htmlparser.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

    if (node->isHTMLElement()) {
        HTMLElement* e = static_cast<HTMLElement*>(node);
        if (e->hasLocalName(aTag) || e->hasLocalName(fontTag) || e->hasLocalName(ttTag) ||
            e->hasLocalName(uTag) || e->hasLocalName(bTag) || e->hasLocalName(iTag) ||
            e->hasLocalName(sTag) || e->hasLocalName(strikeTag) || e->hasLocalName(bigTag) ||
            e->hasLocalName(smallTag) || e->hasLocalName(emTag) || e->hasLocalName(strongTag) ||
            e->hasLocalName(dfnTag) || e->hasLocalName(codeTag) || e->hasLocalName(sampTag) ||
            e->hasLocalName(kbdTag) || e->hasLocalName(varTag) || e->hasLocalName(citeTag) ||
            e->hasLocalName(abbrTag) || e->hasLocalName(acronymTag) || e->hasLocalName(subTag) ||
            e->hasLocalName(supTag) || e->hasLocalName(spanTag) || e->hasLocalName(nobrTag) ||
            e->hasLocalName(noframesTag) || e->hasLocalName(nolayerTag) ||
            e->hasLocalName(noembedTag))
            return true;
        if (e->hasLocalName(noscriptTag) && !m_isParsingFragment) {
            Settings* settings = m_document->settings();
            if (settings && settings->isJavaScriptEnabled())
                return true;
        }
    }
    
    return false;
}

bool HTMLParser::isResidualStyleTag(const AtomicString& tagName)
{
    DEFINE_STATIC_LOCAL(HashSet<AtomicStringImpl*>, residualStyleTags, ());
    if (residualStyleTags.isEmpty()) {
        residualStyleTags.add(aTag.localName().impl());
        residualStyleTags.add(fontTag.localName().impl());
        residualStyleTags.add(ttTag.localName().impl());
        residualStyleTags.add(uTag.localName().impl());
        residualStyleTags.add(bTag.localName().impl());
        residualStyleTags.add(iTag.localName().impl());
        residualStyleTags.add(sTag.localName().impl());
        residualStyleTags.add(strikeTag.localName().impl());
        residualStyleTags.add(bigTag.localName().impl());
        residualStyleTags.add(smallTag.localName().impl());
        residualStyleTags.add(emTag.localName().impl());
        residualStyleTags.add(strongTag.localName().impl());
        residualStyleTags.add(dfnTag.localName().impl());
        residualStyleTags.add(codeTag.localName().impl());
        residualStyleTags.add(sampTag.localName().impl());
        residualStyleTags.add(kbdTag.localName().impl());
        residualStyleTags.add(varTag.localName().impl());
        residualStyleTags.add(nobrTag.localName().impl());
    }
    
    return residualStyleTags.contains(tagName.impl());
}

bool HTMLParser::isAffectedByResidualStyle(const AtomicString& tagName)
{
    DEFINE_STATIC_LOCAL(HashSet<AtomicStringImpl*>, unaffectedTags, ());
    if (unaffectedTags.isEmpty()) {
        unaffectedTags.add(bodyTag.localName().impl());
        unaffectedTags.add(tableTag.localName().impl());
        unaffectedTags.add(theadTag.localName().impl());
        unaffectedTags.add(tbodyTag.localName().impl());
        unaffectedTags.add(tfootTag.localName().impl());
        unaffectedTags.add(trTag.localName().impl());
        unaffectedTags.add(thTag.localName().impl());
        unaffectedTags.add(tdTag.localName().impl());
        unaffectedTags.add(captionTag.localName().impl());
        unaffectedTags.add(colgroupTag.localName().impl());
        unaffectedTags.add(colTag.localName().impl());
        unaffectedTags.add(optionTag.localName().impl());
        unaffectedTags.add(optgroupTag.localName().impl());
        unaffectedTags.add(selectTag.localName().impl());
        unaffectedTags.add(objectTag.localName().impl());
    }
    
    return !unaffectedTags.contains(tagName.impl());
}

void HTMLParser::handleResidualStyleCloseTagAcrossBlocks(HTMLStackElem* elem)
{
    HTMLStackElem* maxElem = 0;
    bool finished = false;
    bool strayTableContent = elem->strayTableContent;

    m_handlingResidualStyleAcrossBlocks = true;
    while (!finished) {
        // Find the outermost element that crosses over to a higher level. If there exists another higher-level
        // element, we will do another pass, until we have corrected the innermost one.
        ExceptionCode ec = 0;
        HTMLStackElem* curr = m_blockStack;
        HTMLStackElem* prev = 0;
        HTMLStackElem* prevMaxElem = 0;
        maxElem = 0;
        finished = true;
        while (curr && curr != elem) {
            if (curr->level > elem->level) {
                if (!isAffectedByResidualStyle(curr->tagName))
                    return;
                if (maxElem)
                    // We will need another pass.
                    finished = false;
                maxElem = curr;
                prevMaxElem = prev;
            }

            prev = curr;
            curr = curr->next;
        }

        if (!curr || !maxElem)
            return;

        Node* residualElem = prev->node;
        Node* blockElem = prevMaxElem ? prevMaxElem->node : m_current;
        Node* parentElem = elem->node;

        // Check to see if the reparenting that is going to occur is allowed according to the DOM.
        // FIXME: We should either always allow it or perform an additional fixup instead of
        // just bailing here.
        // Example: <p><font><center>blah</font></center></p> isn't doing a fixup right now.
        if (!parentElem->childAllowed(blockElem))
            return;

        m_hasPElementInScope = Unknown;

        if (maxElem->node->parentNode() != elem->node) {
            // Walk the stack and remove any elements that aren't residual style tags.  These
            // are basically just being closed up.  Example:
            // <font><span>Moo<p>Goo</font></p>.
            // In the above example, the <span> doesn't need to be reopened.  It can just close.
            HTMLStackElem* currElem = maxElem->next;
            HTMLStackElem* prevElem = maxElem;
            while (currElem != elem) {
                HTMLStackElem* nextElem = currElem->next;
                if (!isResidualStyleTag(currElem->tagName)) {
                    prevElem->next = nextElem;
                    prevElem->derefNode();
                    prevElem->node = currElem->node;
                    prevElem->didRefNode = currElem->didRefNode;
                    delete currElem;
                }
                else
                    prevElem = currElem;
                currElem = nextElem;
            }

            // We have to reopen residual tags in between maxElem and elem.  An example of this case is:
            // <font><i>Moo<p>Foo</font>.
            // In this case, we need to transform the part before the <p> into:
            // <font><i>Moo</i></font><i>
            // so that the <i> will remain open.  This involves the modification of elements
            // in the block stack.
            // This will also affect how we ultimately reparent the block, since we want it to end up
            // under the reopened residual tags (e.g., the <i> in the above example.)
            RefPtr<Node> prevNode = 0;
            currElem = maxElem;
            while (currElem->node != residualElem) {
                if (isResidualStyleTag(currElem->node->localName())) {
                    // Create a clone of this element.
                    // We call releaseRef to get a raw pointer since we plan to hand over ownership to currElem.
                    Node* currNode = currElem->node->cloneNode(false).releaseRef();
                    reportError(ResidualStyleError, &currNode->localName());
    
                    // Change the stack element's node to point to the clone.
                    // The stack element adopts the reference we obtained above by calling release().
                    currElem->derefNode();
                    currElem->node = currNode;
                    currElem->didRefNode = true;

                    // Attach the previous node as a child of this new node.
                    if (prevNode)
                        currNode->appendChild(prevNode, ec);
                    else // The new parent for the block element is going to be the innermost clone.
                        parentElem = currNode;  // FIXME: We shifted parentElem to be a residual inline.  We never checked to see if blockElem could be legally placed inside the inline though.

                    prevNode = currNode;
                }

                currElem = currElem->next;
            }

            // Now append the chain of new residual style elements if one exists.
            if (prevNode)
                elem->node->appendChild(prevNode, ec);  // FIXME: This append can result in weird stuff happening, like an inline chain being put into a table section.
        }

        // Check if the block is still in the tree. If it isn't, then we don't
        // want to remove it from its parent (that would crash) or insert it into
        // a new parent later. See http://bugs.webkit.org/show_bug.cgi?id=6778
        bool isBlockStillInTree = blockElem->parentNode();

        // We need to make a clone of |residualElem| and place it just inside |blockElem|.
        // All content of |blockElem| is reparented to be under this clone.  We then
        // reparent |blockElem| using real DOM calls so that attachment/detachment will
        // be performed to fix up the rendering tree.
        // So for this example: <b>...<p>Foo</b>Goo</p>
        // The end result will be: <b>...</b><p><b>Foo</b>Goo</p>
        //
        // Step 1: Remove |blockElem| from its parent, doing a batch detach of all the kids.
        if (isBlockStillInTree)
            blockElem->parentNode()->removeChild(blockElem, ec);

        Node* newNodePtr = 0;
        if (blockElem->firstChild()) {
            // Step 2: Clone |residualElem|.
            RefPtr<Node> newNode = residualElem->cloneNode(false); // Shallow clone. We don't pick up the same kids.
            newNodePtr = newNode.get();
            reportError(ResidualStyleError, &newNode->localName());

            // Step 3: Place |blockElem|'s children under |newNode|.  Remove all of the children of |blockElem|
            // before we've put |newElem| into the document.  That way we'll only do one attachment of all
            // the new content (instead of a bunch of individual attachments).
            Node* currNode = blockElem->firstChild();
            while (currNode) {
                Node* nextNode = currNode->nextSibling();
                newNode->appendChild(currNode, ec);
                currNode = nextNode;
            }

            // Step 4: Place |newNode| under |blockElem|.  |blockElem| is still out of the document, so no
            // attachment can occur yet.
            blockElem->appendChild(newNode.release(), ec);
        } else
            finished = true;

        // Step 5: Reparent |blockElem|.  Now the full attachment of the fixed up tree takes place.
        if (isBlockStillInTree)
            parentElem->appendChild(blockElem, ec);

        // Step 6: Pull |elem| out of the stack, since it is no longer enclosing us.  Also update
        // the node associated with the previous stack element so that when it gets popped,
        // it doesn't make the residual element the next current node.
        HTMLStackElem* currElem = maxElem;
        HTMLStackElem* prevElem = 0;
        while (currElem != elem) {
            prevElem = currElem;
            currElem = currElem->next;
        }
        prevElem->next = elem->next;
        prevElem->derefNode();
        prevElem->node = elem->node;
        prevElem->didRefNode = elem->didRefNode;
        if (!finished) {
            // Repurpose |elem| to represent |newNode| and insert it at the appropriate position
            // in the stack. We do not do this for the innermost block, because in that case the new
            // node is effectively no longer open.
            elem->next = maxElem;
            elem->node = prevMaxElem->node;
            elem->didRefNode = prevMaxElem->didRefNode;
            elem->strayTableContent = false;
            prevMaxElem->next = elem;
            ASSERT(newNodePtr);
            prevMaxElem->node = newNodePtr;
            prevMaxElem->didRefNode = false;
        } else
            delete elem;
    }

    // FIXME: If we ever make a case like this work:
    // <table><b><i><form></b></form></i></table>
    // Then this check will be too simplistic.  Right now the <i><form> chain will end up inside the <tbody>, which is pretty crazy.
    if (strayTableContent)
        m_inStrayTableContent--;

    // Step 7: Reopen intermediate inlines, e.g., <b><p><i>Foo</b>Goo</p>.
    // In the above example, Goo should stay italic.
    // We cap the number of tags we're willing to reopen based off cResidualStyleMaxDepth.
    
    HTMLStackElem* curr = m_blockStack;
    HTMLStackElem* residualStyleStack = 0;
    unsigned stackDepth = 1;
    unsigned redundantStyleCount = 0;
    while (curr && curr != maxElem) {
        // We will actually schedule this tag for reopening
        // after we complete the close of this entire block.
        if (isResidualStyleTag(curr->tagName) && stackDepth++ < cResidualStyleMaxDepth) {
            // We've overloaded the use of stack elements and are just reusing the
            // struct with a slightly different meaning to the variables.  Instead of chaining
            // from innermost to outermost, we build up a list of all the tags we need to reopen
            // from the outermost to the innermost, i.e., residualStyleStack will end up pointing
            // to the outermost tag we need to reopen.
            // We also set curr->node to be the actual element that corresponds to the ID stored in
            // curr->id rather than the node that you should pop to when the element gets pulled off
            // the stack.
            if (residualStyleStack && curr->tagName == residualStyleStack->tagName && curr->node->attributes()->mapsEquivalent(residualStyleStack->node->attributes()))
                redundantStyleCount++;
            else
                redundantStyleCount = 0;

            if (redundantStyleCount < cMaxRedundantTagDepth)
                moveOneBlockToStack(residualStyleStack);
            else
                popOneBlock();
        } else
            popOneBlock();

        curr = m_blockStack;
    }

    reopenResidualStyleTags(residualStyleStack, 0); // Stray table content can't be an issue here, since some element above will always become the root of new stray table content.

    m_handlingResidualStyleAcrossBlocks = false;
}

void HTMLParser::reopenResidualStyleTags(HTMLStackElem* elem, Node* malformedTableParent)
{
    // Loop for each tag that needs to be reopened.
    while (elem) {
        // Create a shallow clone of the DOM node for this element.
        RefPtr<Node> newNode = elem->node->cloneNode(false); 
        reportError(ResidualStyleError, &newNode->localName());

        // Append the new node. In the malformed table case, we need to insert before the table,
        // which will be the last child.
        ExceptionCode ec = 0;
        if (malformedTableParent)
            malformedTableParent->insertBefore(newNode, malformedTableParent->lastChild(), ec);
        else
            m_current->appendChild(newNode, ec);
        // FIXME: Is it really OK to ignore the exceptions here?

        // Now push a new stack element for this node we just created.
        pushBlock(elem->tagName, elem->level);
        newNode->beginParsingChildren();