You are currently viewing a snapshot of www.mozilla.org taken on April 21, 2008. Most of this content is highly out of date (some pages haven't been updated since the project began in 1998) and exists for historical purposes only. If there are any pages on this archive site that you think should be added back to www.mozilla.org, please file a bug.



WebShell Design

By Travis Bogard
and
Steve Clark
Last Modified: 10-14-99 (Travis Bogard)


    Welcome to the WebShell Design page.  This page is being written pretty much after the fact.  What I mean is WebShell is on it's way out with a new and improved design of nsWebBrowser and DocShell taking it's place.  This page is being provided to show a brief look at the state of the world prior to the redesign.  Some explanation on the sub-systems of the WebShell will be provided, but in general they will simply get high level mention.  You will note that this document will talk mostly about what is wrong with WebShell.  This is to provide explanation of why it must be re-written.  [If you have expertise on the WebShell and wish to provide information about it's design for historical reasons, please feel free to contact me about updating portions of the doc, but readers, don't expect it to happen.]
 

Index:

1. WebShell Design

This is a basic diagram of the composition of WebShell:

2. WebShell Sub-Systems

What follows is a high level description of the major objects involved with the WebShell as it exists today (10/10/99). The WebShell sub-systems are divided into two categories:  content and presentation.

A. Content

Content Model
Document content is represented a tree of content nodes.  Each content node implements nsIContent.  In addition, each content node may implement one of the stardard W3C DOM content interfaces.  Content nodes are entirely independent of a presentation.

Document
The Document is the top-level content object.  It owns the content model and all subdocument.  It owns state common to all presentations of the document, like the style sheet list and whether selection is displayed or not [selection display should not be state on the document.]

DocumentViewer
The DocumentViewer is the root object for the presentation of documents.  It provides a small number of window-like features.

DocumentLoader
The DocumentLoader subsystem is responsible for filling a document instance with content from a URI.  It includes logic for mapping a MIME type the right DocumentLoader.

B. Presentation

Frame Model
The presentation of the content model is represented by a tree of  layout objects called frames.  There is a many:1 relationship between frames and content nodes.  Frames contain all of the presentation data about a content node, including its size, position, relationship to other frames, etc.  The frame tree is a visual reflection of the underlying content tree.

Style Model
The Style System resolves stylistic attributes for every node in the content tree.  It creates a tree of style contexts in a 1:many relationship with nodes in the frame tree.  This is the style model.  Presentation is largely driven by style. The Style System is responsible for mapping each content node to 0 or more frames.  Each frame has an associated style context, which it may share with other frames.

PresShell
The PresShell is the root object for a single presentation of the document.  The PresShell owns the frame model for the presentation and all associated presentation data.  Reflow operations (the layout of the content by the gecko layout system) are initiated by the PresShell.

PresContext
The PresContext holds data about the capabilities of the presentation device and the values of various presentation variables.  For example, the PresContext determines whether the display is paginated or galley, the resolution of the device, etc.  Currently, the PresContext also holds some variable state, but it is our intention to make this an entirely stateless immutable object, so a single instance of the PresContext can be shared throughout the presentation.

WebShell
The WebShell handles embedding browser capabilities in an application.  It also handles embedding subdocuments recursively within a document.  This feature is currently used for HTML frames and text input form elements.
The current WebShell interface is monolithic, handling a large number of responsibilities for both browsing and embedding.  It is our intention to break this interface into several independent interfaces.

3. Problems with WebShell

  1. The API is irrational.  WebShell is a total mess right now.  It's been a dumping ground for anything having to do with high level browser operation, external embedding of gecko, and internal embedding of subdocument.  That's too much for any one object to reasonably handle.  Furthermore, nsIWebShell has become monolithic and needs to be broken into several manageable interfaces.
  2. WebShell is huge. It requires loading lots of services and allocating lots of memory for each instance.  Since most of this state is only required by the top level WebShell, if at all, re-allocating everything for embedded subdocuments is a huge waste of memory and allocation time.
  3. WebShell is slow in some important cases.  In the common case of loading a blank document, WebShell requires an expensive trip through the network library for loading about:blank which is very slow compared to simply constructing a minimal html document.  Embedded documents should not be forced to go through the network library.
  4. WebShell creates it's own PresContext on initialization. WebShell creates it's own PresContext, one per WebShell.   In some cases, this imposes the need to re-resolve style for the entire document on load since there is currently no way to pass in custom style data.  Since PresContext will soon be an immutable object, per-WebShell allocation is a big waste of time and space.
  5. WebShell consumes too many system resources.  The current WebShell implementation always creates at least one and often several native windows per WebShell.  This has huge performance and system resource overhead.
  6. Webshell does not work for applications wishing to embed it.  There are too many un-related callback methods on container interfaces and functionality missing from interfaces embedding apps wish to call.

4. WebShell Solutions

See nsWebBrowser Design for what's being done about this recommondation.
  1. WebShell needs to be split into 2 objects, a topmost webshell used for embedding gecko, and a lightweight DocShell used for embedding subdocs recursively in the webshell.
  2. The DocShell must have an API for the caller to supply a document, rather than going through the network library to load a URL
  3. The DocShell must have an API for the caller to supply a PresContext, rather than creating one itself.
  4. WebShells must not assume a need for a native window, and instead allow the caller to have control over whether a native window is needed or not.  A native window would be useful for when the subdocument needs to be sandboxed:  logically isolated within the presentation of the document.  Otherwise, it is unnecessary overhead.