gecko-dev/editor/txtsvc/nsTextServicesDocument.cpp

3500 строки
91 KiB
C++

/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/* 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 <stddef.h> // for nullptr
#include "mozilla/Assertions.h" // for MOZ_ASSERT, etc
#include "mozilla/dom/Selection.h"
#include "mozilla/mozalloc.h" // for operator new, etc
#include "nsAString.h" // for nsAString::Length, etc
#include "nsContentUtils.h" // for nsContentUtils
#include "nsDebug.h" // for NS_ENSURE_TRUE, etc
#include "nsDependentSubstring.h" // for Substring
#include "nsError.h" // for NS_OK, NS_ERROR_FAILURE, etc
#include "nsFilteredContentIterator.h" // for nsFilteredContentIterator
#include "nsIContent.h" // for nsIContent, etc
#include "nsIContentIterator.h" // for nsIContentIterator
#include "nsID.h" // for NS_GET_IID
#include "nsIDOMDocument.h" // for nsIDOMDocument
#include "nsIDOMElement.h" // for nsIDOMElement
#include "nsIDOMHTMLDocument.h" // for nsIDOMHTMLDocument
#include "nsIDOMHTMLElement.h" // for nsIDOMHTMLElement
#include "nsIDOMNode.h" // for nsIDOMNode, etc
#include "nsIDOMRange.h" // for nsIDOMRange, etc
#include "nsIEditor.h" // for nsIEditor, etc
#include "nsINode.h" // for nsINode
#include "nsIPlaintextEditor.h" // for nsIPlaintextEditor
#include "nsISelection.h" // for nsISelection
#include "nsISelectionController.h" // for nsISelectionController, etc
#include "nsISupportsBase.h" // for nsISupports
#include "nsISupportsUtils.h" // for NS_IF_ADDREF, NS_ADDREF, etc
#include "nsITextServicesFilter.h" // for nsITextServicesFilter
#include "nsIWordBreaker.h" // for nsWordRange, nsIWordBreaker
#include "nsRange.h" // for nsRange
#include "nsStaticAtom.h" // for NS_STATIC_ATOM, etc
#include "nsString.h" // for nsString, nsAutoString
#include "nsTextServicesDocument.h"
#include "nscore.h" // for nsresult, NS_IMETHODIMP, etc
#define LOCK_DOC(doc)
#define UNLOCK_DOC(doc)
using namespace mozilla;
using namespace mozilla::dom;
class OffsetEntry
{
public:
OffsetEntry(nsIDOMNode *aNode, int32_t aOffset, int32_t aLength)
: mNode(aNode), mNodeOffset(0), mStrOffset(aOffset), mLength(aLength),
mIsInsertedText(false), mIsValid(true)
{
if (mStrOffset < 1) {
mStrOffset = 0;
}
if (mLength < 1) {
mLength = 0;
}
}
virtual ~OffsetEntry()
{
mNode = 0;
mNodeOffset = 0;
mStrOffset = 0;
mLength = 0;
mIsValid = false;
}
nsIDOMNode *mNode;
int32_t mNodeOffset;
int32_t mStrOffset;
int32_t mLength;
bool mIsInsertedText;
bool mIsValid;
};
#define TS_ATOM(name_, value_) nsIAtom* nsTextServicesDocument::name_ = 0;
#include "nsTSAtomList.h" // IWYU pragma: keep
#undef TS_ATOM
nsTextServicesDocument::nsTextServicesDocument()
{
mSelStartIndex = -1;
mSelStartOffset = -1;
mSelEndIndex = -1;
mSelEndOffset = -1;
mIteratorStatus = eIsDone;
}
nsTextServicesDocument::~nsTextServicesDocument()
{
ClearOffsetTable(&mOffsetTable);
}
#define TS_ATOM(name_, value_) NS_STATIC_ATOM_BUFFER(name_##_buffer, value_)
#include "nsTSAtomList.h" // IWYU pragma: keep
#undef TS_ATOM
/* static */
void
nsTextServicesDocument::RegisterAtoms()
{
static const nsStaticAtom ts_atoms[] = {
#define TS_ATOM(name_, value_) NS_STATIC_ATOM(name_##_buffer, &name_),
#include "nsTSAtomList.h" // IWYU pragma: keep
#undef TS_ATOM
};
NS_RegisterStaticAtoms(ts_atoms);
}
NS_IMPL_CYCLE_COLLECTING_ADDREF(nsTextServicesDocument)
NS_IMPL_CYCLE_COLLECTING_RELEASE(nsTextServicesDocument)
NS_INTERFACE_MAP_BEGIN(nsTextServicesDocument)
NS_INTERFACE_MAP_ENTRY(nsITextServicesDocument)
NS_INTERFACE_MAP_ENTRY(nsIEditActionListener)
NS_INTERFACE_MAP_ENTRY_AMBIGUOUS(nsISupports, nsITextServicesDocument)
NS_INTERFACE_MAP_ENTRIES_CYCLE_COLLECTION(nsTextServicesDocument)
NS_INTERFACE_MAP_END
NS_IMPL_CYCLE_COLLECTION(nsTextServicesDocument,
mDOMDocument,
mSelCon,
mIterator,
mPrevTextBlock,
mNextTextBlock,
mExtent,
mTxtSvcFilter)
NS_IMETHODIMP
nsTextServicesDocument::InitWithEditor(nsIEditor *aEditor)
{
nsCOMPtr<nsISelectionController> selCon;
nsCOMPtr<nsIDOMDocument> doc;
NS_ENSURE_TRUE(aEditor, NS_ERROR_NULL_POINTER);
LOCK_DOC(this);
// Check to see if we already have an mSelCon. If we do, it
// better be the same one the editor uses!
nsresult rv = aEditor->GetSelectionController(getter_AddRefs(selCon));
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
if (!selCon || (mSelCon && selCon != mSelCon)) {
UNLOCK_DOC(this);
return NS_ERROR_FAILURE;
}
if (!mSelCon) {
mSelCon = selCon;
}
// Check to see if we already have an mDOMDocument. If we do, it
// better be the same one the editor uses!
rv = aEditor->GetDocument(getter_AddRefs(doc));
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
if (!doc || (mDOMDocument && doc != mDOMDocument)) {
UNLOCK_DOC(this);
return NS_ERROR_FAILURE;
}
if (!mDOMDocument) {
mDOMDocument = doc;
rv = CreateDocumentContentIterator(getter_AddRefs(mIterator));
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
mIteratorStatus = nsTextServicesDocument::eIsDone;
rv = FirstBlock();
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
}
mEditor = do_GetWeakReference(aEditor);
rv = aEditor->AddEditActionListener(this);
UNLOCK_DOC(this);
return rv;
}
NS_IMETHODIMP
nsTextServicesDocument::GetDocument(nsIDOMDocument **aDoc)
{
NS_ENSURE_TRUE(aDoc, NS_ERROR_NULL_POINTER);
*aDoc = nullptr; // init out param
NS_ENSURE_TRUE(mDOMDocument, NS_ERROR_NOT_INITIALIZED);
*aDoc = mDOMDocument;
NS_ADDREF(*aDoc);
return NS_OK;
}
NS_IMETHODIMP
nsTextServicesDocument::SetExtent(nsIDOMRange* aDOMRange)
{
NS_ENSURE_ARG_POINTER(aDOMRange);
NS_ENSURE_TRUE(mDOMDocument, NS_ERROR_FAILURE);
LOCK_DOC(this);
// We need to store a copy of aDOMRange since we don't
// know where it came from.
mExtent = static_cast<nsRange*>(aDOMRange)->CloneRange();
// Create a new iterator based on our new extent range.
nsresult rv = CreateContentIterator(mExtent, getter_AddRefs(mIterator));
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
// Now position the iterator at the start of the first block
// in the range.
mIteratorStatus = nsTextServicesDocument::eIsDone;
rv = FirstBlock();
UNLOCK_DOC(this);
return rv;
}
NS_IMETHODIMP
nsTextServicesDocument::ExpandRangeToWordBoundaries(nsIDOMRange *aRange)
{
NS_ENSURE_ARG_POINTER(aRange);
RefPtr<nsRange> range = static_cast<nsRange*>(aRange);
// Get the end points of the range.
nsCOMPtr<nsIDOMNode> rngStartNode, rngEndNode;
int32_t rngStartOffset, rngEndOffset;
nsresult rv = GetRangeEndPoints(range, getter_AddRefs(rngStartNode),
&rngStartOffset,
getter_AddRefs(rngEndNode),
&rngEndOffset);
NS_ENSURE_SUCCESS(rv, rv);
// Create a content iterator based on the range.
nsCOMPtr<nsIContentIterator> iter;
rv = CreateContentIterator(range, getter_AddRefs(iter));
NS_ENSURE_SUCCESS(rv, rv);
// Find the first text node in the range.
TSDIteratorStatus iterStatus;
rv = FirstTextNode(iter, &iterStatus);
NS_ENSURE_SUCCESS(rv, rv);
if (iterStatus == nsTextServicesDocument::eIsDone) {
// No text was found so there's no adjustment necessary!
return NS_OK;
}
nsINode *firstText = iter->GetCurrentNode();
NS_ENSURE_TRUE(firstText, NS_ERROR_FAILURE);
// Find the last text node in the range.
rv = LastTextNode(iter, &iterStatus);
NS_ENSURE_SUCCESS(rv, rv);
if (iterStatus == nsTextServicesDocument::eIsDone) {
// We should never get here because a first text block
// was found above.
NS_ASSERTION(false, "Found a first without a last!");
return NS_ERROR_FAILURE;
}
nsINode *lastText = iter->GetCurrentNode();
NS_ENSURE_TRUE(lastText, NS_ERROR_FAILURE);
// Now make sure our end points are in terms of text nodes in the range!
nsCOMPtr<nsIDOMNode> firstTextNode = do_QueryInterface(firstText);
NS_ENSURE_TRUE(firstTextNode, NS_ERROR_FAILURE);
if (rngStartNode != firstTextNode) {
// The range includes the start of the first text node!
rngStartNode = firstTextNode;
rngStartOffset = 0;
}
nsCOMPtr<nsIDOMNode> lastTextNode = do_QueryInterface(lastText);
NS_ENSURE_TRUE(lastTextNode, NS_ERROR_FAILURE);
if (rngEndNode != lastTextNode) {
// The range includes the end of the last text node!
rngEndNode = lastTextNode;
nsAutoString str;
lastTextNode->GetNodeValue(str);
rngEndOffset = str.Length();
}
// Create a doc iterator so that we can scan beyond
// the bounds of the extent range.
nsCOMPtr<nsIContentIterator> docIter;
rv = CreateDocumentContentIterator(getter_AddRefs(docIter));
NS_ENSURE_SUCCESS(rv, rv);
// Grab all the text in the block containing our
// first text node.
rv = docIter->PositionAt(firstText);
NS_ENSURE_SUCCESS(rv, rv);
iterStatus = nsTextServicesDocument::eValid;
nsTArray<OffsetEntry*> offsetTable;
nsAutoString blockStr;
rv = CreateOffsetTable(&offsetTable, docIter, &iterStatus,
nullptr, &blockStr);
if (NS_FAILED(rv)) {
ClearOffsetTable(&offsetTable);
return rv;
}
nsCOMPtr<nsIDOMNode> wordStartNode, wordEndNode;
int32_t wordStartOffset, wordEndOffset;
rv = FindWordBounds(&offsetTable, &blockStr,
rngStartNode, rngStartOffset,
getter_AddRefs(wordStartNode), &wordStartOffset,
getter_AddRefs(wordEndNode), &wordEndOffset);
ClearOffsetTable(&offsetTable);
NS_ENSURE_SUCCESS(rv, rv);
rngStartNode = wordStartNode;
rngStartOffset = wordStartOffset;
// Grab all the text in the block containing our
// last text node.
rv = docIter->PositionAt(lastText);
NS_ENSURE_SUCCESS(rv, rv);
iterStatus = nsTextServicesDocument::eValid;
rv = CreateOffsetTable(&offsetTable, docIter, &iterStatus,
nullptr, &blockStr);
if (NS_FAILED(rv)) {
ClearOffsetTable(&offsetTable);
return rv;
}
rv = FindWordBounds(&offsetTable, &blockStr,
rngEndNode, rngEndOffset,
getter_AddRefs(wordStartNode), &wordStartOffset,
getter_AddRefs(wordEndNode), &wordEndOffset);
ClearOffsetTable(&offsetTable);
NS_ENSURE_SUCCESS(rv, rv);
// To prevent expanding the range too much, we only change
// rngEndNode and rngEndOffset if it isn't already at the start of the
// word and isn't equivalent to rngStartNode and rngStartOffset.
if (rngEndNode != wordStartNode ||
rngEndOffset != wordStartOffset ||
(rngEndNode == rngStartNode && rngEndOffset == rngStartOffset)) {
rngEndNode = wordEndNode;
rngEndOffset = wordEndOffset;
}
// Now adjust the range so that it uses our new
// end points.
rv = range->SetEnd(rngEndNode, rngEndOffset);
NS_ENSURE_SUCCESS(rv, rv);
return range->SetStart(rngStartNode, rngStartOffset);
}
NS_IMETHODIMP
nsTextServicesDocument::SetFilter(nsITextServicesFilter *aFilter)
{
// Hang on to the filter so we can set it into the filtered iterator.
mTxtSvcFilter = aFilter;
return NS_OK;
}
NS_IMETHODIMP
nsTextServicesDocument::GetCurrentTextBlock(nsString *aStr)
{
NS_ENSURE_TRUE(aStr, NS_ERROR_NULL_POINTER);
aStr->Truncate();
NS_ENSURE_TRUE(mIterator, NS_ERROR_FAILURE);
LOCK_DOC(this);
nsresult rv = CreateOffsetTable(&mOffsetTable, mIterator, &mIteratorStatus,
mExtent, aStr);
UNLOCK_DOC(this);
return rv;
}
NS_IMETHODIMP
nsTextServicesDocument::FirstBlock()
{
NS_ENSURE_TRUE(mIterator, NS_ERROR_FAILURE);
LOCK_DOC(this);
nsresult rv = FirstTextNode(mIterator, &mIteratorStatus);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
// Keep track of prev and next blocks, just in case
// the text service blows away the current block.
if (mIteratorStatus == nsTextServicesDocument::eValid) {
mPrevTextBlock = nullptr;
rv = GetFirstTextNodeInNextBlock(getter_AddRefs(mNextTextBlock));
} else {
// There's no text block in the document!
mPrevTextBlock = nullptr;
mNextTextBlock = nullptr;
}
UNLOCK_DOC(this);
// XXX Result of FirstTextNode() or GetFirstTextNodeInNextBlock().
return rv;
}
NS_IMETHODIMP
nsTextServicesDocument::LastSelectedBlock(TSDBlockSelectionStatus *aSelStatus,
int32_t *aSelOffset,
int32_t *aSelLength)
{
NS_ENSURE_TRUE(aSelStatus && aSelOffset && aSelLength, NS_ERROR_NULL_POINTER);
LOCK_DOC(this);
mIteratorStatus = nsTextServicesDocument::eIsDone;
*aSelStatus = nsITextServicesDocument::eBlockNotFound;
*aSelOffset = *aSelLength = -1;
if (!mSelCon || !mIterator) {
UNLOCK_DOC(this);
return NS_ERROR_FAILURE;
}
nsCOMPtr<nsISelection> domSelection;
nsresult rv = mSelCon->GetSelection(nsISelectionController::SELECTION_NORMAL,
getter_AddRefs(domSelection));
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
RefPtr<Selection> selection = domSelection->AsSelection();
bool isCollapsed = selection->IsCollapsed();
nsCOMPtr<nsIContentIterator> iter;
RefPtr<nsRange> range;
nsCOMPtr<nsIDOMNode> parent;
int32_t rangeCount, offset;
if (isCollapsed) {
// We have a caret. Check if the caret is in a text node.
// If it is, make the text node's block the current block.
// If the caret isn't in a text node, search forwards in
// the document, till we find a text node.
range = selection->GetRangeAt(0);
if (!range) {
UNLOCK_DOC(this);
return NS_ERROR_FAILURE;
}
rv = range->GetStartContainer(getter_AddRefs(parent));
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
if (!parent) {
UNLOCK_DOC(this);
return NS_ERROR_FAILURE;
}
rv = range->GetStartOffset(&offset);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
if (IsTextNode(parent)) {
// The caret is in a text node. Find the beginning
// of the text block containing this text node and
// return.
nsCOMPtr<nsIContent> content(do_QueryInterface(parent));
if (!content) {
UNLOCK_DOC(this);
return NS_ERROR_FAILURE;
}
rv = mIterator->PositionAt(content);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
rv = FirstTextNodeInCurrentBlock(mIterator);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
mIteratorStatus = nsTextServicesDocument::eValid;
rv = CreateOffsetTable(&mOffsetTable, mIterator, &mIteratorStatus,
mExtent, nullptr);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
rv = GetSelection(aSelStatus, aSelOffset, aSelLength);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
if (*aSelStatus == nsITextServicesDocument::eBlockContains) {
rv = SetSelectionInternal(*aSelOffset, *aSelLength, false);
}
} else {
// The caret isn't in a text node. Create an iterator
// based on a range that extends from the current caret
// position to the end of the document, then walk forwards
// till you find a text node, then find the beginning of it's block.
rv = CreateDocumentContentRootToNodeOffsetRange(parent, offset, false,
getter_AddRefs(range));
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
rv = range->GetCollapsed(&isCollapsed);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
if (isCollapsed) {
// If we get here, the range is collapsed because there is nothing after
// the caret! Just return NS_OK;
UNLOCK_DOC(this);
return NS_OK;
}
rv = CreateContentIterator(range, getter_AddRefs(iter));
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
iter->First();
nsCOMPtr<nsIContent> content;
while (!iter->IsDone()) {
content = do_QueryInterface(iter->GetCurrentNode());
if (IsTextNode(content)) {
break;
}
content = nullptr;
iter->Next();
}
if (!content) {
UNLOCK_DOC(this);
return NS_OK;
}
rv = mIterator->PositionAt(content);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
rv = FirstTextNodeInCurrentBlock(mIterator);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
mIteratorStatus = nsTextServicesDocument::eValid;
rv = CreateOffsetTable(&mOffsetTable, mIterator, &mIteratorStatus,
mExtent, nullptr);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
rv = GetSelection(aSelStatus, aSelOffset, aSelLength);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
}
UNLOCK_DOC(this);
// Result of SetSelectionInternal() in the |if| block or NS_OK.
return rv;
}
// If we get here, we have an uncollapsed selection!
// Look backwards through each range in the selection till you
// find the first text node. If you find one, find the
// beginning of its text block, and make it the current
// block.
rv = selection->GetRangeCount(&rangeCount);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
NS_ASSERTION(rangeCount > 0, "Unexpected range count!");
if (rangeCount <= 0) {
UNLOCK_DOC(this);
return NS_OK;
}
// XXX: We may need to add some code here to make sure
// the ranges are sorted in document appearance order!
for (int32_t i = rangeCount - 1; i >= 0; i--) {
// Get the i'th range from the selection.
range = selection->GetRangeAt(i);
if (!range) {
UNLOCK_DOC(this);
return NS_OK; // XXX Really?
}
// Create an iterator for the range.
rv = CreateContentIterator(range, getter_AddRefs(iter));
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
iter->Last();
// Now walk through the range till we find a text node.
while (!iter->IsDone()) {
if (iter->GetCurrentNode()->NodeType() == nsIDOMNode::TEXT_NODE) {
// We found a text node, so position the document's
// iterator at the beginning of the block, then get
// the selection in terms of the string offset.
nsCOMPtr<nsIContent> content = iter->GetCurrentNode()->AsContent();
rv = mIterator->PositionAt(content);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
rv = FirstTextNodeInCurrentBlock(mIterator);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
mIteratorStatus = nsTextServicesDocument::eValid;
rv = CreateOffsetTable(&mOffsetTable, mIterator, &mIteratorStatus,
mExtent, nullptr);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
rv = GetSelection(aSelStatus, aSelOffset, aSelLength);
UNLOCK_DOC(this);
return rv;
}
iter->Prev();
}
}
// If we get here, we didn't find any text node in the selection!
// Create a range that extends from the end of the selection,
// to the end of the document, then iterate forwards through
// it till you find a text node!
range = selection->GetRangeAt(rangeCount - 1);
if (!range) {
UNLOCK_DOC(this);
return NS_ERROR_FAILURE;
}
rv = range->GetEndContainer(getter_AddRefs(parent));
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
if (!parent) {
UNLOCK_DOC(this);
return NS_ERROR_FAILURE;
}
rv = range->GetEndOffset(&offset);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
rv = CreateDocumentContentRootToNodeOffsetRange(parent, offset, false,
getter_AddRefs(range));
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
rv = range->GetCollapsed(&isCollapsed);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
if (isCollapsed) {
// If we get here, the range is collapsed because there is nothing after
// the current selection! Just return NS_OK;
UNLOCK_DOC(this);
return NS_OK;
}
rv = CreateContentIterator(range, getter_AddRefs(iter));
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
iter->First();
while (!iter->IsDone()) {
if (iter->GetCurrentNode()->NodeType() == nsIDOMNode::TEXT_NODE) {
// We found a text node! Adjust the document's iterator to point
// to the beginning of its text block, then get the current selection.
nsCOMPtr<nsIContent> content = iter->GetCurrentNode()->AsContent();
rv = mIterator->PositionAt(content);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
rv = FirstTextNodeInCurrentBlock(mIterator);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
mIteratorStatus = nsTextServicesDocument::eValid;
rv = CreateOffsetTable(&mOffsetTable, mIterator, &mIteratorStatus,
mExtent, nullptr);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
rv = GetSelection(aSelStatus, aSelOffset, aSelLength);
UNLOCK_DOC(this);
return rv;
}
iter->Next();
}
// If we get here, we didn't find any block before or inside
// the selection! Just return OK.
UNLOCK_DOC(this);
return NS_OK;
}
NS_IMETHODIMP
nsTextServicesDocument::PrevBlock()
{
NS_ENSURE_TRUE(mIterator, NS_ERROR_FAILURE);
LOCK_DOC(this);
if (mIteratorStatus == nsTextServicesDocument::eIsDone) {
return NS_OK;
}
switch (mIteratorStatus) {
case nsTextServicesDocument::eValid:
case nsTextServicesDocument::eNext: {
nsresult rv = FirstTextNodeInPrevBlock(mIterator);
if (NS_FAILED(rv)) {
mIteratorStatus = nsTextServicesDocument::eIsDone;
UNLOCK_DOC(this);
return rv;
}
if (mIterator->IsDone()) {
mIteratorStatus = nsTextServicesDocument::eIsDone;
UNLOCK_DOC(this);
return NS_OK;
}
mIteratorStatus = nsTextServicesDocument::eValid;
break;
}
case nsTextServicesDocument::ePrev:
// The iterator already points to the previous
// block, so don't do anything.
mIteratorStatus = nsTextServicesDocument::eValid;
break;
default:
mIteratorStatus = nsTextServicesDocument::eIsDone;
break;
}
// Keep track of prev and next blocks, just in case
// the text service blows away the current block.
nsresult rv = NS_OK;
if (mIteratorStatus == nsTextServicesDocument::eValid) {
GetFirstTextNodeInPrevBlock(getter_AddRefs(mPrevTextBlock));
rv = GetFirstTextNodeInNextBlock(getter_AddRefs(mNextTextBlock));
} else {
// We must be done!
mPrevTextBlock = nullptr;
mNextTextBlock = nullptr;
}
UNLOCK_DOC(this);
// XXX The result of GetFirstTextNodeInNextBlock() or NS_OK.
return rv;
}
NS_IMETHODIMP
nsTextServicesDocument::NextBlock()
{
NS_ENSURE_TRUE(mIterator, NS_ERROR_FAILURE);
LOCK_DOC(this);
if (mIteratorStatus == nsTextServicesDocument::eIsDone) {
return NS_OK;
}
switch (mIteratorStatus) {
case nsTextServicesDocument::eValid: {
// Advance the iterator to the next text block.
nsresult rv = FirstTextNodeInNextBlock(mIterator);
if (NS_FAILED(rv)) {
mIteratorStatus = nsTextServicesDocument::eIsDone;
UNLOCK_DOC(this);
return rv;
}
if (mIterator->IsDone()) {
mIteratorStatus = nsTextServicesDocument::eIsDone;
UNLOCK_DOC(this);
return NS_OK;
}
mIteratorStatus = nsTextServicesDocument::eValid;
break;
}
case nsTextServicesDocument::eNext:
// The iterator already points to the next block,
// so don't do anything to it!
mIteratorStatus = nsTextServicesDocument::eValid;
break;
case nsTextServicesDocument::ePrev:
// If the iterator is pointing to the previous block,
// we know that there is no next text block! Just
// fall through to the default case!
default:
mIteratorStatus = nsTextServicesDocument::eIsDone;
break;
}
// Keep track of prev and next blocks, just in case
// the text service blows away the current block.
nsresult rv = NS_OK;
if (mIteratorStatus == nsTextServicesDocument::eValid) {
GetFirstTextNodeInPrevBlock(getter_AddRefs(mPrevTextBlock));
rv = GetFirstTextNodeInNextBlock(getter_AddRefs(mNextTextBlock));
} else {
// We must be done.
mPrevTextBlock = nullptr;
mNextTextBlock = nullptr;
}
UNLOCK_DOC(this);
// The result of GetFirstTextNodeInNextBlock() or NS_OK.
return rv;
}
NS_IMETHODIMP
nsTextServicesDocument::IsDone(bool *aIsDone)
{
NS_ENSURE_TRUE(aIsDone, NS_ERROR_NULL_POINTER);
*aIsDone = false;
NS_ENSURE_TRUE(mIterator, NS_ERROR_FAILURE);
LOCK_DOC(this);
*aIsDone = (mIteratorStatus == nsTextServicesDocument::eIsDone) ? true : false;
UNLOCK_DOC(this);
return NS_OK;
}
NS_IMETHODIMP
nsTextServicesDocument::SetSelection(int32_t aOffset, int32_t aLength)
{
NS_ENSURE_TRUE(mSelCon && aOffset >= 0 && aLength >= 0, NS_ERROR_FAILURE);
LOCK_DOC(this);
nsresult rv = SetSelectionInternal(aOffset, aLength, true);
UNLOCK_DOC(this);
//**** KDEBUG ****
// printf("\n * Sel: (%2d, %4d) (%2d, %4d)\n", mSelStartIndex, mSelStartOffset, mSelEndIndex, mSelEndOffset);
//**** KDEBUG ****
return rv;
}
NS_IMETHODIMP
nsTextServicesDocument::ScrollSelectionIntoView()
{
NS_ENSURE_TRUE(mSelCon, NS_ERROR_FAILURE);
LOCK_DOC(this);
// After ScrollSelectionIntoView(), the pending notifications might be flushed
// and PresShell/PresContext/Frames may be dead. See bug 418470.
nsresult rv =
mSelCon->ScrollSelectionIntoView(
nsISelectionController::SELECTION_NORMAL,
nsISelectionController::SELECTION_FOCUS_REGION,
nsISelectionController::SCROLL_SYNCHRONOUS);
UNLOCK_DOC(this);
return rv;
}
NS_IMETHODIMP
nsTextServicesDocument::DeleteSelection()
{
// We don't allow deletion during a collapsed selection!
nsCOMPtr<nsIEditor> editor (do_QueryReferent(mEditor));
NS_ASSERTION(editor, "DeleteSelection called without an editor present!");
NS_ASSERTION(SelectionIsValid(), "DeleteSelection called without a valid selection!");
if (!editor || !SelectionIsValid()) {
return NS_ERROR_FAILURE;
}
if (SelectionIsCollapsed()) {
return NS_OK;
}
LOCK_DOC(this);
//**** KDEBUG ****
// printf("\n---- Before Delete\n");
// printf("Sel: (%2d, %4d) (%2d, %4d)\n", mSelStartIndex, mSelStartOffset, mSelEndIndex, mSelEndOffset);
// PrintOffsetTable();
//**** KDEBUG ****
// If we have an mExtent, save off its current set of
// end points so we can compare them against mExtent's
// set after the deletion of the content.
nsCOMPtr<nsIDOMNode> origStartNode, origEndNode;
int32_t origStartOffset = 0, origEndOffset = 0;
if (mExtent) {
nsresult rv =
GetRangeEndPoints(mExtent,
getter_AddRefs(origStartNode), &origStartOffset,
getter_AddRefs(origEndNode), &origEndOffset);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
}
int32_t selLength;
OffsetEntry *entry, *newEntry;
for (int32_t i = mSelStartIndex; i <= mSelEndIndex; i++) {
entry = mOffsetTable[i];
if (i == mSelStartIndex) {
// Calculate the length of the selection. Note that the
// selection length can be zero if the start of the selection
// is at the very end of a text node entry.
if (entry->mIsInsertedText) {
// Inserted text offset entries have no width when
// talking in terms of string offsets! If the beginning
// of the selection is in an inserted text offset entry,
// the caret is always at the end of the entry!
selLength = 0;
} else {
selLength = entry->mLength - (mSelStartOffset - entry->mStrOffset);
}
if (selLength > 0 && mSelStartOffset > entry->mStrOffset) {
// Selection doesn't start at the beginning of the
// text node entry. We need to split this entry into
// two pieces, the piece before the selection, and
// the piece inside the selection.
nsresult rv = SplitOffsetEntry(i, selLength);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
// Adjust selection indexes to account for new entry:
++mSelStartIndex;
++mSelEndIndex;
++i;
entry = mOffsetTable[i];
}
if (selLength > 0 && mSelStartIndex < mSelEndIndex) {
// The entire entry is contained in the selection. Mark the
// entry invalid.
entry->mIsValid = false;
}
}
//**** KDEBUG ****
// printf("\n---- Middle Delete\n");
// printf("Sel: (%2d, %4d) (%2d, %4d)\n", mSelStartIndex, mSelStartOffset, mSelEndIndex, mSelEndOffset);
// PrintOffsetTable();
//**** KDEBUG ****
if (i == mSelEndIndex) {
if (entry->mIsInsertedText) {
// Inserted text offset entries have no width when
// talking in terms of string offsets! If the end
// of the selection is in an inserted text offset entry,
// the selection includes the entire entry!
entry->mIsValid = false;
} else {
// Calculate the length of the selection. Note that the
// selection length can be zero if the end of the selection
// is at the very beginning of a text node entry.
selLength = mSelEndOffset - entry->mStrOffset;
if (selLength > 0 &&
mSelEndOffset < entry->mStrOffset + entry->mLength) {
// mStrOffset is guaranteed to be inside the selection, even
// when mSelStartIndex == mSelEndIndex.
nsresult rv = SplitOffsetEntry(i, entry->mLength - selLength);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
// Update the entry fields:
newEntry = mOffsetTable[i+1];
newEntry->mNodeOffset = entry->mNodeOffset;
}
if (selLength > 0 &&
mSelEndOffset == entry->mStrOffset + entry->mLength) {
// The entire entry is contained in the selection. Mark the
// entry invalid.
entry->mIsValid = false;
}
}
}
if (i != mSelStartIndex && i != mSelEndIndex) {
// The entire entry is contained in the selection. Mark the
// entry invalid.
entry->mIsValid = false;
}
}
// Make sure mIterator always points to something valid!
AdjustContentIterator();
// Now delete the actual content!
nsresult rv =
editor->DeleteSelection(nsIEditor::ePrevious, nsIEditor::eStrip);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
// Now that we've actually deleted the selected content,
// check to see if our mExtent has changed, if so, then
// we have to create a new content iterator!
if (origStartNode && origEndNode) {
nsCOMPtr<nsIDOMNode> curStartNode, curEndNode;
int32_t curStartOffset = 0, curEndOffset = 0;
rv = GetRangeEndPoints(mExtent,
getter_AddRefs(curStartNode), &curStartOffset,
getter_AddRefs(curEndNode), &curEndOffset);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
if (origStartNode != curStartNode || origEndNode != curEndNode) {
// The range has changed, so we need to create a new content
// iterator based on the new range.
nsCOMPtr<nsIContent> curContent;
if (mIteratorStatus != nsTextServicesDocument::eIsDone) {
// The old iterator is still pointing to something valid,
// so get its current node so we can restore it after we
// create the new iterator!
curContent = mIterator->GetCurrentNode()
? mIterator->GetCurrentNode()->AsContent()
: nullptr;
}
// Create the new iterator.
rv = CreateContentIterator(mExtent, getter_AddRefs(mIterator));
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
// Now make the new iterator point to the content node
// the old one was pointing at.
if (curContent) {
rv = mIterator->PositionAt(curContent);
if (NS_FAILED(rv)) {
mIteratorStatus = eIsDone;
} else {
mIteratorStatus = eValid;
}
}
}
}
entry = 0;
// Move the caret to the end of the first valid entry.
// Start with mSelStartIndex since it may still be valid.
for (int32_t i = mSelStartIndex; !entry && i >= 0; i--) {
entry = mOffsetTable[i];
if (!entry->mIsValid) {
entry = 0;
} else {
mSelStartIndex = mSelEndIndex = i;
mSelStartOffset = mSelEndOffset = entry->mStrOffset + entry->mLength;
}
}
// If we still don't have a valid entry, move the caret
// to the next valid entry after the selection:
for (int32_t i = mSelEndIndex;
!entry && i < static_cast<int32_t>(mOffsetTable.Length()); i++) {
entry = mOffsetTable[i];
if (!entry->mIsValid) {
entry = 0;
} else {
mSelStartIndex = mSelEndIndex = i;
mSelStartOffset = mSelEndOffset = entry->mStrOffset;
}
}
if (entry) {
SetSelection(mSelStartOffset, 0);
} else {
// Uuughh we have no valid offset entry to place our
// caret ... just mark the selection invalid.
mSelStartIndex = mSelEndIndex = -1;
mSelStartOffset = mSelEndOffset = -1;
}
// Now remove any invalid entries from the offset table.
rv = RemoveInvalidOffsetEntries();
//**** KDEBUG ****
// printf("\n---- After Delete\n");
// printf("Sel: (%2d, %4d) (%2d, %4d)\n", mSelStartIndex, mSelStartOffset, mSelEndIndex, mSelEndOffset);
// PrintOffsetTable();
//**** KDEBUG ****
UNLOCK_DOC(this);
return rv;
}
NS_IMETHODIMP
nsTextServicesDocument::InsertText(const nsString *aText)
{
nsCOMPtr<nsIEditor> editor (do_QueryReferent(mEditor));
NS_ASSERTION(editor, "InsertText called without an editor present!");
if (!editor || !SelectionIsValid()) {
return NS_ERROR_FAILURE;
}
NS_ENSURE_TRUE(aText, NS_ERROR_NULL_POINTER);
// If the selection is not collapsed, we need to save
// off the selection offsets so we can restore the
// selection and delete the selected content after we've
// inserted the new text. This is necessary to try and
// retain as much of the original style of the content
// being deleted.
bool collapsedSelection = SelectionIsCollapsed();
int32_t savedSelOffset = mSelStartOffset;
int32_t savedSelLength = mSelEndOffset - mSelStartOffset;
if (!collapsedSelection) {
// Collapse to the start of the current selection
// for the insert!
nsresult rv = SetSelection(mSelStartOffset, 0);
NS_ENSURE_SUCCESS(rv, rv);
}
LOCK_DOC(this);
nsresult rv = editor->BeginTransaction();
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
nsCOMPtr<nsIPlaintextEditor> textEditor (do_QueryInterface(editor, &rv));
if (textEditor) {
rv = textEditor->InsertText(*aText);
}
if (NS_FAILED(rv)) {
editor->EndTransaction();
UNLOCK_DOC(this);
return rv;
}
//**** KDEBUG ****
// printf("\n---- Before Insert\n");
// printf("Sel: (%2d, %4d) (%2d, %4d)\n", mSelStartIndex, mSelStartOffset, mSelEndIndex, mSelEndOffset);
// PrintOffsetTable();
//**** KDEBUG ****
int32_t strLength = aText->Length();
nsCOMPtr<nsISelection> selection;
OffsetEntry *itEntry;
OffsetEntry *entry = mOffsetTable[mSelStartIndex];
void *node = entry->mNode;
NS_ASSERTION((entry->mIsValid), "Invalid insertion point!");
if (entry->mStrOffset == mSelStartOffset) {
if (entry->mIsInsertedText) {
// If the caret is in an inserted text offset entry,
// we simply insert the text at the end of the entry.
entry->mLength += strLength;
} else {
// Insert an inserted text offset entry before the current
// entry!
itEntry = new OffsetEntry(entry->mNode, entry->mStrOffset, strLength);
if (!itEntry) {
editor->EndTransaction();
UNLOCK_DOC(this);
return NS_ERROR_OUT_OF_MEMORY;
}
itEntry->mIsInsertedText = true;
itEntry->mNodeOffset = entry->mNodeOffset;
if (!mOffsetTable.InsertElementAt(mSelStartIndex, itEntry)) {
editor->EndTransaction();
UNLOCK_DOC(this);
return NS_ERROR_FAILURE;
}
}
} else if (entry->mStrOffset + entry->mLength == mSelStartOffset) {
// We are inserting text at the end of the current offset entry.
// Look at the next valid entry in the table. If it's an inserted
// text entry, add to its length and adjust its node offset. If
// it isn't, add a new inserted text entry.
// XXX Rename this!
uint32_t i = mSelStartIndex + 1;
itEntry = 0;
if (mOffsetTable.Length() > i) {
itEntry = mOffsetTable[i];
if (!itEntry) {
editor->EndTransaction();
UNLOCK_DOC(this);
return NS_ERROR_FAILURE;
}
// Check if the entry is a match. If it isn't, set
// iEntry to zero.
if (!itEntry->mIsInsertedText || itEntry->mStrOffset != mSelStartOffset) {
itEntry = 0;
}
}
if (!itEntry) {
// We didn't find an inserted text offset entry, so
// create one.
itEntry = new OffsetEntry(entry->mNode, mSelStartOffset, 0);
if (!itEntry) {
editor->EndTransaction();
UNLOCK_DOC(this);
return NS_ERROR_OUT_OF_MEMORY;
}
itEntry->mNodeOffset = entry->mNodeOffset + entry->mLength;
itEntry->mIsInsertedText = true;
if (!mOffsetTable.InsertElementAt(i, itEntry)) {
delete itEntry;
return NS_ERROR_FAILURE;
}
}
// We have a valid inserted text offset entry. Update its
// length, adjust the selection indexes, and make sure the
// caret is properly placed!
itEntry->mLength += strLength;
mSelStartIndex = mSelEndIndex = i;
rv = mSelCon->GetSelection(nsISelectionController::SELECTION_NORMAL,
getter_AddRefs(selection));
if (NS_FAILED(rv)) {
editor->EndTransaction();
UNLOCK_DOC(this);
return rv;
}
rv = selection->Collapse(itEntry->mNode,
itEntry->mNodeOffset + itEntry->mLength);
if (NS_FAILED(rv)) {
editor->EndTransaction();
UNLOCK_DOC(this);
return rv;
}
} else if (entry->mStrOffset + entry->mLength > mSelStartOffset) {
// We are inserting text into the middle of the current offset entry.
// split the current entry into two parts, then insert an inserted text
// entry between them!
// XXX Rename this!
uint32_t i = entry->mLength - (mSelStartOffset - entry->mStrOffset);
rv = SplitOffsetEntry(mSelStartIndex, i);
if (NS_FAILED(rv)) {
editor->EndTransaction();
UNLOCK_DOC(this);
return rv;
}
itEntry = new OffsetEntry(entry->mNode, mSelStartOffset, strLength);
if (!itEntry) {
editor->EndTransaction();
UNLOCK_DOC(this);
return NS_ERROR_OUT_OF_MEMORY;
}
itEntry->mIsInsertedText = true;
itEntry->mNodeOffset = entry->mNodeOffset + entry->mLength;
if (!mOffsetTable.InsertElementAt(mSelStartIndex + 1, itEntry)) {
editor->EndTransaction();
UNLOCK_DOC(this);
return NS_ERROR_FAILURE;
}
mSelEndIndex = ++mSelStartIndex;
}
// We've just finished inserting an inserted text offset entry.
// update all entries with the same mNode pointer that follow
// it in the table!
for (size_t i = mSelStartIndex + 1; i < mOffsetTable.Length(); i++) {
entry = mOffsetTable[i];
if (entry->mNode != node) {
break;
}
if (entry->mIsValid) {
entry->mNodeOffset += strLength;
}
}
//**** KDEBUG ****
// printf("\n---- After Insert\n");
// printf("Sel: (%2d, %4d) (%2d, %4d)\n", mSelStartIndex, mSelStartOffset, mSelEndIndex, mSelEndOffset);
// PrintOffsetTable();
//**** KDEBUG ****
if (!collapsedSelection) {
rv = SetSelection(savedSelOffset, savedSelLength);
if (NS_FAILED(rv)) {
editor->EndTransaction();
UNLOCK_DOC(this);
return rv;
}
rv = DeleteSelection();
if (NS_FAILED(rv)) {
editor->EndTransaction();
UNLOCK_DOC(this);
return rv;
}
}
rv = editor->EndTransaction();
UNLOCK_DOC(this);
return rv;
}
NS_IMETHODIMP
nsTextServicesDocument::DidInsertNode(nsIDOMNode *aNode,
nsIDOMNode *aParent,
int32_t aPosition,
nsresult aResult)
{
return NS_OK;
}
NS_IMETHODIMP
nsTextServicesDocument::DidDeleteNode(nsIDOMNode *aChild, nsresult aResult)
{
NS_ENSURE_SUCCESS(aResult, NS_OK);
NS_ENSURE_TRUE(mIterator, NS_ERROR_FAILURE);
//**** KDEBUG ****
// printf("** DeleteNode: 0x%.8x\n", aChild);
// fflush(stdout);
//**** KDEBUG ****
LOCK_DOC(this);
int32_t nodeIndex = 0;
bool hasEntry = false;
OffsetEntry *entry;
nsresult rv =
NodeHasOffsetEntry(&mOffsetTable, aChild, &hasEntry, &nodeIndex);
if (NS_FAILED(rv)) {
UNLOCK_DOC(this);
return rv;
}
if (!hasEntry) {
// It's okay if the node isn't in the offset table, the
// editor could be cleaning house.
UNLOCK_DOC(this);
return NS_OK;
}
nsCOMPtr<nsIDOMNode> node = do_QueryInterface(mIterator->GetCurrentNode());
if (node && node == aChild &&
mIteratorStatus != nsTextServicesDocument::eIsDone) {
// XXX: This should never really happen because
// AdjustContentIterator() should have been called prior
// to the delete to try and position the iterator on the
// next valid text node in the offset table, and if there
// wasn't a next, it would've set mIteratorStatus to eIsDone.
NS_ERROR("DeleteNode called for current iterator node.");
}
int32_t tcount = mOffsetTable.Length();
while (nodeIndex < tcount) {
entry = mOffsetTable[nodeIndex];
if (!entry) {
UNLOCK_DOC(this);
return NS_ERROR_FAILURE;
}
if (entry->mNode == aChild) {
entry->mIsValid = false;
}
nodeIndex++;
}
UNLOCK_DOC(this);
return NS_OK;
}
NS_IMETHODIMP
nsTextServicesDocument::DidSplitNode(nsIDOMNode *aExistingRightNode,
int32_t aOffset,
nsIDOMNode *aNewLeftNode,
nsresult aResult)
{
//**** KDEBUG ****
// printf("** SplitNode: 0x%.8x %d 0x%.8x\n", aExistingRightNode, aOffset, aNewLeftNode);
// fflush(stdout);
//**** KDEBUG ****
return NS_OK;
}
NS_IMETHODIMP
nsTextServicesDocument::DidJoinNodes(nsIDOMNode *aLeftNode,
nsIDOMNode *aRightNode,
nsIDOMNode *aParent,
nsresult aResult)
{
NS_ENSURE_SUCCESS(aResult, NS_OK);
//**** KDEBUG ****
// printf("** JoinNodes: 0x%.8x 0x%.8x 0x%.8x\n", aLeftNode, aRightNode, aParent);
// fflush(stdout);
//**** KDEBUG ****
// Make sure that both nodes are text nodes -- otherwise we don't care.
uint16_t type;
nsresult rv = aLeftNode->GetNodeType(&type);
NS_ENSURE_SUCCESS(rv, NS_OK);
if (nsIDOMNode::TEXT_NODE != type) {
return NS_OK;
}
rv = aRightNode->GetNodeType(&type);
NS_ENSURE_SUCCESS(rv, NS_OK);
if (nsIDOMNode::TEXT_NODE != type) {
return NS_OK;
}
// Note: The editor merges the contents of the left node into the
// contents of the right.
int32_t leftIndex = 0;
int32_t rightIndex = 0;
bool leftHasEntry = false;
bool rightHasEntry = false;
rv = NodeHasOffsetEntry(&mOffsetTable, aLeftNode, &leftHasEntry, &leftIndex);
NS_ENSURE_SUCCESS(rv, rv);
if (!leftHasEntry) {
// It's okay if the node isn't in the offset table, the
// editor could be cleaning house.
return NS_OK;
}
rv = NodeHasOffsetEntry(&mOffsetTable, aRightNode,
&rightHasEntry, &rightIndex);
NS_ENSURE_SUCCESS(rv, rv);
if (!rightHasEntry) {
// It's okay if the node isn't in the offset table, the
// editor could be cleaning house.
return NS_OK;
}
NS_ASSERTION(leftIndex < rightIndex, "Indexes out of order.");
if (leftIndex > rightIndex) {
// Don't know how to handle this situation.
return NS_ERROR_FAILURE;
}
LOCK_DOC(this);
OffsetEntry *entry = mOffsetTable[rightIndex];
NS_ASSERTION(entry->mNodeOffset == 0, "Unexpected offset value for rightIndex.");
// Run through the table and change all entries referring to
// the left node so that they now refer to the right node:
nsAutoString str;
aLeftNode->GetNodeValue(str);
int32_t nodeLength = str.Length();
for (int32_t i = leftIndex; i < rightIndex; i++) {
entry = mOffsetTable[i];
if (entry->mNode != aLeftNode) {
break;
}
if (entry->mIsValid) {
entry->mNode = aRightNode;
}
}
// Run through the table and adjust the node offsets
// for all entries referring to the right node.
for (int32_t i = rightIndex;
i < static_cast<int32_t>(mOffsetTable.Length()); i++) {
entry = mOffsetTable[i];
if (entry->mNode != aRightNode) {
break;
}
if (entry->mIsValid) {
entry->mNodeOffset += nodeLength;
}
}
// Now check to see if the iterator is pointing to the
// left node. If it is, make it point to the right node!
nsCOMPtr<nsIContent> leftContent = do_QueryInterface(aLeftNode);
nsCOMPtr<nsIContent> rightContent = do_QueryInterface(aRightNode);
if (!leftContent || !rightContent) {
UNLOCK_DOC(this);
return NS_ERROR_FAILURE;
}
if (mIterator->GetCurrentNode() == leftContent) {
mIterator->PositionAt(rightContent);
}
UNLOCK_DOC(this);
return NS_OK;
}
nsresult
nsTextServicesDocument::CreateContentIterator(nsRange* aRange,
nsIContentIterator** aIterator)
{
NS_ENSURE_TRUE(aRange && aIterator, NS_ERROR_NULL_POINTER);
*aIterator = nullptr;
// Create a nsFilteredContentIterator
// This class wraps the ContentIterator in order to give itself a chance
// to filter out certain content nodes
RefPtr<nsFilteredContentIterator> filter = new nsFilteredContentIterator(mTxtSvcFilter);
nsresult rv = filter->Init(aRange);
if (NS_FAILED(rv)) {
return rv;
}
filter.forget(aIterator);
return NS_OK;
}
nsresult
nsTextServicesDocument::GetDocumentContentRootNode(nsIDOMNode **aNode)
{
NS_ENSURE_TRUE(aNode, NS_ERROR_NULL_POINTER);
*aNode = 0;
NS_ENSURE_TRUE(mDOMDocument, NS_ERROR_FAILURE);
nsCOMPtr<nsIDOMHTMLDocument> htmlDoc = do_QueryInterface(mDOMDocument);
if (htmlDoc) {
// For HTML documents, the content root node is the body.
nsCOMPtr<nsIDOMHTMLElement> bodyElement;
nsresult rv = htmlDoc->GetBody(getter_AddRefs(bodyElement));
NS_ENSURE_SUCCESS(rv, rv);
NS_ENSURE_TRUE(bodyElement, NS_ERROR_FAILURE);
bodyElement.forget(aNode);
} else {
// For non-HTML documents, the content root node will be the document element.
nsCOMPtr<nsIDOMElement> docElement;
nsresult rv = mDOMDocument->GetDocumentElement(getter_AddRefs(docElement));
NS_ENSURE_SUCCESS(rv, rv);
NS_ENSURE_TRUE(docElement, NS_ERROR_FAILURE);
docElement.forget(aNode);
}
return NS_OK;
}
nsresult
nsTextServicesDocument::CreateDocumentContentRange(nsRange** aRange)
{
*aRange = nullptr;
nsCOMPtr<nsIDOMNode> node;
nsresult rv = GetDocumentContentRootNode(getter_AddRefs(node));
NS_ENSURE_SUCCESS(rv, rv);
NS_ENSURE_TRUE(node, NS_ERROR_NULL_POINTER);
nsCOMPtr<nsINode> nativeNode = do_QueryInterface(node);
NS_ENSURE_STATE(nativeNode);
RefPtr<nsRange> range = new nsRange(nativeNode);
rv = range->SelectNodeContents(node);
NS_ENSURE_SUCCESS(rv, rv);
range.forget(aRange);
return NS_OK;
}
nsresult
nsTextServicesDocument::CreateDocumentContentRootToNodeOffsetRange(
nsIDOMNode* aParent, int32_t aOffset, bool aToStart, nsRange** aRange)
{
NS_ENSURE_TRUE(aParent && aRange, NS_ERROR_NULL_POINTER);
*aRange = 0;
NS_ASSERTION(aOffset >= 0, "Invalid offset!");
if (aOffset < 0) {
return NS_ERROR_FAILURE;
}
nsCOMPtr<nsIDOMNode> bodyNode;
nsresult rv = GetDocumentContentRootNode(getter_AddRefs(bodyNode));
NS_ENSURE_SUCCESS(rv, rv);
NS_ENSURE_TRUE(bodyNode, NS_ERROR_NULL_POINTER);
nsCOMPtr<nsIDOMNode> startNode;
nsCOMPtr<nsIDOMNode> endNode;
int32_t startOffset, endOffset;
if (aToStart) {
// The range should begin at the start of the document
// and extend up until (aParent, aOffset).
startNode = bodyNode;
startOffset = 0;
endNode = aParent;
endOffset = aOffset;
} else {
// The range should begin at (aParent, aOffset) and
// extend to the end of the document.
startNode = aParent;
startOffset = aOffset;
endNode = bodyNode;
nsCOMPtr<nsINode> body = do_QueryInterface(bodyNode);
endOffset = body ? int32_t(body->GetChildCount()) : 0;
}
return nsRange::CreateRange(startNode, startOffset, endNode, endOffset,
aRange);
}
nsresult
nsTextServicesDocument::CreateDocumentContentIterator(nsIContentIterator **aIterator)
{
NS_ENSURE_TRUE(aIterator, NS_ERROR_NULL_POINTER);
RefPtr<nsRange> range;
nsresult rv = CreateDocumentContentRange(getter_AddRefs(range));
NS_ENSURE_SUCCESS(rv, rv);
return CreateContentIterator(range, aIterator);
}
nsresult
nsTextServicesDocument::AdjustContentIterator()
{
NS_ENSURE_TRUE(mIterator, NS_ERROR_FAILURE);
nsCOMPtr<nsIDOMNode> node(do_QueryInterface(mIterator->GetCurrentNode()));
NS_ENSURE_TRUE(node, NS_ERROR_FAILURE);
nsIDOMNode *nodePtr = node.get();
int32_t tcount = mOffsetTable.Length();
nsIDOMNode *prevValidNode = 0;
nsIDOMNode *nextValidNode = 0;
bool foundEntry = false;
OffsetEntry *entry;
for (int32_t i = 0; i < tcount && !nextValidNode; i++) {
entry = mOffsetTable[i];
NS_ENSURE_TRUE(entry, NS_ERROR_FAILURE);
if (entry->mNode == nodePtr) {
if (entry->mIsValid) {
// The iterator is still pointing to something valid!
// Do nothing!
return NS_OK;
}
// We found an invalid entry that points to
// the current iterator node. Stop looking for
// a previous valid node!
foundEntry = true;
}
if (entry->mIsValid) {
if (!foundEntry) {
prevValidNode = entry->mNode;
} else {
nextValidNode = entry->mNode;
}
}
}
nsCOMPtr<nsIContent> content;
if (prevValidNode) {
content = do_QueryInterface(prevValidNode);
} else if (nextValidNode) {
content = do_QueryInterface(nextValidNode);
}
if (content) {
nsresult rv = mIterator->PositionAt(content);
if (NS_FAILED(rv)) {
mIteratorStatus = eIsDone;
} else {
mIteratorStatus = eValid;
}
return rv;
}
// If we get here, there aren't any valid entries
// in the offset table! Try to position the iterator
// on the next text block first, then previous if
// one doesn't exist!
if (mNextTextBlock) {
nsresult rv = mIterator->PositionAt(mNextTextBlock);
if (NS_FAILED(rv)) {
mIteratorStatus = eIsDone;
return rv;
}
mIteratorStatus = eNext;
} else if (mPrevTextBlock) {
nsresult rv = mIterator->PositionAt(mPrevTextBlock);
if (NS_FAILED(rv)) {
mIteratorStatus = eIsDone;
return rv;
}
mIteratorStatus = ePrev;
} else {
mIteratorStatus = eIsDone;
}
return NS_OK;
}
bool
nsTextServicesDocument::DidSkip(nsIContentIterator* aFilteredIter)
{
// We can assume here that the Iterator is a nsFilteredContentIterator because
// all the iterator are created in CreateContentIterator which create a
// nsFilteredContentIterator
// So if the iterator bailed on one of the "filtered" content nodes then we
// consider that to be a block and bail with true
if (aFilteredIter) {
nsFilteredContentIterator* filter = static_cast<nsFilteredContentIterator *>(aFilteredIter);
if (filter && filter->DidSkip()) {
return true;
}
}
return false;
}
void
nsTextServicesDocument::ClearDidSkip(nsIContentIterator* aFilteredIter)
{
// Clear filter's skip flag
if (aFilteredIter) {
nsFilteredContentIterator* filter = static_cast<nsFilteredContentIterator *>(aFilteredIter);
filter->ClearDidSkip();
}
}
bool
nsTextServicesDocument::IsBlockNode(nsIContent *aContent)
{
if (!aContent) {
NS_ERROR("How did a null pointer get passed to IsBlockNode?");
return false;
}
nsIAtom *atom = aContent->NodeInfo()->NameAtom();
return (sAAtom != atom &&
sAddressAtom != atom &&
sBigAtom != atom &&
sBAtom != atom &&
sCiteAtom != atom &&
sCodeAtom != atom &&
sDfnAtom != atom &&
sEmAtom != atom &&
sFontAtom != atom &&
sIAtom != atom &&
sKbdAtom != atom &&
sKeygenAtom != atom &&
sNobrAtom != atom &&
sSAtom != atom &&
sSampAtom != atom &&
sSmallAtom != atom &&
sSpacerAtom != atom &&
sSpanAtom != atom &&
sStrikeAtom != atom &&
sStrongAtom != atom &&
sSubAtom != atom &&
sSupAtom != atom &&
sTtAtom != atom &&
sUAtom != atom &&
sVarAtom != atom &&
sWbrAtom != atom);
}
bool
nsTextServicesDocument::HasSameBlockNodeParent(nsIContent *aContent1, nsIContent *aContent2)
{
nsIContent* p1 = aContent1->GetParent();
nsIContent* p2 = aContent2->GetParent();
// Quick test:
if (p1 == p2) {
return true;
}
// Walk up the parent hierarchy looking for closest block boundary node:
while (p1 && !IsBlockNode(p1)) {
p1 = p1->GetParent();
}
while (p2 && !IsBlockNode(p2)) {
p2 = p2->GetParent();
}
return p1 == p2;
}
bool
nsTextServicesDocument::IsTextNode(nsIContent *aContent)
{
NS_ENSURE_TRUE(aContent, false);
return nsIDOMNode::TEXT_NODE == aContent->NodeType();
}
bool
nsTextServicesDocument::IsTextNode(nsIDOMNode *aNode)
{
NS_ENSURE_TRUE(aNode, false);
nsCOMPtr<nsIContent> content = do_QueryInterface(aNode);
return IsTextNode(content);
}
nsresult
nsTextServicesDocument::SetSelectionInternal(int32_t aOffset, int32_t aLength, bool aDoUpdate)
{
NS_ENSURE_TRUE(mSelCon && aOffset >= 0 && aLength >= 0, NS_ERROR_FAILURE);
nsIDOMNode *sNode = 0, *eNode = 0;
int32_t sOffset = 0, eOffset = 0;
OffsetEntry *entry;
// Find start of selection in node offset terms:
for (size_t i = 0; !sNode && i < mOffsetTable.Length(); i++) {
entry = mOffsetTable[i];
if (entry->mIsValid) {
if (entry->mIsInsertedText) {
// Caret can only be placed at the end of an
// inserted text offset entry, if the offsets
// match exactly!
if (entry->mStrOffset == aOffset) {
sNode = entry->mNode;
sOffset = entry->mNodeOffset + entry->mLength;
}
} else if (aOffset >= entry->mStrOffset) {
bool foundEntry = false;
int32_t strEndOffset = entry->mStrOffset + entry->mLength;
if (aOffset < strEndOffset) {
foundEntry = true;
} else if (aOffset == strEndOffset) {
// Peek after this entry to see if we have any
// inserted text entries belonging to the same
// entry->mNode. If so, we have to place the selection
// after it!
if (i + 1 < mOffsetTable.Length()) {
OffsetEntry *nextEntry = mOffsetTable[i+1];
if (!nextEntry->mIsValid || nextEntry->mStrOffset != aOffset) {
// Next offset entry isn't an exact match, so we'll
// just use the current entry.
foundEntry = true;
}
}
}
if (foundEntry) {
sNode = entry->mNode;
sOffset = entry->mNodeOffset + aOffset - entry->mStrOffset;
}
}
if (sNode) {
mSelStartIndex = static_cast<int32_t>(i);
mSelStartOffset = aOffset;
}
}
}
NS_ENSURE_TRUE(sNode, NS_ERROR_FAILURE);
// XXX: If we ever get a SetSelection() method in nsIEditor, we should
// use it.
nsCOMPtr<nsISelection> selection;
if (aDoUpdate) {
nsresult rv =
mSelCon->GetSelection(nsISelectionController::SELECTION_NORMAL,
getter_AddRefs(selection));
NS_ENSURE_SUCCESS(rv, rv);
rv = selection->Collapse(sNode, sOffset);
NS_ENSURE_SUCCESS(rv, rv);
}
if (aLength <= 0) {
// We have a collapsed selection. (Caret)
mSelEndIndex = mSelStartIndex;
mSelEndOffset = mSelStartOffset;
//**** KDEBUG ****
// printf("\n* Sel: (%2d, %4d) (%2d, %4d)\n", mSelStartIndex, mSelStartOffset, mSelEndIndex, mSelEndOffset);
//**** KDEBUG ****
return NS_OK;
}
// Find the end of the selection in node offset terms:
int32_t endOffset = aOffset + aLength;
for (int32_t i = mOffsetTable.Length() - 1; !eNode && i >= 0; i--) {
entry = mOffsetTable[i];
if (entry->mIsValid) {
if (entry->mIsInsertedText) {
if (entry->mStrOffset == eOffset) {
// If the selection ends on an inserted text offset entry,
// the selection includes the entire entry!
eNode = entry->mNode;
eOffset = entry->mNodeOffset + entry->mLength;
}
} else if (endOffset >= entry->mStrOffset &&
endOffset <= entry->mStrOffset + entry->mLength) {
eNode = entry->mNode;
eOffset = entry->mNodeOffset + endOffset - entry->mStrOffset;
}
if (eNode) {
mSelEndIndex = i;
mSelEndOffset = endOffset;
}
}
}
if (aDoUpdate && eNode) {
nsresult rv = selection->Extend(eNode, eOffset);
NS_ENSURE_SUCCESS(rv, rv);
}
//**** KDEBUG ****
// printf("\n * Sel: (%2d, %4d) (%2d, %4d)\n", mSelStartIndex, mSelStartOffset, mSelEndIndex, mSelEndOffset);
//**** KDEBUG ****
return NS_OK;
}
nsresult
nsTextServicesDocument::GetSelection(nsITextServicesDocument::TSDBlockSelectionStatus *aSelStatus, int32_t *aSelOffset, int32_t *aSelLength)
{
NS_ENSURE_TRUE(aSelStatus && aSelOffset && aSelLength, NS_ERROR_NULL_POINTER);
*aSelStatus = nsITextServicesDocument::eBlockNotFound;
*aSelOffset = -1;
*aSelLength = -1;
NS_ENSURE_TRUE(mDOMDocument && mSelCon, NS_ERROR_FAILURE);
if (mIteratorStatus == nsTextServicesDocument::eIsDone) {
return NS_OK;
}
nsCOMPtr<nsISelection> selection;
bool isCollapsed;
nsresult rv = mSelCon->GetSelection(nsISelectionController::SELECTION_NORMAL,
getter_AddRefs(selection));
NS_ENSURE_SUCCESS(rv, rv);
NS_ENSURE_TRUE(selection, NS_ERROR_FAILURE);
rv = selection->GetIsCollapsed(&isCollapsed);
NS_ENSURE_SUCCESS(rv, rv);
// XXX: If we expose this method publicly, we need to
// add LOCK_DOC/UNLOCK_DOC calls!
// LOCK_DOC(this);
if (isCollapsed) {
rv = GetCollapsedSelection(aSelStatus, aSelOffset, aSelLength);
} else {
rv = GetUncollapsedSelection(aSelStatus, aSelOffset, aSelLength);
}
// UNLOCK_DOC(this);
// XXX The result of GetCollapsedSelection() or GetUncollapsedSelection().
return rv;
}
nsresult
nsTextServicesDocument::GetCollapsedSelection(nsITextServicesDocument::TSDBlockSelectionStatus *aSelStatus, int32_t *aSelOffset, int32_t *aSelLength)
{
nsCOMPtr<nsISelection> domSelection;
nsresult rv =
mSelCon->GetSelection(nsISelectionController::SELECTION_NORMAL,
getter_AddRefs(domSelection));
NS_ENSURE_SUCCESS(rv, rv);
NS_ENSURE_TRUE(domSelection, NS_ERROR_FAILURE);
RefPtr<Selection> selection = domSelection->AsSelection();
// The calling function should have done the GetIsCollapsed()
// check already. Just assume it's collapsed!
*aSelStatus = nsITextServicesDocument::eBlockOutside;
*aSelOffset = *aSelLength = -1;
int32_t tableCount = mOffsetTable.Length();
if (!tableCount) {
return NS_OK;
}
// Get pointers to the first and last offset entries
// in the table.
OffsetEntry* eStart = mOffsetTable[0];
OffsetEntry* eEnd;
if (tableCount > 1) {
eEnd = mOffsetTable[tableCount - 1];
} else {
eEnd = eStart;
}
int32_t eStartOffset = eStart->mNodeOffset;
int32_t eEndOffset = eEnd->mNodeOffset + eEnd->mLength;
RefPtr<nsRange> range = selection->GetRangeAt(0);
NS_ENSURE_STATE(range);
nsCOMPtr<nsIDOMNode> domParent;
rv = range->GetStartContainer(getter_AddRefs(domParent));
NS_ENSURE_SUCCESS(rv, rv);
nsCOMPtr<nsINode> parent = do_QueryInterface(domParent);
MOZ_ASSERT(parent);
int32_t offset;
rv = range->GetStartOffset(&offset);
NS_ENSURE_SUCCESS(rv, rv);
int32_t e1s1 = nsContentUtils::ComparePoints(eStart->mNode, eStartOffset,
domParent, offset);
int32_t e2s1 = nsContentUtils::ComparePoints(eEnd->mNode, eEndOffset,
domParent, offset);
if (e1s1 > 0 || e2s1 < 0) {
// We're done if the caret is outside the current text block.
return NS_OK;
}
if (parent->NodeType() == nsIDOMNode::TEXT_NODE) {
// Good news, the caret is in a text node. Look
// through the offset table for the entry that
// matches its parent and offset.
for (int32_t i = 0; i < tableCount; i++) {
OffsetEntry* entry = mOffsetTable[i];
NS_ENSURE_TRUE(entry, NS_ERROR_FAILURE);
if (entry->mNode == domParent.get() &&
entry->mNodeOffset <= offset &&
offset <= entry->mNodeOffset + entry->mLength) {
*aSelStatus = nsITextServicesDocument::eBlockContains;
*aSelOffset = entry->mStrOffset + (offset - entry->mNodeOffset);
*aSelLength = 0;
return NS_OK;
}
}
// If we get here, we didn't find a text node entry
// in our offset table that matched.
return NS_ERROR_FAILURE;
}
// The caret is in our text block, but it's positioned in some
// non-text node (ex. <b>). Create a range based on the start
// and end of the text block, then create an iterator based on
// this range, with its initial position set to the closest
// child of this non-text node. Then look for the closest text
// node.
rv = CreateRange(eStart->mNode, eStartOffset, eEnd->mNode, eEndOffset,
getter_AddRefs(range));
NS_ENSURE_SUCCESS(rv, rv);
nsCOMPtr<nsIContentIterator> iter;
rv = CreateContentIterator(range, getter_AddRefs(iter));
NS_ENSURE_SUCCESS(rv, rv);
nsIContent* saveNode;
if (parent->HasChildren()) {
// XXX: We need to make sure that all of parent's
// children are in the text block.
// If the parent has children, position the iterator
// on the child that is to the left of the offset.
uint32_t childIndex = (uint32_t)offset;
if (childIndex > 0) {
uint32_t numChildren = parent->GetChildCount();
NS_ASSERTION(childIndex <= numChildren, "Invalid selection offset!");
if (childIndex > numChildren) {
childIndex = numChildren;
}
childIndex -= 1;
}
nsIContent* content = parent->GetChildAt(childIndex);
NS_ENSURE_TRUE(content, NS_ERROR_FAILURE);
rv = iter->PositionAt(content);
NS_ENSURE_SUCCESS(rv, rv);
saveNode = content;
} else {
// The parent has no children, so position the iterator
// on the parent.
NS_ENSURE_TRUE(parent->IsContent(), NS_ERROR_FAILURE);
nsCOMPtr<nsIContent> content = parent->AsContent();
rv = iter->PositionAt(content);
NS_ENSURE_SUCCESS(rv, rv);
saveNode = content;
}
// Now iterate to the left, towards the beginning of
// the text block, to find the first text node you
// come across.
nsIContent* node = nullptr;
while (!iter->IsDone()) {
nsINode* current = iter->GetCurrentNode();
if (current->NodeType() == nsIDOMNode::TEXT_NODE) {
node = static_cast<nsIContent*>(current);
break;
}
iter->Prev();
}
if (node) {
// We found a node, now set the offset to the end
// of the text node.
offset = node->TextLength();
} else {
// We should never really get here, but I'm paranoid.
// We didn't find a text node above, so iterate to
// the right, towards the end of the text block, looking
// for a text node.
rv = iter->PositionAt(saveNode);
NS_ENSURE_SUCCESS(rv, rv);
node = nullptr;
while (!iter->IsDone()) {
nsINode* current = iter->GetCurrentNode();
if (current->NodeType() == nsIDOMNode::TEXT_NODE) {
node = static_cast<nsIContent*>(current);
break;
}
iter->Next();
}
NS_ENSURE_TRUE(node, NS_ERROR_FAILURE);
// We found a text node, so set the offset to
// the beginning of the node.
offset = 0;
}
for (int32_t i = 0; i < tableCount; i++) {
OffsetEntry* entry = mOffsetTable[i];
NS_ENSURE_TRUE(entry, NS_ERROR_FAILURE);
if (entry->mNode == node->AsDOMNode() &&
entry->mNodeOffset <= offset &&
offset <= entry->mNodeOffset + entry->mLength) {
*aSelStatus = nsITextServicesDocument::eBlockContains;
*aSelOffset = entry->mStrOffset + (offset - entry->mNodeOffset);
*aSelLength = 0;
// Now move the caret so that it is actually in the text node.
// We do this to keep things in sync.
//
// In most cases, the user shouldn't see any movement in the caret
// on screen.
return SetSelectionInternal(*aSelOffset, *aSelLength, true);
}
}
return NS_ERROR_FAILURE;
}
nsresult
nsTextServicesDocument::GetUncollapsedSelection(nsITextServicesDocument::TSDBlockSelectionStatus *aSelStatus, int32_t *aSelOffset, int32_t *aSelLength)
{
RefPtr<nsRange> range;
OffsetEntry *entry;
nsCOMPtr<nsISelection> domSelection;
nsresult rv = mSelCon->GetSelection(nsISelectionController::SELECTION_NORMAL,
getter_AddRefs(domSelection));
NS_ENSURE_SUCCESS(rv, rv);
NS_ENSURE_TRUE(domSelection, NS_ERROR_FAILURE);
RefPtr<Selection> selection = domSelection->AsSelection();
// It is assumed that the calling function has made sure that the
// selection is not collapsed, and that the input params to this
// method are initialized to some defaults.
nsCOMPtr<nsIDOMNode> startParent, endParent;
int32_t startOffset, endOffset;
int32_t rangeCount, tableCount;
int32_t e1s1 = 0, e1s2 = 0, e2s1 = 0, e2s2 = 0;
OffsetEntry *eStart, *eEnd;
int32_t eStartOffset, eEndOffset;
tableCount = mOffsetTable.Length();
// Get pointers to the first and last offset entries
// in the table.
eStart = mOffsetTable[0];
if (tableCount > 1) {
eEnd = mOffsetTable[tableCount - 1];
} else {
eEnd = eStart;
}
eStartOffset = eStart->mNodeOffset;
eEndOffset = eEnd->mNodeOffset + eEnd->mLength;
rv = selection->GetRangeCount(&rangeCount);
NS_ENSURE_SUCCESS(rv, rv);
// Find the first range in the selection that intersects
// the current text block.
for (int32_t i = 0; i < rangeCount; i++) {
range = selection->GetRangeAt(i);
NS_ENSURE_STATE(range);
rv = GetRangeEndPoints(range,
getter_AddRefs(startParent), &startOffset,
getter_AddRefs(endParent), &endOffset);
NS_ENSURE_SUCCESS(rv, rv);
e1s2 = nsContentUtils::ComparePoints(eStart->mNode, eStartOffset,
endParent, endOffset);
e2s1 = nsContentUtils::ComparePoints(eEnd->mNode, eEndOffset,
startParent, startOffset);
// Break out of the loop if the text block intersects the current range.
if (e1s2 <= 0 && e2s1 >= 0) {
break;
}
}
// We're done if we didn't find an intersecting range.
if (rangeCount < 1 || e1s2 > 0 || e2s1 < 0) {
*aSelStatus = nsITextServicesDocument::eBlockOutside;
*aSelOffset = *aSelLength = -1;
return NS_OK;
}
// Now that we have an intersecting range, find out more info:
e1s1 = nsContentUtils::ComparePoints(eStart->mNode, eStartOffset,
startParent, startOffset);
e2s2 = nsContentUtils::ComparePoints(eEnd->mNode, eEndOffset,
endParent, endOffset);
if (rangeCount > 1) {
// There are multiple selection ranges, we only deal
// with the first one that intersects the current,
// text block, so mark this a as a partial.
*aSelStatus = nsITextServicesDocument::eBlockPartial;
} else if (e1s1 > 0 && e2s2 < 0) {
// The range extends beyond the start and
// end of the current text block.
*aSelStatus = nsITextServicesDocument::eBlockInside;
} else if (e1s1 <= 0 && e2s2 >= 0) {
// The current text block contains the entire
// range.
*aSelStatus = nsITextServicesDocument::eBlockContains;
} else {
// The range partially intersects the block.
*aSelStatus = nsITextServicesDocument::eBlockPartial;
}
// Now create a range based on the intersection of the
// text block and range:
nsCOMPtr<nsIDOMNode> p1, p2;
int32_t o1, o2;
// The start of the range will be the rightmost
// start node.
if (e1s1 >= 0) {
p1 = do_QueryInterface(eStart->mNode);
o1 = eStartOffset;
} else {
p1 = startParent;
o1 = startOffset;
}
// The end of the range will be the leftmost
// end node.
if (e2s2 <= 0) {
p2 = do_QueryInterface(eEnd->mNode);
o2 = eEndOffset;
} else {
p2 = endParent;
o2 = endOffset;
}
rv = CreateRange(p1, o1, p2, o2, getter_AddRefs(range));
NS_ENSURE_SUCCESS(rv, rv);
// Now iterate over this range to figure out the selection's
// block offset and length.
nsCOMPtr<nsIContentIterator> iter;
rv = CreateContentIterator(range, getter_AddRefs(iter));
NS_ENSURE_SUCCESS(rv, rv);
// Find the first text node in the range.
bool found;
nsCOMPtr<nsIContent> content;
iter->First();
if (!IsTextNode(p1)) {
found = false;
while (!iter->IsDone()) {
content = do_QueryInterface(iter->GetCurrentNode());
if (IsTextNode(content)) {
p1 = do_QueryInterface(content);
NS_ENSURE_TRUE(p1, NS_ERROR_FAILURE);
o1 = 0;
found = true;
break;
}
iter->Next();
}
NS_ENSURE_TRUE(found, NS_ERROR_FAILURE);
}
// Find the last text node in the range.
iter->Last();
if (!IsTextNode(p2)) {
found = false;
while (!iter->IsDone()) {
content = do_QueryInterface(iter->GetCurrentNode());
if (IsTextNode(content)) {
p2 = do_QueryInterface(content);
NS_ENSURE_TRUE(p2, NS_ERROR_FAILURE);
nsString str;
rv = p2->GetNodeValue(str);
NS_ENSURE_SUCCESS(rv, rv);
o2 = str.Length();
found = true;
break;
}
iter->Prev();
}
NS_ENSURE_TRUE(found, NS_ERROR_FAILURE);
}
found = false;
*aSelLength = 0;
for (int32_t i = 0; i < tableCount; i++) {
entry = mOffsetTable[i];
NS_ENSURE_TRUE(entry, NS_ERROR_FAILURE);
if (!found) {
if (entry->mNode == p1.get() &&
entry->mNodeOffset <= o1 &&
o1 <= entry->mNodeOffset + entry->mLength) {
*aSelOffset = entry->mStrOffset + (o1 - entry->mNodeOffset);
if (p1 == p2 &&
entry->mNodeOffset <= o2 &&
o2 <= entry->mNodeOffset + entry->mLength) {
// The start and end of the range are in the same offset
// entry. Calculate the length of the range then we're done.
*aSelLength = o2 - o1;
break;
}
// Add the length of the sub string in this offset entry
// that follows the start of the range.
*aSelLength = entry->mLength - (o1 - entry->mNodeOffset);
found = true;
}
} else { // Found.
if (entry->mNode == p2.get() &&
entry->mNodeOffset <= o2 &&
o2 <= entry->mNodeOffset + entry->mLength) {
// We found the end of the range. Calculate the length of the
// sub string that is before the end of the range, then we're done.
*aSelLength += o2 - entry->mNodeOffset;
break;
}
// The entire entry must be in the range.
*aSelLength += entry->mLength;
}
}
return NS_OK;
}
bool
nsTextServicesDocument::SelectionIsCollapsed()
{
return(mSelStartIndex == mSelEndIndex && mSelStartOffset == mSelEndOffset);
}
bool
nsTextServicesDocument::SelectionIsValid()
{
return(mSelStartIndex >= 0);
}
nsresult
nsTextServicesDocument::GetRangeEndPoints(nsRange* aRange,
nsIDOMNode **aStartParent, int32_t *aStartOffset,
nsIDOMNode **aEndParent, int32_t *aEndOffset)
{
NS_ENSURE_TRUE(aRange && aStartParent && aStartOffset && aEndParent && aEndOffset, NS_ERROR_NULL_POINTER);
nsresult rv = aRange->GetStartContainer(aStartParent);
NS_ENSURE_SUCCESS(rv, rv);
NS_ENSURE_TRUE(aStartParent, NS_ERROR_FAILURE);
rv = aRange->GetStartOffset(aStartOffset);
NS_ENSURE_SUCCESS(rv, rv);
rv = aRange->GetEndContainer(aEndParent);
NS_ENSURE_SUCCESS(rv, rv);
NS_ENSURE_TRUE(aEndParent, NS_ERROR_FAILURE);
return aRange->GetEndOffset(aEndOffset);
}
nsresult
nsTextServicesDocument::CreateRange(nsIDOMNode *aStartParent, int32_t aStartOffset,
nsIDOMNode *aEndParent, int32_t aEndOffset,
nsRange** aRange)
{
return nsRange::CreateRange(aStartParent, aStartOffset, aEndParent,
aEndOffset, aRange);
}
nsresult
nsTextServicesDocument::FirstTextNode(nsIContentIterator *aIterator,
TSDIteratorStatus *aIteratorStatus)
{
if (aIteratorStatus) {
*aIteratorStatus = nsTextServicesDocument::eIsDone;
}
aIterator->First();
while (!aIterator->IsDone()) {
if (aIterator->GetCurrentNode()->NodeType() == nsIDOMNode::TEXT_NODE) {
if (aIteratorStatus) {
*aIteratorStatus = nsTextServicesDocument::eValid;
}
break;
}
aIterator->Next();
}
return NS_OK;
}
nsresult
nsTextServicesDocument::LastTextNode(nsIContentIterator *aIterator,
TSDIteratorStatus *aIteratorStatus)
{
if (aIteratorStatus) {
*aIteratorStatus = nsTextServicesDocument::eIsDone;
}
aIterator->Last();
while (!aIterator->IsDone()) {
if (aIterator->GetCurrentNode()->NodeType() == nsIDOMNode::TEXT_NODE) {
if (aIteratorStatus) {
*aIteratorStatus = nsTextServicesDocument::eValid;
}
break;
}
aIterator->Prev();
}
return NS_OK;
}
nsresult
nsTextServicesDocument::FirstTextNodeInCurrentBlock(nsIContentIterator *iter)
{
NS_ENSURE_TRUE(iter, NS_ERROR_NULL_POINTER);
ClearDidSkip(iter);
nsCOMPtr<nsIContent> last;
// Walk backwards over adjacent text nodes until
// we hit a block boundary:
while (!iter->IsDone()) {
nsCOMPtr<nsIContent> content = iter->GetCurrentNode()->IsContent()
? iter->GetCurrentNode()->AsContent()
: nullptr;
if (last && IsBlockNode(content)) {
break;
}
if (IsTextNode(content)) {
if (last && !HasSameBlockNodeParent(content, last)) {
// We're done, the current text node is in a
// different block.
break;
}
last = content;
}
iter->Prev();
if (DidSkip(iter)) {
break;
}
}
if (last) {
iter->PositionAt(last);
}
// XXX: What should we return if last is null?
return NS_OK;
}
nsresult
nsTextServicesDocument::FirstTextNodeInPrevBlock(nsIContentIterator *aIterator)
{
NS_ENSURE_TRUE(aIterator, NS_ERROR_NULL_POINTER);
// XXX: What if mIterator is not currently on a text node?
// Make sure mIterator is pointing to the first text node in the
// current block:
nsresult rv = FirstTextNodeInCurrentBlock(aIterator);
NS_ENSURE_SUCCESS(rv, NS_ERROR_FAILURE);
// Point mIterator to the first node before the first text node:
aIterator->Prev();
if (aIterator->IsDone()) {
return NS_ERROR_FAILURE;
}
// Now find the first text node of the next block:
return FirstTextNodeInCurrentBlock(aIterator);
}
nsresult
nsTextServicesDocument::FirstTextNodeInNextBlock(nsIContentIterator *aIterator)
{
nsCOMPtr<nsIContent> prev;
bool crossedBlockBoundary = false;
NS_ENSURE_TRUE(aIterator, NS_ERROR_NULL_POINTER);
ClearDidSkip(aIterator);
while (!aIterator->IsDone()) {
nsCOMPtr<nsIContent> content = aIterator->GetCurrentNode()->IsContent()
? aIterator->GetCurrentNode()->AsContent()
: nullptr;
if (IsTextNode(content)) {
if (crossedBlockBoundary ||
(prev && !HasSameBlockNodeParent(prev, content))) {
break;
}
prev = content;
} else if (!crossedBlockBoundary && IsBlockNode(content)) {
crossedBlockBoundary = true;
}
aIterator->Next();
if (!crossedBlockBoundary && DidSkip(aIterator)) {
crossedBlockBoundary = true;
}
}
return NS_OK;
}
nsresult
nsTextServicesDocument::GetFirstTextNodeInPrevBlock(nsIContent **aContent)
{
NS_ENSURE_TRUE(aContent, NS_ERROR_NULL_POINTER);
*aContent = 0;
// Save the iterator's current content node so we can restore
// it when we are done:
nsINode* node = mIterator->GetCurrentNode();
nsresult rv = FirstTextNodeInPrevBlock(mIterator);
if (NS_FAILED(rv)) {
// Try to restore the iterator before returning.
mIterator->PositionAt(node);
return rv;
}
if (!mIterator->IsDone()) {
nsCOMPtr<nsIContent> current = mIterator->GetCurrentNode()->IsContent()
? mIterator->GetCurrentNode()->AsContent()
: nullptr;
current.forget(aContent);
}
// Restore the iterator:
return mIterator->PositionAt(node);
}
nsresult
nsTextServicesDocument::GetFirstTextNodeInNextBlock(nsIContent **aContent)
{
NS_ENSURE_TRUE(aContent, NS_ERROR_NULL_POINTER);
*aContent = 0;
// Save the iterator's current content node so we can restore
// it when we are done:
nsINode* node = mIterator->GetCurrentNode();
nsresult rv = FirstTextNodeInNextBlock(mIterator);
if (NS_FAILED(rv)) {
// Try to restore the iterator before returning.
mIterator->PositionAt(node);
return rv;
}
if (!mIterator->IsDone()) {
nsCOMPtr<nsIContent> current = mIterator->GetCurrentNode()->IsContent()
? mIterator->GetCurrentNode()->AsContent()
: nullptr;
current.forget(aContent);
}
// Restore the iterator:
return mIterator->PositionAt(node);
}
nsresult
nsTextServicesDocument::CreateOffsetTable(nsTArray<OffsetEntry*> *aOffsetTable,
nsIContentIterator *aIterator,
TSDIteratorStatus *aIteratorStatus,
nsRange* aIterRange, nsString* aStr)
{
nsCOMPtr<nsIContent> first;
nsCOMPtr<nsIContent> prev;
NS_ENSURE_TRUE(aIterator, NS_ERROR_NULL_POINTER);
ClearOffsetTable(aOffsetTable);
if (aStr) {
aStr->Truncate();
}
if (*aIteratorStatus == nsTextServicesDocument::eIsDone) {
return NS_OK;
}
// If we have an aIterRange, retrieve the endpoints so
// they can be used in the while loop below to trim entries
// for text nodes that are partially selected by aIterRange.
nsCOMPtr<nsIDOMNode> rngStartNode, rngEndNode;
int32_t rngStartOffset = 0, rngEndOffset = 0;
if (aIterRange) {
nsresult rv =
GetRangeEndPoints(aIterRange,
getter_AddRefs(rngStartNode), &rngStartOffset,
getter_AddRefs(rngEndNode), &rngEndOffset);
NS_ENSURE_SUCCESS(rv, rv);
}
// The text service could have added text nodes to the beginning
// of the current block and called this method again. Make sure
// we really are at the beginning of the current block:
nsresult rv = FirstTextNodeInCurrentBlock(aIterator);
NS_ENSURE_SUCCESS(rv, rv);
int32_t offset = 0;
ClearDidSkip(aIterator);
while (!aIterator->IsDone()) {
nsCOMPtr<nsIContent> content = aIterator->GetCurrentNode()->IsContent()
? aIterator->GetCurrentNode()->AsContent()
: nullptr;
if (IsTextNode(content)) {
if (prev && !HasSameBlockNodeParent(prev, content)) {
break;
}
nsCOMPtr<nsIDOMNode> node = do_QueryInterface(content);
if (node) {
nsString str;
rv = node->GetNodeValue(str);
NS_ENSURE_SUCCESS(rv, rv);
// Add an entry for this text node into the offset table:
OffsetEntry *entry = new OffsetEntry(node, offset, str.Length());
aOffsetTable->AppendElement(entry);
// If one or both of the endpoints of the iteration range
// are in the text node for this entry, make sure the entry
// only accounts for the portion of the text node that is
// in the range.
int32_t startOffset = 0;
int32_t endOffset = str.Length();
bool adjustStr = false;
if (entry->mNode == rngStartNode) {
entry->mNodeOffset = startOffset = rngStartOffset;
adjustStr = true;
}
if (entry->mNode == rngEndNode) {
endOffset = rngEndOffset;
adjustStr = true;
}
if (adjustStr) {
entry->mLength = endOffset - startOffset;
str = Substring(str, startOffset, entry->mLength);
}
offset += str.Length();
if (aStr) {
// Append the text node's string to the output string:
if (!first) {
*aStr = str;
} else {
*aStr += str;
}
}
}
prev = content;
if (!first) {
first = content;
}
}
// XXX This should be checked before IsTextNode(), but IsBlockNode() returns
// true even if content is a text node. See bug 1311934.
else if (IsBlockNode(content)) {
break;
}
aIterator->Next();
if (DidSkip(aIterator)) {
break;
}
}
if (first) {
// Always leave the iterator pointing at the first
// text node of the current block!
aIterator->PositionAt(first);
} else {
// If we never ran across a text node, the iterator
// might have been pointing to something invalid to
// begin with.
*aIteratorStatus = nsTextServicesDocument::eIsDone;
}
return NS_OK;
}
nsresult
nsTextServicesDocument::RemoveInvalidOffsetEntries()
{
for (size_t i = 0; i < mOffsetTable.Length(); ) {
OffsetEntry* entry = mOffsetTable[i];
if (!entry->mIsValid) {
mOffsetTable.RemoveElementAt(i);
if (mSelStartIndex >= 0 && static_cast<size_t>(mSelStartIndex) >= i) {
// We are deleting an entry that comes before
// mSelStartIndex, decrement mSelStartIndex so
// that it points to the correct entry!
NS_ASSERTION(i != static_cast<size_t>(mSelStartIndex),
"Invalid selection index.");
--mSelStartIndex;
--mSelEndIndex;
}
} else {
i++;
}
}
return NS_OK;
}
nsresult
nsTextServicesDocument::ClearOffsetTable(nsTArray<OffsetEntry*> *aOffsetTable)
{
for (size_t i = 0; i < aOffsetTable->Length(); i++) {
delete aOffsetTable->ElementAt(i);
}
aOffsetTable->Clear();
return NS_OK;
}
nsresult
nsTextServicesDocument::SplitOffsetEntry(int32_t aTableIndex, int32_t aNewEntryLength)
{
OffsetEntry *entry = mOffsetTable[aTableIndex];
NS_ASSERTION((aNewEntryLength > 0), "aNewEntryLength <= 0");
NS_ASSERTION((aNewEntryLength < entry->mLength), "aNewEntryLength >= mLength");
if (aNewEntryLength < 1 || aNewEntryLength >= entry->mLength) {
return NS_ERROR_FAILURE;
}
int32_t oldLength = entry->mLength - aNewEntryLength;
OffsetEntry *newEntry = new OffsetEntry(entry->mNode,
entry->mStrOffset + oldLength,
aNewEntryLength);
if (!mOffsetTable.InsertElementAt(aTableIndex + 1, newEntry)) {
delete newEntry;
return NS_ERROR_FAILURE;
}
// Adjust entry fields:
entry->mLength = oldLength;
newEntry->mNodeOffset = entry->mNodeOffset + oldLength;
return NS_OK;
}
nsresult
nsTextServicesDocument::NodeHasOffsetEntry(nsTArray<OffsetEntry*> *aOffsetTable, nsIDOMNode *aNode, bool *aHasEntry, int32_t *aEntryIndex)
{
NS_ENSURE_TRUE(aNode && aHasEntry && aEntryIndex, NS_ERROR_NULL_POINTER);
for (size_t i = 0; i < aOffsetTable->Length(); i++) {
OffsetEntry* entry = (*aOffsetTable)[i];
NS_ENSURE_TRUE(entry, NS_ERROR_FAILURE);
if (entry->mNode == aNode) {
*aHasEntry = true;
*aEntryIndex = i;
return NS_OK;
}
}
*aHasEntry = false;
*aEntryIndex = -1;
return NS_OK;
}
// Spellchecker code has this. See bug 211343
#define IS_NBSP_CHAR(c) (((unsigned char)0xa0)==(c))
nsresult
nsTextServicesDocument::FindWordBounds(nsTArray<OffsetEntry*> *aOffsetTable,
nsString *aBlockStr,
nsIDOMNode *aNode,
int32_t aNodeOffset,
nsIDOMNode **aWordStartNode,
int32_t *aWordStartOffset,
nsIDOMNode **aWordEndNode,
int32_t *aWordEndOffset)
{
// Initialize return values.
if (aWordStartNode) {
*aWordStartNode = nullptr;
}
if (aWordStartOffset) {
*aWordStartOffset = 0;
}
if (aWordEndNode) {
*aWordEndNode = nullptr;
}
if (aWordEndOffset) {
*aWordEndOffset = 0;
}
int32_t entryIndex = 0;
bool hasEntry = false;
// It's assumed that aNode is a text node. The first thing
// we do is get its index in the offset table so we can
// calculate the dom point's string offset.
nsresult rv = NodeHasOffsetEntry(aOffsetTable, aNode, &hasEntry, &entryIndex);
NS_ENSURE_SUCCESS(rv, rv);
NS_ENSURE_TRUE(hasEntry, NS_ERROR_FAILURE);
// Next we map aNodeOffset into a string offset.
OffsetEntry *entry = (*aOffsetTable)[entryIndex];
uint32_t strOffset = entry->mStrOffset + aNodeOffset - entry->mNodeOffset;
// Now we use the word breaker to find the beginning and end
// of the word from our calculated string offset.
const char16_t *str = aBlockStr->get();
uint32_t strLen = aBlockStr->Length();
nsIWordBreaker* wordBreaker = nsContentUtils::WordBreaker();
nsWordRange res = wordBreaker->FindWord(str, strLen, strOffset);
if (res.mBegin > strLen) {
return str ? NS_ERROR_ILLEGAL_VALUE : NS_ERROR_NULL_POINTER;
}
// Strip out the NBSPs at the ends
while (res.mBegin <= res.mEnd && IS_NBSP_CHAR(str[res.mBegin])) {
res.mBegin++;
}
if (str[res.mEnd] == (unsigned char)0x20) {
uint32_t realEndWord = res.mEnd - 1;
while (realEndWord > res.mBegin && IS_NBSP_CHAR(str[realEndWord])) {
realEndWord--;
}
if (realEndWord < res.mEnd - 1) {
res.mEnd = realEndWord + 1;
}
}
// Now that we have the string offsets for the beginning
// and end of the word, run through the offset table and
// convert them back into dom points.
size_t lastIndex = aOffsetTable->Length() - 1;
for (size_t i = 0; i <= lastIndex; i++) {
entry = (*aOffsetTable)[i];
int32_t strEndOffset = entry->mStrOffset + entry->mLength;
// Check to see if res.mBegin is within the range covered
// by this entry. Note that if res.mBegin is after the last
// character covered by this entry, we will use the next
// entry if there is one.
if (uint32_t(entry->mStrOffset) <= res.mBegin &&
(res.mBegin < static_cast<uint32_t>(strEndOffset) ||
(res.mBegin == static_cast<uint32_t>(strEndOffset) &&
i == lastIndex))) {
if (aWordStartNode) {
*aWordStartNode = entry->mNode;
NS_IF_ADDREF(*aWordStartNode);
}
if (aWordStartOffset) {
*aWordStartOffset = entry->mNodeOffset + res.mBegin - entry->mStrOffset;
}
if (!aWordEndNode && !aWordEndOffset) {
// We've found our start entry, but if we're not looking
// for end entries, we're done.
break;
}
}
// Check to see if res.mEnd is within the range covered
// by this entry.
if (static_cast<uint32_t>(entry->mStrOffset) <= res.mEnd &&
res.mEnd <= static_cast<uint32_t>(strEndOffset)) {
if (res.mBegin == res.mEnd &&
res.mEnd == static_cast<uint32_t>(strEndOffset) &&
i != lastIndex) {
// Wait for the next round so that we use the same entry
// we did for aWordStartNode.
continue;
}
if (aWordEndNode) {
*aWordEndNode = entry->mNode;
NS_IF_ADDREF(*aWordEndNode);
}
if (aWordEndOffset) {
*aWordEndOffset = entry->mNodeOffset + res.mEnd - entry->mStrOffset;
}
break;
}
}
return NS_OK;
}
NS_IMETHODIMP
nsTextServicesDocument::WillInsertNode(nsIDOMNode *aNode,
nsIDOMNode *aParent,
int32_t aPosition)
{
return NS_OK;
}
NS_IMETHODIMP
nsTextServicesDocument::WillDeleteNode(nsIDOMNode *aChild)
{
return NS_OK;
}
NS_IMETHODIMP
nsTextServicesDocument::WillSplitNode(nsIDOMNode *aExistingRightNode,
int32_t aOffset)
{
return NS_OK;
}
NS_IMETHODIMP
nsTextServicesDocument::WillJoinNodes(nsIDOMNode *aLeftNode,
nsIDOMNode *aRightNode,
nsIDOMNode *aParent)
{
return NS_OK;
}
// -------------------------------
// stubs for unused listen methods
// -------------------------------
NS_IMETHODIMP
nsTextServicesDocument::WillCreateNode(const nsAString& aTag, nsIDOMNode *aParent, int32_t aPosition)
{
return NS_OK;
}
NS_IMETHODIMP
nsTextServicesDocument::DidCreateNode(const nsAString& aTag, nsIDOMNode *aNode, nsIDOMNode *aParent, int32_t aPosition, nsresult aResult)
{
return NS_OK;
}
NS_IMETHODIMP
nsTextServicesDocument::WillInsertText(nsIDOMCharacterData *aTextNode, int32_t aOffset, const nsAString &aString)
{
return NS_OK;
}
NS_IMETHODIMP
nsTextServicesDocument::DidInsertText(nsIDOMCharacterData *aTextNode, int32_t aOffset, const nsAString &aString, nsresult aResult)
{
return NS_OK;
}
NS_IMETHODIMP
nsTextServicesDocument::WillDeleteText(nsIDOMCharacterData *aTextNode, int32_t aOffset, int32_t aLength)
{
return NS_OK;
}
NS_IMETHODIMP
nsTextServicesDocument::DidDeleteText(nsIDOMCharacterData *aTextNode, int32_t aOffset, int32_t aLength, nsresult aResult)
{
return NS_OK;
}
NS_IMETHODIMP
nsTextServicesDocument::WillDeleteSelection(nsISelection *aSelection)
{
return NS_OK;
}
NS_IMETHODIMP
nsTextServicesDocument::DidDeleteSelection(nsISelection *aSelection)
{
return NS_OK;
}