Line data Source code
1 : // Copyright 2011 the V8 project authors. All rights reserved.
2 : // Use of this source code is governed by a BSD-style license that can be
3 : // found in the LICENSE file.
4 :
5 : #include "src/parsing/scanner-character-streams.h"
6 :
7 : #include "include/v8.h"
8 : #include "src/counters.h"
9 : #include "src/globals.h"
10 : #include "src/handles.h"
11 : #include "src/objects-inl.h"
12 : #include "src/parsing/scanner.h"
13 : #include "src/unicode-inl.h"
14 :
15 : namespace v8 {
16 : namespace internal {
17 :
18 : namespace {
19 : const unibrow::uchar kUtf8Bom = 0xfeff;
20 : } // namespace
21 :
22 : // ----------------------------------------------------------------------------
23 : // BufferedUtf16CharacterStreams
24 : //
25 : // A buffered character stream based on a random access character
26 : // source (ReadBlock can be called with pos() pointing to any position,
27 : // even positions before the current).
28 4402128 : class BufferedUtf16CharacterStream : public Utf16CharacterStream {
29 : public:
30 : BufferedUtf16CharacterStream();
31 :
32 : protected:
33 : static const size_t kBufferSize = 512;
34 :
35 : bool ReadBlock() override;
36 :
37 : // FillBuffer should read up to kBufferSize characters at position and store
38 : // them into buffer_[0..]. It returns the number of characters stored.
39 : virtual size_t FillBuffer(size_t position) = 0;
40 :
41 : // Fixed sized buffer that this class reads from.
42 : // The base class' buffer_start_ should always point to buffer_.
43 : uc16 buffer_[kBufferSize];
44 : };
45 :
46 0 : BufferedUtf16CharacterStream::BufferedUtf16CharacterStream()
47 8804274 : : Utf16CharacterStream(buffer_, buffer_, buffer_, 0) {}
48 :
49 12781097 : bool BufferedUtf16CharacterStream::ReadBlock() {
50 : DCHECK_EQ(buffer_start_, buffer_);
51 :
52 : size_t position = pos();
53 12781097 : buffer_pos_ = position;
54 12781097 : buffer_cursor_ = buffer_;
55 12781097 : buffer_end_ = buffer_ + FillBuffer(position);
56 : DCHECK_EQ(pos(), position);
57 : DCHECK_LE(buffer_end_, buffer_start_ + kBufferSize);
58 12781112 : return buffer_cursor_ < buffer_end_;
59 : }
60 :
61 : // ----------------------------------------------------------------------------
62 : // GenericStringUtf16CharacterStream.
63 : //
64 : // A stream w/ a data source being a (flattened) Handle<String>.
65 :
66 7871900 : class GenericStringUtf16CharacterStream : public BufferedUtf16CharacterStream {
67 : public:
68 : GenericStringUtf16CharacterStream(Handle<String> data, size_t start_position,
69 : size_t end_position);
70 :
71 : protected:
72 : size_t FillBuffer(size_t position) override;
73 :
74 : Handle<String> string_;
75 : size_t length_;
76 : };
77 :
78 0 : GenericStringUtf16CharacterStream::GenericStringUtf16CharacterStream(
79 : Handle<String> data, size_t start_position, size_t end_position)
80 3935957 : : string_(data), length_(end_position) {
81 : DCHECK_GE(end_position, start_position);
82 : DCHECK_GE(static_cast<size_t>(string_->length()),
83 : end_position - start_position);
84 3935957 : buffer_pos_ = start_position;
85 0 : }
86 :
87 10991702 : size_t GenericStringUtf16CharacterStream::FillBuffer(size_t from_pos) {
88 10991702 : if (from_pos >= length_) return 0;
89 :
90 7808877 : size_t length = i::Min(kBufferSize, length_ - from_pos);
91 : String::WriteToFlat<uc16>(*string_, buffer_, static_cast<int>(from_pos),
92 15617754 : static_cast<int>(from_pos + length));
93 7808882 : return length;
94 : }
95 :
96 : // ----------------------------------------------------------------------------
97 : // ExternalTwoByteStringUtf16CharacterStream.
98 : //
99 : // A stream whose data source is a Handle<ExternalTwoByteString>. It avoids
100 : // all data copying.
101 :
102 164 : class ExternalTwoByteStringUtf16CharacterStream : public Utf16CharacterStream {
103 : public:
104 : ExternalTwoByteStringUtf16CharacterStream(Handle<ExternalTwoByteString> data,
105 : size_t start_position,
106 : size_t end_position);
107 :
108 : private:
109 : bool ReadBlock() override;
110 :
111 : const uc16* raw_data_; // Pointer to the actual array of characters.
112 : size_t start_pos_;
113 : size_t end_pos_;
114 : };
115 :
116 82 : ExternalTwoByteStringUtf16CharacterStream::
117 : ExternalTwoByteStringUtf16CharacterStream(
118 : Handle<ExternalTwoByteString> data, size_t start_position,
119 : size_t end_position)
120 164 : : raw_data_(data->GetTwoByteData(static_cast<int>(start_position))),
121 : start_pos_(start_position),
122 164 : end_pos_(end_position) {
123 82 : buffer_start_ = raw_data_;
124 82 : buffer_cursor_ = raw_data_;
125 82 : buffer_end_ = raw_data_ + (end_pos_ - start_pos_);
126 82 : buffer_pos_ = start_pos_;
127 82 : }
128 :
129 182 : bool ExternalTwoByteStringUtf16CharacterStream::ReadBlock() {
130 : size_t position = pos();
131 182 : bool have_data = start_pos_ <= position && position < end_pos_;
132 182 : if (have_data) {
133 35 : buffer_pos_ = start_pos_;
134 35 : buffer_cursor_ = raw_data_ + (position - start_pos_),
135 35 : buffer_end_ = raw_data_ + (end_pos_ - start_pos_);
136 : } else {
137 147 : buffer_pos_ = position;
138 147 : buffer_cursor_ = raw_data_;
139 147 : buffer_end_ = raw_data_;
140 : }
141 182 : return have_data;
142 : }
143 :
144 : // ----------------------------------------------------------------------------
145 : // ExternalOneByteStringUtf16CharacterStream
146 : //
147 : // A stream whose data source is a Handle<ExternalOneByteString>.
148 :
149 931308 : class ExternalOneByteStringUtf16CharacterStream
150 : : public BufferedUtf16CharacterStream {
151 : public:
152 : ExternalOneByteStringUtf16CharacterStream(Handle<ExternalOneByteString> data,
153 : size_t start_position,
154 : size_t end_position);
155 :
156 : // For testing:
157 : ExternalOneByteStringUtf16CharacterStream(const char* data, size_t length);
158 :
159 : protected:
160 : size_t FillBuffer(size_t position) override;
161 :
162 : const uint8_t* raw_data_; // Pointer to the actual array of characters.
163 : size_t length_;
164 : };
165 :
166 0 : ExternalOneByteStringUtf16CharacterStream::
167 : ExternalOneByteStringUtf16CharacterStream(
168 : Handle<ExternalOneByteString> data, size_t start_position,
169 : size_t end_position)
170 927233 : : raw_data_(data->GetChars()), length_(end_position) {
171 : DCHECK(end_position >= start_position);
172 463615 : buffer_pos_ = start_position;
173 0 : }
174 :
175 0 : ExternalOneByteStringUtf16CharacterStream::
176 : ExternalOneByteStringUtf16CharacterStream(const char* data, size_t length)
177 2038 : : raw_data_(reinterpret_cast<const uint8_t*>(data)), length_(length) {}
178 :
179 1702942 : size_t ExternalOneByteStringUtf16CharacterStream::FillBuffer(size_t from_pos) {
180 1702942 : if (from_pos >= length_) return 0;
181 :
182 1282523 : size_t length = Min(kBufferSize, length_ - from_pos);
183 1282523 : i::CopyCharsUnsigned(buffer_, raw_data_ + from_pos, length);
184 : return length;
185 : }
186 :
187 : // ----------------------------------------------------------------------------
188 : // Utf8ExternalStreamingStream - chunked streaming of Utf-8 data.
189 : //
190 : // This implementation is fairly complex, since data arrives in chunks which
191 : // may 'cut' arbitrarily into utf-8 characters. Also, seeking to a given
192 : // character position is tricky because the byte position cannot be dericed
193 : // from the character position.
194 :
195 : class Utf8ExternalStreamingStream : public BufferedUtf16CharacterStream {
196 : public:
197 : Utf8ExternalStreamingStream(
198 : ScriptCompiler::ExternalSourceStream* source_stream,
199 : RuntimeCallStats* stats)
200 : : current_({0, {0, 0, unibrow::Utf8::Utf8IncrementalBuffer(0)}}),
201 : source_stream_(source_stream),
202 800 : stats_(stats) {}
203 800 : ~Utf8ExternalStreamingStream() override {
204 4628 : for (size_t i = 0; i < chunks_.size(); i++) delete[] chunks_[i].data;
205 800 : }
206 :
207 : protected:
208 : size_t FillBuffer(size_t position) override;
209 :
210 : private:
211 : // A position within the data stream. It stores:
212 : // - The 'physical' position (# of bytes in the stream),
213 : // - the 'logical' position (# of ucs-2 characters, also within the stream),
214 : // - a possibly incomplete utf-8 char at the current 'physical' position.
215 : struct StreamPosition {
216 : size_t bytes;
217 : size_t chars;
218 : unibrow::Utf8::Utf8IncrementalBuffer incomplete_char;
219 : };
220 :
221 : // Position contains a StreamPosition and the index of the chunk the position
222 : // points into. (The chunk_no could be derived from pos, but that'd be
223 : // an expensive search through all chunks.)
224 : struct Position {
225 : size_t chunk_no;
226 : StreamPosition pos;
227 : };
228 :
229 : // A chunk in the list of chunks, containing:
230 : // - The chunk data (data pointer and length), and
231 : // - the position at the first byte of the chunk.
232 : struct Chunk {
233 : const uint8_t* data;
234 : size_t length;
235 : StreamPosition start;
236 : };
237 :
238 : // Within the current chunk, skip forward from current_ towards position.
239 : bool SkipToPosition(size_t position);
240 : // Within the current chunk, fill the buffer_ (while it has capacity).
241 : void FillBufferFromCurrentChunk();
242 : // Fetch a new chunk (assuming current_ is at the end of the current data).
243 : bool FetchChunk();
244 : // Search through the chunks and set current_ to point to the given position.
245 : // (This call is potentially expensive.)
246 : void SearchPosition(size_t position);
247 :
248 : std::vector<Chunk> chunks_;
249 : Position current_;
250 : ScriptCompiler::ExternalSourceStream* source_stream_;
251 : RuntimeCallStats* stats_;
252 : };
253 :
254 21469 : bool Utf8ExternalStreamingStream::SkipToPosition(size_t position) {
255 : DCHECK_LE(current_.pos.chars, position); // We can only skip forward.
256 :
257 : // Already there? Then return immediately.
258 21469 : if (current_.pos.chars == position) return true;
259 :
260 21462 : const Chunk& chunk = chunks_[current_.chunk_no];
261 : DCHECK(current_.pos.bytes >= chunk.start.bytes);
262 :
263 : unibrow::Utf8::Utf8IncrementalBuffer incomplete_char =
264 21462 : chunk.start.incomplete_char;
265 21462 : size_t it = current_.pos.bytes - chunk.start.bytes;
266 21462 : size_t chars = chunk.start.chars;
267 54867701 : while (it < chunk.length && chars < position) {
268 : unibrow::uchar t =
269 54824777 : unibrow::Utf8::ValueOfIncremental(chunk.data[it], &incomplete_char);
270 54824777 : if (t == kUtf8Bom && current_.pos.chars == 0) {
271 : // BOM detected at beginning of the stream. Don't copy it.
272 54824770 : } else if (t != unibrow::Utf8::kIncomplete) {
273 54824735 : chars++;
274 54824735 : if (t > unibrow::Utf16::kMaxNonSurrogateCharCode) chars++;
275 : }
276 54824777 : it++;
277 : }
278 :
279 21462 : current_.pos.bytes += it;
280 21462 : current_.pos.chars = chars;
281 21462 : current_.pos.incomplete_char = incomplete_char;
282 21462 : current_.chunk_no += (it == chunk.length);
283 :
284 21462 : return current_.pos.chars == position;
285 : }
286 :
287 45307 : void Utf8ExternalStreamingStream::FillBufferFromCurrentChunk() {
288 : DCHECK_LT(current_.chunk_no, chunks_.size());
289 : DCHECK_EQ(buffer_start_, buffer_cursor_);
290 : DCHECK_LT(buffer_end_ + 1, buffer_start_ + kBufferSize);
291 :
292 45307 : const Chunk& chunk = chunks_[current_.chunk_no];
293 :
294 : // The buffer_ is writable, but buffer_*_ members are const. So we get a
295 : // non-const pointer into buffer that points to the same char as buffer_end_.
296 45307 : uint16_t* cursor = buffer_ + (buffer_end_ - buffer_start_);
297 : DCHECK_EQ(cursor, buffer_end_);
298 :
299 : // If the current chunk is the last (empty) chunk we'll have to process
300 : // any left-over, partial characters.
301 45307 : if (chunk.length == 0) {
302 : unibrow::uchar t =
303 380 : unibrow::Utf8::ValueOfIncrementalFinish(¤t_.pos.incomplete_char);
304 380 : if (t != unibrow::Utf8::kBufferEmpty) {
305 : DCHECK(t < unibrow::Utf16::kMaxNonSurrogateCharCode);
306 21 : *cursor = static_cast<uc16>(t);
307 21 : buffer_end_++;
308 21 : current_.pos.chars++;
309 : }
310 380 : return;
311 : }
312 :
313 : unibrow::Utf8::Utf8IncrementalBuffer incomplete_char =
314 44927 : current_.pos.incomplete_char;
315 : size_t it;
316 21979730 : for (it = current_.pos.bytes - chunk.start.bytes;
317 21952821 : it < chunk.length && cursor + 1 < buffer_start_ + kBufferSize; it++) {
318 : unibrow::uchar t =
319 10944938 : unibrow::Utf8::ValueOfIncremental(chunk.data[it], &incomplete_char);
320 10944938 : if (t == unibrow::Utf8::kIncomplete) continue;
321 10943250 : if (V8_LIKELY(t < kUtf8Bom)) {
322 10942960 : *(cursor++) = static_cast<uc16>(t); // The by most frequent case.
323 290 : } else if (t == kUtf8Bom && current_.pos.bytes + it == 2) {
324 : // BOM detected at beginning of the stream. Don't copy it.
325 262 : } else if (t <= unibrow::Utf16::kMaxNonSurrogateCharCode) {
326 38 : *(cursor++) = static_cast<uc16>(t);
327 : } else {
328 224 : *(cursor++) = unibrow::Utf16::LeadSurrogate(t);
329 448 : *(cursor++) = unibrow::Utf16::TrailSurrogate(t);
330 : }
331 : }
332 :
333 44927 : current_.pos.bytes = chunk.start.bytes + it;
334 44927 : current_.pos.chars += (cursor - buffer_end_);
335 44927 : current_.pos.incomplete_char = incomplete_char;
336 44927 : current_.chunk_no += (it == chunk.length);
337 :
338 44927 : buffer_end_ = cursor;
339 : }
340 :
341 1914 : bool Utf8ExternalStreamingStream::FetchChunk() {
342 1914 : RuntimeCallTimerScope scope(stats_, &RuntimeCallStats::GetMoreDataCallback);
343 : DCHECK_EQ(current_.chunk_no, chunks_.size());
344 : DCHECK(chunks_.empty() || chunks_.back().length != 0);
345 :
346 1914 : const uint8_t* chunk = nullptr;
347 1914 : size_t length = source_stream_->GetMoreData(&chunk);
348 3828 : chunks_.push_back({chunk, length, current_.pos});
349 3828 : return length > 0;
350 : }
351 :
352 45442 : void Utf8ExternalStreamingStream::SearchPosition(size_t position) {
353 : // If current_ already points to the right position, we're done.
354 : //
355 : // This is expected to be the common case, since we typically call
356 : // FillBuffer right after the current buffer.
357 45442 : if (current_.pos.chars == position) return;
358 :
359 : // No chunks. Fetch at least one, so we can assume !chunks_.empty() below.
360 86478 : if (chunks_.empty()) {
361 : DCHECK_EQ(current_.chunk_no, 0u);
362 : DCHECK_EQ(current_.pos.bytes, 0u);
363 : DCHECK_EQ(current_.pos.chars, 0u);
364 0 : FetchChunk();
365 : }
366 :
367 : // Search for the last chunk whose start position is less or equal to
368 : // position.
369 43239 : size_t chunk_no = chunks_.size() - 1;
370 996352 : while (chunk_no > 0 && chunks_[chunk_no].start.chars > position) {
371 444094 : chunk_no--;
372 : }
373 :
374 : // Did we find the terminating (zero-length) chunk? Then we're seeking
375 : // behind the end of the data, and position does not exist.
376 : // Set current_ to point to the terminating chunk.
377 43239 : if (chunks_[chunk_no].length == 0) {
378 140 : current_ = {chunk_no, chunks_[chunk_no].start};
379 140 : return;
380 : }
381 :
382 : // Did we find the non-last chunk? Then our position must be within chunk_no.
383 43099 : if (chunk_no + 1 < chunks_.size()) {
384 : // Fancy-pants optimization for ASCII chunks within a utf-8 stream.
385 : // (Many web sites declare utf-8 encoding, but use only (or almost only) the
386 : // ASCII subset for their JavaScript sources. We can exploit this, by
387 : // checking whether the # bytes in a chunk are equal to the # chars, and if
388 : // so avoid the expensive SkipToPosition.)
389 : bool ascii_only_chunk =
390 21644 : (chunks_[chunk_no + 1].start.bytes - chunks_[chunk_no].start.bytes) ==
391 21644 : (chunks_[chunk_no + 1].start.chars - chunks_[chunk_no].start.chars);
392 21644 : if (ascii_only_chunk) {
393 21630 : size_t skip = position - chunks_[chunk_no].start.chars;
394 : current_ = {chunk_no,
395 21630 : {chunks_[chunk_no].start.bytes + skip,
396 21630 : chunks_[chunk_no].start.chars + skip,
397 21630 : unibrow::Utf8::Utf8IncrementalBuffer(0)}};
398 : } else {
399 14 : current_ = {chunk_no, chunks_[chunk_no].start};
400 14 : SkipToPosition(position);
401 : }
402 :
403 : // Since position was within the chunk, SkipToPosition should have found
404 : // something.
405 : DCHECK_EQ(position, current_.pos.chars);
406 : return;
407 : }
408 :
409 : // What's left: We're in the last, non-terminating chunk. Our position
410 : // may be in the chunk, but it may also be in 'future' chunks, which we'll
411 : // have to obtain.
412 : DCHECK_EQ(chunk_no, chunks_.size() - 1);
413 21455 : current_ = {chunk_no, chunks_[chunk_no].start};
414 : bool have_more_data = true;
415 21455 : bool found = SkipToPosition(position);
416 42910 : while (have_more_data && !found) {
417 : DCHECK_EQ(current_.chunk_no, chunks_.size());
418 0 : have_more_data = FetchChunk();
419 0 : found = have_more_data && SkipToPosition(position);
420 : }
421 :
422 : // We'll return with a postion != the desired position only if we're out
423 : // of data. In that case, we'll point to the terminating chunk.
424 : DCHECK_EQ(found, current_.pos.chars == position);
425 : DCHECK_EQ(have_more_data, chunks_.back().length != 0);
426 : DCHECK_IMPLIES(!found, !have_more_data);
427 : DCHECK_IMPLIES(!found, current_.chunk_no == chunks_.size() - 1);
428 : }
429 :
430 45442 : size_t Utf8ExternalStreamingStream::FillBuffer(size_t position) {
431 45442 : buffer_cursor_ = buffer_;
432 45442 : buffer_end_ = buffer_;
433 :
434 45442 : SearchPosition(position);
435 179794 : bool out_of_data = current_.chunk_no != chunks_.size() &&
436 43603 : chunks_[current_.chunk_no].length == 0;
437 45442 : if (out_of_data) return 0;
438 :
439 : // Fill the buffer, until we have at least one char (or are out of data).
440 : // (The embedder might give us 1-byte blocks within a utf-8 char, so we
441 : // can't guarantee progress with one chunk. Thus we iterate.)
442 90539 : while (!out_of_data && buffer_cursor_ == buffer_end_) {
443 : // At end of current data, but there might be more? Then fetch it.
444 90614 : if (current_.chunk_no == chunks_.size()) {
445 1914 : out_of_data = !FetchChunk();
446 : }
447 45307 : FillBufferFromCurrentChunk();
448 : }
449 :
450 : DCHECK_EQ(current_.pos.chars - position,
451 : static_cast<size_t>(buffer_end_ - buffer_cursor_));
452 45232 : return buffer_end_ - buffer_cursor_;
453 : }
454 :
455 : // ----------------------------------------------------------------------------
456 : // Chunks - helper for One- + TwoByteExternalStreamingStream
457 : namespace {
458 :
459 : struct Chunk {
460 : const uint8_t* data;
461 : size_t byte_length;
462 : size_t byte_pos;
463 : };
464 :
465 : typedef std::vector<struct Chunk> Chunks;
466 :
467 2469 : void DeleteChunks(Chunks& chunks) {
468 4550 : for (size_t i = 0; i < chunks.size(); i++) delete[] chunks[i].data;
469 194 : }
470 :
471 : // Return the chunk index for the chunk containing position.
472 : // If position is behind the end of the stream, the index of the last,
473 : // zero-length chunk is returned.
474 87854 : size_t FindChunk(Chunks& chunks, ScriptCompiler::ExternalSourceStream* source,
475 : size_t position, RuntimeCallStats* stats) {
476 : size_t end_pos =
477 43927 : chunks.empty() ? 0 : (chunks.back().byte_pos + chunks.back().byte_length);
478 :
479 : // Get more data if needed. We usually won't enter the loop body.
480 43927 : bool out_of_data = !chunks.empty() && chunks.back().byte_length == 0;
481 : {
482 43927 : RuntimeCallTimerScope scope(stats, &RuntimeCallStats::GetMoreDataCallback);
483 89935 : while (!out_of_data && end_pos <= position + 1) {
484 2081 : const uint8_t* chunk = nullptr;
485 2081 : size_t len = source->GetMoreData(&chunk);
486 :
487 4162 : chunks.push_back({chunk, len, end_pos});
488 2081 : end_pos += len;
489 2081 : out_of_data = (len == 0);
490 : }
491 : }
492 :
493 : // Here, we should always have at least one chunk, and we either have the
494 : // chunk we were looking for, or we're out of data. Also, out_of_data and
495 : // end_pos are current (and designate whether we have exhausted the stream,
496 : // and the length of data received so far, respectively).
497 : DCHECK(!chunks.empty());
498 : DCHECK_EQ(end_pos, chunks.back().byte_pos + chunks.back().byte_length);
499 : DCHECK_EQ(out_of_data, chunks.back().byte_length == 0);
500 : DCHECK(position < end_pos || out_of_data);
501 :
502 : // Edge case: position is behind the end of stream: Return the last (length 0)
503 : // chunk to indicate the end of the stream.
504 43927 : if (position >= end_pos) {
505 : DCHECK(out_of_data);
506 552 : return chunks.size() - 1;
507 : }
508 :
509 : // We almost always 'stream', meaning we want data from the last chunk, so
510 : // let's look at chunks back-to-front.
511 43375 : size_t chunk_no = chunks.size() - 1;
512 585023 : while (chunks[chunk_no].byte_pos > position) {
513 : DCHECK_NE(chunk_no, 0u);
514 498273 : chunk_no--;
515 : }
516 : DCHECK_LE(chunks[chunk_no].byte_pos, position);
517 : DCHECK_LT(position, chunks[chunk_no].byte_pos + chunks[chunk_no].byte_length);
518 : return chunk_no;
519 : }
520 :
521 : } // anonymous namespace
522 :
523 : // ----------------------------------------------------------------------------
524 : // OneByteExternalStreamingStream
525 : //
526 : // A stream of latin-1 encoded, chunked data.
527 :
528 : class OneByteExternalStreamingStream : public BufferedUtf16CharacterStream {
529 : public:
530 : explicit OneByteExternalStreamingStream(
531 : ScriptCompiler::ExternalSourceStream* source, RuntimeCallStats* stats)
532 248 : : source_(source), stats_(stats) {}
533 372 : ~OneByteExternalStreamingStream() override { DeleteChunks(chunks_); }
534 :
535 : protected:
536 : size_t FillBuffer(size_t position) override;
537 :
538 : private:
539 : Chunks chunks_;
540 : ScriptCompiler::ExternalSourceStream* source_;
541 : RuntimeCallStats* stats_;
542 : };
543 :
544 41015 : size_t OneByteExternalStreamingStream::FillBuffer(size_t position) {
545 41015 : const Chunk& chunk = chunks_[FindChunk(chunks_, source_, position, stats_)];
546 41015 : if (chunk.byte_length == 0) return 0;
547 :
548 40715 : size_t start_pos = position - chunk.byte_pos;
549 40715 : size_t len = i::Min(kBufferSize, chunk.byte_length - start_pos);
550 40715 : i::CopyCharsUnsigned(buffer_, chunk.data + start_pos, len);
551 : return len;
552 : }
553 :
554 : #if !(V8_TARGET_ARCH_MIPS || V8_TARGET_ARCH_MIPS64)
555 : // ----------------------------------------------------------------------------
556 : // TwoByteExternalStreamingStream
557 : //
558 : // A stream of ucs-2 data, delivered in chunks. Chunks may be 'cut' into the
559 : // middle of characters (or even contain only one byte), which adds a bit
560 : // of complexity. This stream avoid all data copying, except for characters
561 : // that cross chunk boundaries.
562 :
563 : class TwoByteExternalStreamingStream : public Utf16CharacterStream {
564 : public:
565 : explicit TwoByteExternalStreamingStream(
566 : ScriptCompiler::ExternalSourceStream* source, RuntimeCallStats* stats);
567 : ~TwoByteExternalStreamingStream() override;
568 :
569 : protected:
570 : bool ReadBlock() override;
571 :
572 : Chunks chunks_;
573 : ScriptCompiler::ExternalSourceStream* source_;
574 : RuntimeCallStats* stats_;
575 : uc16 one_char_buffer_;
576 : };
577 :
578 0 : TwoByteExternalStreamingStream::TwoByteExternalStreamingStream(
579 : ScriptCompiler::ExternalSourceStream* source, RuntimeCallStats* stats)
580 : : Utf16CharacterStream(&one_char_buffer_, &one_char_buffer_,
581 : &one_char_buffer_, 0),
582 : source_(source),
583 : stats_(stats),
584 210 : one_char_buffer_(0) {}
585 :
586 140 : TwoByteExternalStreamingStream::~TwoByteExternalStreamingStream() {
587 70 : DeleteChunks(chunks_);
588 140 : }
589 :
590 2912 : bool TwoByteExternalStreamingStream::ReadBlock() {
591 : size_t position = pos();
592 :
593 : // We'll search for the 2nd byte of our character, to make sure we
594 : // have enough data for at least one character.
595 2912 : size_t chunk_no = FindChunk(chunks_, source_, 2 * position + 1, stats_);
596 :
597 : // Out of data? Return 0.
598 2912 : if (chunks_[chunk_no].byte_length == 0) {
599 252 : buffer_cursor_ = buffer_start_;
600 252 : buffer_end_ = buffer_start_;
601 252 : return false;
602 : }
603 :
604 : Chunk& current = chunks_[chunk_no];
605 :
606 : // Annoying edge case: Chunks may not be 2-byte aligned, meaning that a
607 : // character may be split between the previous and the current chunk.
608 : // If we find such a lonely byte at the beginning of the chunk, we'll use
609 : // one_char_buffer_ to hold the full character.
610 2660 : bool lonely_byte = (chunks_[chunk_no].byte_pos == (2 * position + 1));
611 2660 : if (lonely_byte) {
612 : DCHECK_NE(chunk_no, 0u);
613 889 : Chunk& previous_chunk = chunks_[chunk_no - 1];
614 : #ifdef V8_TARGET_BIG_ENDIAN
615 : uc16 character = current.data[0] |
616 : previous_chunk.data[previous_chunk.byte_length - 1] << 8;
617 : #else
618 889 : uc16 character = previous_chunk.data[previous_chunk.byte_length - 1] |
619 889 : current.data[0] << 8;
620 : #endif
621 :
622 889 : one_char_buffer_ = character;
623 889 : buffer_pos_ = position;
624 889 : buffer_start_ = &one_char_buffer_;
625 889 : buffer_cursor_ = &one_char_buffer_;
626 889 : buffer_end_ = &one_char_buffer_ + 1;
627 889 : return true;
628 : }
629 :
630 : // Common case: character is in current chunk.
631 : DCHECK_LE(current.byte_pos, 2 * position);
632 : DCHECK_LT(2 * position + 1, current.byte_pos + current.byte_length);
633 :
634 : // Determine # of full ucs-2 chars in stream, and whether we started on an odd
635 : // byte boundary.
636 1771 : bool odd_start = (current.byte_pos % 2) == 1;
637 1771 : size_t number_chars = (current.byte_length - odd_start) / 2;
638 :
639 : // Point the buffer_*_ members into the current chunk and set buffer_cursor_
640 : // to point to position. Be careful when converting the byte positions (in
641 : // Chunk) to the ucs-2 character positions (in buffer_*_ members).
642 1771 : buffer_start_ = reinterpret_cast<const uint16_t*>(current.data + odd_start);
643 1771 : buffer_end_ = buffer_start_ + number_chars;
644 1771 : buffer_pos_ = (current.byte_pos + odd_start) / 2;
645 1771 : buffer_cursor_ = buffer_start_ + (position - buffer_pos_);
646 : DCHECK_EQ(position, pos());
647 1771 : return true;
648 : }
649 :
650 : #else
651 :
652 : // ----------------------------------------------------------------------------
653 : // TwoByteExternalBufferedStream
654 : //
655 : // This class is made specifically to address unaligned access to 16-bit data
656 : // in MIPS and ARM architectures. It replaces class
657 : // TwoByteExternalStreamingStream which in some cases does have unaligned
658 : // accesse to 16-bit data
659 :
660 : class TwoByteExternalBufferedStream : public Utf16CharacterStream {
661 : public:
662 : explicit TwoByteExternalBufferedStream(
663 : ScriptCompiler::ExternalSourceStream* source, RuntimeCallStats* stats);
664 : ~TwoByteExternalBufferedStream();
665 :
666 : protected:
667 : static const size_t kBufferSize = 512;
668 :
669 : bool ReadBlock() override;
670 :
671 : // FillBuffer should read up to kBufferSize characters at position and store
672 : // them into buffer_[0..]. It returns the number of characters stored.
673 : size_t FillBuffer(size_t position, size_t chunk_no);
674 :
675 : // Fixed sized buffer that this class reads from.
676 : // The base class' buffer_start_ should always point to buffer_.
677 : uc16 buffer_[kBufferSize];
678 :
679 : Chunks chunks_;
680 : ScriptCompiler::ExternalSourceStream* source_;
681 : RuntimeCallStats* stats_;
682 : };
683 :
684 : TwoByteExternalBufferedStream::TwoByteExternalBufferedStream(
685 : ScriptCompiler::ExternalSourceStream* source, RuntimeCallStats* stats)
686 : : Utf16CharacterStream(buffer_, buffer_, buffer_, 0),
687 : source_(source),
688 : stats_(stats) {}
689 :
690 : TwoByteExternalBufferedStream::~TwoByteExternalBufferedStream() {
691 : DeleteChunks(chunks_);
692 : }
693 :
694 : bool TwoByteExternalBufferedStream::ReadBlock() {
695 : size_t position = pos();
696 : // Find chunk in which the position belongs
697 : size_t chunk_no = FindChunk(chunks_, source_, 2 * position + 1, stats_);
698 :
699 : // Out of data? Return 0.
700 : if (chunks_[chunk_no].byte_length == 0) {
701 : buffer_cursor_ = buffer_start_;
702 : buffer_end_ = buffer_start_;
703 : return false;
704 : }
705 :
706 : Chunk& current = chunks_[chunk_no];
707 :
708 : bool odd_start = current.byte_pos % 2;
709 : // Common case: character is in current chunk.
710 : DCHECK_LE(current.byte_pos, 2 * position + odd_start);
711 : DCHECK_LT(2 * position + 1, current.byte_pos + current.byte_length);
712 :
713 : // If character starts on odd address copy text in buffer so there is always
714 : // aligned access to characters. This is important on MIPS and ARM
715 : // architectures. Otherwise read characters from memory directly.
716 : if (!odd_start) {
717 : buffer_start_ = reinterpret_cast<const uint16_t*>(current.data);
718 : size_t number_chars = current.byte_length / 2;
719 : buffer_end_ = buffer_start_ + number_chars;
720 : buffer_pos_ = current.byte_pos / 2;
721 : buffer_cursor_ = buffer_start_ + (position - buffer_pos_);
722 : DCHECK_EQ(position, pos());
723 : return true;
724 : } else {
725 : buffer_start_ = buffer_;
726 : buffer_pos_ = position;
727 : buffer_cursor_ = buffer_;
728 : buffer_end_ = buffer_ + FillBuffer(position, chunk_no);
729 : DCHECK_EQ(pos(), position);
730 : DCHECK_LE(buffer_end_, buffer_start_ + kBufferSize);
731 : return buffer_cursor_ < buffer_end_;
732 : }
733 : }
734 :
735 : size_t TwoByteExternalBufferedStream::FillBuffer(size_t position,
736 : size_t chunk_no) {
737 : DCHECK_EQ(chunks_[chunk_no].byte_pos % 2, 1u);
738 : bool odd_start = true;
739 : // Align buffer_pos_ to the size of the buffer.
740 : {
741 : size_t new_pos = position / kBufferSize * kBufferSize;
742 : if (new_pos != position) {
743 : chunk_no = FindChunk(chunks_, source_, 2 * new_pos + 1, stats_);
744 : buffer_pos_ = new_pos;
745 : buffer_cursor_ = buffer_start_ + (position - buffer_pos_);
746 : position = new_pos;
747 : odd_start = chunks_[chunk_no].byte_pos % 2;
748 : }
749 : }
750 :
751 : Chunk* current = &chunks_[chunk_no];
752 :
753 : // Annoying edge case: Chunks may not be 2-byte aligned, meaning that a
754 : // character may be split between the previous and the current chunk.
755 : // If we find such a lonely byte at the beginning of the chunk, we'll copy
756 : // it to the first byte in buffer_.
757 : size_t totalLength = 0;
758 : bool lonely_byte = (current->byte_pos == (2 * position + 1));
759 : if (lonely_byte) {
760 : DCHECK_NE(chunk_no, 0u);
761 : Chunk& previous_chunk = chunks_[chunk_no - 1];
762 : *reinterpret_cast<uint8_t*>(buffer_) =
763 : previous_chunk.data[previous_chunk.byte_length - 1];
764 : totalLength++;
765 : }
766 :
767 : // Common case: character is in current chunk.
768 : DCHECK_LE(current->byte_pos, 2 * position + odd_start);
769 : DCHECK_LT(2 * position + 1, current->byte_pos + current->byte_length);
770 :
771 : // Copy characters from current chunk starting from chunk_pos to the end of
772 : // buffer or chunk.
773 : size_t chunk_pos = position - current->byte_pos / 2;
774 : size_t start_offset = odd_start && chunk_pos != 0;
775 : size_t bytes_to_move =
776 : i::Min(2 * kBufferSize - lonely_byte,
777 : current->byte_length - 2 * chunk_pos + start_offset);
778 : i::MemMove(reinterpret_cast<uint8_t*>(buffer_) + lonely_byte,
779 : current->data + 2 * chunk_pos - start_offset, bytes_to_move);
780 :
781 : // Fill up the rest of the buffer if there is space and data left.
782 : totalLength += bytes_to_move;
783 : position = (current->byte_pos + current->byte_length) / 2;
784 : if (position - buffer_pos_ < kBufferSize) {
785 : chunk_no = FindChunk(chunks_, source_, 2 * position + 1, stats_);
786 : current = &chunks_[chunk_no];
787 : odd_start = current->byte_pos % 2;
788 : bytes_to_move = i::Min(2 * kBufferSize - totalLength, current->byte_length);
789 : while (bytes_to_move) {
790 : // Common case: character is in current chunk.
791 : DCHECK_LE(current->byte_pos, 2 * position + odd_start);
792 : DCHECK_LT(2 * position + 1, current->byte_pos + current->byte_length);
793 :
794 : i::MemMove(reinterpret_cast<uint8_t*>(buffer_) + totalLength,
795 : current->data, bytes_to_move);
796 : totalLength += bytes_to_move;
797 : position = (current->byte_pos + current->byte_length) / 2;
798 : chunk_no = FindChunk(chunks_, source_, 2 * position + 1, stats_);
799 : current = &chunks_[chunk_no];
800 : odd_start = current->byte_pos % 2;
801 : bytes_to_move =
802 : i::Min(2 * kBufferSize - totalLength, current->byte_length);
803 : }
804 : }
805 : return totalLength / 2;
806 : }
807 : #endif
808 :
809 : // ----------------------------------------------------------------------------
810 : // ScannerStream: Create stream instances.
811 :
812 2744060 : Utf16CharacterStream* ScannerStream::For(Handle<String> data) {
813 2744060 : return ScannerStream::For(data, 0, data->length());
814 : }
815 :
816 4399643 : Utf16CharacterStream* ScannerStream::For(Handle<String> data, int start_pos,
817 : int end_pos) {
818 : DCHECK(start_pos >= 0);
819 : DCHECK(end_pos <= data->length());
820 4399652 : if (data->IsExternalOneByteString()) {
821 : return new ExternalOneByteStringUtf16CharacterStream(
822 463616 : Handle<ExternalOneByteString>::cast(data), start_pos, end_pos);
823 3936036 : } else if (data->IsExternalTwoByteString()) {
824 : return new ExternalTwoByteStringUtf16CharacterStream(
825 82 : Handle<ExternalTwoByteString>::cast(data), start_pos, end_pos);
826 : } else {
827 : // TODO(vogelheim): Maybe call data.Flatten() first?
828 3935954 : return new GenericStringUtf16CharacterStream(data, start_pos, end_pos);
829 : }
830 : }
831 :
832 556 : std::unique_ptr<Utf16CharacterStream> ScannerStream::ForTesting(
833 : const char* data) {
834 556 : return ScannerStream::ForTesting(data, strlen(data));
835 : }
836 :
837 2038 : std::unique_ptr<Utf16CharacterStream> ScannerStream::ForTesting(
838 : const char* data, size_t length) {
839 : return std::unique_ptr<Utf16CharacterStream>(
840 4076 : new ExternalOneByteStringUtf16CharacterStream(data, length));
841 : }
842 :
843 594 : Utf16CharacterStream* ScannerStream::For(
844 : ScriptCompiler::ExternalSourceStream* source_stream,
845 : v8::ScriptCompiler::StreamedSource::Encoding encoding,
846 : RuntimeCallStats* stats) {
847 594 : switch (encoding) {
848 : case v8::ScriptCompiler::StreamedSource::TWO_BYTE:
849 : #if !(V8_TARGET_ARCH_MIPS || V8_TARGET_ARCH_MIPS64)
850 140 : return new TwoByteExternalStreamingStream(source_stream, stats);
851 : #else
852 : return new TwoByteExternalBufferedStream(source_stream, stats);
853 : #endif
854 : case v8::ScriptCompiler::StreamedSource::ONE_BYTE:
855 248 : return new OneByteExternalStreamingStream(source_stream, stats);
856 : case v8::ScriptCompiler::StreamedSource::UTF8:
857 800 : return new Utf8ExternalStreamingStream(source_stream, stats);
858 : }
859 0 : UNREACHABLE();
860 : return nullptr;
861 : }
862 :
863 : } // namespace internal
864 : } // namespace v8
|