NS_IMETHODIMP
nsHtml5Parser::Parse(const nsAString& aSourceBuffer,
void* aKey,
const nsACString& aContentType,
bool aLastCall,
nsDTDMode aMode) // ignored
{
if (mExecutor->IsBroken()) {
return NS_ERROR_OUT_OF_MEMORY;
}
if (aSourceBuffer.Length() > PR_INT32_MAX) {
mExecutor->MarkAsBroken();
return NS_ERROR_OUT_OF_MEMORY;
}
// Maintain a reference to ourselves so we don't go away
// till we're completely done. The old parser grips itself in this method.
nsCOMPtr<nsIParser> kungFuDeathGrip(this);
// Gripping the other objects just in case, since the other old grip
// required grips to these, too.
nsRefPtr<nsHtml5StreamParser> streamKungFuDeathGrip(mStreamParser);
nsRefPtr<nsHtml5TreeOpExecutor> treeOpKungFuDeathGrip(mExecutor);
if (!mExecutor->HasStarted()) {
NS_ASSERTION(!mStreamParser,
"Had stream parser but document.write started life cycle.");
// This is the first document.write() on a document.open()ed document
mExecutor->SetParser(this);
mTreeBuilder->setScriptingEnabled(mExecutor->IsScriptEnabled());
mTokenizer->start();
mExecutor->Start();
if (!aContentType.EqualsLiteral("text/html")) {
mTreeBuilder->StartPlainText();
mTokenizer->StartPlainText();
}
/*
* If you move the following line, be very careful not to cause
* WillBuildModel to be called before the document has had its
* script global object set.
*/
mExecutor->WillBuildModel(eDTDMode_unknown);
}
// Return early if the parser has processed EOF
if (mExecutor->IsComplete()) {
return NS_OK;
}
if (aLastCall && aSourceBuffer.IsEmpty() && !aKey) {
// document.close()
NS_ASSERTION(!mStreamParser,
"Had stream parser but got document.close().");
if (mDocumentClosed) {
// already closed
return NS_OK;
}
mDocumentClosed = true;
if (!mBlocked && !mInDocumentWrite) {
ParseUntilBlocked();
}
return NS_OK;
}
// If we got this far, we are dealing with a document.write or
// document.writeln call--not document.close().
NS_ASSERTION(IsInsertionPointDefined(),
"Doc.write reached parser with undefined insertion point.");
NS_ASSERTION(!(mStreamParser && !aKey),
"Got a null key in a non-script-created parser");
// XXX is this optimization bogus?
if (aSourceBuffer.IsEmpty()) {
return NS_OK;
}
// This guard is here to prevent document.close from tokenizing synchronously
// while a document.write (that wrote the script that called document.close!)
// is still on the call stack.
mozilla::AutoRestore<bool> guard(mInDocumentWrite);
mInDocumentWrite = true;
// The script is identified by aKey. If there's nothing in the buffer
// chain for that key, we'll insert at the head of the queue.
// When the script leaves something in the queue, a zero-length
// key-holder "buffer" is inserted in the queue. If the same script
// leaves something in the chain again, it will be inserted immediately
// before the old key holder belonging to the same script.
//
// We don't do the actual data insertion yet in the hope that the data gets
// tokenized and there no data or less data to copy to the heap after
// tokenization. Also, this way, we avoid inserting one empty data buffer
// per document.write, which matters for performance when the parser isn't
// blocked and a badly-authored script calls document.write() once per
// input character. (As seen in a benchmark!)
//
// The insertion into the input stream happens conceptually before anything
// gets tokenized. To make sure multi-level document.write works right,
// it's necessary to establish the location of our parser key up front
// in case this is the first write with this key.
//
// In a document.open() case, the first write level has a null key, so that
// case is handled separately, because normal buffers containing data
// have null keys.
// These don't need to be owning references, because they always point to
// the buffer queue and buffers can't be removed from the buffer queue
// before document.write() returns. The buffer queue clean-up happens the
// next time ParseUntilBlocked() is called.
// However, they are made owning just in case the reasoning above is flawed
// and a flaw would lead to worse problems with plain pointers. If this
// turns out to be a perf problem, it's worthwhile to consider making
// prevSearchbuf a plain pointer again.
nsRefPtr<nsHtml5OwningUTF16Buffer> prevSearchBuf;
nsRefPtr<nsHtml5OwningUTF16Buffer> firstLevelMarker;
if (aKey) {
if (mFirstBuffer == mLastBuffer) {
nsHtml5OwningUTF16Buffer* keyHolder = new nsHtml5OwningUTF16Buffer(aKey);
keyHolder->next = mLastBuffer;
mFirstBuffer = keyHolder;
} else if (mFirstBuffer->key != aKey) {
prevSearchBuf = mFirstBuffer;
for (;;) {
if (prevSearchBuf->next == mLastBuffer) {
// key was not found
nsHtml5OwningUTF16Buffer* keyHolder =
new nsHtml5OwningUTF16Buffer(aKey);
keyHolder->next = mFirstBuffer;
mFirstBuffer = keyHolder;
prevSearchBuf = nsnull;
break;
}
if (prevSearchBuf->next->key == aKey) {
// found a key holder
break;
}
prevSearchBuf = prevSearchBuf->next;
}
} // else mFirstBuffer is the keyholder
// prevSearchBuf is the previous buffer before the keyholder or null if
// there isn't one.
} else {
// We have a first-level write in the document.open() case. We insert
// before mLastBuffer. We need to put a marker there, because otherwise
// additional document.writes from nested event loops would insert in the
// wrong place. Sigh.
firstLevelMarker = new nsHtml5OwningUTF16Buffer((void*)nsnull);
if (mFirstBuffer == mLastBuffer) {
firstLevelMarker->next = mLastBuffer;
mFirstBuffer = firstLevelMarker;
} else {
prevSearchBuf = mFirstBuffer;
while (prevSearchBuf->next != mLastBuffer) {
prevSearchBuf = prevSearchBuf->next;
}
firstLevelMarker->next = mLastBuffer;
prevSearchBuf->next = firstLevelMarker;
}
}
nsHtml5DependentUTF16Buffer stackBuffer(aSourceBuffer);
while (!mBlocked && stackBuffer.hasMore()) {
stackBuffer.adjust(mLastWasCR);
mLastWasCR = false;
if (stackBuffer.hasMore()) {
PRInt32 lineNumberSave;
bool inRootContext = (!mStreamParser && !aKey);
if (inRootContext) {
mTokenizer->setLineNumber(mRootContextLineNumber);
} else {
// we aren't the root context, so save the line number on the
// *stack* so that we can restore it.
lineNumberSave = mTokenizer->getLineNumber();
}
mLastWasCR = mTokenizer->tokenizeBuffer(&stackBuffer);
if (inRootContext) {
mRootContextLineNumber = mTokenizer->getLineNumber();
} else {
mTokenizer->setLineNumber(lineNumberSave);
}
if (mTreeBuilder->HasScript()) {
mTreeBuilder->Flush(); // Move ops to the executor
mExecutor->FlushDocumentWrite(); // run the ops
// Flushing tree ops can cause all sorts of things.
// Return early if the parser got terminated.
if (mExecutor->IsComplete()) {
return NS_OK;
}
}
// Ignore suspension requests
}
}
nsRefPtr<nsHtml5OwningUTF16Buffer> heapBuffer;
if (stackBuffer.hasMore()) {
// The buffer wasn't tokenized to completion. Create a copy of the tail
// on the heap.
heapBuffer = stackBuffer.FalliblyCopyAsOwningBuffer();
if (!heapBuffer) {
// Allocation failed. The parser is now broken.
mExecutor->MarkAsBroken();
return NS_ERROR_OUT_OF_MEMORY;
}
}
if (heapBuffer) {
// We have something to insert before the keyholder holding in the non-null
// aKey case and we have something to swap into firstLevelMarker in the
// null aKey case.
if (aKey) {
NS_ASSERTION(mFirstBuffer != mLastBuffer,
"Where's the keyholder?");
// the key holder is still somewhere further down the list from
// prevSearchBuf (which may be null)
if (mFirstBuffer->key == aKey) {
NS_ASSERTION(!prevSearchBuf,
"Non-null prevSearchBuf when mFirstBuffer is the key holder?");
heapBuffer->next = mFirstBuffer;
mFirstBuffer = heapBuffer;
} else {
if (!prevSearchBuf) {
prevSearchBuf = mFirstBuffer;
}
// We created a key holder earlier, so we will find it without walking
// past the end of the list.
while (prevSearchBuf->next->key != aKey) {
prevSearchBuf = prevSearchBuf->next;
}
heapBuffer->next = prevSearchBuf->next;
prevSearchBuf->next = heapBuffer;
}
} else {
NS_ASSERTION(firstLevelMarker, "How come we don't have a marker.");
firstLevelMarker->Swap(heapBuffer);
}
}
if (!mBlocked) { // buffer was tokenized to completion
NS_ASSERTION(!stackBuffer.hasMore(),
"Buffer wasn't tokenized to completion?");
// Scripting semantics require a forced tree builder flush here
mTreeBuilder->Flush(); // Move ops to the executor
mExecutor->FlushDocumentWrite(); // run the ops
} else if (stackBuffer.hasMore()) {
// The buffer wasn't tokenized to completion. Tokenize the untokenized
// content in order to preload stuff. This content will be retokenized
// later for normal parsing.
if (!mDocWriteSpeculatorActive) {
mDocWriteSpeculatorActive = true;
if (!mDocWriteSpeculativeTreeBuilder) {
// Lazily initialize if uninitialized
mDocWriteSpeculativeTreeBuilder =
new nsHtml5TreeBuilder(nsnull, mExecutor->GetStage());
mDocWriteSpeculativeTreeBuilder->setScriptingEnabled(
mTreeBuilder->isScriptingEnabled());
mDocWriteSpeculativeTokenizer =
new nsHtml5Tokenizer(mDocWriteSpeculativeTreeBuilder, false);
mDocWriteSpeculativeTokenizer->setInterner(&mAtomTable);
mDocWriteSpeculativeTokenizer->start();
}
mDocWriteSpeculativeTokenizer->resetToDataState();
mDocWriteSpeculativeTreeBuilder->loadState(mTreeBuilder, &mAtomTable);
mDocWriteSpeculativeLastWasCR = false;
}
// Note that with multilevel document.write if we didn't just activate the
// speculator, it's possible that the speculator is now in the wrong state.
// That's OK for the sake of simplicity. The worst that can happen is
// that the speculative loads aren't exactly right. The content will be
// reparsed anyway for non-preload purposes.
// The buffer position for subsequent non-speculative parsing now lives
// in heapBuffer, so it's ok to let the buffer position of stackBuffer
// to be overwritten and not restored below.
while (stackBuffer.hasMore()) {
stackBuffer.adjust(mDocWriteSpeculativeLastWasCR);
if (stackBuffer.hasMore()) {
mDocWriteSpeculativeLastWasCR =
mDocWriteSpeculativeTokenizer->tokenizeBuffer(&stackBuffer);
}
}
mDocWriteSpeculativeTreeBuilder->Flush();
mDocWriteSpeculativeTreeBuilder->DropHandles();
mExecutor->FlushSpeculativeLoads();
}
return NS_OK;
}
NS_IMETHODIMP
nsHtml5Parser::Parse(const nsAString& aSourceBuffer,
void* aKey,
const nsACString& aContentType, // ignored
bool aLastCall,
nsDTDMode aMode) // ignored
{
NS_PRECONDITION(!mExecutor->IsFragmentMode(),
"Document.write called in fragment mode!");
if (mExecutor->IsBroken()) {
return NS_ERROR_OUT_OF_MEMORY;
}
if (aSourceBuffer.Length() > PR_INT32_MAX) {
mExecutor->MarkAsBroken();
return NS_ERROR_OUT_OF_MEMORY;
}
// Maintain a reference to ourselves so we don't go away
// till we're completely done. The old parser grips itself in this method.
nsCOMPtr<nsIParser> kungFuDeathGrip(this);
// Gripping the other objects just in case, since the other old grip
// required grips to these, too.
nsRefPtr<nsHtml5StreamParser> streamKungFuDeathGrip(mStreamParser);
nsRefPtr<nsHtml5TreeOpExecutor> treeOpKungFuDeathGrip(mExecutor);
if (!mExecutor->HasStarted()) {
NS_ASSERTION(!mStreamParser,
"Had stream parser but document.write started life cycle.");
// This is the first document.write() on a document.open()ed document
mExecutor->SetParser(this);
mTreeBuilder->setScriptingEnabled(mExecutor->IsScriptEnabled());
mTokenizer->start();
mExecutor->Start();
/*
* If you move the following line, be very careful not to cause
* WillBuildModel to be called before the document has had its
* script global object set.
*/
mExecutor->WillBuildModel(eDTDMode_unknown);
}
// Return early if the parser has processed EOF
if (mExecutor->IsComplete()) {
return NS_OK;
}
if (aLastCall && aSourceBuffer.IsEmpty() && aKey == GetRootContextKey()) {
// document.close()
NS_ASSERTION(!mStreamParser,
"Had stream parser but got document.close().");
mDocumentClosed = true;
if (!mBlocked) {
ParseUntilBlocked();
}
return NS_OK;
}
NS_ASSERTION(IsInsertionPointDefined(),
"Doc.write reached parser with undefined insertion point.");
NS_ASSERTION(!(mStreamParser && !aKey),
"Got a null key in a non-script-created parser");
if (aSourceBuffer.IsEmpty()) {
return NS_OK;
}
nsHtml5DependentUTF16Buffer stackBuffer(aSourceBuffer);
while (!mBlocked && stackBuffer.hasMore()) {
stackBuffer.adjust(mLastWasCR);
mLastWasCR = false;
if (stackBuffer.hasMore()) {
PRInt32 lineNumberSave;
bool inRootContext = (!mStreamParser && (aKey == mRootContextKey));
if (inRootContext) {
mTokenizer->setLineNumber(mRootContextLineNumber);
} else {
// we aren't the root context, so save the line number on the
// *stack* so that we can restore it.
lineNumberSave = mTokenizer->getLineNumber();
}
mLastWasCR = mTokenizer->tokenizeBuffer(&stackBuffer);
if (inRootContext) {
mRootContextLineNumber = mTokenizer->getLineNumber();
} else {
mTokenizer->setLineNumber(lineNumberSave);
}
if (mTreeBuilder->HasScript()) {
mTreeBuilder->Flush(); // Move ops to the executor
mExecutor->FlushDocumentWrite(); // run the ops
}
// Ignore suspension requests
}
}
nsRefPtr<nsHtml5OwningUTF16Buffer> heapBuffer;
if (stackBuffer.hasMore()) {
// The buffer wasn't tokenized to completion. Create a copy of the tail
// on the heap.
heapBuffer = stackBuffer.FalliblyCopyAsOwningBuffer();
if (!heapBuffer) {
// Allocation failed. The parser is now broken.
mExecutor->MarkAsBroken();
return NS_ERROR_OUT_OF_MEMORY;
}
}
// The buffer is inserted to the stream here in case it won't be parsed
// to completion.
// The script is identified by aKey. If there's nothing in the buffer
// chain for that key, we'll insert at the head of the queue.
// When the script leaves something in the queue, a zero-length
// key-holder "buffer" is inserted in the queue. If the same script
// leaves something in the chain again, it will be inserted immediately
// before the old key holder belonging to the same script.
nsHtml5OwningUTF16Buffer* prevSearchBuf = nsnull;
nsHtml5OwningUTF16Buffer* searchBuf = mFirstBuffer;
// after document.open, the first level of document.write has null key
if (aKey) {
while (searchBuf != mLastBuffer) {
if (searchBuf->key == aKey) {
// found a key holder
// now insert the new buffer between the previous buffer
// and the key holder if we have a buffer left.
if (heapBuffer) {
heapBuffer->next = searchBuf;
if (prevSearchBuf) {
prevSearchBuf->next = heapBuffer;
} else {
mFirstBuffer = heapBuffer;
}
}
break;
}
prevSearchBuf = searchBuf;
searchBuf = searchBuf->next;
}
if (searchBuf == mLastBuffer) {
// key was not found
nsHtml5OwningUTF16Buffer* keyHolder = new nsHtml5OwningUTF16Buffer(aKey);
keyHolder->next = mFirstBuffer;
if (heapBuffer) {
heapBuffer->next = keyHolder;
mFirstBuffer = heapBuffer;
} else {
mFirstBuffer = keyHolder;
}
}
} else if (heapBuffer) {
// we have a first level document.write after document.open()
// insert immediately before mLastBuffer
while (searchBuf != mLastBuffer) {
prevSearchBuf = searchBuf;
searchBuf = searchBuf->next;
}
heapBuffer->next = mLastBuffer;
if (prevSearchBuf) {
prevSearchBuf->next = heapBuffer;
} else {
mFirstBuffer = heapBuffer;
}
}
if (!mBlocked) { // buffer was tokenized to completion
NS_ASSERTION(!stackBuffer.hasMore(),
"Buffer wasn't tokenized to completion?");
// Scripting semantics require a forced tree builder flush here
mTreeBuilder->Flush(); // Move ops to the executor
mExecutor->FlushDocumentWrite(); // run the ops
} else if (stackBuffer.hasMore()) {
// The buffer wasn't tokenized to completion. Tokenize the untokenized
// content in order to preload stuff. This content will be retokenized
// later for normal parsing.
if (!mDocWriteSpeculatorActive) {
mDocWriteSpeculatorActive = true;
if (!mDocWriteSpeculativeTreeBuilder) {
// Lazily initialize if uninitialized
mDocWriteSpeculativeTreeBuilder =
new nsHtml5TreeBuilder(nsnull, mExecutor->GetStage());
mDocWriteSpeculativeTreeBuilder->setScriptingEnabled(
mTreeBuilder->isScriptingEnabled());
mDocWriteSpeculativeTokenizer =
new nsHtml5Tokenizer(mDocWriteSpeculativeTreeBuilder);
mDocWriteSpeculativeTokenizer->setInterner(&mAtomTable);
mDocWriteSpeculativeTokenizer->start();
}
mDocWriteSpeculativeTokenizer->resetToDataState();
mDocWriteSpeculativeTreeBuilder->loadState(mTreeBuilder, &mAtomTable);
mDocWriteSpeculativeLastWasCR = false;
}
// Note that with multilevel document.write if we didn't just activate the
// speculator, it's possible that the speculator is now in the wrong state.
// That's OK for the sake of simplicity. The worst that can happen is
// that the speculative loads aren't exactly right. The content will be
// reparsed anyway for non-preload purposes.
// The buffer position for subsequent non-speculative parsing now lives
// in heapBuffer, so it's ok to let the buffer position of stackBuffer
// to be overwritten and not restored below.
while (stackBuffer.hasMore()) {
stackBuffer.adjust(mDocWriteSpeculativeLastWasCR);
if (stackBuffer.hasMore()) {
mDocWriteSpeculativeLastWasCR =
mDocWriteSpeculativeTokenizer->tokenizeBuffer(&stackBuffer);
}
}
mDocWriteSpeculativeTreeBuilder->Flush();
mDocWriteSpeculativeTreeBuilder->DropHandles();
mExecutor->FlushSpeculativeLoads();
}
return NS_OK;
}
