gecko/dom/base/nsContentIterator.cpp

1540 lines
42 KiB
C++

/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/* vim: set ts=8 sts=2 et sw=2 tw=80: */
/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
#include "nsISupports.h"
#include "nsIDOMNodeList.h"
#include "nsIContentIterator.h"
#include "nsRange.h"
#include "nsIContent.h"
#include "nsCOMPtr.h"
#include "nsTArray.h"
#include "nsContentUtils.h"
#include "nsINode.h"
#include "nsCycleCollectionParticipant.h"
// couple of utility static functs
///////////////////////////////////////////////////////////////////////////
// NodeToParentOffset: returns the node's parent and offset.
//
static nsINode*
NodeToParentOffset(nsINode* aNode, int32_t* aOffset)
{
*aOffset = 0;
nsINode* parent = aNode->GetParentNode();
if (parent) {
*aOffset = parent->IndexOf(aNode);
NS_WARN_IF(*aOffset < 0);
}
return parent;
}
///////////////////////////////////////////////////////////////////////////
// NodeIsInTraversalRange: returns true if content is visited during
// the traversal of the range in the specified mode.
//
static bool
NodeIsInTraversalRange(nsINode* aNode, bool aIsPreMode,
nsINode* aStartNode, int32_t aStartOffset,
nsINode* aEndNode, int32_t aEndOffset)
{
if (NS_WARN_IF(!aStartNode) || NS_WARN_IF(!aEndNode) || NS_WARN_IF(!aNode)) {
return false;
}
// If a leaf node contains an end point of the traversal range, it is
// always in the traversal range.
if (aNode == aStartNode || aNode == aEndNode) {
if (aNode->IsNodeOfType(nsINode::eDATA_NODE)) {
return true; // text node or something
}
if (!aNode->HasChildren()) {
MOZ_ASSERT(aNode != aStartNode || !aStartOffset,
"aStartNode doesn't have children and not a data node, "
"aStartOffset should be 0");
MOZ_ASSERT(aNode != aEndNode || !aEndOffset,
"aStartNode doesn't have children and not a data node, "
"aStartOffset should be 0");
return true;
}
}
nsINode* parent = aNode->GetParentNode();
if (!parent) {
return false;
}
int32_t indx = parent->IndexOf(aNode);
NS_WARN_IF(indx == -1);
if (!aIsPreMode) {
++indx;
}
return nsContentUtils::ComparePoints(aStartNode, aStartOffset,
parent, indx) <= 0 &&
nsContentUtils::ComparePoints(aEndNode, aEndOffset,
parent, indx) >= 0;
}
/*
* A simple iterator class for traversing the content in "close tag" order
*/
class nsContentIterator : public nsIContentIterator
{
public:
NS_DECL_CYCLE_COLLECTING_ISUPPORTS
NS_DECL_CYCLE_COLLECTION_CLASS(nsContentIterator)
explicit nsContentIterator(bool aPre);
// nsIContentIterator interface methods ------------------------------
virtual nsresult Init(nsINode* aRoot) override;
virtual nsresult Init(nsIDOMRange* aRange) override;
virtual void First() override;
virtual void Last() override;
virtual void Next() override;
virtual void Prev() override;
virtual nsINode* GetCurrentNode() override;
virtual bool IsDone() override;
virtual nsresult PositionAt(nsINode* aCurNode) override;
protected:
virtual ~nsContentIterator();
// Recursively get the deepest first/last child of aRoot. This will return
// aRoot itself if it has no children.
nsINode* GetDeepFirstChild(nsINode* aRoot,
nsTArray<int32_t>* aIndexes = nullptr);
nsIContent* GetDeepFirstChild(nsIContent* aRoot,
nsTArray<int32_t>* aIndexes = nullptr);
nsINode* GetDeepLastChild(nsINode* aRoot,
nsTArray<int32_t>* aIndexes = nullptr);
nsIContent* GetDeepLastChild(nsIContent* aRoot,
nsTArray<int32_t>* aIndexes = nullptr);
// Get the next/previous sibling of aNode, or its parent's, or grandparent's,
// etc. Returns null if aNode and all its ancestors have no next/previous
// sibling.
nsIContent* GetNextSibling(nsINode* aNode,
nsTArray<int32_t>* aIndexes = nullptr);
nsIContent* GetPrevSibling(nsINode* aNode,
nsTArray<int32_t>* aIndexes = nullptr);
nsINode* NextNode(nsINode* aNode, nsTArray<int32_t>* aIndexes = nullptr);
nsINode* PrevNode(nsINode* aNode, nsTArray<int32_t>* aIndexes = nullptr);
// WARNING: This function is expensive
nsresult RebuildIndexStack();
void MakeEmpty();
virtual void LastRelease();
nsCOMPtr<nsINode> mCurNode;
nsCOMPtr<nsINode> mFirst;
nsCOMPtr<nsINode> mLast;
nsCOMPtr<nsINode> mCommonParent;
// used by nsContentIterator to cache indices
nsAutoTArray<int32_t, 8> mIndexes;
// used by nsSubtreeIterator to cache indices. Why put them in the base
// class? Because otherwise I have to duplicate the routines GetNextSibling
// etc across both classes, with slight variations for caching. Or
// alternately, create a base class for the cache itself and have all the
// cache manipulation go through a vptr. I think this is the best space and
// speed combo, even though it's ugly.
int32_t mCachedIndex;
// another note about mCachedIndex: why should the subtree iterator use a
// trivial cached index instead of the mre robust array of indicies (which is
// what the basic content iterator uses)? The reason is that subtree
// iterators do not do much transitioning between parents and children. They
// tend to stay at the same level. In fact, you can prove (though I won't
// attempt it here) that they change levels at most n+m times, where n is the
// height of the parent hierarchy from the range start to the common
// ancestor, and m is the the height of the parent hierarchy from the range
// end to the common ancestor. If we used the index array, we would pay the
// price up front for n, and then pay the cost for m on the fly later on.
// With the simple cache, we only "pay as we go". Either way, we call
// IndexOf() once for each change of level in the hierarchy. Since a trivial
// index is much simpler, we use it for the subtree iterator.
bool mIsDone;
bool mPre;
private:
// no copies or assigns FIX ME
nsContentIterator(const nsContentIterator&);
nsContentIterator& operator=(const nsContentIterator&);
};
/******************************************************
* repository cruft
******************************************************/
already_AddRefed<nsIContentIterator>
NS_NewContentIterator()
{
nsCOMPtr<nsIContentIterator> iter = new nsContentIterator(false);
return iter.forget();
}
already_AddRefed<nsIContentIterator>
NS_NewPreContentIterator()
{
nsCOMPtr<nsIContentIterator> iter = new nsContentIterator(true);
return iter.forget();
}
/******************************************************
* XPCOM cruft
******************************************************/
NS_IMPL_CYCLE_COLLECTING_ADDREF(nsContentIterator)
NS_IMPL_CYCLE_COLLECTING_RELEASE_WITH_LAST_RELEASE(nsContentIterator,
LastRelease())
NS_INTERFACE_MAP_BEGIN(nsContentIterator)
NS_INTERFACE_MAP_ENTRY(nsIContentIterator)
NS_INTERFACE_MAP_ENTRY_AMBIGUOUS(nsISupports, nsIContentIterator)
NS_INTERFACE_MAP_ENTRIES_CYCLE_COLLECTION(nsContentIterator)
NS_INTERFACE_MAP_END
NS_IMPL_CYCLE_COLLECTION(nsContentIterator,
mCurNode,
mFirst,
mLast,
mCommonParent)
void
nsContentIterator::LastRelease()
{
mCurNode = nullptr;
mFirst = nullptr;
mLast = nullptr;
mCommonParent = nullptr;
}
/******************************************************
* constructor/destructor
******************************************************/
nsContentIterator::nsContentIterator(bool aPre) :
// don't need to explicitly initialize |nsCOMPtr|s, they will automatically
// be nullptr
mCachedIndex(0), mIsDone(false), mPre(aPre)
{
}
nsContentIterator::~nsContentIterator()
{
}
/******************************************************
* Init routines
******************************************************/
nsresult
nsContentIterator::Init(nsINode* aRoot)
{
if (NS_WARN_IF(!aRoot)) {
return NS_ERROR_NULL_POINTER;
}
mIsDone = false;
mIndexes.Clear();
if (mPre) {
mFirst = aRoot;
mLast = GetDeepLastChild(aRoot);
NS_WARN_IF(!mLast);
} else {
mFirst = GetDeepFirstChild(aRoot);
NS_WARN_IF(!mFirst);
mLast = aRoot;
}
mCommonParent = aRoot;
mCurNode = mFirst;
RebuildIndexStack();
return NS_OK;
}
nsresult
nsContentIterator::Init(nsIDOMRange* aDOMRange)
{
if (NS_WARN_IF(!aDOMRange)) {
return NS_ERROR_INVALID_ARG;
}
nsRange* range = static_cast<nsRange*>(aDOMRange);
mIsDone = false;
// get common content parent
mCommonParent = range->GetCommonAncestor();
if (NS_WARN_IF(!mCommonParent)) {
return NS_ERROR_FAILURE;
}
// get the start node and offset
int32_t startIndx = range->StartOffset();
NS_WARN_IF(startIndx < 0);
nsINode* startNode = range->GetStartParent();
if (NS_WARN_IF(!startNode)) {
return NS_ERROR_FAILURE;
}
// get the end node and offset
int32_t endIndx = range->EndOffset();
NS_WARN_IF(endIndx < 0);
nsINode* endNode = range->GetEndParent();
if (NS_WARN_IF(!endNode)) {
return NS_ERROR_FAILURE;
}
bool startIsData = startNode->IsNodeOfType(nsINode::eDATA_NODE);
// short circuit when start node == end node
if (startNode == endNode) {
// Check to see if we have a collapsed range, if so, there is nothing to
// iterate over.
//
// XXX: CharacterDataNodes (text nodes) are currently an exception, since
// we always want to be able to iterate text nodes at the end points
// of a range.
if (!startIsData && startIndx == endIndx) {
MakeEmpty();
return NS_OK;
}
if (startIsData) {
// It's a character data node.
mFirst = startNode->AsContent();
mLast = mFirst;
mCurNode = mFirst;
nsresult rv = RebuildIndexStack();
NS_WARN_IF(NS_FAILED(rv));
return NS_OK;
}
}
// Find first node in range.
nsIContent* cChild = nullptr;
if (!startIsData && startNode->HasChildren()) {
cChild = startNode->GetChildAt(startIndx);
NS_WARN_IF(!cChild);
}
if (!cChild) {
// no children, must be a text node
//
// XXXbz no children might also just mean no children. So I'm not
// sure what that comment above is talking about.
if (mPre) {
// XXX: In the future, if start offset is after the last
// character in the cdata node, should we set mFirst to
// the next sibling?
// If the node has no child, the child may be <br> or something.
// So, we shouldn't skip the empty node if the start offset is 0.
// In other words, if the offset is 1, the node should be ignored.
if (!startIsData && startIndx) {
mFirst = GetNextSibling(startNode);
NS_WARN_IF(!mFirst);
// Does mFirst node really intersect the range? The range could be
// 'degenerate', i.e., not collapsed but still contain no content.
if (mFirst &&
NS_WARN_IF(!NodeIsInTraversalRange(mFirst, mPre, startNode,
startIndx, endNode, endIndx))) {
mFirst = nullptr;
}
} else {
mFirst = startNode->AsContent();
}
} else {
// post-order
if (NS_WARN_IF(!startNode->IsContent())) {
// What else can we do?
mFirst = nullptr;
} else {
mFirst = startNode->AsContent();
}
}
} else {
if (mPre) {
mFirst = cChild;
} else {
// post-order
mFirst = GetDeepFirstChild(cChild);
NS_WARN_IF(!mFirst);
// Does mFirst node really intersect the range? The range could be
// 'degenerate', i.e., not collapsed but still contain no content.
if (mFirst &&
NS_WARN_IF(!NodeIsInTraversalRange(mFirst, mPre, startNode, startIndx,
endNode, endIndx))) {
mFirst = nullptr;
}
}
}
// Find last node in range.
bool endIsData = endNode->IsNodeOfType(nsINode::eDATA_NODE);
if (endIsData || !endNode->HasChildren() || endIndx == 0) {
if (mPre) {
if (NS_WARN_IF(!endNode->IsContent())) {
// Not much else to do here...
mLast = nullptr;
} else {
// If the end node is an empty element and the end offset is 0,
// the last element should be the previous node (i.e., shouldn't
// include the end node in the range).
if (!endIsData && !endNode->HasChildren() && !endIndx) {
mLast = GetPrevSibling(endNode);
NS_WARN_IF(!mLast);
if (NS_WARN_IF(!NodeIsInTraversalRange(mLast, mPre,
startNode, startIndx,
endNode, endIndx))) {
mLast = nullptr;
}
} else {
mLast = endNode->AsContent();
}
}
} else {
// post-order
//
// XXX: In the future, if end offset is before the first character in the
// cdata node, should we set mLast to the prev sibling?
if (!endIsData) {
mLast = GetPrevSibling(endNode);
NS_WARN_IF(!mLast);
if (NS_WARN_IF(!NodeIsInTraversalRange(mLast, mPre,
startNode, startIndx,
endNode, endIndx))) {
mLast = nullptr;
}
} else {
mLast = endNode->AsContent();
}
}
} else {
int32_t indx = endIndx;
cChild = endNode->GetChildAt(--indx);
if (NS_WARN_IF(!cChild)) {
// No child at offset!
NS_NOTREACHED("nsContentIterator::nsContentIterator");
return NS_ERROR_FAILURE;
}
if (mPre) {
mLast = GetDeepLastChild(cChild);
NS_WARN_IF(!mLast);
if (NS_WARN_IF(!NodeIsInTraversalRange(mLast, mPre,
startNode, startIndx,
endNode, endIndx))) {
mLast = nullptr;
}
} else {
// post-order
mLast = cChild;
}
}
// If either first or last is null, they both have to be null!
if (NS_WARN_IF(!mFirst) || NS_WARN_IF(!mLast)) {
mFirst = nullptr;
mLast = nullptr;
}
mCurNode = mFirst;
mIsDone = !mCurNode;
if (!mCurNode) {
mIndexes.Clear();
} else {
nsresult rv = RebuildIndexStack();
NS_WARN_IF(NS_FAILED(rv));
}
return NS_OK;
}
/******************************************************
* Helper routines
******************************************************/
// WARNING: This function is expensive
nsresult
nsContentIterator::RebuildIndexStack()
{
// Make sure we start at the right indexes on the stack! Build array up
// to common parent of start and end. Perhaps it's too many entries, but
// that's far better than too few.
nsINode* parent;
nsINode* current;
mIndexes.Clear();
current = mCurNode;
if (!current) {
return NS_OK;
}
while (current != mCommonParent) {
parent = current->GetParentNode();
if (NS_WARN_IF(!parent)) {
return NS_ERROR_FAILURE;
}
mIndexes.InsertElementAt(0, parent->IndexOf(current));
current = parent;
}
return NS_OK;
}
void
nsContentIterator::MakeEmpty()
{
mCurNode = nullptr;
mFirst = nullptr;
mLast = nullptr;
mCommonParent = nullptr;
mIsDone = true;
mIndexes.Clear();
}
nsINode*
nsContentIterator::GetDeepFirstChild(nsINode* aRoot,
nsTArray<int32_t>* aIndexes)
{
if (NS_WARN_IF(!aRoot) || !aRoot->HasChildren()) {
return aRoot;
}
// We can't pass aRoot itself to the full GetDeepFirstChild, because that
// will only take nsIContent and aRoot might be a document. Pass aRoot's
// child, but be sure to preserve aIndexes.
if (aIndexes) {
aIndexes->AppendElement(0);
}
return GetDeepFirstChild(aRoot->GetFirstChild(), aIndexes);
}
nsIContent*
nsContentIterator::GetDeepFirstChild(nsIContent* aRoot,
nsTArray<int32_t>* aIndexes)
{
if (NS_WARN_IF(!aRoot)) {
return nullptr;
}
nsIContent* node = aRoot;
nsIContent* child = node->GetFirstChild();
while (child) {
if (aIndexes) {
// Add this node to the stack of indexes
aIndexes->AppendElement(0);
}
node = child;
child = node->GetFirstChild();
}
return node;
}
nsINode*
nsContentIterator::GetDeepLastChild(nsINode* aRoot,
nsTArray<int32_t>* aIndexes)
{
if (NS_WARN_IF(!aRoot) || !aRoot->HasChildren()) {
return aRoot;
}
// We can't pass aRoot itself to the full GetDeepLastChild, because that will
// only take nsIContent and aRoot might be a document. Pass aRoot's child,
// but be sure to preserve aIndexes.
if (aIndexes) {
aIndexes->AppendElement(aRoot->GetChildCount() - 1);
}
return GetDeepLastChild(aRoot->GetLastChild(), aIndexes);
}
nsIContent*
nsContentIterator::GetDeepLastChild(nsIContent* aRoot,
nsTArray<int32_t>* aIndexes)
{
if (NS_WARN_IF(!aRoot)) {
return nullptr;
}
nsIContent* node = aRoot;
int32_t numChildren = node->GetChildCount();
while (numChildren) {
nsIContent* child = node->GetChildAt(--numChildren);
if (aIndexes) {
// Add this node to the stack of indexes
aIndexes->AppendElement(numChildren);
}
numChildren = child->GetChildCount();
node = child;
}
return node;
}
// Get the next sibling, or parent's next sibling, or grandpa's next sibling...
nsIContent*
nsContentIterator::GetNextSibling(nsINode* aNode,
nsTArray<int32_t>* aIndexes)
{
if (NS_WARN_IF(!aNode)) {
return nullptr;
}
nsINode* parent = aNode->GetParentNode();
if (NS_WARN_IF(!parent)) {
return nullptr;
}
int32_t indx = 0;
NS_ASSERTION(!aIndexes || !aIndexes->IsEmpty(),
"ContentIterator stack underflow");
if (aIndexes && !aIndexes->IsEmpty()) {
// use the last entry on the Indexes array for the current index
indx = (*aIndexes)[aIndexes->Length()-1];
} else {
indx = mCachedIndex;
}
NS_WARN_IF(indx < 0);
// reverify that the index of the current node hasn't changed.
// not super cheap, but a lot cheaper than IndexOf(), and still O(1).
// ignore result this time - the index may now be out of range.
nsIContent* sib = parent->GetChildAt(indx);
if (sib != aNode) {
// someone changed our index - find the new index the painful way
indx = parent->IndexOf(aNode);
NS_WARN_IF(indx < 0);
}
// indx is now canonically correct
if ((sib = parent->GetChildAt(++indx))) {
// update index cache
if (aIndexes && !aIndexes->IsEmpty()) {
aIndexes->ElementAt(aIndexes->Length()-1) = indx;
} else {
mCachedIndex = indx;
}
} else {
if (parent != mCommonParent) {
if (aIndexes) {
// pop node off the stack, go up one level and return parent or fail.
// Don't leave the index empty, especially if we're
// returning nullptr. This confuses other parts of the code.
if (aIndexes->Length() > 1) {
aIndexes->RemoveElementAt(aIndexes->Length()-1);
}
}
}
// ok to leave cache out of date here if parent == mCommonParent?
sib = GetNextSibling(parent, aIndexes);
}
return sib;
}
// Get the prev sibling, or parent's prev sibling, or grandpa's prev sibling...
nsIContent*
nsContentIterator::GetPrevSibling(nsINode* aNode,
nsTArray<int32_t>* aIndexes)
{
if (NS_WARN_IF(!aNode)) {
return nullptr;
}
nsINode* parent = aNode->GetParentNode();
if (NS_WARN_IF(!parent)) {
return nullptr;
}
int32_t indx = 0;
NS_ASSERTION(!aIndexes || !aIndexes->IsEmpty(),
"ContentIterator stack underflow");
if (aIndexes && !aIndexes->IsEmpty()) {
// use the last entry on the Indexes array for the current index
indx = (*aIndexes)[aIndexes->Length()-1];
} else {
indx = mCachedIndex;
}
// reverify that the index of the current node hasn't changed
// ignore result this time - the index may now be out of range.
nsIContent* sib = parent->GetChildAt(indx);
if (sib != aNode) {
// someone changed our index - find the new index the painful way
indx = parent->IndexOf(aNode);
NS_WARN_IF(indx < 0);
}
// indx is now canonically correct
if (indx > 0 && (sib = parent->GetChildAt(--indx))) {
// update index cache
if (aIndexes && !aIndexes->IsEmpty()) {
aIndexes->ElementAt(aIndexes->Length()-1) = indx;
} else {
mCachedIndex = indx;
}
} else if (parent != mCommonParent) {
if (aIndexes && !aIndexes->IsEmpty()) {
// pop node off the stack, go up one level and try again.
aIndexes->RemoveElementAt(aIndexes->Length()-1);
}
return GetPrevSibling(parent, aIndexes);
}
return sib;
}
nsINode*
nsContentIterator::NextNode(nsINode* aNode, nsTArray<int32_t>* aIndexes)
{
nsINode* node = aNode;
// if we are a Pre-order iterator, use pre-order
if (mPre) {
// if it has children then next node is first child
if (node->HasChildren()) {
nsIContent* firstChild = node->GetFirstChild();
MOZ_ASSERT(firstChild);
// update cache
if (aIndexes) {
// push an entry on the index stack
aIndexes->AppendElement(0);
} else {
mCachedIndex = 0;
}
return firstChild;
}
// else next sibling is next
return GetNextSibling(node, aIndexes);
}
// post-order
nsINode* parent = node->GetParentNode();
NS_WARN_IF(!parent);
nsIContent* sibling = nullptr;
int32_t indx = 0;
// get the cached index
NS_ASSERTION(!aIndexes || !aIndexes->IsEmpty(),
"ContentIterator stack underflow");
if (aIndexes && !aIndexes->IsEmpty()) {
// use the last entry on the Indexes array for the current index
indx = (*aIndexes)[aIndexes->Length()-1];
} else {
indx = mCachedIndex;
}
// reverify that the index of the current node hasn't changed. not super
// cheap, but a lot cheaper than IndexOf(), and still O(1). ignore result
// this time - the index may now be out of range.
if (indx >= 0) {
sibling = parent->GetChildAt(indx);
}
if (sibling != node) {
// someone changed our index - find the new index the painful way
indx = parent->IndexOf(node);
NS_WARN_IF(indx < 0);
}
// indx is now canonically correct
sibling = parent->GetChildAt(++indx);
if (sibling) {
// update cache
if (aIndexes && !aIndexes->IsEmpty()) {
// replace an entry on the index stack
aIndexes->ElementAt(aIndexes->Length()-1) = indx;
} else {
mCachedIndex = indx;
}
// next node is sibling's "deep left" child
return GetDeepFirstChild(sibling, aIndexes);
}
// else it's the parent, update cache
if (aIndexes) {
// Pop an entry off the index stack. Don't leave the index empty,
// especially if we're returning nullptr. This confuses other parts of the
// code.
if (aIndexes->Length() > 1) {
aIndexes->RemoveElementAt(aIndexes->Length()-1);
}
} else {
// this might be wrong, but we are better off guessing
mCachedIndex = 0;
}
return parent;
}
nsINode*
nsContentIterator::PrevNode(nsINode* aNode, nsTArray<int32_t>* aIndexes)
{
nsINode* node = aNode;
// if we are a Pre-order iterator, use pre-order
if (mPre) {
nsINode* parent = node->GetParentNode();
NS_WARN_IF(!parent);
nsIContent* sibling = nullptr;
int32_t indx = 0;
// get the cached index
NS_ASSERTION(!aIndexes || !aIndexes->IsEmpty(),
"ContentIterator stack underflow");
if (aIndexes && !aIndexes->IsEmpty()) {
// use the last entry on the Indexes array for the current index
indx = (*aIndexes)[aIndexes->Length()-1];
} else {
indx = mCachedIndex;
}
// reverify that the index of the current node hasn't changed. not super
// cheap, but a lot cheaper than IndexOf(), and still O(1). ignore result
// this time - the index may now be out of range.
if (indx >= 0) {
sibling = parent->GetChildAt(indx);
NS_WARN_IF(!sibling);
}
if (sibling != node) {
// someone changed our index - find the new index the painful way
indx = parent->IndexOf(node);
NS_WARN_IF(indx < 0);
}
// indx is now canonically correct
if (indx && (sibling = parent->GetChildAt(--indx))) {
// update cache
if (aIndexes && !aIndexes->IsEmpty()) {
// replace an entry on the index stack
aIndexes->ElementAt(aIndexes->Length()-1) = indx;
} else {
mCachedIndex = indx;
}
// prev node is sibling's "deep right" child
return GetDeepLastChild(sibling, aIndexes);
}
// else it's the parent, update cache
if (aIndexes && !aIndexes->IsEmpty()) {
// pop an entry off the index stack
aIndexes->RemoveElementAt(aIndexes->Length()-1);
} else {
// this might be wrong, but we are better off guessing
mCachedIndex = 0;
}
return parent;
}
// post-order
int32_t numChildren = node->GetChildCount();
NS_WARN_IF(numChildren < 0);
// if it has children then prev node is last child
if (numChildren) {
nsIContent* lastChild = node->GetLastChild();
NS_WARN_IF(!lastChild);
numChildren--;
// update cache
if (aIndexes) {
// push an entry on the index stack
aIndexes->AppendElement(numChildren);
} else {
mCachedIndex = numChildren;
}
return lastChild;
}
// else prev sibling is previous
return GetPrevSibling(node, aIndexes);
}
/******************************************************
* ContentIterator routines
******************************************************/
void
nsContentIterator::First()
{
if (mFirst) {
DebugOnly<nsresult> rv = PositionAt(mFirst);
NS_ASSERTION(NS_SUCCEEDED(rv), "Failed to position iterator!");
}
mIsDone = mFirst == nullptr;
}
void
nsContentIterator::Last()
{
NS_ASSERTION(mLast, "No last node!");
if (mLast) {
DebugOnly<nsresult> rv = PositionAt(mLast);
NS_ASSERTION(NS_SUCCEEDED(rv), "Failed to position iterator!");
}
mIsDone = mLast == nullptr;
}
void
nsContentIterator::Next()
{
if (mIsDone || NS_WARN_IF(!mCurNode)) {
return;
}
if (mCurNode == mLast) {
mIsDone = true;
return;
}
mCurNode = NextNode(mCurNode, &mIndexes);
}
void
nsContentIterator::Prev()
{
if (NS_WARN_IF(mIsDone) || NS_WARN_IF(!mCurNode)) {
return;
}
if (mCurNode == mFirst) {
mIsDone = true;
return;
}
mCurNode = PrevNode(mCurNode, &mIndexes);
}
bool
nsContentIterator::IsDone()
{
return mIsDone;
}
// Keeping arrays of indexes for the stack of nodes makes PositionAt
// interesting...
nsresult
nsContentIterator::PositionAt(nsINode* aCurNode)
{
if (NS_WARN_IF(!aCurNode)) {
return NS_ERROR_NULL_POINTER;
}
nsINode* newCurNode = aCurNode;
nsINode* tempNode = mCurNode;
mCurNode = aCurNode;
// take an early out if this doesn't actually change the position
if (mCurNode == tempNode) {
mIsDone = false; // paranoia
return NS_OK;
}
// Check to see if the node falls within the traversal range.
nsINode* firstNode = mFirst;
nsINode* lastNode = mLast;
int32_t firstOffset = 0, lastOffset = 0;
if (firstNode && lastNode) {
if (mPre) {
firstNode = NodeToParentOffset(mFirst, &firstOffset);
NS_WARN_IF(!firstNode);
NS_WARN_IF(firstOffset < 0);
if (lastNode->GetChildCount()) {
lastOffset = 0;
} else {
lastNode = NodeToParentOffset(mLast, &lastOffset);
NS_WARN_IF(!lastNode);
NS_WARN_IF(lastOffset < 0);
++lastOffset;
}
} else {
uint32_t numChildren = firstNode->GetChildCount();
if (numChildren) {
firstOffset = numChildren;
NS_WARN_IF(firstOffset < 0);
} else {
firstNode = NodeToParentOffset(mFirst, &firstOffset);
NS_WARN_IF(!firstNode);
NS_WARN_IF(firstOffset < 0);
}
lastNode = NodeToParentOffset(mLast, &lastOffset);
NS_WARN_IF(!lastNode);
NS_WARN_IF(lastOffset < 0);
++lastOffset;
}
}
// The end positions are always in the range even if it has no parent. We
// need to allow that or 'iter->Init(root)' would assert in Last() or First()
// for example, bug 327694.
if (mFirst != mCurNode && mLast != mCurNode &&
(NS_WARN_IF(!firstNode) || NS_WARN_IF(!lastNode) ||
NS_WARN_IF(!NodeIsInTraversalRange(mCurNode, mPre,
firstNode, firstOffset,
lastNode, lastOffset)))) {
mIsDone = true;
return NS_ERROR_FAILURE;
}
// We can be at ANY node in the sequence. Need to regenerate the array of
// indexes back to the root or common parent!
nsAutoTArray<nsINode*, 8> oldParentStack;
nsAutoTArray<int32_t, 8> newIndexes;
// Get a list of the parents up to the root, then compare the new node with
// entries in that array until we find a match (lowest common ancestor). If
// no match, use IndexOf, take the parent, and repeat. This avoids using
// IndexOf() N times on possibly large arrays. We still end up doing it a
// fair bit. It's better to use Clone() if possible.
// we know the depth we're down (though we may not have started at the top).
oldParentStack.SetCapacity(mIndexes.Length() + 1);
// We want to loop mIndexes.Length() + 1 times here, because we want to make
// sure we include mCommonParent in the oldParentStack, for use in the next
// for loop, and mIndexes only has entries for nodes from tempNode up through
// an ancestor of tempNode that's a child of mCommonParent.
for (int32_t i = mIndexes.Length() + 1; i > 0 && tempNode; i--) {
// Insert at head since we're walking up
oldParentStack.InsertElementAt(0, tempNode);
nsINode* parent = tempNode->GetParentNode();
if (NS_WARN_IF(!parent)) {
// this node has no parent, and thus no index
break;
}
if (parent == mCurNode) {
// The position was moved to a parent of the current position. All we
// need to do is drop some indexes. Shortcut here.
mIndexes.RemoveElementsAt(mIndexes.Length() - oldParentStack.Length(),
oldParentStack.Length());
mIsDone = false;
return NS_OK;
}
tempNode = parent;
}
// Ok. We have the array of old parents. Look for a match.
while (newCurNode) {
nsINode* parent = newCurNode->GetParentNode();
if (NS_WARN_IF(!parent)) {
// this node has no parent, and thus no index
break;
}
int32_t indx = parent->IndexOf(newCurNode);
NS_WARN_IF(indx < 0);
// insert at the head!
newIndexes.InsertElementAt(0, indx);
// look to see if the parent is in the stack
indx = oldParentStack.IndexOf(parent);
if (indx >= 0) {
// ok, the parent IS on the old stack! Rework things. We want
// newIndexes to replace all nodes equal to or below the match. Note
// that index oldParentStack.Length() - 1 is the last node, which is one
// BELOW the last index in the mIndexes stack. In other words, we want
// to remove elements starting at index (indx + 1).
int32_t numToDrop = oldParentStack.Length() - (1 + indx);
if (numToDrop > 0) {
mIndexes.RemoveElementsAt(mIndexes.Length() - numToDrop, numToDrop);
}
mIndexes.AppendElements(newIndexes);
break;
}
newCurNode = parent;
}
// phew!
mIsDone = false;
return NS_OK;
}
nsINode*
nsContentIterator::GetCurrentNode()
{
if (mIsDone) {
return nullptr;
}
NS_ASSERTION(mCurNode, "Null current node in an iterator that's not done!");
return mCurNode;
}
/*====================================================================================*/
/*====================================================================================*/
/******************************************************
* nsContentSubtreeIterator
******************************************************/
/*
* A simple iterator class for traversing the content in "top subtree" order
*/
class nsContentSubtreeIterator : public nsContentIterator
{
public:
nsContentSubtreeIterator() : nsContentIterator(false) {}
NS_DECL_ISUPPORTS_INHERITED
NS_DECL_CYCLE_COLLECTION_CLASS_INHERITED(nsContentSubtreeIterator, nsContentIterator)
// nsContentIterator overrides ------------------------------
virtual nsresult Init(nsINode* aRoot) override;
virtual nsresult Init(nsIDOMRange* aRange) override;
virtual void Next() override;
virtual void Prev() override;
virtual nsresult PositionAt(nsINode* aCurNode) override;
// Must override these because we don't do PositionAt
virtual void First() override;
// Must override these because we don't do PositionAt
virtual void Last() override;
protected:
virtual ~nsContentSubtreeIterator() {}
// Returns the highest inclusive ancestor of aNode that's in the range
// (possibly aNode itself). Returns null if aNode is null, or is not itself
// in the range. A node is in the range if (node, 0) comes strictly after
// the range endpoint, and (node, node.length) comes strictly before it, so
// the range's start and end nodes will never be considered "in" it.
nsIContent* GetTopAncestorInRange(nsINode* aNode);
// no copy's or assigns FIX ME
nsContentSubtreeIterator(const nsContentSubtreeIterator&);
nsContentSubtreeIterator& operator=(const nsContentSubtreeIterator&);
virtual void LastRelease() override;
RefPtr<nsRange> mRange;
// these arrays all typically are used and have elements
nsAutoTArray<nsIContent*, 8> mEndNodes;
nsAutoTArray<int32_t, 8> mEndOffsets;
};
NS_IMPL_ADDREF_INHERITED(nsContentSubtreeIterator, nsContentIterator)
NS_IMPL_RELEASE_INHERITED(nsContentSubtreeIterator, nsContentIterator)
NS_INTERFACE_MAP_BEGIN_CYCLE_COLLECTION_INHERITED(nsContentSubtreeIterator)
NS_INTERFACE_MAP_END_INHERITING(nsContentIterator)
NS_IMPL_CYCLE_COLLECTION_INHERITED(nsContentSubtreeIterator, nsContentIterator,
mRange)
void
nsContentSubtreeIterator::LastRelease()
{
mRange = nullptr;
nsContentIterator::LastRelease();
}
/******************************************************
* repository cruft
******************************************************/
already_AddRefed<nsIContentIterator>
NS_NewContentSubtreeIterator()
{
nsCOMPtr<nsIContentIterator> iter = new nsContentSubtreeIterator();
return iter.forget();
}
/******************************************************
* Init routines
******************************************************/
nsresult
nsContentSubtreeIterator::Init(nsINode* aRoot)
{
return NS_ERROR_NOT_IMPLEMENTED;
}
nsresult
nsContentSubtreeIterator::Init(nsIDOMRange* aRange)
{
MOZ_ASSERT(aRange);
mIsDone = false;
mRange = static_cast<nsRange*>(aRange);
// get the start node and offset, convert to nsINode
mCommonParent = mRange->GetCommonAncestor();
nsINode* startParent = mRange->GetStartParent();
int32_t startOffset = mRange->StartOffset();
nsINode* endParent = mRange->GetEndParent();
int32_t endOffset = mRange->EndOffset();
MOZ_ASSERT(mCommonParent && startParent && endParent);
// Bug 767169
MOZ_ASSERT(uint32_t(startOffset) <= startParent->Length() &&
uint32_t(endOffset) <= endParent->Length());
// short circuit when start node == end node
if (startParent == endParent) {
nsINode* child = startParent->GetFirstChild();
if (!child || startOffset == endOffset) {
// Text node, empty container, or collapsed
MakeEmpty();
return NS_OK;
}
}
// cache ancestors
nsContentUtils::GetAncestorsAndOffsets(endParent->AsDOMNode(), endOffset,
&mEndNodes, &mEndOffsets);
nsIContent* firstCandidate = nullptr;
nsIContent* lastCandidate = nullptr;
// find first node in range
int32_t offset = mRange->StartOffset();
nsINode* node;
if (!startParent->GetChildCount()) {
// no children, start at the node itself
node = startParent;
} else {
nsIContent* child = startParent->GetChildAt(offset);
if (!child) {
// offset after last child
node = startParent;
} else {
firstCandidate = child;
}
}
if (!firstCandidate) {
// then firstCandidate is next node after node
firstCandidate = GetNextSibling(node);
if (!firstCandidate) {
MakeEmpty();
return NS_OK;
}
}
firstCandidate = GetDeepFirstChild(firstCandidate);
// confirm that this first possible contained node is indeed contained. Else
// we have a range that does not fully contain any node.
bool nodeBefore, nodeAfter;
MOZ_ALWAYS_TRUE(NS_SUCCEEDED(
nsRange::CompareNodeToRange(firstCandidate, mRange, &nodeBefore, &nodeAfter)));
if (nodeBefore || nodeAfter) {
MakeEmpty();
return NS_OK;
}
// cool, we have the first node in the range. Now we walk up its ancestors
// to find the most senior that is still in the range. That's the real first
// node.
mFirst = GetTopAncestorInRange(firstCandidate);
// now to find the last node
offset = mRange->EndOffset();
int32_t numChildren = endParent->GetChildCount();
if (offset > numChildren) {
// Can happen for text nodes
offset = numChildren;
}
if (!offset || !numChildren) {
node = endParent;
} else {
lastCandidate = endParent->GetChildAt(--offset);
NS_ASSERTION(lastCandidate,
"tree traversal trouble in nsContentSubtreeIterator::Init");
}
if (!lastCandidate) {
// then lastCandidate is prev node before node
lastCandidate = GetPrevSibling(node);
}
if (!lastCandidate) {
MakeEmpty();
return NS_OK;
}
lastCandidate = GetDeepLastChild(lastCandidate);
// confirm that this last possible contained node is indeed contained. Else
// we have a range that does not fully contain any node.
MOZ_ALWAYS_TRUE(NS_SUCCEEDED(
nsRange::CompareNodeToRange(lastCandidate, mRange, &nodeBefore, &nodeAfter)));
if (nodeBefore || nodeAfter) {
MakeEmpty();
return NS_OK;
}
// cool, we have the last node in the range. Now we walk up its ancestors to
// find the most senior that is still in the range. That's the real first
// node.
mLast = GetTopAncestorInRange(lastCandidate);
mCurNode = mFirst;
return NS_OK;
}
/****************************************************************
* nsContentSubtreeIterator overrides of ContentIterator routines
****************************************************************/
// we can't call PositionAt in a subtree iterator...
void
nsContentSubtreeIterator::First()
{
mIsDone = mFirst == nullptr;
mCurNode = mFirst;
}
// we can't call PositionAt in a subtree iterator...
void
nsContentSubtreeIterator::Last()
{
mIsDone = mLast == nullptr;
mCurNode = mLast;
}
void
nsContentSubtreeIterator::Next()
{
if (mIsDone || !mCurNode) {
return;
}
if (mCurNode == mLast) {
mIsDone = true;
return;
}
nsINode* nextNode = GetNextSibling(mCurNode);
NS_ASSERTION(nextNode, "No next sibling!?! This could mean deadlock!");
int32_t i = mEndNodes.IndexOf(nextNode);
while (i != -1) {
// as long as we are finding ancestors of the endpoint of the range,
// dive down into their children
nextNode = nextNode->GetFirstChild();
NS_ASSERTION(nextNode, "Iterator error, expected a child node!");
// should be impossible to get a null pointer. If we went all the way
// down the child chain to the bottom without finding an interior node,
// then the previous node should have been the last, which was
// was tested at top of routine.
i = mEndNodes.IndexOf(nextNode);
}
mCurNode = nextNode;
// This shouldn't be needed, but since our selection code can put us
// in a situation where mLast is in generated content, we need this
// to stop the iterator when we've walked past past the last node!
mIsDone = mCurNode == nullptr;
}
void
nsContentSubtreeIterator::Prev()
{
// Prev should be optimized to use the mStartNodes, just as Next
// uses mEndNodes.
if (mIsDone || !mCurNode) {
return;
}
if (mCurNode == mFirst) {
mIsDone = true;
return;
}
// If any of these function calls return null, so will all succeeding ones,
// so mCurNode will wind up set to null.
nsINode* prevNode = GetDeepFirstChild(mCurNode);
prevNode = PrevNode(prevNode);
prevNode = GetDeepLastChild(prevNode);
mCurNode = GetTopAncestorInRange(prevNode);
// This shouldn't be needed, but since our selection code can put us
// in a situation where mFirst is in generated content, we need this
// to stop the iterator when we've walked past past the first node!
mIsDone = mCurNode == nullptr;
}
nsresult
nsContentSubtreeIterator::PositionAt(nsINode* aCurNode)
{
NS_ERROR("Not implemented!");
return NS_ERROR_NOT_IMPLEMENTED;
}
/****************************************************************
* nsContentSubtreeIterator helper routines
****************************************************************/
nsIContent*
nsContentSubtreeIterator::GetTopAncestorInRange(nsINode* aNode)
{
if (!aNode || !aNode->GetParentNode()) {
return nullptr;
}
// aNode has a parent, so it must be content.
nsIContent* content = aNode->AsContent();
// sanity check: aNode is itself in the range
bool nodeBefore, nodeAfter;
nsresult res = nsRange::CompareNodeToRange(aNode, mRange,
&nodeBefore, &nodeAfter);
NS_ASSERTION(NS_SUCCEEDED(res) && !nodeBefore && !nodeAfter,
"aNode isn't in mRange, or something else weird happened");
if (NS_FAILED(res) || nodeBefore || nodeAfter) {
return nullptr;
}
while (content) {
nsIContent* parent = content->GetParent();
// content always has a parent. If its parent is the root, however --
// i.e., either it's not content, or it is content but its own parent is
// null -- then we're finished, since we don't go up to the root.
//
// We have to special-case this because CompareNodeToRange treats the root
// node differently -- see bug 765205.
if (!parent || !parent->GetParentNode()) {
return content;
}
MOZ_ALWAYS_TRUE(NS_SUCCEEDED(
nsRange::CompareNodeToRange(parent, mRange, &nodeBefore, &nodeAfter)));
if (nodeBefore || nodeAfter) {
return content;
}
content = parent;
}
MOZ_CRASH("This should only be possible if aNode was null");
}