | |
1438 | state = 1; | |
1439 | } | |
1440 | } else if (x == 'e' || x == 'E') { | |
1441 | if (state >= 2 || int_only) { | |
1442 | break; | |
1443 | } else { | |
1444 | state = 2; | |
1445 | } | |
1446 | } else if (x == '-' || x == '+') { | |
1447 | if (state != 2) { | |
1448 | break; | |
1449 | } else { | |
1450 | state = 3; | |
1451 | } | |
1452 | } else { | |
1453 | break; | |
1454 | } | |
1455 | DUK__APPENDBUFFER(lex_ctx, x); | |
1456 | DUK__ADVANCECHARS(lex_ctx, 1); | |
1457 | } | |
1458 | ||
1459 | /* XXX: better coercion */ | |
1460 | duk__internbuffer(lex_ctx, lex_ctx->slot1_idx); | |
1461 | ||
1462 | s2n_flags = DUK_S2N_FLAG_ALLOW_EXP | | |
1463 | DUK_S2N_FLAG_ALLOW_FRAC | | |
1464 | DUK_S2N_FLAG_ALLOW_NAKED_FRAC | | |
1465 | DUK_S2N_FLAG_ALLOW_EMPTY_FRAC | | |
1466 | #if defined(DUK_USE_OCTAL_SUPPORT) | |
1467 | (strict_mode ? 0 : DUK_S2N_FLAG_ALLOW_AUTO_OCT_INT) | | |
1468 | #endif | |
1469 | DUK_S2N_FLAG_ALLOW_AUTO_HEX_INT; | |
1470 | ||
1471 | duk_dup((duk_context *) lex_ctx->thr, lex_ctx->slot1_idx); | |
1472 | duk_numconv_parse((duk_context *) lex_ctx->thr, 10 /*radix*/, s2n_flags); | |
1473 | val = duk_to_number((duk_context *) lex_ctx->thr, -1); | |
1474 | if (DUK_ISNAN(val)) { | |
1475 | DUK_ERROR_SYNTAX(lex_ctx->thr, "invalid numeric literal"); | |
1476 | } | |
1477 | duk_replace((duk_context *) lex_ctx->thr, lex_ctx->slot1_idx); /* could also just pop? */ | |
1478 | ||
1479 | DUK__INITBUFFER(lex_ctx); /* free some memory */ | |
1480 | ||
1481 | /* Section 7.8.3 (note): NumericLiteral must be followed by something other than | |
1482 | * IdentifierStart or DecimalDigit. | |
1483 | */ | |
1484 | ||
1485 | if (DUK__ISDIGIT(DUK__L0()) || duk_unicode_is_identifier_start(DUK__L0())) { | |
1486 | DUK_ERROR_SYNTAX(lex_ctx->thr, "invalid numeric literal"); | |
1487 | } | |
1488 | ||
1489 | out_token->num = val; | |
1490 | advtok = DUK__ADVTOK(0, DUK_TOK_NUMBER); | |
1491 | } else if (duk_unicode_is_whitespace(DUK__LOOKUP(lex_ctx, 0))) { | |
1492 | DUK__ADVANCECHARS(lex_ctx, 1); | |
1493 | goto restart; | |
1494 | } else if (x < 0) { | |
1495 | advtok = DUK__ADVTOK(0, DUK_TOK_EOF); | |
1496 | } else { | |
1497 | DUK_ERROR_SYNTAX(lex_ctx->thr, "invalid token"); | |
1498 | } | |
1499 | skip_slow_path: | |
1500 | ||
1501 | /* | |
1502 | * Shared exit path | |
1503 | */ | |
1504 | ||
1505 | DUK__ADVANCEBYTES(lex_ctx, advtok >> 8); | |
1506 | out_token->t = advtok & 0xff; | |
1507 | if (out_token->t_nores < 0) { | |
1508 | out_token->t_nores = out_token->t; | |
1509 | } | |
1510 | out_token->lineterm = got_lineterm; | |
1511 | ||
1512 | /* Automatic semicolon insertion is allowed if a token is preceded | |
1513 | * by line terminator(s), or terminates a statement list (right curly | |
1514 | * or EOF). | |
1515 | */ | |
1516 | if (got_lineterm || out_token->t == DUK_TOK_RCURLY || out_token->t == DUK_TOK_EOF) { | |
1517 | out_token->allow_auto_semi = 1; | |
1518 | } else { | |
1519 | out_token->allow_auto_semi = 0; | |
1520 | } | |
1521 | } | |
1522 | ||
1523 | #if defined(DUK_USE_REGEXP_SUPPORT) | |
1524 | ||
1525 | /* | |
1526 | * Parse a RegExp token. The grammar is described in E5 Section 15.10. | |
1527 | * Terminal constructions (such as quantifiers) are parsed directly here. | |
1528 | * | |
1529 | * 0xffffffffU is used as a marker for "infinity" in quantifiers. Further, | |
1530 | * DUK__MAX_RE_QUANT_DIGITS limits the maximum number of digits that | |
1531 | * will be accepted for a quantifier. | |
1532 | */ | |
1533 | ||
1534 | DUK_INTERNAL void duk_lexer_parse_re_token(duk_lexer_ctx *lex_ctx, duk_re_token *out_token) { | |
1535 | duk_small_int_t advtok = 0; /* init is unnecessary but suppresses "may be used uninitialized" warnings */ | |
1536 | duk_codepoint_t x, y; | |
1537 | ||
1538 | if (++lex_ctx->token_count >= lex_ctx->token_limit) { | |
1539 | DUK_ERROR_RANGE(lex_ctx->thr, "token limit"); | |
1540 | return; /* unreachable */ | |
1541 | } | |
1542 | ||
1543 | DUK_MEMZERO(out_token, sizeof(*out_token)); | |
1544 | ||
1545 | x = DUK__L0(); | |
1546 | y = DUK__L1(); | |
1547 | ||
1548 | DUK_DDD(DUK_DDDPRINT("parsing regexp token, L0=%ld, L1=%ld", (long) x, (long) y)); | |
1549 | ||
1550 | switch (x) { | |
1551 | case '|': { | |
1552 | advtok = DUK__ADVTOK(1, DUK_RETOK_DISJUNCTION); | |
1553 | break; | |
1554 | } | |
1555 | case '^': { | |
1556 | advtok = DUK__ADVTOK(1, DUK_RETOK_ASSERT_START); | |
1557 | break; | |
1558 | } | |
1559 | case '$': { | |
1560 | advtok = DUK__ADVTOK(1, DUK_RETOK_ASSERT_END); | |
1561 | break; | |
1562 | } | |
1563 | case '?': { | |
1564 | out_token->qmin = 0; | |
1565 | out_token->qmax = 1; | |
1566 | if (y == '?') { | |
1567 | advtok = DUK__ADVTOK(2, DUK_RETOK_QUANTIFIER); | |
1568 | out_token->greedy = 0; | |
1569 | } else { | |
1570 | advtok = DUK__ADVTOK(1, DUK_RETOK_QUANTIFIER); | |
1571 | out_token->greedy = 1; | |
1572 | } | |
1573 | break; | |
1574 | } | |
1575 | case '*': { | |
1576 | out_token->qmin = 0; | |
1577 | out_token->qmax = DUK_RE_QUANTIFIER_INFINITE; | |
1578 | if (y == '?') { | |
1579 | advtok = DUK__ADVTOK(2, DUK_RETOK_QUANTIFIER); | |
1580 | out_token->greedy = 0; | |
1581 | } else { | |
1582 | advtok = DUK__ADVTOK(1, DUK_RETOK_QUANTIFIER); | |
1583 | out_token->greedy = 1; | |
1584 | } | |
1585 | break; | |
1586 | } | |
1587 | case '+': { | |
1588 | out_token->qmin = 1; | |
1589 | out_token->qmax = DUK_RE_QUANTIFIER_INFINITE; | |
1590 | if (y == '?') { | |
1591 | advtok = DUK__ADVTOK(2, DUK_RETOK_QUANTIFIER); | |
1592 | out_token->greedy = 0; | |
1593 | } else { | |
1594 | advtok = DUK__ADVTOK(1, DUK_RETOK_QUANTIFIER); | |
1595 | out_token->greedy = 1; | |
1596 | } | |
1597 | break; | |
1598 | } | |
1599 | case '{': { | |
1600 | /* Production allows 'DecimalDigits', including leading zeroes */ | |
1601 | duk_uint_fast32_t val1 = 0; | |
1602 | duk_uint_fast32_t val2 = DUK_RE_QUANTIFIER_INFINITE; | |
1603 | duk_small_int_t digits = 0; | |
1604 | #if defined(DUK_USE_ES6_REGEXP_BRACES) | |
1605 | duk_lexer_point lex_pt; | |
1606 | #endif | |
1607 | ||
1608 | #if defined(DUK_USE_ES6_REGEXP_BRACES) | |
1609 | /* Store lexer position, restoring if quantifier is invalid. */ | |
1610 | DUK_LEXER_GETPOINT(lex_ctx, &lex_pt); | |
1611 | #endif | |
1612 | ||
1613 | for (;;) { | |
1614 | DUK__ADVANCECHARS(lex_ctx, 1); /* eat '{' on entry */ | |
1615 | x = DUK__L0(); | |
1616 | if (DUK__ISDIGIT(x)) { | |
1617 | digits++; | |
1618 | val1 = val1 * 10 + (duk_uint_fast32_t) duk__hexval(lex_ctx, x); | |
1619 | } else if (x == ',') { | |
1620 | if (digits > DUK__MAX_RE_QUANT_DIGITS) { | |
1621 | goto invalid_quantifier; | |
1622 | } | |
1623 | if (val2 != DUK_RE_QUANTIFIER_INFINITE) { | |
1624 | goto invalid_quantifier; | |
1625 | } | |
1626 | if (DUK__L1() == '}') { | |
1627 | /* form: { DecimalDigits , }, val1 = min count */ | |
1628 | if (digits == 0) { | |
1629 | goto invalid_quantifier; | |
1630 | } | |
1631 | out_token->qmin = val1; | |
1632 | out_token->qmax = DUK_RE_QUANTIFIER_INFINITE; | |
1633 | DUK__ADVANCECHARS(lex_ctx, 2); | |
1634 | break; | |
1635 | } | |
1636 | val2 = val1; | |
1637 | val1 = 0; | |
1638 | digits = 0; /* not strictly necessary because of lookahead '}' above */ | |
1639 | } else if (x == '}') { | |
1640 | if (digits > DUK__MAX_RE_QUANT_DIGITS) { | |
1641 | goto invalid_quantifier; | |
1642 | } | |
1643 | if (digits == 0) { | |
1644 | goto invalid_quantifier; | |
1645 | } | |
1646 | if (val2 != DUK_RE_QUANTIFIER_INFINITE) { | |
1647 | /* val2 = min count, val1 = max count */ | |
1648 | out_token->qmin = val2; | |
1649 | out_token->qmax = val1; | |
1650 | } else { | |
1651 | /* val1 = count */ | |
1652 | out_token->qmin = val1; | |
1653 | out_token->qmax = val1; | |
1654 | } | |
1655 | DUK__ADVANCECHARS(lex_ctx, 1); | |
1656 | break; | |
1657 | } else { | |
1658 | goto invalid_quantifier; | |
1659 | } | |
1660 | } | |
1661 | if (DUK__L0() == '?') { | |
1662 | out_token->greedy = 0; | |
1663 | DUK__ADVANCECHARS(lex_ctx, 1); | |
1664 | } else { | |
1665 | out_token->greedy = 1; | |
1666 | } | |
1667 | advtok = DUK__ADVTOK(0, DUK_RETOK_QUANTIFIER); | |
1668 | break; | |
1669 | invalid_quantifier: | |
1670 | #if defined(DUK_USE_ES6_REGEXP_BRACES) | |
1671 | /* Failed to match the quantifier, restore lexer and parse | |
1672 | * opening brace as a literal. | |
1673 | */ | |
1674 | DUK_LEXER_SETPOINT(lex_ctx, &lex_pt); | |
1675 | advtok = DUK__ADVTOK(1, DUK_RETOK_ATOM_CHAR); | |
1676 | out_token->num = '{'; | |
1677 | #else | |
1678 | DUK_ERROR_SYNTAX(lex_ctx->thr, "invalid regexp quantifier"); | |
1679 | #endif | |
1680 | break; | |
1681 | } | |
1682 | case '.': { | |
1683 | advtok = DUK__ADVTOK(1, DUK_RETOK_ATOM_PERIOD); | |
1684 | break; | |
1685 | } | |
1686 | case '\\': { | |
1687 | /* The E5.1 specification does not seem to allow IdentifierPart characters | |
1688 | * to be used as identity escapes. Unfortunately this includes '$', which | |
1689 | * cannot be escaped as '\$'; it needs to be escaped e.g. as '\u0024'. | |
1690 | * Many other implementations (including V8 and Rhino, for instance) do | |
1691 | * accept '\$' as a valid identity escape, which is quite pragmatic. | |
1692 | * See: test-regexp-identity-escape-dollar.js. | |
1693 | */ | |
1694 | ||
1695 | advtok = DUK__ADVTOK(2, DUK_RETOK_ATOM_CHAR); /* default: char escape (two chars) */ | |
1696 | if (y == 'b') { | |
1697 | advtok = DUK__ADVTOK(2, DUK_RETOK_ASSERT_WORD_BOUNDARY); | |
1698 | } else if (y == 'B') { | |
1699 | advtok = DUK__ADVTOK(2, DUK_RETOK_ASSERT_NOT_WORD_BOUNDARY); | |
1700 | } else if (y == 'f') { | |
1701 | out_token->num = 0x000c; | |
1702 | } else if (y == 'n') { | |
1703 | out_token->num = 0x000a; | |
1704 | } else if (y == 't') { | |
1705 | out_token->num = 0x0009; | |
1706 | } else if (y == 'r') { | |
1707 | out_token->num = 0x000d; | |
1708 | } else if (y == 'v') { | |
1709 | out_token->num = 0x000b; | |
1710 | } else if (y == 'c') { | |
1711 | x = DUK__L2(); | |
1712 | if ((x >= 'a' && x <= 'z') || | |
1713 | (x >= 'A' && x <= 'Z')) { | |
1714 | out_token->num = (x % 32); | |
1715 | advtok = DUK__ADVTOK(3, DUK_RETOK_ATOM_CHAR); | |
1716 | } else { | |
1717 | DUK_ERROR_SYNTAX(lex_ctx->thr, "invalid regexp escape"); | |
1718 | } | |
1719 | } else if (y == 'x') { | |
1720 | out_token->num = duk__decode_hexesc_from_window(lex_ctx, 2); | |
1721 | advtok = DUK__ADVTOK(4, DUK_RETOK_ATOM_CHAR); | |
1722 | } else if (y == 'u') { | |
1723 | out_token->num = duk__decode_uniesc_from_window(lex_ctx, 2); | |
1724 | advtok = DUK__ADVTOK(6, DUK_RETOK_ATOM_CHAR); | |
1725 | } else if (y == 'd') { | |
1726 | advtok = DUK__ADVTOK(2, DUK_RETOK_ATOM_DIGIT); | |
1727 | } else if (y == 'D') { | |
1728 | advtok = DUK__ADVTOK(2, DUK_RETOK_ATOM_NOT_DIGIT); | |
1729 | } else if (y == 's') { | |
1730 | advtok = DUK__ADVTOK(2, DUK_RETOK_ATOM_WHITE); | |
1731 | } else if (y == 'S') { | |
1732 | advtok = DUK__ADVTOK(2, DUK_RETOK_ATOM_NOT_WHITE); | |
1733 | } else if (y == 'w') { | |
1734 | advtok = DUK__ADVTOK(2, DUK_RETOK_ATOM_WORD_CHAR); | |
1735 | } else if (y == 'W') { | |
1736 | advtok = DUK__ADVTOK(2, DUK_RETOK_ATOM_NOT_WORD_CHAR); | |
1737 | } else if (DUK__ISDIGIT(y)) { | |
1738 | /* E5 Section 15.10.2.11 */ | |
1739 | if (y == '0') { | |
1740 | if (DUK__ISDIGIT(DUK__L2())) { | |
1741 | DUK_ERROR_SYNTAX(lex_ctx->thr, "invalid regexp escape"); | |
1742 | } | |
1743 | out_token->num = 0x0000; | |
1744 | advtok = DUK__ADVTOK(2, DUK_RETOK_ATOM_CHAR); | |
1745 | } else { | |
1746 | /* XXX: shared parsing? */ | |
1747 | duk_uint_fast32_t val = 0; | |
1748 | duk_small_int_t i; | |
1749 | for (i = 0; ; i++) { | |
1750 | if (i >= DUK__MAX_RE_DECESC_DIGITS) { | |
1751 | DUK_ERROR_SYNTAX(lex_ctx->thr, "invalid regexp escape"); | |
1752 | } | |
1753 | DUK__ADVANCECHARS(lex_ctx, 1); /* eat backslash on entry */ | |
1754 | x = DUK__L0(); | |
1755 | if (!DUK__ISDIGIT(x)) { | |
1756 | break; | |
1757 | } | |
1758 | val = val * 10 + (duk_uint_fast32_t) duk__hexval(lex_ctx, x); | |
1759 | } | |
1760 | /* DUK__L0() cannot be a digit, because the loop doesn't terminate if it is */ | |
1761 | advtok = DUK__ADVTOK(0, DUK_RETOK_ATOM_BACKREFERENCE); | |
1762 | out_token->num = val; | |
1763 | } | |
1764 | } else if ((y >= 0 && !duk_unicode_is_identifier_part(y)) || | |
1765 | #if defined(DUK_USE_NONSTD_REGEXP_DOLLAR_ESCAPE) | |
1766 | y == '$' || | |
1767 | #endif | |
1768 | y == DUK_UNICODE_CP_ZWNJ || | |
1769 | y == DUK_UNICODE_CP_ZWJ) { | |
1770 | /* IdentityEscape, with dollar added as a valid additional | |
1771 | * non-standard escape (see test-regexp-identity-escape-dollar.js). | |
1772 | * Careful not to match end-of-buffer (<0) here. | |
1773 | */ | |
1774 | out_token->num = y; | |
1775 | } else { | |
1776 | DUK_ERROR_SYNTAX(lex_ctx->thr, "invalid regexp escape"); | |
1777 | } | |
1778 | break; | |
1779 | } | |
1780 | case '(': { | |
1781 | /* XXX: naming is inconsistent: ATOM_END_GROUP ends an ASSERT_START_LOOKAHEAD */ | |
1782 | ||
1783 | if (y == '?') { | |
1784 | if (DUK__L2() == '=') { | |
1785 | /* (?= */ | |
1786 | advtok = DUK__ADVTOK(3, DUK_RETOK_ASSERT_START_POS_LOOKAHEAD); | |
1787 | } else if (DUK__L2() == '!') { | |
1788 | /* (?! */ | |
1789 | advtok = DUK__ADVTOK(3, DUK_RETOK_ASSERT_START_NEG_LOOKAHEAD); | |
1790 | } else if (DUK__L2() == ':') { | |
1791 | /* (?: */ | |
1792 | advtok = DUK__ADVTOK(3, DUK_RETOK_ATOM_START_NONCAPTURE_GROUP); | |
1793 | } else { | |
1794 | DUK_ERROR_SYNTAX(lex_ctx->thr, "invalid regexp group"); | |
1795 | return; | |
1796 | } | |
1797 | } else { | |
1798 | /* ( */ | |
1799 | advtok = DUK__ADVTOK(1, DUK_RETOK_ATOM_START_CAPTURE_GROUP); | |
1800 | } | |
1801 | break; | |
1802 | } | |
1803 | case ')': { | |
1804 | advtok = DUK__ADVTOK(1, DUK_RETOK_ATOM_END_GROUP); | |
1805 | break; | |
1806 | } | |
1807 | case '[': { | |
1808 | /* | |
1809 | * To avoid creating a heavy intermediate value for the list of ranges, | |
1810 | * only the start token ('[' or '[^') is parsed here. The regexp | |
1811 | * compiler parses the ranges itself. | |
1812 | */ | |
1813 | advtok = DUK__ADVTOK(1, DUK_RETOK_ATOM_START_CHARCLASS); | |
1814 | if (y == '^') { | |
1815 | advtok = DUK__ADVTOK(2, DUK_RETOK_ATOM_START_CHARCLASS_INVERTED); | |
1816 | } | |
1817 | break; | |
1818 | } | |
1819 | #if !defined(DUK_USE_ES6_REGEXP_BRACES) | |
1820 | case '}': | |
1821 | #endif | |
1822 | case ']': { | |
1823 | /* Although these could be parsed as PatternCharacters unambiguously (here), | |
1824 | * E5 Section 15.10.1 grammar explicitly forbids these as PatternCharacters. | |
1825 | */ | |
1826 | DUK_ERROR_SYNTAX(lex_ctx->thr, "invalid regexp character"); | |
1827 | break; | |
1828 | } | |
1829 | case -1: { | |
1830 | /* EOF */ | |
1831 | advtok = DUK__ADVTOK(0, DUK_TOK_EOF); | |
1832 | break; | |
1833 | } | |
1834 | default: { | |
1835 | /* PatternCharacter, all excluded characters are matched by cases above */ | |
1836 | advtok = DUK__ADVTOK(1, DUK_RETOK_ATOM_CHAR); | |
1837 | out_token->num = x; | |
1838 | break; | |
1839 | } | |
1840 | } | |
1841 | ||
1842 | /* | |
1843 | * Shared exit path | |
1844 | */ | |
1845 | ||
1846 | DUK__ADVANCEBYTES(lex_ctx, advtok >> 8); | |
1847 | out_token->t = advtok & 0xff; | |
1848 | } | |
1849 | ||
1850 | /* | |
1851 | * Special parser for character classes; calls callback for every | |
1852 | * range parsed and returns the number of ranges present. | |
1853 | */ | |
1854 | ||
1855 | /* XXX: this duplicates functionality in duk_regexp.c where a similar loop is | |
1856 | * required anyway. We could use that BUT we need to update the regexp compiler | |
1857 | * 'nranges' too. Work this out a bit more cleanly to save space. | |
1858 | */ | |
1859 | ||
1860 | /* XXX: the handling of character range detection is a bit convoluted. | |
1861 | * Try to simplify and make smaller. | |
1862 | */ | |
1863 | ||
1864 | /* XXX: logic for handling character ranges is now incorrect, it will accept | |
1865 | * e.g. [\d-z] whereas it should croak from it? SMJS accepts this too, though. | |
1866 | * | |
1867 | * Needs a read through and a lot of additional tests. | |
1868 | */ | |
1869 | ||
1870 | DUK_LOCAL | |
1871 | void duk__emit_u16_direct_ranges(duk_lexer_ctx *lex_ctx, | |
1872 | duk_re_range_callback gen_range, | |
1873 | void *userdata, | |
1874 | const duk_uint16_t *ranges, | |
1875 | duk_small_int_t num) { | |
1876 | const duk_uint16_t *ranges_end; | |
1877 | ||
1878 | DUK_UNREF(lex_ctx); | |
1879 | ||
1880 | ranges_end = ranges + num; | |
1881 | while (ranges < ranges_end) { | |
1882 | /* mark range 'direct', bypass canonicalization (see Wiki) */ | |
1883 | gen_range(userdata, (duk_codepoint_t) ranges[0], (duk_codepoint_t) ranges[1], 1); | |
1884 | ranges += 2; | |
1885 | } | |
1886 | } | |
1887 | ||
1888 | DUK_INTERNAL void duk_lexer_parse_re_ranges(duk_lexer_ctx *lex_ctx, duk_re_range_callback gen_range, void *userdata) { | |
1889 | duk_codepoint_t start = -1; | |
1890 | duk_codepoint_t ch; | |
1891 | duk_codepoint_t x; | |
1892 | duk_bool_t dash = 0; | |
1893 | ||
1894 | DUK_DD(DUK_DDPRINT("parsing regexp ranges")); | |
1895 | ||
1896 | for (;;) { | |
1897 | x = DUK__L0(); | |
1898 | DUK__ADVANCECHARS(lex_ctx, 1); | |
1899 | ||
1900 | ch = -1; /* not strictly necessary, but avoids "uninitialized variable" warnings */ | |
1901 | DUK_UNREF(ch); | |
1902 | ||
1903 | if (x < 0) { | |
1904 | DUK_ERROR_SYNTAX(lex_ctx->thr, "eof in character class"); | |
1905 | } else if (x == ']') { | |
1906 | if (start >= 0) { | |
1907 | gen_range(userdata, start, start, 0); | |
1908 | } | |
1909 | break; | |
1910 | } else if (x == '-') { | |
1911 | if (start >= 0 && !dash && DUK__L0() != ']') { | |
1912 | /* '-' as a range indicator */ | |
1913 | dash = 1; | |
1914 | continue; | |
1915 | } else { | |
1916 | /* '-' verbatim */ | |
1917 | ch = x; | |
1918 | } | |
1919 | } else if (x == '\\') { | |
1920 | /* | |
1921 | * The escapes are same as outside a character class, except that \b has a | |
1922 | * different meaning, and \B and backreferences are prohibited (see E5 | |
1923 | * Section 15.10.2.19). However, it's difficult to share code because we | |
1924 | * handle e.g. "\n" very differently: here we generate a single character | |
1925 | * range for it. | |
1926 | */ | |
1927 | ||
1928 | x = DUK__L0(); | |
1929 | DUK__ADVANCECHARS(lex_ctx, 1); | |
1930 | ||
1931 | if (x == 'b') { | |
1932 | /* Note: '\b' in char class is different than outside (assertion), | |
1933 | * '\B' is not allowed and is caught by the duk_unicode_is_identifier_part() | |
1934 | * check below. | |
1935 | */ | |
1936 | ch = 0x0008; | |
1937 | } else if (x == 'f') { | |
1938 | ch = 0x000c; | |
1939 | } else if (x == 'n') { | |
1940 | ch = 0x000a; | |
1941 | } else if (x == 't') { | |
1942 | ch = 0x0009; | |
1943 | } else if (x == 'r') { | |
1944 | ch = 0x000d; | |
1945 | } else if (x == 'v') { | |
1946 | ch = 0x000b; | |
1947 | } else if (x == 'c') { | |
1948 | x = DUK__L0(); | |
1949 | DUK__ADVANCECHARS(lex_ctx, 1); | |
1950 | if ((x >= 'a' && x <= 'z') || | |
1951 | (x >= 'A' && x <= 'Z')) { | |
1952 | ch = (x % 32); | |
1953 | } else { | |
1954 | DUK_ERROR_SYNTAX(lex_ctx->thr, "invalid regexp escape"); | |
1955 | return; /* never reached, but avoids warnings of | |
1956 | * potentially unused variables. | |
1957 | */ | |
1958 | } | |
1959 | } else if (x == 'x') { | |
1960 | ch = duk__decode_hexesc_from_window(lex_ctx, 0); | |
1961 | DUK__ADVANCECHARS(lex_ctx, 2); | |
1962 | } else if (x == 'u') { | |
1963 | ch = duk__decode_uniesc_from_window(lex_ctx, 0); | |
1964 | DUK__ADVANCECHARS(lex_ctx, 4); | |
1965 | } else if (x == 'd') { | |
1966 | duk__emit_u16_direct_ranges(lex_ctx, | |
1967 | gen_range, | |
1968 | userdata, | |
1969 | duk_unicode_re_ranges_digit, | |
1970 | sizeof(duk_unicode_re_ranges_digit) / sizeof(duk_uint16_t)); | |
1971 | ch = -1; | |
1972 | } else if (x == 'D') { | |
1973 | duk__emit_u16_direct_ranges(lex_ctx, | |
1974 | gen_range, | |
1975 | userdata, | |
1976 | duk_unicode_re_ranges_not_digit, | |
1977 | sizeof(duk_unicode_re_ranges_not_digit) / sizeof(duk_uint16_t)); | |
1978 | ch = -1; | |
1979 | } else if (x == 's') { | |
1980 | duk__emit_u16_direct_ranges(lex_ctx, | |
1981 | gen_range, | |
1982 | userdata, | |
1983 | duk_unicode_re_ranges_white, | |
1984 | sizeof(duk_unicode_re_ranges_white) / sizeof(duk_uint16_t)); | |
1985 | ch = -1; | |
1986 | } else if (x == 'S') { | |
1987 | duk__emit_u16_direct_ranges(lex_ctx, | |
1988 | gen_range, | |
1989 | userdata, | |
1990 | duk_unicode_re_ranges_not_white, | |
1991 | sizeof(duk_unicode_re_ranges_not_white) / sizeof(duk_uint16_t)); | |
1992 | ch = -1; | |
1993 | } else if (x == 'w') { | |
1994 | duk__emit_u16_direct_ranges(lex_ctx, | |
1995 | gen_range, | |
1996 | userdata, | |
1997 | duk_unicode_re_ranges_wordchar, | |
1998 | sizeof(duk_unicode_re_ranges_wordchar) / sizeof(duk_uint16_t)); | |
1999 | ch = -1; | |
2000 | } else if (x == 'W') { | |
2001 | duk__emit_u16_direct_ranges(lex_ctx, | |
2002 | gen_range, | |
2003 | userdata, | |
2004 | duk_unicode_re_ranges_not_wordchar, | |
2005 | sizeof(duk_unicode_re_ranges_not_wordchar) / sizeof(duk_uint16_t)); | |
2006 | ch = -1; | |
2007 | } else if (DUK__ISDIGIT(x)) { | |
2008 | /* DecimalEscape, only \0 is allowed, no leading zeroes are allowed */ | |
2009 | if (x == '0' && !DUK__ISDIGIT(DUK__L0())) { | |
2010 | ch = 0x0000; | |
2011 | } else { | |
2012 | DUK_ERROR_SYNTAX(lex_ctx->thr, "invalid regexp escape"); | |
2013 | } | |
2014 | } else if (!duk_unicode_is_identifier_part(x) | |
2015 | #if defined(DUK_USE_NONSTD_REGEXP_DOLLAR_ESCAPE) | |
2016 | || x == '$' | |
2017 | #endif | |
2018 | ) { | |
2019 | /* IdentityEscape */ | |
2020 | ch = x; | |
2021 | } else { | |
2022 | DUK_ERROR_SYNTAX(lex_ctx->thr, "invalid regexp escape"); | |
2023 | } | |
2024 | } else { | |
2025 | /* character represents itself */ | |
2026 | ch = x; | |
2027 | } | |
2028 | ||
2029 | /* ch is a literal character here or -1 if parsed entity was | |
2030 | * an escape such as "\s". | |
2031 | */ | |
2032 | ||
2033 | if (ch < 0) { | |
2034 | /* multi-character sets not allowed as part of ranges, see | |
2035 | * E5 Section 15.10.2.15, abstract operation CharacterRange. | |
2036 | */ | |
2037 | if (start >= 0) { | |
2038 | if (dash) { | |
2039 | DUK_ERROR_SYNTAX(lex_ctx->thr, "invalid range"); | |
2040 | } else { | |
2041 | gen_range(userdata, start, start, 0); | |
2042 | start = -1; | |
2043 | /* dash is already 0 */ | |
2044 | } | |
2045 | } | |
2046 | } else { | |
2047 | if (start >= 0) { | |
2048 | if (dash) { | |
2049 | if (start > ch) { | |
2050 | DUK_ERROR_SYNTAX(lex_ctx->thr, "invalid range"); | |
2051 | } | |
2052 | gen_range(userdata, start, ch, 0); | |
2053 | start = -1; | |
2054 | dash = 0; | |
2055 | } else { | |
2056 | gen_range(userdata, start, start, 0); | |
2057 | start = ch; | |
2058 | /* dash is already 0 */ | |
2059 | } | |
2060 | } else { | |
2061 | start = ch; | |
2062 | } | |
2063 | } | |
2064 | } | |
2065 | ||
2066 | return; | |
2067 | } | |
2068 | ||
2069 | #endif /* DUK_USE_REGEXP_SUPPORT */ |