1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/parser/html/nsHtml5Parser.cpp Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,731 @@
1.4 +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
1.5 +/* vim: set sw=2 ts=2 et tw=79: */
1.6 +/* This Source Code Form is subject to the terms of the Mozilla Public
1.7 + * License, v. 2.0. If a copy of the MPL was not distributed with this
1.8 + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
1.9 +
1.10 +#include "nsHtml5Parser.h"
1.11 +
1.12 +#include "mozilla/AutoRestore.h"
1.13 +#include "nsContentUtils.h" // for kLoadAsData
1.14 +#include "nsHtml5Tokenizer.h"
1.15 +#include "nsHtml5TreeBuilder.h"
1.16 +#include "nsHtml5AtomTable.h"
1.17 +#include "nsHtml5DependentUTF16Buffer.h"
1.18 +#include "nsNetUtil.h"
1.19 +
1.20 +NS_INTERFACE_TABLE_HEAD(nsHtml5Parser)
1.21 + NS_INTERFACE_TABLE(nsHtml5Parser, nsIParser, nsISupportsWeakReference)
1.22 + NS_INTERFACE_TABLE_TO_MAP_SEGUE_CYCLE_COLLECTION(nsHtml5Parser)
1.23 +NS_INTERFACE_MAP_END
1.24 +
1.25 +NS_IMPL_CYCLE_COLLECTING_ADDREF(nsHtml5Parser)
1.26 +NS_IMPL_CYCLE_COLLECTING_RELEASE(nsHtml5Parser)
1.27 +
1.28 +NS_IMPL_CYCLE_COLLECTION_CLASS(nsHtml5Parser)
1.29 +
1.30 +NS_IMPL_CYCLE_COLLECTION_TRAVERSE_BEGIN(nsHtml5Parser)
1.31 + NS_IMPL_CYCLE_COLLECTION_TRAVERSE(mExecutor)
1.32 + NS_IMPL_CYCLE_COLLECTION_TRAVERSE_RAWPTR(GetStreamParser())
1.33 +NS_IMPL_CYCLE_COLLECTION_TRAVERSE_END
1.34 +
1.35 +NS_IMPL_CYCLE_COLLECTION_UNLINK_BEGIN(nsHtml5Parser)
1.36 + NS_IMPL_CYCLE_COLLECTION_UNLINK(mExecutor)
1.37 + tmp->DropStreamParser();
1.38 +NS_IMPL_CYCLE_COLLECTION_UNLINK_END
1.39 +
1.40 +nsHtml5Parser::nsHtml5Parser()
1.41 + : mFirstBuffer(new nsHtml5OwningUTF16Buffer((void*)nullptr))
1.42 + , mLastBuffer(mFirstBuffer)
1.43 + , mExecutor(new nsHtml5TreeOpExecutor())
1.44 + , mTreeBuilder(new nsHtml5TreeBuilder(mExecutor, nullptr))
1.45 + , mTokenizer(new nsHtml5Tokenizer(mTreeBuilder, false))
1.46 + , mRootContextLineNumber(1)
1.47 +{
1.48 + mTokenizer->setInterner(&mAtomTable);
1.49 + // There's a zeroing operator new for everything else
1.50 +}
1.51 +
1.52 +nsHtml5Parser::~nsHtml5Parser()
1.53 +{
1.54 + mTokenizer->end();
1.55 + if (mDocWriteSpeculativeTokenizer) {
1.56 + mDocWriteSpeculativeTokenizer->end();
1.57 + }
1.58 +}
1.59 +
1.60 +NS_IMETHODIMP_(void)
1.61 +nsHtml5Parser::SetContentSink(nsIContentSink* aSink)
1.62 +{
1.63 + NS_ASSERTION(aSink == static_cast<nsIContentSink*> (mExecutor),
1.64 + "Attempt to set a foreign sink.");
1.65 +}
1.66 +
1.67 +NS_IMETHODIMP_(nsIContentSink*)
1.68 +nsHtml5Parser::GetContentSink()
1.69 +{
1.70 + return static_cast<nsIContentSink*> (mExecutor);
1.71 +}
1.72 +
1.73 +NS_IMETHODIMP_(void)
1.74 +nsHtml5Parser::GetCommand(nsCString& aCommand)
1.75 +{
1.76 + aCommand.Assign("view");
1.77 +}
1.78 +
1.79 +NS_IMETHODIMP_(void)
1.80 +nsHtml5Parser::SetCommand(const char* aCommand)
1.81 +{
1.82 + NS_ASSERTION(!strcmp(aCommand, "view") ||
1.83 + !strcmp(aCommand, "view-source") ||
1.84 + !strcmp(aCommand, "external-resource") ||
1.85 + !strcmp(aCommand, kLoadAsData),
1.86 + "Unsupported parser command");
1.87 +}
1.88 +
1.89 +NS_IMETHODIMP_(void)
1.90 +nsHtml5Parser::SetCommand(eParserCommands aParserCommand)
1.91 +{
1.92 + NS_ASSERTION(aParserCommand == eViewNormal,
1.93 + "Parser command was not eViewNormal.");
1.94 +}
1.95 +
1.96 +NS_IMETHODIMP_(void)
1.97 +nsHtml5Parser::SetDocumentCharset(const nsACString& aCharset,
1.98 + int32_t aCharsetSource)
1.99 +{
1.100 + NS_PRECONDITION(!mExecutor->HasStarted(),
1.101 + "Document charset set too late.");
1.102 + NS_PRECONDITION(GetStreamParser(), "Setting charset on a script-only parser.");
1.103 + nsAutoCString trimmed;
1.104 + trimmed.Assign(aCharset);
1.105 + trimmed.Trim(" \t\r\n\f");
1.106 + GetStreamParser()->SetDocumentCharset(trimmed, aCharsetSource);
1.107 + mExecutor->SetDocumentCharsetAndSource(trimmed,
1.108 + aCharsetSource);
1.109 +}
1.110 +
1.111 +NS_IMETHODIMP
1.112 +nsHtml5Parser::GetChannel(nsIChannel** aChannel)
1.113 +{
1.114 + if (GetStreamParser()) {
1.115 + return GetStreamParser()->GetChannel(aChannel);
1.116 + } else {
1.117 + return NS_ERROR_NOT_AVAILABLE;
1.118 + }
1.119 +}
1.120 +
1.121 +NS_IMETHODIMP
1.122 +nsHtml5Parser::GetDTD(nsIDTD** aDTD)
1.123 +{
1.124 + *aDTD = nullptr;
1.125 + return NS_OK;
1.126 +}
1.127 +
1.128 +nsIStreamListener*
1.129 +nsHtml5Parser::GetStreamListener()
1.130 +{
1.131 + return mStreamListener;
1.132 +}
1.133 +
1.134 +NS_IMETHODIMP
1.135 +nsHtml5Parser::ContinueInterruptedParsing()
1.136 +{
1.137 + NS_NOTREACHED("Don't call. For interface compat only.");
1.138 + return NS_ERROR_NOT_IMPLEMENTED;
1.139 +}
1.140 +
1.141 +NS_IMETHODIMP_(void)
1.142 +nsHtml5Parser::BlockParser()
1.143 +{
1.144 + mBlocked = true;
1.145 +}
1.146 +
1.147 +NS_IMETHODIMP_(void)
1.148 +nsHtml5Parser::UnblockParser()
1.149 +{
1.150 + mBlocked = false;
1.151 + mExecutor->ContinueInterruptedParsingAsync();
1.152 +}
1.153 +
1.154 +NS_IMETHODIMP_(void)
1.155 +nsHtml5Parser::ContinueInterruptedParsingAsync()
1.156 +{
1.157 + mExecutor->ContinueInterruptedParsingAsync();
1.158 +}
1.159 +
1.160 +NS_IMETHODIMP_(bool)
1.161 +nsHtml5Parser::IsParserEnabled()
1.162 +{
1.163 + return !mBlocked;
1.164 +}
1.165 +
1.166 +NS_IMETHODIMP_(bool)
1.167 +nsHtml5Parser::IsComplete()
1.168 +{
1.169 + return mExecutor->IsComplete();
1.170 +}
1.171 +
1.172 +NS_IMETHODIMP
1.173 +nsHtml5Parser::Parse(nsIURI* aURL,
1.174 + nsIRequestObserver* aObserver,
1.175 + void* aKey, // legacy; ignored
1.176 + nsDTDMode aMode) // legacy; ignored
1.177 +{
1.178 + /*
1.179 + * Do NOT cause WillBuildModel to be called synchronously from here!
1.180 + * The document won't be ready for it until OnStartRequest!
1.181 + */
1.182 + NS_PRECONDITION(!mExecutor->HasStarted(),
1.183 + "Tried to start parse without initializing the parser.");
1.184 + NS_PRECONDITION(GetStreamParser(),
1.185 + "Can't call this Parse() variant on script-created parser");
1.186 + GetStreamParser()->SetObserver(aObserver);
1.187 + GetStreamParser()->SetViewSourceTitle(aURL); // In case we're viewing source
1.188 + mExecutor->SetStreamParser(GetStreamParser());
1.189 + mExecutor->SetParser(this);
1.190 + return NS_OK;
1.191 +}
1.192 +
1.193 +NS_IMETHODIMP
1.194 +nsHtml5Parser::Parse(const nsAString& aSourceBuffer,
1.195 + void* aKey,
1.196 + const nsACString& aContentType,
1.197 + bool aLastCall,
1.198 + nsDTDMode aMode) // ignored
1.199 +{
1.200 + nsresult rv;
1.201 + if (NS_FAILED(rv = mExecutor->IsBroken())) {
1.202 + return rv;
1.203 + }
1.204 + if (aSourceBuffer.Length() > INT32_MAX) {
1.205 + return mExecutor->MarkAsBroken(NS_ERROR_OUT_OF_MEMORY);
1.206 + }
1.207 +
1.208 + // Maintain a reference to ourselves so we don't go away
1.209 + // till we're completely done. The old parser grips itself in this method.
1.210 + nsCOMPtr<nsIParser> kungFuDeathGrip(this);
1.211 +
1.212 + // Gripping the other objects just in case, since the other old grip
1.213 + // required grips to these, too.
1.214 + nsRefPtr<nsHtml5StreamParser> streamKungFuDeathGrip(GetStreamParser());
1.215 + nsRefPtr<nsHtml5TreeOpExecutor> treeOpKungFuDeathGrip(mExecutor);
1.216 +
1.217 + if (!mExecutor->HasStarted()) {
1.218 + NS_ASSERTION(!GetStreamParser(),
1.219 + "Had stream parser but document.write started life cycle.");
1.220 + // This is the first document.write() on a document.open()ed document
1.221 + mExecutor->SetParser(this);
1.222 + mTreeBuilder->setScriptingEnabled(mExecutor->IsScriptEnabled());
1.223 +
1.224 + bool isSrcdoc = false;
1.225 + nsCOMPtr<nsIChannel> channel;
1.226 + rv = GetChannel(getter_AddRefs(channel));
1.227 + if (NS_SUCCEEDED(rv)) {
1.228 + isSrcdoc = NS_IsSrcdocChannel(channel);
1.229 + }
1.230 + mTreeBuilder->setIsSrcdocDocument(isSrcdoc);
1.231 +
1.232 + mTokenizer->start();
1.233 + mExecutor->Start();
1.234 + if (!aContentType.EqualsLiteral("text/html")) {
1.235 + mTreeBuilder->StartPlainText();
1.236 + mTokenizer->StartPlainText();
1.237 + }
1.238 + /*
1.239 + * If you move the following line, be very careful not to cause
1.240 + * WillBuildModel to be called before the document has had its
1.241 + * script global object set.
1.242 + */
1.243 + rv = mExecutor->WillBuildModel(eDTDMode_unknown);
1.244 + NS_ENSURE_SUCCESS(rv, rv);
1.245 + }
1.246 +
1.247 + // Return early if the parser has processed EOF
1.248 + if (mExecutor->IsComplete()) {
1.249 + return NS_OK;
1.250 + }
1.251 +
1.252 + if (aLastCall && aSourceBuffer.IsEmpty() && !aKey) {
1.253 + // document.close()
1.254 + NS_ASSERTION(!GetStreamParser(),
1.255 + "Had stream parser but got document.close().");
1.256 + if (mDocumentClosed) {
1.257 + // already closed
1.258 + return NS_OK;
1.259 + }
1.260 + mDocumentClosed = true;
1.261 + if (!mBlocked && !mInDocumentWrite) {
1.262 + return ParseUntilBlocked();
1.263 + }
1.264 + return NS_OK;
1.265 + }
1.266 +
1.267 + // If we got this far, we are dealing with a document.write or
1.268 + // document.writeln call--not document.close().
1.269 +
1.270 + NS_ASSERTION(IsInsertionPointDefined(),
1.271 + "Doc.write reached parser with undefined insertion point.");
1.272 +
1.273 + NS_ASSERTION(!(GetStreamParser() && !aKey),
1.274 + "Got a null key in a non-script-created parser");
1.275 +
1.276 + // XXX is this optimization bogus?
1.277 + if (aSourceBuffer.IsEmpty()) {
1.278 + return NS_OK;
1.279 + }
1.280 +
1.281 + // This guard is here to prevent document.close from tokenizing synchronously
1.282 + // while a document.write (that wrote the script that called document.close!)
1.283 + // is still on the call stack.
1.284 + mozilla::AutoRestore<bool> guard(mInDocumentWrite);
1.285 + mInDocumentWrite = true;
1.286 +
1.287 + // The script is identified by aKey. If there's nothing in the buffer
1.288 + // chain for that key, we'll insert at the head of the queue.
1.289 + // When the script leaves something in the queue, a zero-length
1.290 + // key-holder "buffer" is inserted in the queue. If the same script
1.291 + // leaves something in the chain again, it will be inserted immediately
1.292 + // before the old key holder belonging to the same script.
1.293 + //
1.294 + // We don't do the actual data insertion yet in the hope that the data gets
1.295 + // tokenized and there no data or less data to copy to the heap after
1.296 + // tokenization. Also, this way, we avoid inserting one empty data buffer
1.297 + // per document.write, which matters for performance when the parser isn't
1.298 + // blocked and a badly-authored script calls document.write() once per
1.299 + // input character. (As seen in a benchmark!)
1.300 + //
1.301 + // The insertion into the input stream happens conceptually before anything
1.302 + // gets tokenized. To make sure multi-level document.write works right,
1.303 + // it's necessary to establish the location of our parser key up front
1.304 + // in case this is the first write with this key.
1.305 + //
1.306 + // In a document.open() case, the first write level has a null key, so that
1.307 + // case is handled separately, because normal buffers containing data
1.308 + // have null keys.
1.309 +
1.310 + // These don't need to be owning references, because they always point to
1.311 + // the buffer queue and buffers can't be removed from the buffer queue
1.312 + // before document.write() returns. The buffer queue clean-up happens the
1.313 + // next time ParseUntilBlocked() is called.
1.314 + // However, they are made owning just in case the reasoning above is flawed
1.315 + // and a flaw would lead to worse problems with plain pointers. If this
1.316 + // turns out to be a perf problem, it's worthwhile to consider making
1.317 + // prevSearchbuf a plain pointer again.
1.318 + nsRefPtr<nsHtml5OwningUTF16Buffer> prevSearchBuf;
1.319 + nsRefPtr<nsHtml5OwningUTF16Buffer> firstLevelMarker;
1.320 +
1.321 + if (aKey) {
1.322 + if (mFirstBuffer == mLastBuffer) {
1.323 + nsHtml5OwningUTF16Buffer* keyHolder = new nsHtml5OwningUTF16Buffer(aKey);
1.324 + keyHolder->next = mLastBuffer;
1.325 + mFirstBuffer = keyHolder;
1.326 + } else if (mFirstBuffer->key != aKey) {
1.327 + prevSearchBuf = mFirstBuffer;
1.328 + for (;;) {
1.329 + if (prevSearchBuf->next == mLastBuffer) {
1.330 + // key was not found
1.331 + nsHtml5OwningUTF16Buffer* keyHolder =
1.332 + new nsHtml5OwningUTF16Buffer(aKey);
1.333 + keyHolder->next = mFirstBuffer;
1.334 + mFirstBuffer = keyHolder;
1.335 + prevSearchBuf = nullptr;
1.336 + break;
1.337 + }
1.338 + if (prevSearchBuf->next->key == aKey) {
1.339 + // found a key holder
1.340 + break;
1.341 + }
1.342 + prevSearchBuf = prevSearchBuf->next;
1.343 + }
1.344 + } // else mFirstBuffer is the keyholder
1.345 +
1.346 + // prevSearchBuf is the previous buffer before the keyholder or null if
1.347 + // there isn't one.
1.348 + } else {
1.349 + // We have a first-level write in the document.open() case. We insert before
1.350 + // mLastBuffer, effectively, by making mLastBuffer be a new sentinel object
1.351 + // and redesignating the previous mLastBuffer as our firstLevelMarker. We
1.352 + // need to put a marker there, because otherwise additional document.writes
1.353 + // from nested event loops would insert in the wrong place. Sigh.
1.354 + mLastBuffer->next = new nsHtml5OwningUTF16Buffer((void*)nullptr);
1.355 + firstLevelMarker = mLastBuffer;
1.356 + mLastBuffer = mLastBuffer->next;
1.357 + }
1.358 +
1.359 + nsHtml5DependentUTF16Buffer stackBuffer(aSourceBuffer);
1.360 +
1.361 + while (!mBlocked && stackBuffer.hasMore()) {
1.362 + stackBuffer.adjust(mLastWasCR);
1.363 + mLastWasCR = false;
1.364 + if (stackBuffer.hasMore()) {
1.365 + int32_t lineNumberSave;
1.366 + bool inRootContext = (!GetStreamParser() && !aKey);
1.367 + if (inRootContext) {
1.368 + mTokenizer->setLineNumber(mRootContextLineNumber);
1.369 + } else {
1.370 + // we aren't the root context, so save the line number on the
1.371 + // *stack* so that we can restore it.
1.372 + lineNumberSave = mTokenizer->getLineNumber();
1.373 + }
1.374 +
1.375 + mLastWasCR = mTokenizer->tokenizeBuffer(&stackBuffer);
1.376 +
1.377 + if (inRootContext) {
1.378 + mRootContextLineNumber = mTokenizer->getLineNumber();
1.379 + } else {
1.380 + mTokenizer->setLineNumber(lineNumberSave);
1.381 + }
1.382 +
1.383 + if (mTreeBuilder->HasScript()) {
1.384 + mTreeBuilder->Flush(); // Move ops to the executor
1.385 + rv = mExecutor->FlushDocumentWrite(); // run the ops
1.386 + NS_ENSURE_SUCCESS(rv, rv);
1.387 + // Flushing tree ops can cause all sorts of things.
1.388 + // Return early if the parser got terminated.
1.389 + if (mExecutor->IsComplete()) {
1.390 + return NS_OK;
1.391 + }
1.392 + }
1.393 + // Ignore suspension requests
1.394 + }
1.395 + }
1.396 +
1.397 + nsRefPtr<nsHtml5OwningUTF16Buffer> heapBuffer;
1.398 + if (stackBuffer.hasMore()) {
1.399 + // The buffer wasn't tokenized to completion. Create a copy of the tail
1.400 + // on the heap.
1.401 + heapBuffer = stackBuffer.FalliblyCopyAsOwningBuffer();
1.402 + if (!heapBuffer) {
1.403 + // Allocation failed. The parser is now broken.
1.404 + return mExecutor->MarkAsBroken(NS_ERROR_OUT_OF_MEMORY);
1.405 + }
1.406 + }
1.407 +
1.408 + if (heapBuffer) {
1.409 + // We have something to insert before the keyholder holding in the non-null
1.410 + // aKey case and we have something to swap into firstLevelMarker in the
1.411 + // null aKey case.
1.412 + if (aKey) {
1.413 + NS_ASSERTION(mFirstBuffer != mLastBuffer,
1.414 + "Where's the keyholder?");
1.415 + // the key holder is still somewhere further down the list from
1.416 + // prevSearchBuf (which may be null)
1.417 + if (mFirstBuffer->key == aKey) {
1.418 + NS_ASSERTION(!prevSearchBuf,
1.419 + "Non-null prevSearchBuf when mFirstBuffer is the key holder?");
1.420 + heapBuffer->next = mFirstBuffer;
1.421 + mFirstBuffer = heapBuffer;
1.422 + } else {
1.423 + if (!prevSearchBuf) {
1.424 + prevSearchBuf = mFirstBuffer;
1.425 + }
1.426 + // We created a key holder earlier, so we will find it without walking
1.427 + // past the end of the list.
1.428 + while (prevSearchBuf->next->key != aKey) {
1.429 + prevSearchBuf = prevSearchBuf->next;
1.430 + }
1.431 + heapBuffer->next = prevSearchBuf->next;
1.432 + prevSearchBuf->next = heapBuffer;
1.433 + }
1.434 + } else {
1.435 + NS_ASSERTION(firstLevelMarker, "How come we don't have a marker.");
1.436 + firstLevelMarker->Swap(heapBuffer);
1.437 + }
1.438 + }
1.439 +
1.440 + if (!mBlocked) { // buffer was tokenized to completion
1.441 + NS_ASSERTION(!stackBuffer.hasMore(),
1.442 + "Buffer wasn't tokenized to completion?");
1.443 + // Scripting semantics require a forced tree builder flush here
1.444 + mTreeBuilder->Flush(); // Move ops to the executor
1.445 + rv = mExecutor->FlushDocumentWrite(); // run the ops
1.446 + NS_ENSURE_SUCCESS(rv, rv);
1.447 + } else if (stackBuffer.hasMore()) {
1.448 + // The buffer wasn't tokenized to completion. Tokenize the untokenized
1.449 + // content in order to preload stuff. This content will be retokenized
1.450 + // later for normal parsing.
1.451 + if (!mDocWriteSpeculatorActive) {
1.452 + mDocWriteSpeculatorActive = true;
1.453 + if (!mDocWriteSpeculativeTreeBuilder) {
1.454 + // Lazily initialize if uninitialized
1.455 + mDocWriteSpeculativeTreeBuilder =
1.456 + new nsHtml5TreeBuilder(nullptr, mExecutor->GetStage());
1.457 + mDocWriteSpeculativeTreeBuilder->setScriptingEnabled(
1.458 + mTreeBuilder->isScriptingEnabled());
1.459 + mDocWriteSpeculativeTokenizer =
1.460 + new nsHtml5Tokenizer(mDocWriteSpeculativeTreeBuilder, false);
1.461 + mDocWriteSpeculativeTokenizer->setInterner(&mAtomTable);
1.462 + mDocWriteSpeculativeTokenizer->start();
1.463 + }
1.464 + mDocWriteSpeculativeTokenizer->resetToDataState();
1.465 + mDocWriteSpeculativeTreeBuilder->loadState(mTreeBuilder, &mAtomTable);
1.466 + mDocWriteSpeculativeLastWasCR = false;
1.467 + }
1.468 +
1.469 + // Note that with multilevel document.write if we didn't just activate the
1.470 + // speculator, it's possible that the speculator is now in the wrong state.
1.471 + // That's OK for the sake of simplicity. The worst that can happen is
1.472 + // that the speculative loads aren't exactly right. The content will be
1.473 + // reparsed anyway for non-preload purposes.
1.474 +
1.475 + // The buffer position for subsequent non-speculative parsing now lives
1.476 + // in heapBuffer, so it's ok to let the buffer position of stackBuffer
1.477 + // to be overwritten and not restored below.
1.478 + while (stackBuffer.hasMore()) {
1.479 + stackBuffer.adjust(mDocWriteSpeculativeLastWasCR);
1.480 + if (stackBuffer.hasMore()) {
1.481 + mDocWriteSpeculativeLastWasCR =
1.482 + mDocWriteSpeculativeTokenizer->tokenizeBuffer(&stackBuffer);
1.483 + }
1.484 + }
1.485 +
1.486 + mDocWriteSpeculativeTreeBuilder->Flush();
1.487 + mDocWriteSpeculativeTreeBuilder->DropHandles();
1.488 + mExecutor->FlushSpeculativeLoads();
1.489 + }
1.490 +
1.491 + return NS_OK;
1.492 +}
1.493 +
1.494 +NS_IMETHODIMP
1.495 +nsHtml5Parser::Terminate()
1.496 +{
1.497 + // We should only call DidBuildModel once, so don't do anything if this is
1.498 + // the second time that Terminate has been called.
1.499 + if (mExecutor->IsComplete()) {
1.500 + return NS_OK;
1.501 + }
1.502 + // XXX - [ until we figure out a way to break parser-sink circularity ]
1.503 + // Hack - Hold a reference until we are completely done...
1.504 + nsCOMPtr<nsIParser> kungFuDeathGrip(this);
1.505 + nsRefPtr<nsHtml5StreamParser> streamKungFuDeathGrip(GetStreamParser());
1.506 + nsRefPtr<nsHtml5TreeOpExecutor> treeOpKungFuDeathGrip(mExecutor);
1.507 + if (GetStreamParser()) {
1.508 + GetStreamParser()->Terminate();
1.509 + }
1.510 + return mExecutor->DidBuildModel(true);
1.511 +}
1.512 +
1.513 +NS_IMETHODIMP
1.514 +nsHtml5Parser::ParseFragment(const nsAString& aSourceBuffer,
1.515 + nsTArray<nsString>& aTagStack)
1.516 +{
1.517 + return NS_ERROR_NOT_IMPLEMENTED;
1.518 +}
1.519 +
1.520 +NS_IMETHODIMP
1.521 +nsHtml5Parser::BuildModel()
1.522 +{
1.523 + NS_NOTREACHED("Don't call this!");
1.524 + return NS_ERROR_NOT_IMPLEMENTED;
1.525 +}
1.526 +
1.527 +NS_IMETHODIMP
1.528 +nsHtml5Parser::CancelParsingEvents()
1.529 +{
1.530 + NS_NOTREACHED("Don't call this!");
1.531 + return NS_ERROR_NOT_IMPLEMENTED;
1.532 +}
1.533 +
1.534 +void
1.535 +nsHtml5Parser::Reset()
1.536 +{
1.537 + NS_NOTREACHED("Don't call this!");
1.538 +}
1.539 +
1.540 +bool
1.541 +nsHtml5Parser::CanInterrupt()
1.542 +{
1.543 + // nsContentSink needs this to let nsContentSink::DidProcessATokenImpl
1.544 + // interrupt.
1.545 + return true;
1.546 +}
1.547 +
1.548 +bool
1.549 +nsHtml5Parser::IsInsertionPointDefined()
1.550 +{
1.551 + return !mExecutor->IsFlushing() &&
1.552 + (!GetStreamParser() || mParserInsertedScriptsBeingEvaluated);
1.553 +}
1.554 +
1.555 +void
1.556 +nsHtml5Parser::BeginEvaluatingParserInsertedScript()
1.557 +{
1.558 + ++mParserInsertedScriptsBeingEvaluated;
1.559 +}
1.560 +
1.561 +void
1.562 +nsHtml5Parser::EndEvaluatingParserInsertedScript()
1.563 +{
1.564 + --mParserInsertedScriptsBeingEvaluated;
1.565 +}
1.566 +
1.567 +void
1.568 +nsHtml5Parser::MarkAsNotScriptCreated(const char* aCommand)
1.569 +{
1.570 + NS_PRECONDITION(!mStreamListener, "Must not call this twice.");
1.571 + eParserMode mode = NORMAL;
1.572 + if (!nsCRT::strcmp(aCommand, "view-source")) {
1.573 + mode = VIEW_SOURCE_HTML;
1.574 + } else if (!nsCRT::strcmp(aCommand, "view-source-xml")) {
1.575 + mode = VIEW_SOURCE_XML;
1.576 + } else if (!nsCRT::strcmp(aCommand, "view-source-plain")) {
1.577 + mode = VIEW_SOURCE_PLAIN;
1.578 + } else if (!nsCRT::strcmp(aCommand, "plain-text")) {
1.579 + mode = PLAIN_TEXT;
1.580 + } else if (!nsCRT::strcmp(aCommand, kLoadAsData)) {
1.581 + mode = LOAD_AS_DATA;
1.582 + }
1.583 +#ifdef DEBUG
1.584 + else {
1.585 + NS_ASSERTION(!nsCRT::strcmp(aCommand, "view") ||
1.586 + !nsCRT::strcmp(aCommand, "external-resource"),
1.587 + "Unsupported parser command!");
1.588 + }
1.589 +#endif
1.590 + mStreamListener =
1.591 + new nsHtml5StreamListener(new nsHtml5StreamParser(mExecutor, this, mode));
1.592 +}
1.593 +
1.594 +bool
1.595 +nsHtml5Parser::IsScriptCreated()
1.596 +{
1.597 + return !GetStreamParser();
1.598 +}
1.599 +
1.600 +/* End nsIParser */
1.601 +
1.602 +// not from interface
1.603 +nsresult
1.604 +nsHtml5Parser::ParseUntilBlocked()
1.605 +{
1.606 + nsresult rv = mExecutor->IsBroken();
1.607 + NS_ENSURE_SUCCESS(rv, rv);
1.608 + if (mBlocked || mExecutor->IsComplete()) {
1.609 + return NS_OK;
1.610 + }
1.611 + NS_ASSERTION(mExecutor->HasStarted(), "Bad life cycle.");
1.612 + NS_ASSERTION(!mInDocumentWrite,
1.613 + "ParseUntilBlocked entered while in doc.write!");
1.614 +
1.615 + mDocWriteSpeculatorActive = false;
1.616 +
1.617 + for (;;) {
1.618 + if (!mFirstBuffer->hasMore()) {
1.619 + if (mFirstBuffer == mLastBuffer) {
1.620 + if (mExecutor->IsComplete()) {
1.621 + // something like cache manisfests stopped the parse in mid-flight
1.622 + return NS_OK;
1.623 + }
1.624 + if (mDocumentClosed) {
1.625 + NS_ASSERTION(!GetStreamParser(),
1.626 + "This should only happen with script-created parser.");
1.627 + mTokenizer->eof();
1.628 + mTreeBuilder->StreamEnded();
1.629 + mTreeBuilder->Flush();
1.630 + mExecutor->FlushDocumentWrite();
1.631 + // The below call does memory cleanup, so call it even if the
1.632 + // parser has been marked as broken.
1.633 + mTokenizer->end();
1.634 + return NS_OK;
1.635 + }
1.636 + // never release the last buffer.
1.637 + NS_ASSERTION(!mLastBuffer->getStart() && !mLastBuffer->getEnd(),
1.638 + "Sentinel buffer had its indeces changed.");
1.639 + if (GetStreamParser()) {
1.640 + if (mReturnToStreamParserPermitted &&
1.641 + !mExecutor->IsScriptExecuting()) {
1.642 + mTreeBuilder->Flush();
1.643 + mReturnToStreamParserPermitted = false;
1.644 + GetStreamParser()->ContinueAfterScripts(mTokenizer,
1.645 + mTreeBuilder,
1.646 + mLastWasCR);
1.647 + }
1.648 + } else {
1.649 + // Script-created parser
1.650 + mTreeBuilder->Flush();
1.651 + // No need to flush the executor, because the executor is already
1.652 + // in a flush
1.653 + NS_ASSERTION(mExecutor->IsInFlushLoop(),
1.654 + "How did we come here without being in the flush loop?");
1.655 + }
1.656 + return NS_OK; // no more data for now but expecting more
1.657 + }
1.658 + mFirstBuffer = mFirstBuffer->next;
1.659 + continue;
1.660 + }
1.661 +
1.662 + if (mBlocked || mExecutor->IsComplete()) {
1.663 + return NS_OK;
1.664 + }
1.665 +
1.666 + // now we have a non-empty buffer
1.667 + mFirstBuffer->adjust(mLastWasCR);
1.668 + mLastWasCR = false;
1.669 + if (mFirstBuffer->hasMore()) {
1.670 + bool inRootContext = (!GetStreamParser() && !mFirstBuffer->key);
1.671 + if (inRootContext) {
1.672 + mTokenizer->setLineNumber(mRootContextLineNumber);
1.673 + }
1.674 + mLastWasCR = mTokenizer->tokenizeBuffer(mFirstBuffer);
1.675 + if (inRootContext) {
1.676 + mRootContextLineNumber = mTokenizer->getLineNumber();
1.677 + }
1.678 + if (mTreeBuilder->HasScript()) {
1.679 + mTreeBuilder->Flush();
1.680 + nsresult rv = mExecutor->FlushDocumentWrite();
1.681 + NS_ENSURE_SUCCESS(rv, rv);
1.682 + }
1.683 + if (mBlocked) {
1.684 + return NS_OK;
1.685 + }
1.686 + }
1.687 + continue;
1.688 + }
1.689 +}
1.690 +
1.691 +nsresult
1.692 +nsHtml5Parser::Initialize(nsIDocument* aDoc,
1.693 + nsIURI* aURI,
1.694 + nsISupports* aContainer,
1.695 + nsIChannel* aChannel)
1.696 +{
1.697 + return mExecutor->Init(aDoc, aURI, aContainer, aChannel);
1.698 +}
1.699 +
1.700 +void
1.701 +nsHtml5Parser::StartTokenizer(bool aScriptingEnabled) {
1.702 +
1.703 + bool isSrcdoc = false;
1.704 + nsCOMPtr<nsIChannel> channel;
1.705 + nsresult rv = GetChannel(getter_AddRefs(channel));
1.706 + if (NS_SUCCEEDED(rv)) {
1.707 + isSrcdoc = NS_IsSrcdocChannel(channel);
1.708 + }
1.709 + mTreeBuilder->setIsSrcdocDocument(isSrcdoc);
1.710 +
1.711 + mTreeBuilder->SetPreventScriptExecution(!aScriptingEnabled);
1.712 + mTreeBuilder->setScriptingEnabled(aScriptingEnabled);
1.713 + mTokenizer->start();
1.714 +}
1.715 +
1.716 +void
1.717 +nsHtml5Parser::InitializeDocWriteParserState(nsAHtml5TreeBuilderState* aState,
1.718 + int32_t aLine)
1.719 +{
1.720 + mTokenizer->resetToDataState();
1.721 + mTokenizer->setLineNumber(aLine);
1.722 + mTreeBuilder->loadState(aState, &mAtomTable);
1.723 + mLastWasCR = false;
1.724 + mReturnToStreamParserPermitted = true;
1.725 +}
1.726 +
1.727 +void
1.728 +nsHtml5Parser::ContinueAfterFailedCharsetSwitch()
1.729 +{
1.730 + NS_PRECONDITION(GetStreamParser(),
1.731 + "Tried to continue after failed charset switch without a stream parser");
1.732 + GetStreamParser()->ContinueAfterFailedCharsetSwitch();
1.733 +}
1.734 +
