You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
1641 lines
54 KiB
1641 lines
54 KiB
/**
|
|
* This file is part of the DOM implementation for KDE.
|
|
*
|
|
* (C) 1999-2003 Lars Knoll (knoll@kde.org)
|
|
* (C) 2001-2003 Dirk Mueller (mueller@kde.org)
|
|
* (C) 2000 Gunnstein Lye (gunnstein@netcom.no)
|
|
* (C) 2000 Frederik Holljen (frederik.holljen@hig.no)
|
|
* (C) 2001 Peter Kelly (pmk@post.com)
|
|
* Copyright (C) 2003 Apple Computer, Inc.
|
|
*
|
|
* This library is free software; you can redistribute it and/or
|
|
* modify it under the terms of the GNU Library General Public
|
|
* License as published by the Free Software Foundation; either
|
|
* version 2 of the License, or (at your option) any later version.
|
|
*
|
|
* This library is distributed in the hope that it will be useful,
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
* Library General Public License for more details.
|
|
*
|
|
* You should have received a copy of the GNU Library General Public License
|
|
* along with this library; see the file COPYING.LIB. If not, write to
|
|
* the Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor,
|
|
* Boston, MA 02110-1301, USA.
|
|
*/
|
|
|
|
#include "dom/dom_exception.h"
|
|
#include "dom_docimpl.h"
|
|
#include "dom2_rangeimpl.h"
|
|
#include "dom_textimpl.h"
|
|
#include "dom_xmlimpl.h"
|
|
#include "html/html_elementimpl.h"
|
|
#include "misc/htmltags.h"
|
|
|
|
using namespace DOM;
|
|
|
|
|
|
RangeImpl::RangeImpl(DocumentImpl *_ownerDocument)
|
|
{
|
|
m_ownerDocument = _ownerDocument;
|
|
m_ownerDocument->ref();
|
|
m_startContainer = _ownerDocument;
|
|
m_startContainer->ref();
|
|
m_endContainer = _ownerDocument;
|
|
m_endContainer->ref();
|
|
m_startOffset = 0;
|
|
m_endOffset = 0;
|
|
m_detached = false;
|
|
}
|
|
|
|
RangeImpl::RangeImpl(DocumentImpl *_ownerDocument,
|
|
NodeImpl *_startContainer, long _startOffset,
|
|
NodeImpl *_endContainer, long _endOffset)
|
|
{
|
|
m_ownerDocument = _ownerDocument;
|
|
m_ownerDocument->ref();
|
|
m_startContainer = _startContainer;
|
|
m_startContainer->ref();
|
|
m_startOffset = _startOffset;
|
|
m_endContainer = _endContainer;
|
|
m_endContainer->ref();
|
|
m_endOffset = _endOffset;
|
|
m_detached = false;
|
|
}
|
|
|
|
RangeImpl::~RangeImpl()
|
|
{
|
|
m_ownerDocument->deref();
|
|
int exceptioncode = 0;
|
|
if (!m_detached)
|
|
detach(exceptioncode);
|
|
}
|
|
|
|
NodeImpl *RangeImpl::startContainer(int &exceptioncode) const
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return 0;
|
|
}
|
|
|
|
return m_startContainer;
|
|
}
|
|
|
|
long RangeImpl::startOffset(int &exceptioncode) const
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return 0;
|
|
}
|
|
|
|
return m_startOffset;
|
|
}
|
|
|
|
NodeImpl *RangeImpl::endContainer(int &exceptioncode) const
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return 0;
|
|
}
|
|
|
|
return m_endContainer;
|
|
}
|
|
|
|
long RangeImpl::endOffset(int &exceptioncode) const
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return 0;
|
|
}
|
|
|
|
return m_endOffset;
|
|
}
|
|
|
|
NodeImpl *RangeImpl::commonAncestorContainer(int &exceptioncode)
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return 0;
|
|
}
|
|
|
|
NodeImpl *com = commonAncestorContainer(m_startContainer,m_endContainer);
|
|
if (!com) // should never happen
|
|
exceptioncode = DOMException::WRONG_DOCUMENT_ERR;
|
|
return com;
|
|
}
|
|
|
|
NodeImpl *RangeImpl::commonAncestorContainer(NodeImpl *containerA, NodeImpl *containerB)
|
|
{
|
|
NodeImpl *parentStart;
|
|
|
|
for (parentStart = containerA; parentStart; parentStart = parentStart->parentNode()) {
|
|
NodeImpl *parentEnd = containerB;
|
|
while( parentEnd && (parentStart != parentEnd) )
|
|
parentEnd = parentEnd->parentNode();
|
|
|
|
if(parentStart == parentEnd) break;
|
|
}
|
|
|
|
return parentStart;
|
|
}
|
|
|
|
bool RangeImpl::collapsed(int &exceptioncode) const
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return 0;
|
|
}
|
|
|
|
return (m_startContainer == m_endContainer && m_startOffset == m_endOffset);
|
|
}
|
|
|
|
void RangeImpl::setStart( NodeImpl *refNode, long offset, int &exceptioncode )
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return;
|
|
}
|
|
|
|
if (!refNode) {
|
|
exceptioncode = DOMException::NOT_FOUND_ERR;
|
|
return;
|
|
}
|
|
|
|
if (refNode->getDocument() != m_ownerDocument) {
|
|
exceptioncode = DOMException::WRONG_DOCUMENT_ERR;
|
|
return;
|
|
}
|
|
|
|
checkNodeWOffset( refNode, offset, exceptioncode );
|
|
if (exceptioncode)
|
|
return;
|
|
|
|
setStartContainer(refNode);
|
|
m_startOffset = offset;
|
|
|
|
// check if different root container
|
|
NodeImpl *endRootContainer = m_endContainer;
|
|
while (endRootContainer->parentNode())
|
|
endRootContainer = endRootContainer->parentNode();
|
|
NodeImpl *startRootContainer = m_startContainer;
|
|
while (startRootContainer->parentNode())
|
|
startRootContainer = startRootContainer->parentNode();
|
|
if (startRootContainer != endRootContainer)
|
|
collapse(true,exceptioncode);
|
|
// check if new start after end
|
|
else if (compareBoundaryPoints(m_startContainer,m_startOffset,m_endContainer,m_endOffset) > 0)
|
|
collapse(true,exceptioncode);
|
|
}
|
|
|
|
void RangeImpl::setEnd( NodeImpl *refNode, long offset, int &exceptioncode )
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return;
|
|
}
|
|
|
|
if (!refNode) {
|
|
exceptioncode = DOMException::NOT_FOUND_ERR;
|
|
return;
|
|
}
|
|
|
|
if (refNode->getDocument() != m_ownerDocument) {
|
|
exceptioncode = DOMException::WRONG_DOCUMENT_ERR;
|
|
return;
|
|
}
|
|
|
|
checkNodeWOffset( refNode, offset, exceptioncode );
|
|
if (exceptioncode)
|
|
return;
|
|
|
|
setEndContainer(refNode);
|
|
m_endOffset = offset;
|
|
|
|
// check if different root container
|
|
NodeImpl *endRootContainer = m_endContainer;
|
|
while (endRootContainer->parentNode())
|
|
endRootContainer = endRootContainer->parentNode();
|
|
NodeImpl *startRootContainer = m_startContainer;
|
|
while (startRootContainer->parentNode())
|
|
startRootContainer = startRootContainer->parentNode();
|
|
if (startRootContainer != endRootContainer)
|
|
collapse(false,exceptioncode);
|
|
// check if new end before start
|
|
if (compareBoundaryPoints(m_startContainer,m_startOffset,m_endContainer,m_endOffset) > 0)
|
|
collapse(false,exceptioncode);
|
|
}
|
|
|
|
void RangeImpl::collapse( bool toStart, int &exceptioncode )
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return;
|
|
}
|
|
|
|
if( toStart ) // collapse to start
|
|
{
|
|
setEndContainer(m_startContainer);
|
|
m_endOffset = m_startOffset;
|
|
}
|
|
else // collapse to end
|
|
{
|
|
setStartContainer(m_endContainer);
|
|
m_startOffset = m_endOffset;
|
|
}
|
|
}
|
|
|
|
short RangeImpl::compareBoundaryPoints( Range::CompareHow how, RangeImpl *sourceRange, int &exceptioncode )
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return 0;
|
|
}
|
|
|
|
if (!sourceRange) {
|
|
exceptioncode = DOMException::NOT_FOUND_ERR;
|
|
return 0;
|
|
}
|
|
|
|
NodeImpl *thisCont = commonAncestorContainer(exceptioncode);
|
|
NodeImpl *sourceCont = sourceRange->commonAncestorContainer(exceptioncode);
|
|
if (exceptioncode)
|
|
return 0;
|
|
|
|
if (thisCont->getDocument() != sourceCont->getDocument()) {
|
|
exceptioncode = DOMException::WRONG_DOCUMENT_ERR;
|
|
return 0;
|
|
}
|
|
|
|
NodeImpl *thisTop = thisCont;
|
|
NodeImpl *sourceTop = sourceCont;
|
|
while (thisTop->parentNode())
|
|
thisTop = thisTop->parentNode();
|
|
while (sourceTop->parentNode())
|
|
sourceTop = sourceTop->parentNode();
|
|
if (thisTop != sourceTop) { // in different DocumentFragments
|
|
exceptioncode = DOMException::WRONG_DOCUMENT_ERR;
|
|
return 0;
|
|
}
|
|
|
|
switch(how)
|
|
{
|
|
case Range::START_TO_START:
|
|
return compareBoundaryPoints( m_startContainer, m_startOffset,
|
|
sourceRange->startContainer(exceptioncode), sourceRange->startOffset(exceptioncode) );
|
|
break;
|
|
case Range::START_TO_END:
|
|
return compareBoundaryPoints( m_startContainer, m_startOffset,
|
|
sourceRange->endContainer(exceptioncode), sourceRange->endOffset(exceptioncode) );
|
|
break;
|
|
case Range::END_TO_END:
|
|
return compareBoundaryPoints( m_endContainer, m_endOffset,
|
|
sourceRange->endContainer(exceptioncode), sourceRange->endOffset(exceptioncode) );
|
|
break;
|
|
case Range::END_TO_START:
|
|
return compareBoundaryPoints( m_endContainer, m_endOffset,
|
|
sourceRange->startContainer(exceptioncode), sourceRange->startOffset(exceptioncode) );
|
|
break;
|
|
default:
|
|
exceptioncode = DOMException::SYNTAX_ERR;
|
|
return 0;
|
|
}
|
|
}
|
|
|
|
short RangeImpl::compareBoundaryPoints( NodeImpl *containerA, long offsetA, NodeImpl *containerB, long offsetB )
|
|
{
|
|
// see DOM2 traversal & range section 2.5
|
|
|
|
// case 1: both points have the same container
|
|
if( containerA == containerB )
|
|
{
|
|
if( offsetA == offsetB ) return 0; // A is equal to B
|
|
if( offsetA < offsetB ) return -1; // A is before B
|
|
else return 1; // A is after B
|
|
}
|
|
|
|
// case 2: node C (container B or an ancestor) is a child node of A
|
|
NodeImpl *c = containerB;
|
|
while (c && c->parentNode() != containerA)
|
|
c = c->parentNode();
|
|
if (c) {
|
|
int offsetC = 0;
|
|
NodeImpl* n = containerA->firstChild();
|
|
while (n != c) {
|
|
offsetC++;
|
|
n = n->nextSibling();
|
|
}
|
|
|
|
if( offsetA <= offsetC ) return -1; // A is before B
|
|
else return 1; // A is after B
|
|
}
|
|
|
|
// case 3: node C (container A or an ancestor) is a child node of B
|
|
c = containerA;
|
|
while (c && c->parentNode() != containerB)
|
|
c = c->parentNode();
|
|
if (c) {
|
|
int offsetC = 0;
|
|
NodeImpl* n = containerB->firstChild();
|
|
while (n != c) {
|
|
offsetC++;
|
|
n = n->nextSibling();
|
|
}
|
|
|
|
if( offsetC < offsetB ) return -1; // A is before B
|
|
else return 1; // A is after B
|
|
}
|
|
|
|
// case 4: containers A & B are siblings, or children of siblings
|
|
// ### we need to do a traversal here instead
|
|
NodeImpl *cmnRoot = commonAncestorContainer(containerA,containerB);
|
|
if (!cmnRoot) return -1; // Whatever...
|
|
NodeImpl *childA = containerA;
|
|
while (childA->parentNode() != cmnRoot)
|
|
childA = childA->parentNode();
|
|
NodeImpl *childB = containerB;
|
|
while (childB->parentNode() != cmnRoot)
|
|
childB = childB->parentNode();
|
|
|
|
NodeImpl *n = cmnRoot->firstChild();
|
|
int i = 0;
|
|
int childAOffset = -1;
|
|
int childBOffset = -1;
|
|
while (childAOffset < 0 || childBOffset < 0) {
|
|
if (n == childA)
|
|
childAOffset = i;
|
|
if (n == childB)
|
|
childBOffset = i;
|
|
n = n->nextSibling();
|
|
i++;
|
|
}
|
|
|
|
if( childAOffset == childBOffset ) return 0; // A is equal to B
|
|
if( childAOffset < childBOffset ) return -1; // A is before B
|
|
else return 1; // A is after B
|
|
}
|
|
|
|
bool RangeImpl::boundaryPointsValid( )
|
|
{
|
|
short valid = compareBoundaryPoints( m_startContainer, m_startOffset,
|
|
m_endContainer, m_endOffset );
|
|
if( valid == 1 ) return false;
|
|
else return true;
|
|
|
|
}
|
|
|
|
void RangeImpl::deleteContents( int &exceptioncode ) {
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return;
|
|
}
|
|
|
|
checkDeleteExtract(exceptioncode);
|
|
if (exceptioncode)
|
|
return;
|
|
|
|
processContents(DELETE_CONTENTS,exceptioncode);
|
|
}
|
|
|
|
DocumentFragmentImpl *RangeImpl::processContents ( ActionType action, int &exceptioncode )
|
|
{
|
|
// ### when mutation events are implemented, we will have to take into account
|
|
// situations where the tree is being transformed while we delete - ugh!
|
|
|
|
// ### perhaps disable node deletion notification for this range while we do this?
|
|
|
|
if (collapsed(exceptioncode))
|
|
return 0;
|
|
if (exceptioncode)
|
|
return 0;
|
|
|
|
NodeImpl *cmnRoot = commonAncestorContainer(exceptioncode);
|
|
if (exceptioncode)
|
|
return 0;
|
|
|
|
// what is the highest node that partially selects the start of the range?
|
|
NodeImpl *partialStart = 0;
|
|
if (m_startContainer != cmnRoot) {
|
|
partialStart = m_startContainer;
|
|
while (partialStart->parentNode() != cmnRoot)
|
|
partialStart = partialStart->parentNode();
|
|
}
|
|
|
|
// what is the highest node that partially selects the end of the range?
|
|
NodeImpl *partialEnd = 0;
|
|
if (m_endContainer != cmnRoot) {
|
|
partialEnd = m_endContainer;
|
|
while (partialEnd->parentNode() != cmnRoot)
|
|
partialEnd = partialEnd->parentNode();
|
|
}
|
|
|
|
DocumentFragmentImpl *fragment = 0;
|
|
if (action == EXTRACT_CONTENTS || action == CLONE_CONTENTS)
|
|
fragment = new DocumentFragmentImpl(m_ownerDocument);
|
|
|
|
// Simple case: the start and end containers are the same. We just grab
|
|
// everything >= start offset and < end offset
|
|
if (m_startContainer == m_endContainer) {
|
|
if(m_startContainer->nodeType() == Node::TEXT_NODE ||
|
|
m_startContainer->nodeType() == Node::CDATA_SECTION_NODE ||
|
|
m_startContainer->nodeType() == Node::COMMENT_NODE) {
|
|
|
|
if (action == EXTRACT_CONTENTS || action == CLONE_CONTENTS) {
|
|
CharacterDataImpl *c = static_cast<CharacterDataImpl*>(m_startContainer->cloneNode(true));
|
|
c->deleteData(m_endOffset,static_cast<CharacterDataImpl*>(m_startContainer)->length()-m_endOffset,exceptioncode);
|
|
c->deleteData(0,m_startOffset,exceptioncode);
|
|
fragment->appendChild(c,exceptioncode);
|
|
}
|
|
if (action == EXTRACT_CONTENTS || action == DELETE_CONTENTS)
|
|
static_cast<CharacterDataImpl*>(m_startContainer)->deleteData(m_startOffset,m_endOffset-m_startOffset,exceptioncode);
|
|
}
|
|
else if (m_startContainer->nodeType() == Node::PROCESSING_INSTRUCTION_NODE) {
|
|
// ### operate just on data ?
|
|
}
|
|
else {
|
|
NodeImpl *n = m_startContainer->firstChild();
|
|
unsigned long i;
|
|
for(i = 0; i < m_startOffset; i++) // skip until m_startOffset
|
|
n = n->nextSibling();
|
|
while (n && i < m_endOffset) { // delete until m_endOffset
|
|
NodeImpl *next = n->nextSibling();
|
|
if (action == EXTRACT_CONTENTS)
|
|
fragment->appendChild(n,exceptioncode); // will remove n from its parent
|
|
else if (action == CLONE_CONTENTS)
|
|
fragment->appendChild(n->cloneNode(true),exceptioncode);
|
|
else
|
|
m_startContainer->removeChild(n,exceptioncode);
|
|
n = next;
|
|
i++;
|
|
}
|
|
}
|
|
collapse(true,exceptioncode);
|
|
return fragment;
|
|
}
|
|
|
|
// Complex case: Start and end containers are different.
|
|
// There are three possiblities here:
|
|
// 1. Start container == cmnRoot (End container must be a descendant)
|
|
// 2. End container == cmnRoot (Start container must be a descendant)
|
|
// 3. Neither is cmnRoot, they are both descendants
|
|
//
|
|
// In case 3, we grab everything after the start (up until a direct child
|
|
// of cmnRoot) into leftContents, and everything before the end (up until
|
|
// a direct child of cmnRoot) into rightContents. Then we process all
|
|
// cmnRoot children between leftContents and rightContents
|
|
//
|
|
// In case 1 or 2, we skip either processing of leftContents or rightContents,
|
|
// in which case the last lot of nodes either goes from the first or last
|
|
// child of cmnRoot.
|
|
//
|
|
// These are deleted, cloned, or extracted (i.e. both) depending on action.
|
|
|
|
NodeImpl *leftContents = 0;
|
|
if (m_startContainer != cmnRoot) {
|
|
// process the left-hand side of the range, up until the last ancestor of
|
|
// m_startContainer before cmnRoot
|
|
if(m_startContainer->nodeType() == Node::TEXT_NODE ||
|
|
m_startContainer->nodeType() == Node::CDATA_SECTION_NODE ||
|
|
m_startContainer->nodeType() == Node::COMMENT_NODE) {
|
|
|
|
if (action == EXTRACT_CONTENTS || action == CLONE_CONTENTS) {
|
|
CharacterDataImpl *c = static_cast<CharacterDataImpl*>(m_startContainer->cloneNode(true));
|
|
c->deleteData(0,m_startOffset,exceptioncode);
|
|
leftContents = c;
|
|
}
|
|
if (action == EXTRACT_CONTENTS || action == DELETE_CONTENTS)
|
|
static_cast<CharacterDataImpl*>(m_startContainer)->deleteData(
|
|
m_startOffset,static_cast<CharacterDataImpl*>(m_startContainer)->length()-m_startOffset,exceptioncode);
|
|
}
|
|
else if (m_startContainer->nodeType() == Node::PROCESSING_INSTRUCTION_NODE) {
|
|
// ### operate just on data ?
|
|
// leftContents = ...
|
|
}
|
|
else {
|
|
if (action == EXTRACT_CONTENTS || action == CLONE_CONTENTS)
|
|
leftContents = m_startContainer->cloneNode(false);
|
|
NodeImpl *n = m_startContainer->firstChild();
|
|
unsigned long i;
|
|
for(i = 0; i < m_startOffset; i++) // skip until m_startOffset
|
|
n = n->nextSibling();
|
|
while (n) { // process until end
|
|
NodeImpl *next = n->nextSibling();
|
|
if (action == EXTRACT_CONTENTS)
|
|
leftContents->appendChild(n,exceptioncode); // will remove n from m_startContainer
|
|
else if (action == CLONE_CONTENTS)
|
|
leftContents->appendChild(n->cloneNode(true),exceptioncode);
|
|
else
|
|
m_startContainer->removeChild(n,exceptioncode);
|
|
n = next;
|
|
}
|
|
}
|
|
|
|
NodeImpl *leftParent = m_startContainer->parentNode();
|
|
NodeImpl *n = m_startContainer->nextSibling();
|
|
for (; leftParent != cmnRoot; leftParent = leftParent->parentNode()) {
|
|
if (action == EXTRACT_CONTENTS || action == CLONE_CONTENTS) {
|
|
NodeImpl *leftContentsParent = leftParent->cloneNode(false);
|
|
leftContentsParent->appendChild(leftContents,exceptioncode);
|
|
leftContents = leftContentsParent;
|
|
}
|
|
|
|
NodeImpl *next;
|
|
for (; n; n = next ) {
|
|
next = n->nextSibling();
|
|
if (action == EXTRACT_CONTENTS)
|
|
leftContents->appendChild(n,exceptioncode); // will remove n from leftParent
|
|
else if (action == CLONE_CONTENTS)
|
|
leftContents->appendChild(n->cloneNode(true),exceptioncode);
|
|
else
|
|
leftParent->removeChild(n,exceptioncode);
|
|
}
|
|
n = leftParent->nextSibling();
|
|
}
|
|
}
|
|
|
|
NodeImpl *rightContents = 0;
|
|
if (m_endContainer != cmnRoot) {
|
|
// delete the right-hand side of the range, up until the last ancestor of
|
|
// m_endContainer before cmnRoot
|
|
if(m_endContainer->nodeType() == Node::TEXT_NODE ||
|
|
m_endContainer->nodeType() == Node::CDATA_SECTION_NODE ||
|
|
m_endContainer->nodeType() == Node::COMMENT_NODE) {
|
|
|
|
if (action == EXTRACT_CONTENTS || action == CLONE_CONTENTS) {
|
|
CharacterDataImpl *c = static_cast<CharacterDataImpl*>(m_endContainer->cloneNode(true));
|
|
c->deleteData(m_endOffset,static_cast<CharacterDataImpl*>(m_endContainer)->length()-m_endOffset,exceptioncode);
|
|
rightContents = c;
|
|
}
|
|
if (action == EXTRACT_CONTENTS || action == DELETE_CONTENTS)
|
|
static_cast<CharacterDataImpl*>(m_endContainer)->deleteData(0,m_endOffset,exceptioncode);
|
|
}
|
|
else if (m_startContainer->nodeType() == Node::PROCESSING_INSTRUCTION_NODE) {
|
|
// ### operate just on data ?
|
|
// rightContents = ...
|
|
}
|
|
else {
|
|
if (action == EXTRACT_CONTENTS || action == CLONE_CONTENTS)
|
|
rightContents = m_endContainer->cloneNode(false);
|
|
NodeImpl *n = m_endContainer->firstChild();
|
|
unsigned long i;
|
|
for(i = 0; i+1 < m_endOffset; i++) // skip to m_endOffset
|
|
n = n->nextSibling();
|
|
NodeImpl *prev;
|
|
for (; n; n = prev ) {
|
|
prev = n->previousSibling();
|
|
if (action == EXTRACT_CONTENTS)
|
|
rightContents->insertBefore(n,rightContents->firstChild(),exceptioncode); // will remove n from its parent
|
|
else if (action == CLONE_CONTENTS)
|
|
rightContents->insertBefore(n->cloneNode(true),rightContents->firstChild(),exceptioncode);
|
|
else
|
|
m_endContainer->removeChild(n,exceptioncode);
|
|
}
|
|
}
|
|
|
|
NodeImpl *rightParent = m_endContainer->parentNode();
|
|
NodeImpl *n = m_endContainer->previousSibling();
|
|
for (; rightParent != cmnRoot; rightParent = rightParent->parentNode()) {
|
|
if (action == EXTRACT_CONTENTS || action == CLONE_CONTENTS) {
|
|
NodeImpl *rightContentsParent = rightParent->cloneNode(false);
|
|
rightContentsParent->appendChild(rightContents,exceptioncode);
|
|
rightContents = rightContentsParent;
|
|
}
|
|
|
|
NodeImpl *prev;
|
|
for (; n; n = prev ) {
|
|
prev = n->previousSibling();
|
|
if (action == EXTRACT_CONTENTS)
|
|
rightContents->insertBefore(n,rightContents->firstChild(),exceptioncode); // will remove n from its parent
|
|
else if (action == CLONE_CONTENTS)
|
|
rightContents->insertBefore(n->cloneNode(true),rightContents->firstChild(),exceptioncode);
|
|
else
|
|
rightParent->removeChild(n,exceptioncode);
|
|
|
|
}
|
|
n = rightParent->previousSibling();
|
|
}
|
|
}
|
|
|
|
// delete all children of cmnRoot between the start and end container
|
|
|
|
NodeImpl *processStart; // child of cmnRooot
|
|
if (m_startContainer == cmnRoot) {
|
|
unsigned long i;
|
|
processStart = m_startContainer->firstChild();
|
|
for (i = 0; i < m_startOffset; i++)
|
|
processStart = processStart->nextSibling();
|
|
}
|
|
else {
|
|
processStart = m_startContainer;
|
|
while (processStart->parentNode() != cmnRoot)
|
|
processStart = processStart->parentNode();
|
|
processStart = processStart->nextSibling();
|
|
}
|
|
NodeImpl *processEnd; // child of cmnRooot
|
|
if (m_endContainer == cmnRoot) {
|
|
unsigned long i;
|
|
processEnd = m_endContainer->firstChild();
|
|
for (i = 0; i < m_endOffset; i++)
|
|
processEnd = processEnd->nextSibling();
|
|
}
|
|
else {
|
|
processEnd = m_endContainer;
|
|
while (processEnd->parentNode() != cmnRoot)
|
|
processEnd = processEnd->parentNode();
|
|
}
|
|
|
|
// Now add leftContents, stuff in between, and rightContents to the fragment
|
|
// (or just delete the stuff in between)
|
|
|
|
if ((action == EXTRACT_CONTENTS || action == CLONE_CONTENTS) && leftContents)
|
|
fragment->appendChild(leftContents,exceptioncode);
|
|
|
|
NodeImpl *next;
|
|
NodeImpl *n;
|
|
if (processStart) {
|
|
for (n = processStart; n && n != processEnd; n = next) {
|
|
next = n->nextSibling();
|
|
|
|
if (action == EXTRACT_CONTENTS)
|
|
fragment->appendChild(n,exceptioncode); // will remove from cmnRoot
|
|
else if (action == CLONE_CONTENTS)
|
|
fragment->appendChild(n->cloneNode(true),exceptioncode);
|
|
else
|
|
cmnRoot->removeChild(n,exceptioncode);
|
|
}
|
|
}
|
|
|
|
if ((action == EXTRACT_CONTENTS || action == CLONE_CONTENTS) && rightContents)
|
|
fragment->appendChild(rightContents,exceptioncode);
|
|
|
|
// collapse to the proper position - see spec section 2.6
|
|
if (action == EXTRACT_CONTENTS || action == DELETE_CONTENTS) {
|
|
if (!partialStart && !partialEnd)
|
|
collapse(true,exceptioncode);
|
|
else if (partialStart) {
|
|
setStartContainer(partialStart->parentNode());
|
|
setEndContainer(partialStart->parentNode());
|
|
m_startOffset = m_endOffset = partialStart->nodeIndex()+1;
|
|
}
|
|
else if (partialEnd) {
|
|
setStartContainer(partialEnd->parentNode());
|
|
setEndContainer(partialEnd->parentNode());
|
|
m_startOffset = m_endOffset = partialEnd->nodeIndex();
|
|
}
|
|
}
|
|
return fragment;
|
|
}
|
|
|
|
|
|
DocumentFragmentImpl *RangeImpl::extractContents( int &exceptioncode )
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return 0;
|
|
}
|
|
|
|
checkDeleteExtract(exceptioncode);
|
|
if (exceptioncode)
|
|
return 0;
|
|
|
|
return processContents(EXTRACT_CONTENTS,exceptioncode);
|
|
}
|
|
|
|
DocumentFragmentImpl *RangeImpl::cloneContents( int &exceptioncode )
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return 0;
|
|
}
|
|
|
|
return processContents(CLONE_CONTENTS,exceptioncode);
|
|
}
|
|
|
|
void RangeImpl::insertNode( NodeImpl *newNode, int &exceptioncode )
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return;
|
|
}
|
|
|
|
// NO_MODIFICATION_ALLOWED_ERR: Raised if an ancestor container of either boundary-point of
|
|
// the Range is read-only.
|
|
NodeImpl *n = m_startContainer;
|
|
while (n && !n->isReadOnly())
|
|
n = n->parentNode();
|
|
if (n) {
|
|
exceptioncode = DOMException::NO_MODIFICATION_ALLOWED_ERR;
|
|
return;
|
|
}
|
|
|
|
n = m_endContainer;
|
|
while (n && !n->isReadOnly())
|
|
n = n->parentNode();
|
|
if (n) {
|
|
exceptioncode = DOMException::NO_MODIFICATION_ALLOWED_ERR;
|
|
return;
|
|
}
|
|
|
|
// WRONG_DOCUMENT_ERR: Raised if newParent and the container of the start of the Range were
|
|
// not created from the same document.
|
|
if (newNode->getDocument() != m_startContainer->getDocument()) {
|
|
exceptioncode = DOMException::WRONG_DOCUMENT_ERR;
|
|
return;
|
|
}
|
|
|
|
|
|
// HIERARCHY_REQUEST_ERR: Raised if the container of the start of the Range is of a type that
|
|
// does not allow children of the type of newNode or if newNode is an ancestor of the container.
|
|
|
|
// an extra one here - if a text node is going to split, it must have a parent to insert into
|
|
if (m_startContainer->nodeType() == Node::TEXT_NODE && !m_startContainer->parentNode()) {
|
|
exceptioncode = DOMException::HIERARCHY_REQUEST_ERR;
|
|
return;
|
|
}
|
|
|
|
// In the case where the container is a text node, we check against the container's parent, because
|
|
// text nodes get split up upon insertion.
|
|
NodeImpl *checkAgainst;
|
|
if (m_startContainer->nodeType() == Node::TEXT_NODE)
|
|
checkAgainst = m_startContainer->parentNode();
|
|
else
|
|
checkAgainst = m_startContainer;
|
|
|
|
if (newNode->nodeType() == Node::DOCUMENT_FRAGMENT_NODE) {
|
|
// check each child node, not the DocumentFragment itself
|
|
NodeImpl *c;
|
|
for (c = newNode->firstChild(); c; c = c->nextSibling()) {
|
|
if (!checkAgainst->childTypeAllowed(c->nodeType())) {
|
|
exceptioncode = DOMException::HIERARCHY_REQUEST_ERR;
|
|
return;
|
|
}
|
|
}
|
|
}
|
|
else {
|
|
if (!checkAgainst->childTypeAllowed(newNode->nodeType())) {
|
|
exceptioncode = DOMException::HIERARCHY_REQUEST_ERR;
|
|
return;
|
|
}
|
|
}
|
|
|
|
for (n = m_startContainer; n; n = n->parentNode()) {
|
|
if (n == newNode) {
|
|
exceptioncode = DOMException::HIERARCHY_REQUEST_ERR;
|
|
return;
|
|
}
|
|
}
|
|
|
|
// INVALID_NODE_TYPE_ERR: Raised if newNode is an Attr, Entity, Notation, or Document node.
|
|
if( newNode->nodeType() == Node::ATTRIBUTE_NODE ||
|
|
newNode->nodeType() == Node::ENTITY_NODE ||
|
|
newNode->nodeType() == Node::NOTATION_NODE ||
|
|
newNode->nodeType() == Node::DOCUMENT_NODE) {
|
|
exceptioncode = RangeException::INVALID_NODE_TYPE_ERR + RangeException::_EXCEPTION_OFFSET;
|
|
return;
|
|
}
|
|
|
|
if( m_startContainer->nodeType() == Node::TEXT_NODE ||
|
|
m_startContainer->nodeType() == Node::CDATA_SECTION_NODE )
|
|
{
|
|
TextImpl *newText = static_cast<TextImpl*>(m_startContainer)->splitText(m_startOffset,exceptioncode);
|
|
if (exceptioncode)
|
|
return;
|
|
m_startContainer->parentNode()->insertBefore( newNode, newText, exceptioncode );
|
|
}
|
|
else {
|
|
m_startContainer->insertBefore( newNode, m_startContainer->childNode( m_startOffset ), exceptioncode );
|
|
}
|
|
}
|
|
|
|
DOMString RangeImpl::toString( int &exceptioncode )
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return DOMString();
|
|
}
|
|
|
|
DOMString text = "";
|
|
NodeImpl *n = m_startContainer;
|
|
|
|
/* This function converts a dom range to the plain text string that the user would see in this
|
|
* portion of rendered html.
|
|
*
|
|
* There are several ways ranges can be used.
|
|
*
|
|
* The simplest is the start and endContainer is a text node. The start and end offset is the
|
|
* number of characters into the text to remove/truncate.
|
|
*
|
|
* The next case is the start and endContainer is, well, a container, such a P tag or DIV tag.
|
|
* In this case the start and end offset is the number of children into the container to start
|
|
* from and end at.
|
|
*
|
|
* The other cases are different arrangements of the first two.
|
|
*
|
|
* psuedo code:
|
|
*
|
|
* if start container is not text:
|
|
* count through the children to find where we start (m_startOffset children)
|
|
*
|
|
* loop from the start position:
|
|
* if the current node is text, add the text to our variable 'text', truncating/removing if at the end/start.
|
|
*
|
|
* if the node has children, step to the first child.
|
|
* if the node has no children but does have siblings, step to the next sibling
|
|
* until we find a sibling, go to next the parent but:
|
|
* make sure this sibling isn't past the end of where we are supposed to go. (position > endOffset and the parent is the endContainer)
|
|
*
|
|
*/
|
|
|
|
|
|
if( m_startContainer == m_endContainer && m_startOffset >= m_endOffset)
|
|
return text;
|
|
|
|
|
|
if(n->firstChild()) {
|
|
n = n->firstChild();
|
|
int current_offset = m_startOffset;
|
|
while(current_offset-- && n) {
|
|
n = n->nextSibling();
|
|
}
|
|
}
|
|
|
|
while(n) {
|
|
if(n->nodeType() == DOM::Node::TEXT_NODE ||
|
|
n->nodeType() == DOM::Node::CDATA_SECTION_NODE) {
|
|
|
|
DOMString str;
|
|
str = static_cast<TextImpl *>(n)->string();
|
|
if( n == m_endContainer || n == m_startContainer)
|
|
str = str.copy(); //copy if we are going to modify.
|
|
|
|
if (n == m_endContainer)
|
|
str.truncate(m_endOffset);
|
|
if (n == m_startContainer)
|
|
str.remove(0,m_startOffset);
|
|
text += str;
|
|
if (n == m_endContainer)
|
|
break;
|
|
}
|
|
|
|
|
|
NodeImpl *next = n->firstChild();
|
|
if(!next)
|
|
next = n->nextSibling();
|
|
|
|
while( !next && n->parentNode() ) {
|
|
if (n == m_endContainer) return text;
|
|
n = n->parentNode();
|
|
if (n == m_endContainer) return text;
|
|
next = n->nextSibling();
|
|
}
|
|
|
|
if(n->parentNode() == m_endContainer) {
|
|
if(!next) break;
|
|
unsigned long current_offset = 0;
|
|
NodeImpl *it = n;
|
|
while((it = it->previousSibling())) ++current_offset;
|
|
if(current_offset >= m_endOffset) {
|
|
break;
|
|
}
|
|
}
|
|
|
|
n = next;
|
|
}
|
|
return text;
|
|
}
|
|
|
|
DOMString RangeImpl::toHTML( int &exceptioncode )
|
|
{
|
|
bool hasHtmlTag = false;
|
|
bool hasBodyTag = false;
|
|
//FIXME: What is this section of code below exactly? Do I want it here?
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return DOMString();
|
|
}
|
|
DOMString text = "";
|
|
NodeImpl *n = m_startContainer;
|
|
int num_tables=0;
|
|
bool in_li = false; //whether we have an li in the text, without an ol/ul
|
|
int depth_difference = 0;
|
|
int lowest_depth_difference = 0;
|
|
|
|
if( m_startContainer == m_endContainer && m_startOffset >= m_endOffset)
|
|
return text;
|
|
|
|
while(n) {
|
|
/* First, we could have an tag <tagname key=value>otherstuff</tagname> */
|
|
if(n->nodeType() == DOM::Node::ELEMENT_NODE) {
|
|
int elementId = static_cast<ElementImpl *>(n)->id();
|
|
if(elementId == ID_TABLE) num_tables++;
|
|
if(elementId == ID_BODY) hasBodyTag = true;
|
|
if(elementId == ID_HTML) hasHtmlTag = true;
|
|
if(elementId == ID_LI) in_li=true;
|
|
if(num_tables==0 && ( elementId == ID_TD || elementId == ID_TR || elementId == ID_TH || elementId == ID_TBODY || elementId == ID_TFOOT || elementId == ID_THEAD)) num_tables++;
|
|
if(!( !n->hasChildNodes() && (elementId == ID_H1 || elementId == ID_H2 || elementId == ID_H3 || elementId == ID_H4 || elementId ==ID_H5))) { //Don't add <h1/> etc. Just skip these nodes just to make the output html a bit nicer.
|
|
text += static_cast<ElementImpl *>(n)->openTagStartToString(true /*safely expand img urls*/); // adds "<tagname key=value"
|
|
if(n->hasChildNodes()) {
|
|
depth_difference++;
|
|
text += ">";
|
|
} else {
|
|
text += "/>";
|
|
}
|
|
}
|
|
} else
|
|
if(n->nodeType() == DOM::Node::TEXT_NODE ||
|
|
n->nodeType() == DOM::Node::CDATA_SECTION_NODE) {
|
|
if(n->nodeType() == DOM::Node::CDATA_SECTION_NODE) text += "<![CDATA[ ";
|
|
long long startOffset = (n == m_startContainer)?(long long)m_startOffset:-1;
|
|
long long endOffset = (n == m_endContainer)?(long long) m_endOffset:-1;
|
|
text += static_cast<TextImpl *>(n)->toString(startOffset, endOffset); //Note this should always work since CDataImpl inherits TextImpl
|
|
if(n->nodeType() == DOM::Node::CDATA_SECTION_NODE) text += " ]]>";
|
|
if(n == m_endContainer) {
|
|
break;
|
|
}
|
|
}
|
|
if(n->parentNode() == m_endContainer && !n->nextSibling()) {
|
|
break;
|
|
}
|
|
|
|
//if (n == m_endContainer) break;
|
|
NodeImpl *next = n->firstChild();
|
|
if(next) {
|
|
if(n == m_startContainer) {
|
|
//This is the start of our selection, so we have to move to where we have started selecting.
|
|
//For example, if 'n' is "hello <img src='hello.png'> how are you? <img src='goodbye.png'>"
|
|
//then this has four children. If our selection started on the image, then we need to start from there.
|
|
unsigned long current_offset = 0;
|
|
while(current_offset < m_startOffset && next) {
|
|
next = next->nextSibling();
|
|
++current_offset;
|
|
}
|
|
}
|
|
} else {
|
|
next = n->nextSibling();
|
|
|
|
if(n->parentNode() == m_endContainer) {
|
|
unsigned long current_offset = 1;
|
|
NodeImpl *it = n;
|
|
while((it = it->previousSibling())) ++current_offset;
|
|
|
|
if(current_offset >= m_endOffset) {
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
while( !next && n->parentNode() ) {
|
|
n = n->parentNode();
|
|
if(n->nodeType() == DOM::Node::ELEMENT_NODE) {
|
|
text += "</";
|
|
text += static_cast<ElementImpl *>(n)->tagName();
|
|
int elementId = static_cast<ElementImpl *>(n)->id();
|
|
if(elementId == ID_TABLE) num_tables--;
|
|
depth_difference--;
|
|
if(lowest_depth_difference > depth_difference) lowest_depth_difference=depth_difference;
|
|
if(num_tables==0 && ( elementId == ID_TD || elementId == ID_TR || elementId == ID_TH || elementId == ID_TBODY || elementId == ID_TFOOT || elementId == ID_THEAD)) num_tables--;
|
|
if(elementId == ID_OL || elementId == ID_UL) in_li=false;
|
|
text += ">";
|
|
}
|
|
next = n->nextSibling();
|
|
}
|
|
n = next;
|
|
}
|
|
|
|
//We have the html in the selection. But now we need to properly add the opening and closing tags.
|
|
//For example say we have: "Hello <b>Mr. John</b> How are you?" and we select "John" or even
|
|
//"John</b> How" and copy. We want to return "<b>John</b>" and "<b>John</b> How" respectively
|
|
|
|
//To do this, we need to go up the tree from the start, and prepend those tags.
|
|
//Imagine our selection was this:
|
|
//
|
|
// hello</b></p><p>there
|
|
//
|
|
// The difference in depths between the start and end is -1, and the lowest depth
|
|
// difference from the starting point is -2
|
|
//
|
|
// So from the start of the selection, we want to go down to the lowest_depth_difference
|
|
// and prepend those tags. (<p><b>)
|
|
//
|
|
// From the end of the selection, we want to also go down to the lowest_depth_difference.
|
|
// We know the depth of the end of the selection - i.e. depth_difference.
|
|
//
|
|
//
|
|
n = m_startContainer;
|
|
int startdepth = 0; //by definition - we are counting from zero.
|
|
while((n = n->parentNode()) && startdepth>lowest_depth_difference) {
|
|
if(n->nodeType() == DOM::Node::ELEMENT_NODE) { //This should always be true.. right?
|
|
switch (static_cast<ElementImpl *>(n)->id()) {
|
|
case ID_TABLE:
|
|
num_tables--;
|
|
break;
|
|
case ID_BODY:
|
|
hasBodyTag = true;
|
|
break;
|
|
case ID_HTML:
|
|
hasHtmlTag = true;
|
|
break;
|
|
case ID_LI:
|
|
in_li = true;
|
|
break;
|
|
}
|
|
text = static_cast<ElementImpl *>(n)->openTagStartToString(true /*expand img urls*/)+">" +text; // prepends "<tagname key=value>"
|
|
}
|
|
startdepth--;
|
|
}
|
|
n = m_endContainer;
|
|
while( depth_difference>lowest_depth_difference && (n = n->parentNode())) {
|
|
if(n->nodeType() == DOM::Node::ELEMENT_NODE) { //This should always be true.. right?
|
|
switch (static_cast<ElementImpl *>(n)->id()) {
|
|
case ID_TABLE:
|
|
num_tables++;
|
|
break;
|
|
case ID_OL:
|
|
case ID_UL:
|
|
in_li=false;
|
|
break;
|
|
}
|
|
text += "</";
|
|
text += static_cast<ElementImpl *>(n)->tagName();
|
|
text += ">";
|
|
}
|
|
depth_difference--;
|
|
}
|
|
|
|
// Now our text string is the same depth on both sides, with nothing lower (in other words all the
|
|
// tags in it match up.) This also means that the end value for n in the first loop is a sibling of the
|
|
// end value for n in the second loop.
|
|
//
|
|
// We now need to go down the tree, and for certain tags, add them in on both ends of the text.
|
|
// For example, if have: "<b>hello</b>" and we select "ll", then we want to go down the tree and
|
|
// add "<b>" and "</b>" to it, to produce "<b>ll</b>".
|
|
//
|
|
// I just guessed at which tags you'd want to keep (bold, italic etc) and which you wouldn't (tables etc).
|
|
// It's just wild guessing. feel free to change.
|
|
//
|
|
// Note we can carry on with the value of n
|
|
if(n) {
|
|
while((n = n->parentNode())) {
|
|
if(n->nodeType() == DOM::Node::ELEMENT_NODE) { //This should always be true.. right?
|
|
int elementId = static_cast<ElementImpl *>(n)->id();
|
|
switch (elementId) {
|
|
case ID_TABLE:
|
|
case ID_TD:
|
|
case ID_TR:
|
|
case ID_TH:
|
|
case ID_TBODY:
|
|
case ID_TFOOT:
|
|
case ID_THEAD:
|
|
if(num_tables>0) {
|
|
if(elementId == ID_TABLE) num_tables--;
|
|
text = static_cast<ElementImpl *>(n)->openTagStartToString(true /*expand img urls*/)+">" +text;
|
|
text += "</";
|
|
text += static_cast<ElementImpl *>(n)->tagName();
|
|
text += ">";
|
|
|
|
}
|
|
break;
|
|
|
|
case ID_LI:
|
|
if(!in_li) break;
|
|
text = static_cast<ElementImpl *>(n)->openTagStartToString(true /*expand img urls*/)+">" +text;
|
|
text += "</";
|
|
text += static_cast<ElementImpl *>(n)->tagName();
|
|
text += ">";
|
|
break;
|
|
|
|
case ID_UL:
|
|
case ID_OL:
|
|
if(!in_li) break;
|
|
in_li = false;
|
|
case ID_B:
|
|
case ID_I:
|
|
case ID_U:
|
|
case ID_FONT:
|
|
case ID_S:
|
|
case ID_STRONG:
|
|
case ID_STRIKE:
|
|
case ID_DEL:
|
|
case ID_A:
|
|
case ID_H1:
|
|
case ID_H2:
|
|
case ID_H3:
|
|
case ID_H4:
|
|
case ID_H5:
|
|
//should small, etc be here? so hard to decide. this is such a hack :(
|
|
//There's probably tons of others you'd want here.
|
|
text = static_cast<ElementImpl *>(n)->openTagStartToString(true /*expand img urls*/)+">" +text;
|
|
text += "</";
|
|
text += static_cast<ElementImpl *>(n)->tagName();
|
|
text += ">";
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
if(!hasBodyTag) text = DOMString("<body>") + text + "</body>";
|
|
else if(!hasHtmlTag) {
|
|
text = DOMString("<html xmlns=\"http://www.w3.org/1999/xhtml\">\n"
|
|
"<head>\n"
|
|
"<meta http-equiv=\"Content-Type\" content=\"text/html; charset=UTF-8\" />\n"
|
|
"<meta name=\"Generator\" content=\"TDEHTML, the TDE Web Page Viewer\" />\n"
|
|
"</head>\n") +
|
|
text +
|
|
"</html>";
|
|
}
|
|
text = DOMString("<!DOCTYPE html PUBLIC \"-//W3C//DTD XHTML 1.0 Strict//EN\" \"DTD/xhtml1-strict.dtd\">\n") + text;
|
|
|
|
return text;
|
|
|
|
}
|
|
|
|
DocumentFragment RangeImpl::createContextualFragment ( const DOMString &html, int &exceptioncode )
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return DocumentFragment();
|
|
}
|
|
|
|
if (! m_startContainer->isHTMLElement()) {
|
|
exceptioncode = DOMException::NOT_SUPPORTED_ERR;
|
|
return DocumentFragment();
|
|
}
|
|
|
|
HTMLElementImpl *e = static_cast<HTMLElementImpl *>(m_startContainer);
|
|
DocumentFragment fragment = e->createContextualFragment(html);
|
|
if (fragment.isNull()) {
|
|
exceptioncode = DOMException::NOT_SUPPORTED_ERR;
|
|
return DocumentFragment();
|
|
}
|
|
|
|
return fragment;
|
|
}
|
|
|
|
|
|
void RangeImpl::detach( int &exceptioncode )
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return;
|
|
}
|
|
|
|
if (m_startContainer)
|
|
m_startContainer->deref();
|
|
m_startContainer = 0;
|
|
if (m_endContainer)
|
|
m_endContainer->deref();
|
|
m_endContainer = 0;
|
|
m_detached = true;
|
|
}
|
|
|
|
bool RangeImpl::isDetached() const
|
|
{
|
|
return m_detached;
|
|
}
|
|
|
|
void RangeImpl::checkNodeWOffset( NodeImpl *n, int offset, int &exceptioncode) const
|
|
{
|
|
if( offset < 0 ) {
|
|
exceptioncode = DOMException::INDEX_SIZE_ERR;
|
|
}
|
|
|
|
switch (n->nodeType()) {
|
|
case Node::ENTITY_NODE:
|
|
case Node::NOTATION_NODE:
|
|
case Node::DOCUMENT_TYPE_NODE:
|
|
exceptioncode = RangeException::INVALID_NODE_TYPE_ERR + RangeException::_EXCEPTION_OFFSET;
|
|
break;
|
|
case Node::TEXT_NODE:
|
|
case Node::COMMENT_NODE:
|
|
case Node::CDATA_SECTION_NODE:
|
|
if ( (unsigned long)offset > static_cast<CharacterDataImpl*>(n)->length() )
|
|
exceptioncode = DOMException::INDEX_SIZE_ERR;
|
|
break;
|
|
case Node::PROCESSING_INSTRUCTION_NODE:
|
|
// ### are we supposed to check with just data or the whole contents?
|
|
if ( (unsigned long)offset > static_cast<ProcessingInstructionImpl*>(n)->data().length() )
|
|
exceptioncode = DOMException::INDEX_SIZE_ERR;
|
|
break;
|
|
default:
|
|
if ( (unsigned long)offset > n->childNodeCount() )
|
|
exceptioncode = DOMException::INDEX_SIZE_ERR;
|
|
break;
|
|
}
|
|
}
|
|
|
|
void RangeImpl::checkNodeBA( NodeImpl *n, int &exceptioncode ) const
|
|
{
|
|
// INVALID_NODE_TYPE_ERR: Raised if the root container of refNode is not an
|
|
// Attr, Document or DocumentFragment node or if refNode is a Document,
|
|
// DocumentFragment, Attr, Entity, or Notation node.
|
|
NodeImpl *root = n;
|
|
while (root->parentNode())
|
|
root = root->parentNode();
|
|
if (!(root->nodeType() == Node::ATTRIBUTE_NODE ||
|
|
root->nodeType() == Node::DOCUMENT_NODE ||
|
|
root->nodeType() == Node::DOCUMENT_FRAGMENT_NODE)) {
|
|
exceptioncode = RangeException::INVALID_NODE_TYPE_ERR + RangeException::_EXCEPTION_OFFSET;
|
|
return;
|
|
}
|
|
|
|
if( n->nodeType() == Node::DOCUMENT_NODE ||
|
|
n->nodeType() == Node::DOCUMENT_FRAGMENT_NODE ||
|
|
n->nodeType() == Node::ATTRIBUTE_NODE ||
|
|
n->nodeType() == Node::ENTITY_NODE ||
|
|
n->nodeType() == Node::NOTATION_NODE )
|
|
exceptioncode = RangeException::INVALID_NODE_TYPE_ERR + RangeException::_EXCEPTION_OFFSET;
|
|
|
|
}
|
|
|
|
RangeImpl *RangeImpl::cloneRange(int &exceptioncode)
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return 0;
|
|
}
|
|
|
|
return new RangeImpl(m_ownerDocument,m_startContainer,m_startOffset,m_endContainer,m_endOffset);
|
|
}
|
|
|
|
void RangeImpl::setStartAfter( NodeImpl *refNode, int &exceptioncode )
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return;
|
|
}
|
|
|
|
if (!refNode) {
|
|
exceptioncode = DOMException::NOT_FOUND_ERR;
|
|
return;
|
|
}
|
|
|
|
if (refNode->getDocument() != m_ownerDocument) {
|
|
exceptioncode = DOMException::WRONG_DOCUMENT_ERR;
|
|
return;
|
|
}
|
|
|
|
checkNodeBA( refNode, exceptioncode );
|
|
if (exceptioncode)
|
|
return;
|
|
|
|
setStart( refNode->parentNode(), refNode->nodeIndex()+1, exceptioncode );
|
|
}
|
|
|
|
void RangeImpl::setEndBefore( NodeImpl *refNode, int &exceptioncode )
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return;
|
|
}
|
|
|
|
if (!refNode) {
|
|
exceptioncode = DOMException::NOT_FOUND_ERR;
|
|
return;
|
|
}
|
|
|
|
if (refNode->getDocument() != m_ownerDocument) {
|
|
exceptioncode = DOMException::WRONG_DOCUMENT_ERR;
|
|
return;
|
|
}
|
|
|
|
checkNodeBA( refNode, exceptioncode );
|
|
if (exceptioncode)
|
|
return;
|
|
|
|
setEnd( refNode->parentNode(), refNode->nodeIndex(), exceptioncode );
|
|
}
|
|
|
|
void RangeImpl::setEndAfter( NodeImpl *refNode, int &exceptioncode )
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return;
|
|
}
|
|
|
|
if (!refNode) {
|
|
exceptioncode = DOMException::NOT_FOUND_ERR;
|
|
return;
|
|
}
|
|
|
|
if (refNode->getDocument() != m_ownerDocument) {
|
|
exceptioncode = DOMException::WRONG_DOCUMENT_ERR;
|
|
return;
|
|
}
|
|
|
|
checkNodeBA( refNode, exceptioncode );
|
|
if (exceptioncode)
|
|
return;
|
|
|
|
setEnd( refNode->parentNode(), refNode->nodeIndex()+1, exceptioncode );
|
|
|
|
}
|
|
|
|
void RangeImpl::selectNode( NodeImpl *refNode, int &exceptioncode )
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return;
|
|
}
|
|
|
|
if (!refNode) {
|
|
exceptioncode = DOMException::NOT_FOUND_ERR;
|
|
return;
|
|
}
|
|
|
|
// INVALID_NODE_TYPE_ERR: Raised if an ancestor of refNode is an Entity, Notation or
|
|
// DocumentType node or if refNode is a Document, DocumentFragment, Attr, Entity, or Notation
|
|
// node.
|
|
NodeImpl *anc;
|
|
for (anc = refNode->parentNode(); anc; anc = anc->parentNode()) {
|
|
if (anc->nodeType() == Node::ENTITY_NODE ||
|
|
anc->nodeType() == Node::NOTATION_NODE ||
|
|
anc->nodeType() == Node::DOCUMENT_TYPE_NODE) {
|
|
|
|
exceptioncode = RangeException::INVALID_NODE_TYPE_ERR + RangeException::_EXCEPTION_OFFSET;
|
|
return;
|
|
}
|
|
}
|
|
|
|
if (refNode->nodeType() == Node::DOCUMENT_NODE ||
|
|
refNode->nodeType() == Node::DOCUMENT_FRAGMENT_NODE ||
|
|
refNode->nodeType() == Node::ATTRIBUTE_NODE ||
|
|
refNode->nodeType() == Node::ENTITY_NODE ||
|
|
refNode->nodeType() == Node::NOTATION_NODE) {
|
|
|
|
exceptioncode = RangeException::INVALID_NODE_TYPE_ERR + RangeException::_EXCEPTION_OFFSET;
|
|
return;
|
|
}
|
|
|
|
setStartBefore( refNode, exceptioncode );
|
|
if (exceptioncode)
|
|
return;
|
|
setEndAfter( refNode, exceptioncode );
|
|
}
|
|
|
|
void RangeImpl::selectNodeContents( NodeImpl *refNode, int &exceptioncode )
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return;
|
|
}
|
|
|
|
if (!refNode) {
|
|
exceptioncode = DOMException::NOT_FOUND_ERR;
|
|
return;
|
|
}
|
|
|
|
// INVALID_NODE_TYPE_ERR: Raised if refNode or an ancestor of refNode is an Entity, Notation
|
|
// or DocumentType node.
|
|
NodeImpl *n;
|
|
for (n = refNode; n; n = n->parentNode()) {
|
|
if (n->nodeType() == Node::ENTITY_NODE ||
|
|
n->nodeType() == Node::NOTATION_NODE ||
|
|
n->nodeType() == Node::DOCUMENT_TYPE_NODE) {
|
|
|
|
exceptioncode = RangeException::INVALID_NODE_TYPE_ERR + RangeException::_EXCEPTION_OFFSET;
|
|
return;
|
|
}
|
|
}
|
|
|
|
setStartContainer(refNode);
|
|
m_startOffset = 0;
|
|
setEndContainer(refNode);
|
|
m_endOffset = refNode->childNodeCount();
|
|
}
|
|
|
|
void RangeImpl::surroundContents( NodeImpl *newParent, int &exceptioncode )
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return;
|
|
}
|
|
|
|
if( !newParent ) {
|
|
exceptioncode = DOMException::NOT_FOUND_ERR;
|
|
return;
|
|
}
|
|
|
|
// INVALID_NODE_TYPE_ERR: Raised if node is an Attr, Entity, DocumentType, Notation,
|
|
// Document, or DocumentFragment node.
|
|
if( newParent->nodeType() == Node::ATTRIBUTE_NODE ||
|
|
newParent->nodeType() == Node::ENTITY_NODE ||
|
|
newParent->nodeType() == Node::NOTATION_NODE ||
|
|
newParent->nodeType() == Node::DOCUMENT_TYPE_NODE ||
|
|
newParent->nodeType() == Node::DOCUMENT_NODE ||
|
|
newParent->nodeType() == Node::DOCUMENT_FRAGMENT_NODE) {
|
|
exceptioncode = RangeException::INVALID_NODE_TYPE_ERR + RangeException::_EXCEPTION_OFFSET;
|
|
return;
|
|
}
|
|
|
|
// NO_MODIFICATION_ALLOWED_ERR: Raised if an ancestor container of either boundary-point of
|
|
// the Range is read-only.
|
|
if (readOnly()) {
|
|
exceptioncode = DOMException::NO_MODIFICATION_ALLOWED_ERR;
|
|
return;
|
|
}
|
|
|
|
NodeImpl *n = m_startContainer;
|
|
while (n && !n->isReadOnly())
|
|
n = n->parentNode();
|
|
if (n) {
|
|
exceptioncode = DOMException::NO_MODIFICATION_ALLOWED_ERR;
|
|
return;
|
|
}
|
|
|
|
n = m_endContainer;
|
|
while (n && !n->isReadOnly())
|
|
n = n->parentNode();
|
|
if (n) {
|
|
exceptioncode = DOMException::NO_MODIFICATION_ALLOWED_ERR;
|
|
return;
|
|
}
|
|
|
|
// WRONG_DOCUMENT_ERR: Raised if newParent and the container of the start of the Range were
|
|
// not created from the same document.
|
|
if (newParent->getDocument() != m_startContainer->getDocument()) {
|
|
exceptioncode = DOMException::WRONG_DOCUMENT_ERR;
|
|
return;
|
|
}
|
|
|
|
// HIERARCHY_REQUEST_ERR: Raised if the container of the start of the Range is of a type that
|
|
// does not allow children of the type of newParent or if newParent is an ancestor of the container
|
|
// or if node would end up with a child node of a type not allowed by the type of node.
|
|
if (!m_startContainer->childTypeAllowed(newParent->nodeType())) {
|
|
exceptioncode = DOMException::HIERARCHY_REQUEST_ERR;
|
|
return;
|
|
}
|
|
|
|
for (n = m_startContainer; n; n = n->parentNode()) {
|
|
if (n == newParent) {
|
|
exceptioncode = DOMException::HIERARCHY_REQUEST_ERR;
|
|
return;
|
|
}
|
|
}
|
|
|
|
// ### check if node would end up with a child node of a type not allowed by the type of node
|
|
|
|
// BAD_BOUNDARYPOINTS_ERR: Raised if the Range partially selects a non-text node.
|
|
if (m_startContainer->nodeType() != Node::TEXT_NODE &&
|
|
m_startContainer->nodeType() != Node::COMMENT_NODE &&
|
|
m_startContainer->nodeType() != Node::CDATA_SECTION_NODE &&
|
|
m_startContainer->nodeType() != Node::PROCESSING_INSTRUCTION_NODE) {
|
|
|
|
if (m_startOffset > 0 && m_startOffset < m_startContainer->childNodeCount()) {
|
|
exceptioncode = RangeException::BAD_BOUNDARYPOINTS_ERR + RangeException::_EXCEPTION_OFFSET;
|
|
return;
|
|
}
|
|
}
|
|
|
|
if (m_endContainer->nodeType() != Node::TEXT_NODE &&
|
|
m_endContainer->nodeType() != Node::COMMENT_NODE &&
|
|
m_endContainer->nodeType() != Node::CDATA_SECTION_NODE &&
|
|
m_endContainer->nodeType() != Node::PROCESSING_INSTRUCTION_NODE) {
|
|
|
|
if (m_endOffset > 0 && m_endOffset < m_endContainer->childNodeCount()) {
|
|
exceptioncode = RangeException::BAD_BOUNDARYPOINTS_ERR + RangeException::_EXCEPTION_OFFSET;
|
|
return;
|
|
}
|
|
}
|
|
|
|
while (newParent->firstChild()) {
|
|
newParent->removeChild(newParent->firstChild(),exceptioncode);
|
|
if (exceptioncode)
|
|
return;
|
|
}
|
|
DocumentFragmentImpl *fragment = extractContents(exceptioncode);
|
|
if (exceptioncode)
|
|
return;
|
|
insertNode( newParent, exceptioncode );
|
|
if (exceptioncode)
|
|
return;
|
|
newParent->appendChild( fragment, exceptioncode );
|
|
if (exceptioncode)
|
|
return;
|
|
selectNode( newParent, exceptioncode );
|
|
}
|
|
|
|
void RangeImpl::setStartBefore( NodeImpl *refNode, int &exceptioncode )
|
|
{
|
|
if (m_detached) {
|
|
exceptioncode = DOMException::INVALID_STATE_ERR;
|
|
return;
|
|
}
|
|
|
|
if (!refNode) {
|
|
exceptioncode = DOMException::NOT_FOUND_ERR;
|
|
return;
|
|
}
|
|
|
|
if (refNode->getDocument() != m_ownerDocument) {
|
|
exceptioncode = DOMException::WRONG_DOCUMENT_ERR;
|
|
return;
|
|
}
|
|
|
|
checkNodeBA( refNode, exceptioncode );
|
|
if (exceptioncode)
|
|
return;
|
|
|
|
setStart( refNode->parentNode(), refNode->nodeIndex(), exceptioncode );
|
|
}
|
|
|
|
void RangeImpl::setStartContainer(NodeImpl *_startContainer)
|
|
{
|
|
if (m_startContainer == _startContainer)
|
|
return;
|
|
|
|
if (m_startContainer)
|
|
m_startContainer->deref();
|
|
m_startContainer = _startContainer;
|
|
if (m_startContainer)
|
|
m_startContainer->ref();
|
|
}
|
|
|
|
void RangeImpl::setEndContainer(NodeImpl *_endContainer)
|
|
{
|
|
if (m_endContainer == _endContainer)
|
|
return;
|
|
|
|
if (m_endContainer)
|
|
m_endContainer->deref();
|
|
m_endContainer = _endContainer;
|
|
if (m_endContainer)
|
|
m_endContainer->ref();
|
|
}
|
|
|
|
void RangeImpl::checkDeleteExtract(int &exceptioncode) {
|
|
|
|
NodeImpl *start;
|
|
if (m_startContainer->nodeType() != Node::TEXT_NODE &&
|
|
m_startContainer->nodeType() != Node::CDATA_SECTION_NODE &&
|
|
m_startContainer->nodeType() != Node::COMMENT_NODE &&
|
|
m_startContainer->nodeType() != Node::PROCESSING_INSTRUCTION_NODE) {
|
|
|
|
start = m_startContainer->childNode(m_startOffset);
|
|
if (!start) {
|
|
if (m_startContainer->lastChild())
|
|
start = m_startContainer->lastChild()->traverseNextNode();
|
|
else
|
|
start = m_startContainer->traverseNextNode();
|
|
}
|
|
}
|
|
else
|
|
start = m_startContainer;
|
|
|
|
NodeImpl *end;
|
|
if (m_endContainer->nodeType() != Node::TEXT_NODE &&
|
|
m_endContainer->nodeType() != Node::CDATA_SECTION_NODE &&
|
|
m_endContainer->nodeType() != Node::COMMENT_NODE &&
|
|
m_endContainer->nodeType() != Node::PROCESSING_INSTRUCTION_NODE) {
|
|
|
|
end = m_endContainer->childNode(m_endOffset);
|
|
if (!end) {
|
|
if (m_endContainer->lastChild())
|
|
end = m_endContainer->lastChild()->traverseNextNode();
|
|
else
|
|
end = m_endContainer->traverseNextNode();
|
|
}
|
|
}
|
|
else
|
|
end = m_endContainer;
|
|
|
|
NodeImpl *n;
|
|
for (n = start; n != end; n = n->traverseNextNode()) {
|
|
if (n->isReadOnly()) {
|
|
exceptioncode = DOMException::NO_MODIFICATION_ALLOWED_ERR;
|
|
return;
|
|
}
|
|
if (n->nodeType() == Node::DOCUMENT_TYPE_NODE) { // ### is this for only directly under the DF, or anywhere?
|
|
exceptioncode = DOMException::HIERARCHY_REQUEST_ERR;
|
|
return;
|
|
}
|
|
}
|
|
|
|
if (containedByReadOnly()) {
|
|
exceptioncode = DOMException::NO_MODIFICATION_ALLOWED_ERR;
|
|
return;
|
|
}
|
|
}
|
|
|
|
|
|
bool RangeImpl::containedByReadOnly() {
|
|
NodeImpl *n;
|
|
for (n = m_startContainer; n; n = n->parentNode()) {
|
|
if (n->isReadOnly())
|
|
return true;
|
|
}
|
|
for (n = m_endContainer; n; n = n->parentNode()) {
|
|
if (n->isReadOnly())
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|