textiterator.cpp

linux下开源浏览器WebKit的源码,市面上的很多商用浏览器都是移植自WebKit

        return false;
    }

    m_haveEmitted = true;

    if (m_emitCharactersBetweenAllVisiblePositions) {
        // We want replaced elements to behave like punctuation for boundary 
        // finding, and to simply take up space for the selection preservation 
        // code in moveParagraphs, so we use a comma.
        emitCharacter(',', m_node->parentNode(), m_node, 0, 1);
        return true;
    }

    m_positionNode = m_node->parentNode();
    m_positionOffsetBaseNode = m_node;
    m_positionStartOffset = 0;
    m_positionEndOffset = 1;

    m_textCharacters = 0;
    m_textLength = 0;

    m_lastCharacter = 0;

    return true;
}

static bool shouldEmitTabBeforeNode(Node* node)
{
    RenderObject* r = node->renderer();
    
    // Table cells are delimited by tabs.
    if (!r || !isTableCell(node))
        return false;
    
    // Want a tab before every cell other than the first one
    RenderTableCell* rc = static_cast<RenderTableCell*>(r);
    RenderTable* t = rc->table();
    return t && (t->cellBefore(rc) || t->cellAbove(rc));
}

static bool shouldEmitNewlineForNode(Node* node)
{
    // br elements are represented by a single newline.
    RenderObject* r = node->renderer();
    if (!r)
        return node->hasTagName(brTag);
        
    return r->isBR();
}

static bool shouldEmitNewlinesBeforeAndAfterNode(Node* node)
{
    // Block flow (versus inline flow) is represented by having
    // a newline both before and after the element.
    RenderObject* r = node->renderer();
    if (!r) {
        return (node->hasTagName(blockquoteTag)
                || node->hasTagName(ddTag)
                || node->hasTagName(divTag)
                || node->hasTagName(dlTag)
                || node->hasTagName(dtTag)
                || node->hasTagName(h1Tag)
                || node->hasTagName(h2Tag)
                || node->hasTagName(h3Tag)
                || node->hasTagName(h4Tag)
                || node->hasTagName(h5Tag)
                || node->hasTagName(h6Tag)
                || node->hasTagName(hrTag)
                || node->hasTagName(liTag)
                || node->hasTagName(listingTag)
                || node->hasTagName(olTag)
                || node->hasTagName(pTag)
                || node->hasTagName(preTag)
                || node->hasTagName(trTag)
                || node->hasTagName(ulTag));
    }
    
    // Need to make an exception for table cells, because they are blocks, but we
    // want them tab-delimited rather than having newlines before and after.
    if (isTableCell(node))
        return false;
    
    // Need to make an exception for table row elements, because they are neither
    // "inline" or "RenderBlock", but we want newlines for them.
    if (r->isTableRow()) {
        RenderTable* t = static_cast<RenderTableRow*>(r)->table();
        if (t && !t->isInline())
            return true;
    }
    
    return !r->isInline() && r->isRenderBlock() && !r->isFloatingOrPositioned() && !r->isBody();
}

static bool shouldEmitNewlineAfterNode(Node* node)
{
    // FIXME: It should be better but slower to create a VisiblePosition here.
    if (!shouldEmitNewlinesBeforeAndAfterNode(node))
        return false;
    // Check if this is the very last renderer in the document.
    // If so, then we should not emit a newline.
    while ((node = node->traverseNextSibling()))
        if (node->renderer())
            return true;
    return false;
}

static bool shouldEmitNewlineBeforeNode(Node* node)
{
    return shouldEmitNewlinesBeforeAndAfterNode(node); 
}

static bool shouldEmitExtraNewlineForNode(Node* node)
{
    // When there is a significant collapsed bottom margin, emit an extra
    // newline for a more realistic result.  We end up getting the right
    // result even without margin collapsing. For example: <div><p>text</p></div>
    // will work right even if both the <div> and the <p> have bottom margins.
    RenderObject* r = node->renderer();
    if (!r || !r->isBox())
        return false;
    
    // NOTE: We only do this for a select set of nodes, and fwiw WinIE appears
    // not to do this at all
    if (node->hasTagName(h1Tag)
        || node->hasTagName(h2Tag)
        || node->hasTagName(h3Tag)
        || node->hasTagName(h4Tag)
        || node->hasTagName(h5Tag)
        || node->hasTagName(h6Tag)
        || node->hasTagName(pTag)) {
        RenderStyle* style = r->style();
        if (style) {
            int bottomMargin = toRenderBox(r)->collapsedMarginBottom();
            int fontSize = style->fontDescription().computedPixelSize();
            if (bottomMargin * 2 >= fontSize)
                return true;
        }
    }
    
    return false;
}

// Whether or not we should emit a character as we enter m_node (if it's a container) or as we hit it (if it's atomic).
bool TextIterator::shouldRepresentNodeOffsetZero()
{
    if (m_emitCharactersBetweenAllVisiblePositions && m_node->renderer() && m_node->renderer()->isTable())
        return true;
        
    // Leave element positioned flush with start of a paragraph
    // (e.g. do not insert tab before a table cell at the start of a paragraph)
    if (m_lastCharacter == '\n')
        return false;
    
    // Otherwise, show the position if we have emitted any characters
    if (m_haveEmitted)
        return true;
    
    // We've not emitted anything yet. Generally, there is no need for any positioning then.
    // The only exception is when the element is visually not in the same line as
    // the start of the range (e.g. the range starts at the end of the previous paragraph).
    // NOTE: Creating VisiblePositions and comparing them is relatively expensive, so we
    // make quicker checks to possibly avoid that. Another check that we could make is
    // is whether the inline vs block flow changed since the previous visible element.
    // I think we're already in a special enough case that that won't be needed, tho.

    // No character needed if this is the first node in the range.
    if (m_node == m_startContainer)
        return false;
    
    // If we are outside the start container's subtree, assume we need to emit.
    // FIXME: m_startContainer could be an inline block
    if (!m_node->isDescendantOf(m_startContainer))
        return true;

    // If we started as m_startContainer offset 0 and the current node is a descendant of
    // the start container, we already had enough context to correctly decide whether to
    // emit after a preceding block. We chose not to emit (m_haveEmitted is false),
    // so don't second guess that now.
    // NOTE: Is this really correct when m_node is not a leftmost descendant? Probably
    // immaterial since we likely would have already emitted something by now.
    if (m_startOffset == 0)
        return false;
        
    // If this node is unrendered or invisible the VisiblePosition checks below won't have much meaning.
    // Additionally, if the range we are iterating over contains huge sections of unrendered content, 
    // we would create VisiblePositions on every call to this function without this check.
    if (!m_node->renderer() || m_node->renderer()->style()->visibility() != VISIBLE)
        return false;
    
    // The currPos.isNotNull() check is needed because positions in non-html content
    // (like svg) do not have visible positions, and we don't want to emit for them either.
    VisiblePosition startPos = VisiblePosition(m_startContainer, m_startOffset, DOWNSTREAM);
    VisiblePosition currPos = VisiblePosition(m_node, 0, DOWNSTREAM);
    return currPos.isNotNull() && !inSameLine(startPos, currPos);
}

bool TextIterator::shouldEmitSpaceBeforeAndAfterNode(Node* node)
{
    return node->renderer() && node->renderer()->isTable() && (node->renderer()->isInline() || m_emitCharactersBetweenAllVisiblePositions);
}

void TextIterator::representNodeOffsetZero()
{
    // Emit a character to show the positioning of m_node.
    
    // When we haven't been emitting any characters, shouldRepresentNodeOffsetZero() can 
    // create VisiblePositions, which is expensive.  So, we perform the inexpensive checks
    // on m_node to see if it necessitates emitting a character first and will early return 
    // before encountering shouldRepresentNodeOffsetZero()s worse case behavior.
    if (shouldEmitTabBeforeNode(m_node)) {
        if (shouldRepresentNodeOffsetZero())
            emitCharacter('\t', m_node->parentNode(), m_node, 0, 0);
    } else if (shouldEmitNewlineBeforeNode(m_node)) {
        if (shouldRepresentNodeOffsetZero())
            emitCharacter('\n', m_node->parentNode(), m_node, 0, 0);
    } else if (shouldEmitSpaceBeforeAndAfterNode(m_node)) {
        if (shouldRepresentNodeOffsetZero())
            emitCharacter(' ', m_node->parentNode(), m_node, 0, 0);
    }
}

bool TextIterator::handleNonTextNode()
{
    if (shouldEmitNewlineForNode(m_node))
        emitCharacter('\n', m_node->parentNode(), m_node, 0, 1);
    else if (m_emitCharactersBetweenAllVisiblePositions && m_node->renderer() && m_node->renderer()->isHR())
        emitCharacter(' ', m_node->parentNode(), m_node, 0, 1);
    else
        representNodeOffsetZero();

    return true;
}

void TextIterator::exitNode()
{
    // prevent emitting a newline when exiting a collapsed block at beginning of the range
    // FIXME: !m_haveEmitted does not necessarily mean there was a collapsed block... it could
    // have been an hr (e.g.). Also, a collapsed block could have height (e.g. a table) and
    // therefore look like a blank line.
    if (!m_haveEmitted)
        return;
        
    // Emit with a position *inside* m_node, after m_node's contents, in 
    // case it is a block, because the run should start where the 
    // emitted character is positioned visually.
    Node* baseNode = m_node->lastChild() ? m_node->lastChild() : m_node;
    // FIXME: This shouldn't require the m_lastTextNode to be true, but we can't change that without making
    // the logic in _web_attributedStringFromRange match.  We'll get that for free when we switch to use
    // TextIterator in _web_attributedStringFromRange.
    // See <rdar://problem/5428427> for an example of how this mismatch will cause problems.
    if (m_lastTextNode && shouldEmitNewlineAfterNode(m_node)) {
        // use extra newline to represent margin bottom, as needed
        bool addNewline = shouldEmitExtraNewlineForNode(m_node);
        
        // FIXME: We need to emit a '\n' as we leave an empty block(s) that
        // contain a VisiblePosition when doing selection preservation.
        if (m_lastCharacter != '\n') {
            // insert a newline with a position following this block's contents.
            emitCharacter('\n', baseNode->parentNode(), baseNode, 1, 1);
            // remember whether to later add a newline for the current node
            ASSERT(!m_needAnotherNewline);
            m_needAnotherNewline = addNewline;
        } else if (addNewline)
            // insert a newline with a position following this block's contents.
            emitCharacter('\n', baseNode->parentNode(), baseNode, 1, 1);
    }
    
    // If nothing was emitted, see if we need to emit a space.
    if (!m_positionNode && shouldEmitSpaceBeforeAndAfterNode(m_node))
        emitCharacter(' ', baseNode->parentNode(), baseNode, 1, 1);
}

void TextIterator::emitCharacter(UChar c, Node *textNode, Node *offsetBaseNode, int textStartOffset, int textEndOffset)
{
    m_haveEmitted = true;
    
    // remember information with which to construct the TextIterator::range()
    // NOTE: textNode is often not a text node, so the range will specify child nodes of positionNode
    m_positionNode = textNode;
    m_positionOffsetBaseNode = offsetBaseNode;
    m_positionStartOffset = textStartOffset;
    m_positionEndOffset = textEndOffset;
 
    // remember information with which to construct the TextIterator::characters() and length()
    m_singleCharacterBuffer = c;
    m_textCharacters = &m_singleCharacterBuffer;
    m_textLength = 1;

    // remember some iteration state
    m_lastTextNodeEndedWithCollapsedSpace = false;
    m_lastCharacter = c;
}

void TextIterator::emitText(Node* textNode, int textStartOffset, int textEndOffset)
{
    RenderText* renderer = toRenderText(m_node->renderer());
    String str = renderer->text();
    ASSERT(str.characters());

    m_positionNode = textNode;
    m_positionOffsetBaseNode = 0;
    m_positionStartOffset = textStartOffset;
    m_positionEndOffset = textEndOffset;
    m_textCharacters = str.characters() + textStartOffset;
    m_textLength = textEndOffset - textStartOffset;
    m_lastCharacter = str[textEndOffset - 1];

    m_lastTextNodeEndedWithCollapsedSpace = false;
    m_haveEmitted = true;
}

PassRefPtr<Range> TextIterator::range() const
{
    // use the current run information, if we have it
    if (m_positionNode) {
        if (m_positionOffsetBaseNode) {
            int index = m_positionOffsetBaseNode->nodeIndex();
            m_positionStartOffset += index;
            m_positionEndOffset += index;
            m_positionOffsetBaseNode = 0;
        }
        return Range::create(m_positionNode->document(), m_positionNode, m_positionStartOffset, m_positionNode, m_positionEndOffset);
    }

    // otherwise, return the end of the overall range we were given
    if (m_endContainer)
        return Range::create(m_endContainer->document(), m_endContainer, m_endOffset, m_endContainer, m_endOffset);
        
    return 0;
}
    
Node* TextIterator::node() const
{
    RefPtr<Range> textRange = range();
    if (!textRange)
        return 0;

    Node* node = textRange->startContainer();
    if (!node)
        return 0;
    if (node->offsetInCharacters())
        return node;
    
    return node->childNode(textRange->startOffset());
}

// --------

SimplifiedBackwardsTextIterator::SimplifiedBackwardsTextIterator() : m_positionNode(0)
{
}

SimplifiedBackwardsTextIterator::SimplifiedBackwardsTextIterator(const Range *r)
{
    m_positionNode = 0;

    if (!r)
        return;

    Node* startNode = r->startContainer();
    if (!startNode)
        return;
    Node* endNode = r->endContainer();
    int startOffset = r->startOffset();
    int endOffset = r->endOffset();

    if (!startNode->offsetInCharacters()) {
        if (startOffset >= 0 && startOffset < static_cast<int>(startNode->childNodeCount())) {
            startNode = startNode->childNode(startOffset);
            startOffset = 0;
        }
    }
    if (!endNode->offsetInCharacters()) {
        if (endOffset > 0 && endOffset <= static_cast<int>(endNode->childNodeCount())) {
            endNode = endNode->childNode(endOffset - 1);
            endOffset = endNode->offsetInCharacters() ? endNode->maxCharacterOffset() : endNode->childNodeCount();
        }
    }

    m_node = endNode;
    m_offset = endOffset;
    m_handledNode = false;
    m_handledChildren = endOffset == 0;

    m_startNode = startNode;
    m_startOffset = startOffset;
    m_endNode = endNode;
    m_endOffset = endOffset;
    
#ifndef NDEBUG
    // Need this just because of the assert.
    m_positionNode = endNode;
#endif

    m_lastTextNode = 0;
    m_lastCharacter = '\n';
    
    if (startOffset == 0 || !startNode->firstChild()) {
        m_pastStartNode = startNode->previousSibling();
        while (!m_pastStartNode && startNode->parentNode()) {
            startNode = startNode->parentNode();
            m_pastStartNode = startNode->previousSibling();
        }
    } else
        m_pastStartNode = startNode->childNode(startOffset - 1);

    advance();
}

void SimplifiedBackwardsTextIterator::advance()
{
    ASSERT(m_positionNode);

    m_positionNode = 0;
    m_textLength = 0;

    while (m_node && m_node != m_pastStartNode) {
        // Don't handle node if we start iterating at [node, 0].
        if (!m_handledNode && !(m_node == m_endNode && m_endOffset == 0)) {
            RenderObject *renderer = m_node->renderer();
            if (renderer && renderer->isText() && m_node->nodeType() == Node::TEXT_NODE) {
                // FIXME: What about CDATA_SECTION_NODE?