2007-12-06 Jb Evain <jbevain@novell.com>

[mono.git] / web / xml-classes

* XML Classes Status and Tasks

** Abstract

  XML library is used by several areas of Mono such as ADO.NET and XML 
  Digital Signature (xmldsig). Here I write about System.Xml.dll and
  related tools. This page won't include any classes which are in other
  assemblies such as XmlDataDocument.

  Note that current corlib has its own XML parser class (Mono.Xml.MiniParser).

  Basically System.XML.dll feature is almost finished, so I write this
  document mainly for bugs and improvement hints.

** Status

*** System.Xml namespace


**** Document Object Model (Core)

  DOM implementation has finished and our DOM implementation scores better
  than MS.NET as to the NIST DOM test results (it is ported by Mainsoft
  hackers and in our unit tests).

**** Xml Writer

  Here XmlWriter almost equals to XmlTextWriter. If you want to see
  another implementation, check XmlNodeWriter.cs and DTMXPathDocumentWriter.cs
  in System.XML sources.

  XmlTextWriter is completed, though it looks a bit slower than MS.NET (I
  tried 1.1).

**** XmlResolver

  XmlUrlResolver is implemented.

  XmlSecureResolver, which is introduced in MS .NET Framework 1.1 is basically
  implemented, but it requires CAS (code access security) feature. We need to
  fixup this class after ongoing CAS effort works.

  You might also be interested in some improved implementations which resides
  in Mono.Xml.Ext.dll.

  [2.0] XmlDataSourceResolver <del>is not implemented as yet</del><ins>vanished</ins>.

**** XmlNameTable

  NameTable is implemented, but also needs performance improvement.
  Optimization hackings are welcome. There is also a <a 
  href="http://bugzilla.ximian.com/show_bug.cgi?id=59537">bugzilla entry</a> 
  for this matter.

**** XML Reader

  XmlTextReader, XmlNodeReader and XmlValidatingReader are almost finished.

        <ul>
                * All OASIS conformance test passes as Microsoft does. Some 
                  W3C tests fail, but it looks better than Microsoft.
                * Entity expansion and its well-formedness check is incomplete.
                  It incorrectly allows divided content models. It incorrectly
                  treats its Base URI, so some dtd parse fails.
                * I won't add any XDR support on XmlValidatingReader. (I haven't
                  ever seen XDR used other than Microsoft's BizTalk Server 2000,
                  and Now they have 2002 with XML Schema support). If anyone
                  contributes an implementation, it would be still nice.
        </ul>

  XmlTextReader and XmlValidatingReader should be faster than now. Currently
  XmlTextReader looks nearly twice as slow as MS.NET, and XmlValidatingReader
  (which uses this slow XmlTextReader) looks nearly three times slower. (Note
  that XmlValidatingReader wouldn't be so slow as itself. It uses schema
  validating reader and dtd validating reader.)


**** Some Advantages

  The design of Mono's XmlValidatingReader is radically different from
  that of Microsoft's implementation. Under MS.NET, DTD content validation
  engine is in fact simple replacement of XML Schema validation engine.
  Mono's DTD validation is designed fully separate and does validation
  as normal XML parser does. For example, Mono allows non-deterministic DTD.

  Another advantage of this XmlValidatingReader is support for *any* XmlReader.
  Microsoft supports only XmlTextReader (this bug is fixed in .NET 2.0 beta,
  taking shape of XmlReader.Create()).

  <del>I added extra support interface named "IHasXmlParserContext", which is
  considered in XmlValidatingReader.ResolveEntity(). </del><ins>This is now
  made as internal interface.</ins> Microsoft failed to design XmlReader
  so that XmlReader cannot be subtree-pluggable (i.e. wrapping use of other
  XmlReader) since XmlParserContext shoud be supplied for DTD information
  support (e.g. entity references cannot be expanded) and namespace manager.
  (In .NET 2.0, Microsoft also supported similar to IHasXmlParserContext,
  named IXmlNamespaceResolver, but it still does not provide DTD information.)

  We also have RELAX NG validating reader (described later).


*** System.Xml.Schema

**** Summary

  Basically it is completed. You can test how current schema validation engine
  is complete (incomplete) by using standalone test module (see 
  mcs/class/System.XML/Test/System.Xml.Schema/standalone_tests).
  At least in my box, msxsdtest fails only 30 cases with bugfixed catalog -
  this score is better than that of Microsoft implementation. But instead,
  we need performance boost. There should be many points to improve
  schema compilation and validation.

**** Schema Object Model

  Completed, except for some things to be fixed:

        <ul>
                * Complete facet support. Currently some of them is missing.
                  Recently David Sheldon is doing several fixes on them.
                * ContentTypeParticle for pointless xs:choice is incomplete
                  (fixing this arose another bugs in compilation.
                  Interestingly, MS.NET also fails around here, so it might
                  be nature of ContentTypeParticle design)
                * Some derivation by restriction (DBR) handling is incorrect.
        </ul>

**** Validating Reader

  Basically this is implemented and actually its feature is complete,
  but I have only did validation feature testing. So we have to write more 
  tests on properties, methods, and events (validation errors).


*** System.Xml.Serialization

  Lluis Sanchez is the maintainer of this namespace (as well as
  System.Web.Services).

  XmlSerializer is almost finished and is on bugfix phase.

  However, we appliciate more tests. Please try 
  
        <ul>
                * System.Web.Services to invoke SOAP services.
                * xsd.exe and wsdl.exe to create classes.
        </ul>

  And if any problems were found, please file it to bugzilla.

  Lluis also built interesting standalone test system placed under
  mcs/class/System.Web.Services/Test/standalone.

  You might also interested in "genxs", which enables you to create custom
  XML serializer. See "tools" section discussed later. There is also "sgen"
  that is introduced in .NET 2.0.


*** System.Xml.XPath and System.Xml.Xsl

  There are two XSLT implementations. One and historical implementation is
  based on libxslt (aka Unmanaged XSLT). Now we uses fully implemented and
  managed XSLT by default. To use Unmanaged XSLT, set MONO_UNMANAGED_XSLT
  environment value (any value is acceptable).

  As for Managed XSLT, we support msxsl:script.

  It would be nice if we can support <a href="http://www.exslt.org/">EXSLT</a>.
  <a href="http://msdn.microsoft.com/WebServices/default.aspx?pull=/library/en-us/dnexxml/html/xml05192003.asp">Microsoft has tried to do some of them</a>, 
  but it is not successful because of System.Xml.Xsl design problem:

        <ul>
                * In general, .NET's "extension objects" (including 
                  msxsl:script) is not useful to return node-sets (MS XSLT
                  implementation rejects just overriden XPathNodeIterator, 
                  but accepts only their hidden classes. And are the same 
                  in Mono though classes are different)

                * In .NET's extension object design, extension function name 
                  is a valid method name that cannot contain some characters
                  such as '-'. That is, implementing EXSLT in C# is impossible.
        </ul>

  So if we support EXSLT, it has to be done inside our System.XML.dll.
  Microsoft developers are also aware of this problem and some of them wish
  to have EXSLT support in WinFX (not whidbey). If anyone is interested
  in it, it would be nice.

  Our managed XSLT implementation is slower than MS XSLT for some kind of
  stylesheets, and faster for some.


*** RELAX NG

  I implemented an experimental RelaxngValidatingReader. It is still not
  complete, for example some simplification stuff (see RELAX NG spec 
  chapter 4; especially 4.17-19) and some constraints (especially 7.3).
  See mcs/class/Commons.Xml.Relaxng/README for details.

  Currently we have

        <ul>
                * Custom datatype support. Right now, you can use XML schema
                  datatypes ( http://www.w3.org/2001/XMLSchema-datatypes ) 
                  as well as RELAX NG default datatypes (as used in relaxng.rng).

                * RELAX NG Compact Syntax support, though not yet stable.
                  See Commons.Xml.Relaxng.Rnc.RncParser class.
</ul>


** System.Xml v2.0

  Microsoft released the first public beta version of .NET Framework 2.0,
  available from <a href="http://www.microsoft.com/downloads/details.aspx?familyid=916EC067-8BDC-4737-9430-6CEC9667655C&displaylang=en">MSDN</a>.
  It contains several new classes.

  There are two assemblies related to System.Xml v2.0; System.Xml.dll and
  System.Data.SqlXml.dll. Now that System.Data.SqlXml.dll is little important.
  It just contains only XQueryCommand class inside System.Xml.* namespace.
  Most of the important part are in System.Xml.dll.

  Note that .NET Framework is pre-release version, so they are subject
  to change. Actually many of the pre-released classes vanished.

  System.Xml 2.0 contains several features such as:

        <ul>
                * new XPathNavigator members that supports editing XmlDocument
                * XmlReaderSettings, XmlWriterSettings and factory methods
                * Strongly typed XmlReader, XmlWriter and XPathNavigator.
                * XML Schema design changes
                * XSD Inference
                * Well-documented and improved XmlSerializer.
                * XSLT per-stylesheet assembly generator
        </ul>

*** System.Xml 2.0

**** Summary

  <a href="http://mono.ximian.com/class-status/mono-HEAD-vs-fx-2/class-status-System.Xml.html">
  Here</a> is the latest status of mono's System.Xml 2.0.

  Major remaining tasks are:

        <ul>
                * binary content support on XmlReader
                * XmlValueConverter
                * XslCompiledTransform (not available in .NET FX 2.0 as yet)
        </ul>

  In general, System.Xml 2.0 bits are largely implemented.

  I have some notes on things marked as "missing" in the above class status
  page:

        <ul>
                * XmlReader, XmlTextReader and XmlValidatingReader has
                  many "missing bits", but ReadValueAsXxx() are obsoleted.
                  They were now taken its place by ReadContentAsXxx() and
                  ReadElementContentAsXxx() methods (those differences come
                  from the basis of the methods: ReadElementString() or
                  ReadString()).
                * In System.Xml.XPath - first, new features in XPathDocument
                  are said as reverted. So I just commented out those new
                  features that had been already implemented in mono.
                  Other missing classes such as XPathEditableNavigator are
                  obsolete and will vanish at the time of 2.0 RTM.
                * There are some security attributes. They are either way
                  not applicable as yet, and actually currently corcompare
                  ignores them. (So I already added some, but are ignored.)
        </ul>

**** XmlReader/XmlWrier Factory methods

  In .NET 2.0, XmlTextReader, XmlNodeReader, XmlValidatingReader are
  obsolete and XmlReader.Create() is recommended (there is however no
  alternative way to create XmlNodeReader). Similarly, there are
  XmlWriter.Create() overloads.

  Currently, Microsoft's XmlWriter.Create() is unstable and there will
  be changes. So basically remaining XmlWriter.Create() is supposed to be
  done after the next beta version of .NET 2.0.

  Some of XmlReader.Create() overloads are implemented, with limited
  XmlReaderSettings support.


**** Strongly typed XmlReader/XmlWriter

  In .NET 2.0, XmlReader is supposed to support strongly-typed data reading.
  They are based on W3C "XML Schema Datatypes" Recommendation and "XQuery 1.0
  and XPath 2.0 Data Model" Working Draft.

  Note that this W3C specification is just a Working Draft and there might
  be incompatible changes for those strongly-typed API. For example, in the
  latest version of the spec (as of November 2004), xs:QName is specified as
  to have prefix, local name and namespace URI, while .NET XML classes just
  treat it as XmlQualifiedName which does not contain prefix. So this part of
  API might become obsolete or useless.

  Some of XmlReader.ReadValueAsXxx() and XmlWriter.WriteValue() overloads are
  implemented, though incompletely. They are based on internal XQueryConvert.


**** Sub-tree handling in XmlReader/XmlWriter/XPathNavigator

  Currently XmlReader.ReadSubtree(), XmlWriter.WriteSubtree() and
  XPathNavigator.ReadSubtree() are implemented, though not well-tested.
  They are based on Mono.Xml.SubtreeXmlReader and
  Mono.Xml.XPath.XPathNavigatorReader classes.


*** System.Xml.Schema 2.0

  Since .NET 1.x is not so compliant with W3C XML Schema specification,
  Microsoft had to redesign System.Xml.Schema classes. We also have to
  change many things.

        <ul>
                * It does not expose XmlSchemaDatatype as a type of an
                  element or an attribute (except for obsolete members). 
                  Primitive types are represented as XmlSchemaSimpleType 
                  instances (thus there are ElementSchemaType, 
                  AttributeSchemaType, BaseXmlSchemaType that replace some 
                  existing properties). Note that XmlSchemaDatatype itself is 
                  not deprecated as yet.

                * "XQuery 1.0 and XPath 2.0 Data Model" datatypes (such as
                  xdt:dayTimeDuration) are newly supported. They are partially
                  implemented yet.

                * schema structures are now bound in parent-child
                  relationship. It is not yet implemented. With related to
                  it, there seems bunch of schema compilation bugfixes.

                * XmlSchemaCollection is not used anymore to represent
                  effective set of schemas. Instead, new XmlSchemaSet class
                  is used. It should affect on schema compilation design.
        </ul>


**** XSD Inference

  In .NET 2.0, there is an XML Schema inference implementation. The API (or
  implementation design) is not cool and Microsoft will make changes in
  later version of beta or .NET Framework.

**** XmlSchemaValidator

  It is (somewhat) exposed XML Schema validation functionality. It enables
  developers to get expected particles and attributes. Actually it is
  internally used to implement XmlReader.Create() with XmlSchemaSet.


*** System.Xml.XPath 2.0

**** Editable XPathNavigator

  In System.Xml 2.0, XPathNavigator has many editing API (that used to be
  "XPathEditableNavigator"). It enables developers to edit XmlDocument
  (not for XPathDocument; it is read-only).

  (In the early stage of .NET 2.0, XPathDocument is supposed to be editable.
  Also, the new version of editable XPathNavigator had been already
  implemented in Mono, as Mono.Xml.XPath.XPathEditableDocument.)


*** System.Xml.Xsl 2.0

  There used to be System.Xml.Query.XsltCommand class, and it now disappeared
  and it is said (by Microsoft developers) that there will be a class named
  XslCompiledTransform - so it will be included in the next beta version.
  It is regarded as the identical thing to XsltCommand.

  XslCompiledtransform seems almost the same as System.Xml.Xsl.XslTransform,
  but this class transforms documents twice to four times as fast as
  XslTransform. Instead, stylesheet compilation is much slower, because
  it generates compiled stylesheet assembly.



** External assemblies for Mono 2.0

*** Mono.Xml.Ext

  (<strong>IMPORTANT</strong>: There used to be System.Xml.Query.XQueryCommand
  class that implemented XQuery, but now Microsoft dropped it for .NET 2.0.
  So we moved them out to external assembly named Mono.Xml.Ext.dll, but I
  haven't changed thing so much when I just extracted them. Thus, this
  section is kept as is, and basically old stuff.)

  Mono.Xml.Ext.dll contains mainly XQuery implementation, and some utility 
  classes.

  XQueryCommand implements XQuery. XQuery is a new face XML document
  manipulation language (at least new face in .NET world). It is similar
  to XSLT, but extended to support XML Schema based datatypes (and it is
  not XML based langauge). It is similar to XPath, but it can construct
  XML nodes. It has no complicated template resolution, but works like
  functional languages.

  Under MS.NET, XQuery implementation is mainly in System.Xml.Query and
  MS.Internal.Xml.* namespaces. The implementation is mostly
  in System.Xml.dll. It is also true to our System.Xml.dll. Our XQueryCommand
  in System.Data.SqlXml.dll just invokes the actual XQuery processor
  (Mono.Xml.XPath2.XQueryCommandImpl) which resides in System.Xml.dll via
  reflection.

  Currently we are not implementing MS.Internal.Xml.* classes. MS
  implementation is based on an old version of the W3C specification, and
  our implementation is currently based on 
  <a href="http://www.w3.org/TR/2004/WD-xquery-20040723/">23 July 2004
  version</a> (latest as of now) of the working draft.

  XQuery implementation tasks are:

        <ul>
                * XQuery syntax parser that parses xquery string to AST 
                  (abstract syntax tree). -> partly not done.

                * XQuery AST compiler into static context -> partly not done.

                * XQuery (dynamic context) runtime = XQuery expression evaluator
                  + sequence iterator. -> partly not done.

                * XPathItem data model and (mainly) conversion support.
                  -> partly done.

                * Applied expression classes for XQuery/XPath 2.0 functions and
                  operators. -> partly done.

                * Optimization, and design per-query assembly code generator (later)
        </ul>

  It already handles some queries, while major part implementation is missing
  or buggy (like FLWOR, expressions for sequence type handling, built-in
  function support etc.).


*** Relax NG and DSDL in Mono 1.2

  Currently we support only RELAX NG as one part of ISO DSDL effort. There
  is existing Schematron implementation (NMatrix Project: <a 
  href="http://sourceforge.net/projects/dotnetopensrc/">
  http://sourceforge.net/projects/dotnetopensrc/</a>). With a few changes,
  it can be used with mono.
  
  We don't have multi-language based validation support, namely
  Namespace-based Validation Dispatch Language (NVDL). To support unwrapping,
  one special XmlReader implementation is required (other schema validation
  support can be done by ReadSubtree()). Note that we had seen RELAX
  Namespace, Modular Namespace (MNS) and Namespace Routing Language (NRL)
  - that is, standardization effort is still ongoing (though NVDL looks
  mostly the same as NRL).

  In Mono 2.0, there are some improvements:

        <ul>
                * RelaxngPattern.Compile() now provides error location.

                * RelaxngValidatingReader now supports "expected label
                  collector" API; You can get name list of expected element
                  names and attribute names by simulating validation.

  In Mono 2.0, there might be improvements on Commons.Xml.Relaxng.

        <ul>
                * The type of error should be kind of RelaxngGrammarException.

                * Right now there is no ambiguity detection implementation that
                  would be useful for RelaxngPattern based xml serialization (if
                  there is need).

                * Because of lack of ambiguity detection, there is no way to
                  provide XmlMapping (XmlTypeMapping/XmlMemberMapping). But 
                  If anyone is interested in such effort, integration with
                  XmlSerializer would be interesting task.
        </ul>


** Tools

*** xsd.exe

  See <a href="ado-net.html">ADO.NET page</a>.

  Note that xsd.exe does not support XmlSchemaInference in System.Xml 2.0.

*** genxs.exe

  genxs.exe is a custom XmlSerializer code generator. This is Mono exclusive 
  advantage. See <a 
  href="http://primates.ximian.com/~lluis/blog/archives/000120.html">Lluis's
  description</a> and manpages for details. Code files are in mcs/tools/genxs.

*** sgen.exe

  sgen.exe is an alternative XmlSerializer code generator which is based on
  XmlSerializer.GenerateSerializer() written by Lluis. Code files are in
  mcs/tools/sgen.

*** dtd2xsd.exe

  It is a tiny tool that enables conversion from DTD structure to XML Schema.
  You need an XML instance, not DTD file itself (since basic XmlTextReader has
  no feature to read external DTD resource itself). It runs only under mono.

** Miscellaneous

*** Mutual assembly dependency

  Sometimes I hear complain about System.dll and System.Xml.dll mutual
  dependency: System.dll references to System.Xml.dll (e.g. 
  System.Configuration.ConfigXmlDocument extended from XmlDocument), while 
  System.Xml.dll vice versa (e.g. XmlUrlResolver.ResolveUri takes System.Uri).
  Since they are in public method signatures, so at least we cannot get rid
  of these mutual references.

  Nowadays System.Xml.dll is built using incomplete System.dll (lacking
  System.Xml dependent classes such as ConfigXmlDocument). Full System.dll
  is built after System.Xml.dll is done.

  Note that you still need System.dll to run mcs.


  Atsushi Eno <atsushi@ximian.com>
  last updated 12/16/2004