//---------------------------------------------------------------------------
//
// File: HtmlParser.cs
//
// Copyright (C) Microsoft Corporation. All rights reserved.
//
// Description: Parser for Html-to-Xaml converter
//
//---------------------------------------------------------------------------
using System;
using System.Xml;
using System.Diagnostics;
using System.Collections;
using System.Collections.Generic;
using System.IO;
using System.Text; // StringBuilder
// important TODOS:
// TODO 1. Start tags: The ParseXmlElement function has been modified to be called after both the
// angle bracket < and element name have been read, instead of just the < bracket and some valid name character,
// previously the case. This change was made so that elements with optional closing tags could read a new
// element's start tag and decide whether they were required to close. However, there is a question of whether to
// handle this in the parser or lexical analyzer. It is currently handled in the parser - the lexical analyzer still
// recognizes a start tag opener as a '<' + valid name start char; it is the parser that reads the actual name.
// this is correct behavior assuming that the name is a valid html name, because the lexical analyzer should not know anything
// about optional closing tags, etc. UPDATED: 10/13/2004: I am updating this to read the whole start tag of something
// that is not an HTML, treat it as empty, and add it to the tree. That way the converter will know it's there, but
// it will hvae no content. We could also partially recover by trying to look up and match names if they are similar
// TODO 2. Invalid element names: However, it might make sense to give the lexical analyzer the ability to identify
// a valid html element name and not return something as a start tag otherwise. For example, if we type <good>, should
// the lexical analyzer return that it has found the start of an element when this is not the case in HTML? But this will
// require implementing a lookahead token in the lexical analyzer so that it can treat an invalid element name as text. One
// character of lookahead will not be enough.
// TODO 3. Attributes: The attribute recovery is poor when reading attribute values in quotes - if no closing quotes are found,
// the lexical analyzer just keeps reading and if it eventually reaches the end of file, it would have just skipped everything.
// There are a couple of ways to deal with this: 1) stop reading attributes when we encounter a '>' character - this doesn't allow
// the '>' character to be used in attribute values, but it can still be used as an entity. 2) Maintain a HTML-specific list
// of attributes and their values that each html element can take, and if we find correct attribute namesand values for an
// element we use them regardless of the quotes, this way we could just ignore something invalid. One more option: 3) Read ahead
// in the quoted value and if we find an end of file, we can return to where we were and process as text. However this requires
// a lot of lookahead and a resettable reader.
// TODO 4: elements with optional closing tags: For elements with optional closing tags, we always close the element if we find
// that one of it's ancestors has closed. This condition may be too broad and we should develop a better heuristic. We should also
// improve the heuristics for closing certain elements when the next element starts
// TODO 5. Nesting: Support for unbalanced nesting, e.g. <b> <i> </b> </i>: this is not presently supported. To support it we may need
// to maintain two xml elements, one the element that represents what has already been read and another represents what we are presently reading.
// Then if we encounter an unbalanced nesting tag we could close the element that was supposed to close, save the current element
// and store it in the list of already-read content, and then open a new element to which all tags that are currently open
// can be applied. Is there a better way to do this? Should we do it at all?
// TODO 6. Elements with optional starting tags: there are 4 such elements in the HTML 4 specification - html, tbody, body and head.
// The current recovery doesn;t do anything for any of these elements except the html element, because it's not critical - head
// and body elementscan be contained within html element, and tbody is contained within table. To extend this for XHTML
// extensions, and to recover in case other elements are missing start tags, we would need to insert an extra recursive call
// to ParseXmlElement for the missing start tag. It is suggested to do this by giving ParseXmlElement an argument that specifies
// a name to use. If this argument is null, it assumes its name is the next token from the lexical analyzer and continues
// exactly as it does now. However, if the argument contains a valid html element name then it takes that value as its name
// and continues as before. This way, if the next token is the element that should actually be its child, it will see
// the name in the next step and initiate a recursive call. We would also need to add some logic in the loop for when a start tag
// is found - if the start tag is not compatible with current context and indicates that a start tag has been missed, then we
// can initiate the extra recursive call and give it the name of the missed start tag. The issues are when to insert this logic,
// and if we want to support it over multiple missing start tags. If we insert it at the time a start tag is read in element
// text, then we can support only one missing start tag, since the extra call will read the next start tag and make a recursive
// call without checking the context. This is a conceptual problem, and the check should be made just before a recursive call,
// with the choice being whether we should supply an element name as argument, or leave it as NULL and read from the input
// TODO 7: Context: Is it appropriate to keep track of context here? For example, should we only expect td, tr elements when
// reading a table and ignore them otherwise? This may be too much of a load on the parser, I think it's better if the converter
// deals with it
namespace HTMLConverter
{
/// <summary>
/// HtmlParser class accepts a string of possibly badly formed Html, parses it and returns a string
/// of well-formed Html that is as close to the original string in content as possible
/// </summary>
internal class HtmlParser
{
// ---------------------------------------------------------------------
//
// Constructors
//
// ---------------------------------------------------------------------
#region Constructors
/// <summary>
/// Constructor. Initializes the _htmlLexicalAnalayzer element with the given input string
/// </summary>
/// <param name="inputString">
/// string to parsed into well-formed Html
/// </param>
private HtmlParser(string inputString)
{
// Create an output xml document
_document = new XmlDocument();
// initialize open tag stack
_openedElements = new Stack<XmlElement>();
_pendingInlineElements = new Stack<XmlElement>();
// initialize lexical analyzer
_htmlLexicalAnalyzer = new HtmlLexicalAnalyzer(inputString);
// get first token from input, expecting text
_htmlLexicalAnalyzer.GetNextContentToken();
}
#endregion Constructors
// ---------------------------------------------------------------------
//
// Internal Methods
//
// ---------------------------------------------------------------------
#region Internal Methods
/// <summary>
/// Instantiates an HtmlParser element and calls the parsing function on the given input string
/// </summary>
/// <param name="htmlString">
/// Input string of pssibly badly-formed Html to be parsed into well-formed Html
/// </param>
/// <returns>
/// XmlElement rep
/// </returns>
internal static XmlElement ParseHtml(string htmlString)
{
HtmlParser htmlParser = new HtmlParser(htmlString);
XmlElement htmlRootElement = htmlParser.ParseHtmlContent();
return htmlRootElement;
}
// .....................................................................
//
// Html Header on Clipboard
//
// .....................................................................
// Html header structure.
// Version:1.0
// StartHTML:000000000
// EndHTML:000000000
// StartFragment:000000000
// EndFragment:000000000
// StartSelection:000000000
// EndSelection:000000000
internal const string HtmlHeader = "Version:1.0\r\nStartHTML:{0:D10}\r\nEndHTML:{1:D10}\r\nStartFragment:{2:D10}\r\nEndFragment:{3:D10}\r\nStartSelection:{4:D10}\r\nEndSelection:{5:D10}\r\n";
internal const string HtmlStartFragmentComment = "<!--StartFragment-->";
internal const string HtmlEndFragmentComment = "<!--EndFragment-->";
/// <summary>
/// Extracts Html string from clipboard data by parsing header information in htmlDataString
/// </summary>
/// <param name="htmlDataString">
/// String representing Html clipboard data. This includes Html header
/// </param>
/// <returns>
/// String containing only the Html data part of htmlDataString, without header
/// </returns>
internal static string ExtractHtmlFromClipboardData(string htmlDataString)
{
int startHtmlIndex = htmlDataString.IndexOf("StartHTML:");
if (startHtmlIndex < 0)
{
return "ERROR: Urecognized html header";
}
// TODO: We assume that indices represented by strictly 10 zeros ("0123456789".Length),
// which could be wrong assumption. We need to implement more flrxible parsing here
startHtmlIndex = Int32.Parse(htmlDataString.Substring(startHtmlIndex + "StartHTML:".Length, "0123456789".Length));
if (startHtmlIndex < 0 || startHtmlIndex > htmlDataString.Length)
{
return "ERROR: Urecognized html header";
}
int endHtmlIndex = htmlDataString.IndexOf("EndHTML:");
if (endHtmlIndex < 0)
{
return "ERROR: Urecognized html header";
}
// TODO: We assume that indices represented by strictly 10 zeros ("0123456789".Length),
// which could be wrong assumption. We need to implement more flrxible parsing here
endHtmlIndex = Int32.Parse(htmlDataString.Substring(endHtmlIndex + "EndHTML:".Length, "0123456789".Length));
if (endHtmlIndex > htmlDataString.Length)
{
endHtmlIndex = htmlDataString.Length;
}
return htmlDataString.Substring(startHtmlIndex, endHtmlIndex - startHtmlIndex);
}
/// <summary>
/// Adds Xhtml header information to Html data string so that it can be placed on clipboard
/// </summary>
/// <param name="htmlString">
/// Html string to be placed on clipboard with appropriate header
/// </param>
/// <returns>
/// String wrapping htmlString with appropriate Html header
/// </returns>
internal static string AddHtmlClipboardHeader(string htmlString)
{
StringBuilder stringBuilder = new StringBuilder();
// each of 6 numbers is represented by "{0:D10}" in the format string
// must actually occupy 10 digit positions ("0123456789")
int startHTML = HtmlHeader.Length + 6 * ("0123456789".Length - "{0:D10}".Length);
int endHTML = startHTML + htmlString.Length;
int startFragment = htmlString.IndexOf(HtmlStartFragmentComment, 0);
if (startFragment >= 0)
{
startFragment = startHTML + startFragment + HtmlStartFragmentComment.Length;
}
else
{
startFragment = startHTML;
}
int endFragment = htmlString.IndexOf(HtmlEndFragmentComment, 0);
if (endFragment >= 0)
{
endFragment = startHTML + endFragment;
}
else
{
endFragment = endHTML;
}
// Create HTML clipboard header string
stringBuilder.AppendFormat(HtmlHeader, startHTML, endHTML, startFragment, endFragment, startFragment, endFragment);
// Append HTML body.
stringBuilder.Append(htmlString);
return stringBuilder.ToString();
}
#endregion Internal Methods
// ---------------------------------------------------------------------
//
// Private methods
//
// ---------------------------------------------------------------------
#region Private Methods
private void InvariantAssert(bool condition, string message)
{
if (!condition)
{
throw new Exception("Assertion error: " + message);
}
}
/// <summary>
/// Parses the stream of html tokens starting
/// from the name of top-level element.
/// Returns XmlElement representing the top-level
/// html element
/// </summary>
private XmlElement ParseHtmlContent()
{
// Create artificial root elelemt to be able to group multiple top-level elements
// We create "html" element which may be a duplicate of real HTML element, which is ok, as HtmlConverter will swallow it painlessly..
XmlElement htmlRootElement = _document.CreateElement("html", XhtmlNamespace);
OpenStructuringElement(htmlRootElement);
while (_htmlLexicalAnalyzer.NextTokenType != HtmlTokenType.EOF)
{
if (_htmlLexicalAnalyzer.NextTokenType == HtmlTokenType.OpeningTagStart)
{
_htmlLexicalAnalyzer.GetNextTagToken();
if (_htmlLexicalAnalyzer.NextTokenType == HtmlTokenType.Name)
{
string htmlElementName = _htmlLexicalAnalyzer.NextToken.ToLower();
_htmlLexicalAnalyzer.GetNextTagToken();
// Create an element
XmlElement htmlElement = _document.CreateElement(htmlElementName, XhtmlNamespace);
// Parse element attributes
ParseAttributes(htmlElement);
if (_htmlLexicalAnalyzer.NextTokenType == HtmlTokenType.EmptyTagEnd || HtmlSchema.IsEmptyElement(htmlElementName))
{
// It is an element without content (because of explicit slash or based on implicit knowledge aboout html)
AddEmptyElement(htmlElement);
}
else if (HtmlSchema.IsInlineElement(htmlElementName))
{
// Elements known as formatting are pushed to some special
// pending stack, which allows them to be transferred
// over block tags - by doing this we convert
// overlapping tags into normal heirarchical element structure.
OpenInlineElement(htmlElement);
}
else if (HtmlSchema.IsBlockElement(htmlElementName) || HtmlSchema.IsKnownOpenableElement(htmlElementName))
{
// This includes no-scope elements
OpenStructuringElement(htmlElement);
}
else
{
// Do nothing. Skip the whole opening tag.
// Ignoring all unknown elements on their start tags.
// Thus we will ignore them on closinng tag as well.
// Anyway we don't know what to do withthem on conversion to Xaml.
}
}
else
{
// Note that the token following opening angle bracket must be a name - lexical analyzer must guarantee that.
// Otherwise - we skip the angle bracket and continue parsing the content as if it is just text.
// Add the following asserion here, right? or output "<" as a text run instead?:
// InvariantAssert(false, "Angle bracket without a following name is not expected");
}
}
else if (_htmlLexicalAnalyzer.NextTokenType == HtmlTokenType.ClosingTagStart)
{
_htmlLexicalAnalyzer.GetNextTagToken();
if (_htmlLexicalAnalyzer.NextTokenType == HtmlTokenType.Name)
{
string htmlElementName = _htmlLexicalAnalyzer.NextToken.ToLower();
// Skip the name token. Assume that the following token is end of tag,
// but do not check this. If it is not true, we simply ignore one token
// - this is our recovery from bad xml in this case.
_htmlLexicalAnalyzer.GetNextTagToken();
CloseElement(htmlElementName);
}
}
else if (_htmlLexicalAnalyzer.NextTokenType == HtmlTokenType.Text)
{
AddTextContent(_htmlLexicalAnalyzer.NextToken);
}
else if (_htmlLexicalAnalyzer.NextTokenType == HtmlTokenType.Comment)
{
AddComment(_htmlLexicalAnalyzer.NextToken);
}
_htmlLexicalAnalyzer.GetNextContentToken();
}
// Get rid of the artificial root element
if (htmlRootElement.FirstChild is XmlElement &&
htmlRootElement.FirstChild == htmlRootElement.LastChild &&
htmlRootElement.FirstChild.LocalName.ToLower() == "html")
{
htmlRootElement = (XmlElement)htmlRootElement.FirstChild;
}
return htmlRootElement;
}
private XmlElement CreateElementCopy(XmlElement htmlElement)
{
XmlElement htmlElementCopy = _document.CreateElement(htmlElement.LocalName, XhtmlNamespace);
for (int i = 0; i < htmlElement.Attributes.Count; i++)
{
XmlAttribute attribute = htmlElement.Attributes[i];
htmlElementCopy.SetAttribute(attribute.Name, attribute.Value);
}
return htmlElementCopy;
}
private void AddEmptyElement(XmlElement htmlEmptyElement)
{
InvariantAssert(_openedElements.Count > 0, "AddEmptyElement: Stack of opened elements cannot be empty, as we have at least one artificial root element");
XmlElement htmlParent = _openedElements.Peek();
htmlParent.AppendChild(htmlEmptyElement);
}
private void OpenInlineElement(XmlElement htmlInlineElement)
{
_pendingInlineElements.Push(htmlInlineElement);
}
// Opens structurig element such as Div or Table etc.
private void OpenStructuringElement(XmlElement htmlElement)
{
// Close all pending inline elements
// All block elements are considered as delimiters for inline elements
// which forces all inline elements to be closed and re-opened in the following
// structural element (if any).
// By doing that we guarantee that all inline elements appear only within most nested blocks
if (HtmlSchema.IsBlockElement(htmlElement.LocalName))
{
while (_openedElements.Count > 0 && HtmlSchema.IsInlineElement(_openedElements.Peek().LocalName))
{
XmlElement htmlInlineElement = _openedElements.Pop();
InvariantAssert(_openedElements.Count > 0, "OpenStructuringElement: stack of opened elements cannot become empty here");
_pendingInlineElements.Push(CreateElementCopy(htmlInlineElement));
}
}
// Add this block element to its parent
if (_openedElements.Count > 0)
{
XmlElement htmlParent = _openedElements.Peek();
// Check some known block elements for auto-closing (LI and P)
if (HtmlSchema.ClosesOnNextElementStart(htmlParent.LocalName, htmlElement.LocalName))
{
_openedElements.Pop();
htmlParent = _openedElements.Count > 0 ? _openedElements.Peek() : null;
}
if (htmlParent != null)
{
// NOTE:
// Actually we never expect null - it would mean two top-level P or LI (without a parent).
// In such weird case we will loose all paragraphs except the first one...
htmlParent.AppendChild(htmlElement);
}
}
// Push it onto a stack
_openedElements.Push(htmlElement);
}
private bool IsElementOpened(string htmlElementName)
{
foreach (XmlElement openedElement in _openedElements)
{
if (openedElement.LocalName == htmlElementName)
{
return true;
}
}
return false;
}
private void CloseElement(string htmlElementName)
{
// Check if the element is opened and already added to the parent
InvariantAssert(_openedElements.Count > 0, "CloseElement: Stack of opened elements cannot be empty, as we have at least one artificial root element");
// Check if the element is opened and still waiting to be added to the parent
if (_pendingInlineElements.Count > 0 && _pendingInlineElements.Peek().LocalName == htmlElementName)
{
// Closing an empty inline element.
// Note that HtmlConverter will skip empty inlines, but for completeness we keep them here on parser level.
XmlElement htmlInlineElement = _pendingInlineElements.Pop();
InvariantAssert(_openedElements.Count > 0, "CloseElement: Stack of opened elements cannot be empty, as we have at least one artificial root element");
XmlElement htmlParent = _openedElements.Peek();
htmlParent.AppendChild(htmlInlineElement);
return;
}
else if (IsElementOpened(htmlElementName))
{
while (_openedElements.Count > 1) // we never pop the last element - the artificial root
{
// Close all unbalanced elements.
XmlElement htmlOpenedElement = _openedElements.Pop();
if (htmlOpenedElement.LocalName == htmlElementName)
{
return;
}
if (HtmlSchema.IsInlineElement(htmlOpenedElement.LocalName))
{
// Unbalances Inlines will be transfered to the next element content
_pendingInlineElements.Push(CreateElementCopy(htmlOpenedElement));
}
}
}
// If element was not opened, we simply ignore the unbalanced closing tag
return;
}
private void AddTextContent(string textContent)
{
OpenPendingInlineElements();
InvariantAssert(_openedElements.Count > 0, "AddTextContent: Stack of opened elements cannot be empty, as we have at least one artificial root element");
XmlElement htmlParent = _openedElements.Peek();
XmlText textNode = _document.CreateTextNode(textContent);
htmlParent.AppendChild(textNode);
}
private void AddComment(string comment)
{
OpenPendingInlineElements();
InvariantAssert(_openedElements.Count > 0, "AddComment: Stack of opened elements cannot be empty, as we have at least one artificial root element");
XmlElement htmlParent = _openedElements.Peek();
XmlComment xmlComment = _document.CreateComment(comment);
htmlParent.AppendChild(xmlComment);
}
// Moves all inline elements pending for opening to actual document
// and adds them to current open stack.
private void OpenPendingInlineElements()
{
if (_pendingInlineElements.Count > 0)
{
XmlElement htmlInlineElement = _pendingInlineElements.Pop();
OpenPendingInlineElements();
InvariantAssert(_openedElements.Count > 0, "OpenPendingInlineElements: Stack of opened elements cannot be empty, as we have at least one artificial root element");
XmlElement htmlParent = _openedElements.Peek();
htmlParent.AppendChild(htmlInlineElement);
_openedElements.Push(htmlInlineElement);
}
}
private void ParseAttributes(XmlElement xmlElement)
{
while (_htmlLexicalAnalyzer.NextTokenType != HtmlTokenType.EOF && //
_htmlLexicalAnalyzer.NextTokenType != HtmlTokenType.TagEnd && //
_htmlLexicalAnalyzer.NextTokenType != HtmlTokenType.EmptyTagEnd)
{
// read next attribute (name=value)
if (_htmlLexicalAnalyzer.NextTokenType == HtmlTokenType.Name)
{
string attributeName = _htmlLexicalAnalyzer.NextToken;
_htmlLexicalAnalyzer.GetNextEqualSignToken();
_htmlLexicalAnalyzer.GetNextAtomToken();
string attributeValue = _htmlLexicalAnalyzer.NextToken;
xmlElement.SetAttribute(attributeName, attributeValue);
}
_htmlLexicalAnalyzer.GetNextTagToken();
}
}
#endregion Private Methods
// ---------------------------------------------------------------------
//
// Private Fields
//
// ---------------------------------------------------------------------
#region Private Fields
internal const string XhtmlNamespace = "http://www.w3.org/1999/xhtml";
private HtmlLexicalAnalyzer _htmlLexicalAnalyzer;
// document from which all elements are created
private XmlDocument _document;
// stack for open elements
Stack<XmlElement> _openedElements;
Stack<XmlElement> _pendingInlineElements;
#endregion Private Fields
}
}