summaryrefslogtreecommitdiffstats
path: root/xparts/doc/xparts.html
blob: 0ff872be7d75acb995a26dac79e895e893fd63eb (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
<!DOCTYPE HTML PUBLIC "-//IETF//DTD HTML//EN">
<html>
  <head>
    <title>XParts - Extending KParts</title>
  </head>

  <body>
<center>    
<h1>XParts - Extending KParts</h1>
    <small>Matthias Ettrich, Simon Hausmann, Lars Knoll</small>
</center> 

    <p>This article briefly describe the concepts, architecture and
    reasoning behind the XParts technology.  The purpose of XParts is
    to extend KParts over language, toolkit, process and machine
    bounderies. XParts makes it possible to write KDE components with
    almost any toolkit or language an author prefers or to turn
    existing applications into KDE components quite easily.

    <p>In addition, XParts is also an important glueing technology to
     make KParts available in other component based systems or to
     utilize non-KPart components transparently as KParts.

    <h4>Classic KParts</h4>

    <p>In order to understand, what is extending about XParts, first a
    brief overview on how KParts work.

      <img src="kparts.png">

    <p>Imagine an application - for example the integrated file manager
      "Konqueror" -  wants to utilize a component that handles the
      "text/html" mimetype. It therefore asks the trader of the KIO
      subsystem whether such a service is available and where. The
      trader uses the system configuration cache to localize an
      appropriate service that fits with the user's preferences. The
      system configuration cache is a service type database
      constructed from the desktop files of a KDE setup. In the case
      of "text/html", the trader will very like return KDE's builtin
      HTML viewer dubbed KHtml. This viewer is is most certainly
      available as a KPart component. The application will then - via
      KLibLoader and KLibFactory - load the shared library object that
      implements the component and create a KPart instance.  The
      LibLoader keeps track of any objects created in the loaded
      library and will automatically unload it after all objects have
      been deleted.

    <p>If the application does not only want to display HTML, but
      act as a full featured browser, the plain KPart interface is not
      sufficient. If the user clicks on a link, for example, the HTML
      component has to request a new URL. This kind of interaction is
      defined in the BrowserExtension interface. An application can
      query the KParts for additonal interfaces and get handles to
      them in case those are available. In the example case of KHtml,
      the BrowserExtension interface is exported. In the case of a
      text editor component, it's very likely that the TextEditor
      interface is available.

    <h4>In-process components</h4>

      <p>The beauty of KParts is its simplicity. It's a clean and
      flexible in-process approach with all its advantages:
      <ul>
      <li> lightweight - components share the same application
      context and all its allocated resources.
      <li> synchronious - calls are predictable, there are no
      timeouts to wait for and no events to process in an uncertain
      amount of time.
      <li> stable - neither race conditions nor rare exceptions
      can occur
      <li> extremely powerful - there are virtually no
      limitations to how a component API can look like (including
      passing pointers) or what a plugin can do with an application.
      </ul>

    <p>Those advantages are unvaluable for a lightweight and tightly
    integrated office suite like KOffice. However, there are no silver
    bullets and most certainly there are drawbacks when the system is
    used in settings with different requirements.

    <p>Take the fourth item, it's comprehensive power while
    maintaining simplicity. This was one of the main requirements of
    the KOffice team, and it alone almost determines an in-process
    approach with dynamically loadable shared objects. In a generic
    browser like Konqueror, the requirements for integrated components
    are not as high as with an office suite. In an office suite,
    different components operate on one single document, whereas in a
    browser, the components basically provide different views for
    given Urls.  To illustrate this issue, imagine how far the web
    came with such primitive and inflexible component technology like
    Netscape plugins. They did most of what people wanted to do with
    browser plugins, though, and so became a huge success.

    <h4>Out-of-process components</h4>

    <p>To sum this up: for multi-view applications like a generic
    browser, there's no technical argument why out-of-process
    components could not be sufficient. So let's look closer at the
    specific advantages of such a solution.

    <ul>
    <li>With out-of-process components, it's much easier to provide
    applications as components that do not support being loaded
    dynamically as shared library objects.  Typical examples are
    programs written in interpreted languages. With a pure in-process
    model, one would have to be able to load the interpreter as
    embedded language.

    <li>If a component handles the event loop differently from the
    embedding application, an complete event loop merger is
    required. This glueing code can be tricky and might not work well
    in all cases. It's much easier for out-of-process components to
    provide full toolkit independence.

     <li> components of the same type could share one process
     context. Not sure where this is actually useful, but it has most
     certainly some technical beauty attached to it.
	
    </ul>

    <p>Let's pick a concrete example. Imagine that you - for whatever
    reason - want to offer the Mozilla rendering engine (gecko) as
    KPart, so that users have an an alternative to KDE's builtin
    rendering engine KHtml.

    <p>The first step of such a project is to find out, whether
    Mozilla already is available as a reusable component that could
    form the basis of a KDE integration. And in fact, it is. A small
    library called GtkMozEmbed makes it possible to load the entire
    Mozilla as a single Gtk widget, i.e. the rendering engine gecko,
    the networking protocol implementations, the javascript
    interpreter and whatever else Mozilla.org comes up with. The
    MozEmbed library works pretty similar to KParts. Once
    instantiated, it dynamically loads all libraries required by
    Mozilla. As an interesting side note, all Unix filemanager
    projects that utilize Mozilla (for example the Nautilus
    filemanager) use this library to embed mozilla. This means you are
    in good company using a stock MozEmbed library, as you don't have
    to maintain this code but somebody else will do it for you.

    <p>Now that we have a dynamically loaded Gtk widget, how do we
    turn that into a KPart? Quite straight forward. There is a
    QGtkWidget extension available for Qt, that lets you use Gtk
    widgets in your Qt applications. You simply create a QGtkWidget
    with a pointer to the Gtk widget you get from MozEmbed and insert
    that into your KPart. Then you do a few trivial reimplementations
    of the virtual functions of the BrowserExtension interface that
    map to the corresponding functions of Mozilla and you are
    done. The result is a fully functional Konqueror that uses Mozilla
    as backend - or rather a fully functional Mozilla that uses
    Konqueror as graphical user interface, however you want to look at
    it.

    <h4>Trouble ahead</h4>

    <p> While the skedged solution works, there are some unmentioned
    and ugly details. First of all, Mozilla uses the event loop of
    glib, while Konqueror uses Qt. Unfortunatly, mixing both event
    loops is not possible with the current release of glib, unless one
    want to end up with an application that constantly requires some
    CPU to run, even when being idle. While this seems to be ok for
    today's Java virtual machines, it's not acceptable by KDE's
    quality standards. Until glib 2.0 is released, you need to patch
    glib in order to make the QGtkWidget work properly. No big deal
    for most Linux users, still a hassle. And keep in mind that glib
    is a fairly open system. If the component was written in some
    other toolkit, it might be possible that glueing code is
    impossible to get right, without wasting at least a bit of CPU.

    <p>The second problem is Mozilla's size. It's by no means an
    ordinary component. In fact, it's a magnitude larger than the
    Konqueror framework. And since Mozilla and Konqueror do not share
    the same graphics toolkit, the toolkit's size has to be added to
    that. It seems odd to load and unload such a huge amount of code -
    and it can to lead to all kind of problems when trying to unload
    it again.
      
   <p>To make things worse, Mozilla wasn't even released as final
   version yet. While it is already quite usable, it's stability is
   still far from being production quality. This doesn't matter too
   much for a standalone browser, but can really hurt with a
   component.  A standalone browser usually is supposed to display one
   web page. If it crashes, this page is gone, so the user simply
   tries again. With a generic browser like Konqueror, there is not
   just one component active at a time, but several. There might be
   some directory views, an embedded console, another toplevel window
   window, an imaged preview and much more. A crashing Mozilla would
   take all those component with it - and leave the user with only
   half of its prior desktop.

  <p>Imagine that some users define Mozilla to be the primary
  component to handle text/html in Konqueror. After some testing, all
  works well and they continue using it. A couple of days later, they
  might have forgotten the configuration change they did. Whenever
  they now hit a web page where Mozilla crashes, they will blame
  Konqueror. This we don't want.  No code is perfect, but if a crash
  occurs in our code, at least it's our crash. That means, we can fix
  it and we can provide newer versions.

  <p>Thus, from a maintainance and support point of you, it is not
  acceptable for KDE to run code inprocess that is not actually
  maintained or controlled by the team, at least not in the default
  setup.

  <h4>Out-of-process components</h4>

  <p>For the given reasons, it makes a lot of sense to extend KParts
  over process bounderies. In addition, we also win a high degree of
  toolkit and language independency.

  <p>To make this work, we have to identify the streamable parts of
  the KParts interface and offer them via some kind of middleware. 

  <p>We chose KDE's native desktop middleware, the desktop
  communication protocol (DCOP) to establish the communication. In
  addition to the fact that DCOP was explicitely designed for these
  kind of tasks, there are some more benefits:
   <ul>
      <li> DCOP runs already on the desktop, i.e. there are no additonal costs 
	in terms of  resource consumption.
      <li> Does not put any limitations onto the interfaces as long as
	data types are streamable
      <li> Server architecture makes it easy and robust to detect
	crashes on either side.
    </ul>
    
    There are several DCOP implementations available. The reference
    implementation is the one using C++ and Qt that is used in KDE
    applications. For Mozilla, we would choose a plain ANSI-C
    implementation that uses glib.

    <p>The following picture shows the interface structure:

    <p> <img src="xparts.png">

     <p>The main thing that differs from KParts is the
     <em>XPartHost</em> interface that is responsible for embedding a
     part. The missing link now is a standard KPart component that
     implements the <em>XPartHost</em> interface. Via this
     <b>KXPartHost</b> component, it is possible to use any XPart
     transparently as KPart without changing a single line of code:
    <p>
      <img src="kxparthost.png">

      <p>On the other side of the fence, we need an implementation of
      the <em>XPartManager</em> interface and can serve us with
      <em>XPart</em> interfaces. We provide this through the
      relatively highlevel and generic classes GtkXPartManger and
      GtkXPart, as shown in the next picture:
    <p>
      <img src="gtkxpart.png">
    <p> The GtkXPart is a standard Gtk widget that can have a MozEmbed
    widget as child widget. The only code that is necessary to write
    is the code used to connect the <em>BrowserExtension</em>
    interface to the corresponding functions of Mozilla.

    <h4>External KParts</h4>

    <p>The same technique can now be used to utilize standard KPart
    components in an out-of-process fashion via the XPart system. All
    we need is a KXPartManager that wraps standard KParts in
    KXParts. The KXParts then export the <em>XPart</em> interface. The
    complete structure is shown in the next picture:
   <p><img src="kxpart.png">

    <h4>Conclusion</h4> <p> Although the implementation of the
    external mozilla part is more a proof of concept than a finished
    xpart, we have shown a clean way to realize out of process
    components on top of KParts. It could also be shown that this
    approach is both language and toolkit independent. 

      <p>To accomplish this task, not a <em>single</em> line of code
      in konqueror had to be changed. All we did was providing yet
      another independent KPart component.

    <p>By writing a small wrapper it is possible to embed any kind of
    visual component. In addition, we can provide generic wrappers for
    any kind of visual component model, as long as those models are
    powerful enough to describe their interfaces and GUI requirements
    at runtime. This includes KParts (eg. KOffice components), Bonobo
    components (like the Nautilus MP3 viewer) and Uno components
    provided by OpenOffice (formerly known as StarOffice).

    <hr>
    <address><a href="mailto:ettrich@kde.org">Matthias Ettrich</a></address>
    <address><a href="mailto:hausmann@kde.org">Simon Hausmann</a></address>
    <address><a href="mailto:knoll@kde.org">Lars Knoll</a></address>
<!-- Created: Tue Oct 17 18:08:25 CEST 2000 -->
<!-- hhmts start -->
Last modified: Tue Apr  3 20:39:13 CEST 2001
<!-- hhmts end -->
  </body>
</html>