On 9/14/07, Vincent Massol <vincent@massol.net > wrote:

+1 to all that. So let me summarizes and rephrase to see if I have
understood :)

1) We have 4 types of objects:
* TextProcessors: take text and generate text
* Parsers: take text and generate an internal DOM format (pivot format)
* DomProcessors: take DOM and generate DOM
* Renderers: take DOM and generate anything (text, PDF, RTF, HTML,
XML, etc)

Yes.

2) Document contents are stored in the database in textual format in
the main xwiki syntax (whatever we decide it is - we could
standardize on creole for example)

It can be the "Common Syntax" for the reasons mentioned above :-). Creole syntax is one of the most restrictive syntaxes. And I tried to uses in the CommonSyntax as much markups of the Creole as possible.

An another possibility is to store directly in XML or in XHTML+microformat enhancements (for additional structural elements).
pro:
- it can be exported/imported directly and used by external applications which knows nothing about wikis; just a standard XML or XHTML
- this content can be transformed with XSLT processors directly without usage of the WikiModel
- it can be faster to parse XML than the CommonWiki syntax (I have no comparisons)
con:
- it is more difficult to work with diffs (but for diffs it is *better* to use WkiModel and to generate a specific wiki syntax; for example "Common syntax");
- it is not a "human readable" format; it is difficult to understand what you loads from the DB

3) Use case 1: Viewing a document

a) Get the doc from the DB --> text1 (xwiki text format)
b) Apply TextProcessors --> text2
c) Call XWikiParser --> DOM1 (transforms XWiki text syntax into an
internal DOM)
d) Apply DomProcessors --> DOM2
e) Call the required Renderer --> PDF, XML, HTML, RTF, text, etc

Yes.

4) Use case 2: Editing a document, assuming the user wants to use the
MediaWiki syntax for editing

a) Get the doc from the DB --> text1 (xwiki text format)
b) Call XWikiParser --> DOM1 (transforms XWiki text syntax into an
internal DOM)
c) Call MediaWikiRenderer --> text2 (text in MediaWiki format)
d) the user edits and hits save
e) MediaWikiParser --> DOM2 (transforms MediaWiki text syntax into
the internal DOM)
f) Call XWikiRenderer --> text" (transforms DOM into xwiki textual
format)
g) Save text3 in the database

Yes. (text1 and text3 can be XML, as I said above)

5) In practice this means the following classes:

* TextProcessorManager: to chain several text processors

Yes. But it can be just a composite processor implementing the same ProcessorManager interfaces.

* TextProcessor
- VelocityTextProcessor
- GroovyTextProcessor

Yes.

* WikiParser: Takes wiki syntax and generates a DOM in a XWiki-
specific format (independent of the different wiki syntaxes).
   - LegacyXWikiWikiParser
   - XWikiWikiParser (or simply use CreoleWikiParser if we want our
internal format to be Creole)
   - ConfluenceWikiParser
   - MediaWikiWikiParser
   - JSPWikiWikiParser
   - CreoleWikiParser
   - HTMLParser: I think all parsers above need to support HTML since
the wiki syntaxes can be mixed with HTML. So this HTMLParser is
probably a parent of the other parsers in some regard. Anyway we need
this one for the WYSIWYG editor which may need to transform HTML to
wiki syntax (so we may need a XWikiDomProcessor too to transform into
XWiki syntax). The alternative (much better) is to have the WYSIWYG
editor only use the internal XWiki-specific DOM format for all its
manipulations.

If you want, you can put HTML as a non-interpreted block ("verbatim blocks") and interpret it in the client code. But internally the WikiModel does not support "embedded" (X)HTML. The main reason: in this case I loose control of the document structure. And this control is the main goal of the WikiModel.

* DomProcessorManager: to chain several DOM processors
* DomProcessor
- Don't know yet what we're going to use this for. TOCDomProcessor
as you say above maybe.

DOMProcessor can be used to transform the original DOM object representing the document in the DB into a new (user and query-specific) DOM object which can contain new elements, generated dynamically. Now all dynamic page elements are interpreted as simple Velocity or Groovy scripts and they generate text documents which should be parsed using Radeox and transformed to the final HTML document. Using the DOM representation it is possible to interpret some nodes of this graph as Groovy scripts. In WikiModel they will correspond to Verbatim blocks which are opaque for WikiModel but they can be interpreted as scripts by the DomProcessor(s). And these "Groovy"-nodes can be executed and they will add new DOM elements to the DOM2. For example this approach can be used to generate search results.

The advantages of this approach:
- You can put your parsed document DOM1 in the cache, which will avoid you to to parse the document for each query. It is a slowest step in the page processing. Even if the current version of WikiModel is faster than before and it should be faster than Radeox processor.
- Your Groovy scripts will manipulate with normal java classes (DOM nodes) and it will produce DOM nodes and not a plain text. It seems especially interesting taking into account Groovy's Builders ( http://groovy.codehaus.org/Builders). It is enough to write a very simple builder (see http://groovy.codehaus.org/BuilderSupport) generating DOM nodes and ... voila! Your Groovy node from a wiki page generates search results as DOM nodes! These manipulations with DOM objects should be MUCH faster that process plain text for every request. And all following steps are fast as well - to generate an HTML page it is enough to visit all node with an "XHTMLVisitor".

BTW: do you need Velocity at all? Using only Groovy is much cleaner. It can be used as THE language of XWiki. It can be used as a template *and* programming language at the same time. And if you *really* want it is possible to integrate Jasper (from Tomcat) engine to use it for pure templating. The code from Jetty (th e org.mortbay.jetty.jspc.plugin package) can be used as an example of integration with Jasper (see http://jetty.mortbay.org/xref/index.html).
In this case in templates it will be possible to use:
- JSP tag libraries (including standard ones)
- Multiple scripting languages (like javabeans, javascript, jpython, jruby, groovy,...)

* Renderer
   - XMLRenderer
   - HTMLRenderer
   - PDFRenderer
   - RTFRenderer
   - XWikiRenderer (or simply use CreoleRenderer if we want our
internal format to be Creole)
   - ConfluenceRenderer
   - MediaWikiRenderer
   - JSPWikiRenderer
   - CreoleRenderer

Yes. All these renderers should be written if you want to support all these syntaxes. I think that it should not be very difficult.

WDYT? Do I have it right? :)

Best regards,
Mikhail

Thanks
-Vincent

On Sep 13, 2007, at 6:37 PM, StÃ©phane LauriÃ¨re wrote:

> Hi Vincent, hi everyone,
>
> We discussed the WikiModel integration with Mikhail this afternoon.
> Here
> is below our input.
>
> Vincent Massol wrote:
>> Hi,
>>
>> I've started working on designing the new Rendering/Parsing
>> components and API for XWiki. The implementation will be based on
>> WikiModel but we need some XWiki wrapping interfaces around it. Note
>>  that this is a prerequisite for the new WYSIWYG editor based on GWT
>>  (see http://www.xwiki.org/xwiki/bin/view/Design/
>> NewWysiwygEditorBasedOnGwt).
>>
>> I've updated http://www.xwiki.org/xwiki/bin/view/Design/
>> WikiModelIntegration with the information below, which I'm pasting
>> here so that we can have a discussion about it. I'll consolidate the
>>  results on that wiki page.
>>
>> Componentize the Parsing/Rendering APIs
>> ==================================
>>
>> We need 4 main components:
>>
>> * A Scripting component to manage scripting inside XWiki documents
>> and to evaluate them.
>
> On the topic of scripting we would like to propose a distinction
> between
> scripts that act on text and scripts that act on the DOM.
> Typically, the
> text rendering processing for flow would be the following, for say
> "text1":
>
> text1 =TextProcessor=> text2 =Parser=> dom1 =DomProcessor=> dom2
> => ...
>
> - the scripts contained in text1 are processed in the context of
> user1,
> this results into a new text: text2
> - the parser parses text2 and converts text2 to a DOM tree, dom1
> - dom1 is processed by scripts that work directly on the DOM (example:
> table of content generator), this results in dom2
> - dom2 is made to available as such or is converted to XML, HTML, PDF
> etc. depending on the user request
>
> TextProcessor and DomProcessor would have the following interfaces:
>
> TextProcessor
> - String execute(String content)
>
> DomProcessor
> - DOM execute(DOM content)
>
> That means we should have a syntax to distinguish between scripts that
> generate text content, and scripts that manipulate the DOM.
>
>>      * A Rendering component to manage rendering Wiki syntax into
>> HTML and other (PDF, RTF, etc)
>>      * A Wiki Parser component to offer a typed interface to XWiki
>> content so that it can be manipulated
>>      * A HTML Parser component (for the WYSIWYG editor)
>>
>> Different Syntaxes ===============
>>
>> Two possible solutions:
>>
>>     1. Have a WikiSyntax Object (A simple class with one property: a
>> combox box with different syntaxes: XWiki Legacy, Creole, MediaWiki,
>> Confluence, JSPWiki, etc) that users can attach to pages to tell the
>> Renderers what syntax is used. If no such object is attached then
>> it'll default to XWiki's default syntax (XWiki Legacy or Creole for
>> example).
>>     2. Have some special syntax, independent of the wiki syntaxes to
>> tell the Rendered that such block of content should be rendered with
>> that given syntax. Again there would be a default.
>>
>
> Here's our view regarding the syntax used in wiki edit mode: document
> requested for edition are available from the database in a serialized
> format, for instance XHTML. When entering into the edit action, the
> user
> indicates his preferred syntax. If the text of the requested document
> contains some blocks that are not handled by the chosen syntax, the
> user
> gets a warning (example: the document contains a table as a list item,
> and the user tries to edit the document using JSPWiki syntax). If not,
> WikiModel converts the serialized format into a DOM, the user edits
> the
> DOM and the WikiModel serializer serializes it back when the user
> saves it.
>
> Note that the DOM representation of wiki documents in the latest
> version
> of WikiModel is still pending.
>
>>
>> XWiki Interfaces
>> =============
>>
>>      * ScriptingEngineManager: Manages the different Scripting
>> Engines, calling them in turn.
>>      * ScriptingEngine
>>            o Method: evaluate(String content)
>>            o Implementation: VelocityScriptingEngine
>>            o Implementation: GroovyScriptingEngine
>>      * RenderingEngineManager: Manages the different Rendering
>> Engines, calling them in turn.
>>      * RenderingEngine
>>            o Method: render(String content)
>>            o Implementation: XWikiLegacyRenderingEngine (current
>> rendering engine)
>>            o Implementation: WikiModelRenderingEngine
>>      * Parser: content parsing
>>            o HTMLParser: parses HTML syntax
>>            o WikiParser: parses wiki syntax
>>            o Implementation: WikiModelHTMLParser
>>            o Implementation: WikiModelWikiParser
>>
>> Open Questions:
>>
>>      * Does WikiModel support a generic syntax for macros?
>
> WikiModel generates events for blocks that are not to be parsed
> (typically because they contain scripts).
>
> For example, in the WikiModel syntax currently called "CommonSyntax",
> this looks like the following:
> ==============
> {{{macro:mymacro (String parameters)
> dothis
> dothat
>
> }}}
>
>
> $mymacro(parameters)
> ==============
>
> For each syntax, macro blocks are identified as far as possible (we
> still have to check it's the case for all types of macro blocks inde
> indeed).
>
>
>>      * Is the Rendering also in charge of generating PDF, RTF,
>> XML, etc?
>>            o I think so, need to modify interfaces above to reflect
>> this.
>>      * The WikiParser needs to recognizes scripts since this is
>> needed for the WYSIWYG editor.
>
> the WikiModel parser recognizes scripts indeed.
>
>
> Mikhail and StÃ©phane
>
>>
>> Use cases
>> ========
>>
>>      * View page
>>            o ViewAction -- template ->
>> ScriptingEngineManager.evaluate
>> () -- wiki syntax -> RenderingEngineManager.render() ---> HTML, XML,
>> PDF, RTF, etc
>>      * Edit page in WYSIWYG editor
>>            o Uses the WikiParser to create a "DOM" of the page
>> content and to render it accordingly. NOTE: This is required since
>> rendering in the WYSIWYG editor is different from the final
>> rendering. For example, macros need to be shown in a special way to
>> make them visible, etc.
>>            o Changes done by the user are entered in HTML. Note: it
>> would be better to capture them so that they are entered in the
>> "DOM". Is that possible? If not, then the HTMLParser is used to
>> convert from HTML to Wiki Syntax but they're likely be some loss in
>> the conversion. The advantage is the ability to take any HTML content
>> and generate wiki syntax from it.
>>
>>
>> This is my very earlier thinking but I wanted to make it visible to
>> give everyone the change to 1) know what's happening and 2) suggest
>> ideas.
>>
>> I'll refine this in the coming days and post again on this thread.
>>
>> Thanks
>> -Vincent