koffice/doc/chalk/developers-plugins.docbook

<sect1 id="developers-plugins">
<title>Developing &chalk; Plugins</title>

<sect2 id="developers-plugins-introduction">
<title>Introduction</title>

<para>
&chalk; is infinitely extensible with plugins. Tools, filters, large
chunks of the user interface and even colorspaces are plugins. In fact,
&chalk; recognizes these six types of plugins:
</para>

<itemizedlist>
<listitem><para>colorspaces &mdash; these define the channels that constitute
a single pixel</para></listitem>
<listitem><para>tools &mdash; anything that is done with a mouse or tablet
input device</para></listitem>
<listitem><para>paint operations &mdash; pluggable painting effects for
tools</para></listitem>
<listitem><para>image filters &mdash; change all pixels, or just the selected
pixels in a layer</para></listitem>
<listitem><para>viewplugins &mdash; extend Chalk’s user interface with new
dialog boxes, palettes and operations</para></listitem>
<listitem><para>import/export filters &mdash; read and write all kinds of
image formats</para></listitem>
</itemizedlist>

<para>
&chalk; itself consists of three layered libraries and a directory with some
common support classes: chalkcolor, chalkimage and chalkui. Within
&chalk;, objects can by identified by a <classname>KisID</classname>, that is
the combination of a unique untranslated string (used when saving, for
instance) and a translated string for GUI purposes.
</para><para>
A word on compatibility: &chalk; is still in development. From &chalk; 1.5 to
1.6 not many API changes are expected, but there may be some.
If you develop a plugin for &chalk; and choose to do so in
&chalk;’s subversion repository, chances are excellent that we’ll help you
porting. These changes may also render parts of this document out of date.
Always check with the latest API documentation or the header files installed
on your system.
</para>

<sect3 id="developers-plugins-introduction-chalkcolor">
<title>ChalkColor</title>

<para>
The first library is chalkcolor. This library loads the colorspace plugins.
</para><para>
A colorspace plugin should implement the <classname>KisColorSpace</classname>
abstract class or, if the basic capabilities of the new colorspace will be
implemented by <command>lcms</command> (<ulink url="http://www.littlecms.com/"
/>), extend <classname>KisAbstractColorSpace</classname>. The chalkcolor
library could be used from other applications and does not depend on
&koffice;.
</para>
</sect3>

<sect3 id="developers-plugins-introduction-chalkimage">
<title>ChalkImage</title>

<para>
The libchalkimage library loads the filter and paintop plugins and is
responsible for working with image data: changing pixels, compositing and
painting. Brushes, palettes, gradients and patterns are also loaded by
libchalkimage. It is our stated goal to make libchalkimage independent of
&koffice;, but we currently share the gradient loading code with &koffice;.
</para><para>
It is not easy at the moment to add new types of resources such as brushes,
palettes, gradients or patterns to &chalk;. (Adding new brushes, palettes,
gradients and patterns is easy, of course.) &chalk; follows the guidelines of
the Create project (<ulink url="http://create.freedesktop.org/" />) for these.
Adding support for Photoshop's brush file format needs libchalkimage hacking;
adding more gimp brush data files not.
</para><para>
<classname>ChalkImage</classname> loads the following types of plugins:
</para>

<itemizedlist>
<listitem><para>&chalk; filters must extend and implement the abstract class
<classname>KisFilter</classname>,
<classname>KisFilterConfiguration</classname> and possibly
<classname>KisFilterConfigurationWidget</classname>.
An example of a filter is Unsharp Mask.</para></listitem>
<listitem><para>Paint operations or paintops are the set of operations
painting tools suchs as freehand or circle have access to. Examples of
paintops are pen, airbrush or eraser. Paintops should extend the
<classname>KisPaintop</classname> base class. Examples of new paintops could
be a chalk brush, an oilpaint brush or a complex programmable
brush.</para></listitem>
</itemizedlist>

</sect3>

<sect3 id="developers-plugins-introduction-chalkui">
<title>ChalkUI</title>

<para>
The libchalkui library loads the tool and viewplugins. This library is a
&koffice; Part, but also contains a number of widgets that are useful for
graphics applications. Maybe we will have to split this library in chalkpart
and chalkui in the 2.0 release. For now, script writers are not given access
to this library and plugin writers are only allowed to use this library when
writing tools or viewplugins. <classname>ChalkUI</classname> loads the
following types of plugins:
</para>

<itemizedlist>
<listitem><para>Tools are derived from <classname>KisTool</classname> or one
of the specialized tool base classes such as
<classname>KisToolPaint</classname>, <classname>KisToolNonPaint</classname> or
<classname>KisToolFreehand</classname>. A new tool could be a foreground
object selection tool. Painting tools (and that includes tools that paint on
the selection) can use any paintop to determine the way pixels are
changed.</para></listitem>
<listitem><para>Viewplugins are ordinary KParts that use
<command>kxmlgui</command> to insinuate themselves into &chalk;'s user
interface. Menu options, dialogs, toolbars &mdash; any kind of user interface
extension can be a viewplugin. In fact, important functionality like &chalk;'s
scripting support is written as a viewplugin.</para></listitem>
</itemizedlist>

</sect3>

<sect3 id="developers-plugins-introduction-importexport">
<title>Import/Export filters</title>

<para>
Import/Export filters are &koffice; filters, subclasses of
<classname>KoFilter</classname>. Filters read and write image data in any of
the myriad image formats in existence. And example of a new &chalk;
import/export filter could be a PDF filter. Filters are loaded by the
&koffice; libraries.
</para>

</sect3>

</sect2>

<sect2 id="developers-plugins-creating">
<title>Creating plugins</title>

<para>
Plugins are written in C++ and can use all of &tde; and &TQt; and the &chalk;
developer API. Only viewplugins should use the &koffice; API. Don’t worry:
&chalk;’s API’s are quite clear and rather extensively documented (for free
software) and coding your first filter is really easy.
</para><para>
If you do not want to use C++, you can write scripts in Python or Ruby; that
is a different thing altogether, though, and you cannot currently write tools,
colorspaces, paintops or import/export filters as scripts.
</para><para>
&chalk; plugins use &tde;'s parts mechanism for loading, so the parts
documentation at <ulink url="http://developer.kde.org" /> is relevant here, too.
</para><para>
Your distribution should have either installed the relevant header files with
&chalk; itself, or might have split the header files into either a &koffice;
dev or a &chalk; dev package. You can find the API documentation for &chalk;'s
public API at <ulink url="http://koffice.org/developer/apidocs/chalk/html/" />.
</para>

<sect3 id="developers-plugins-creating-automake">
<title>Automake (and CMake)</title>

<para>
&tde; 3.x and thus &koffice; 1.5 and 1.6 use <command>automake</command>;
&tde; 4.0 and &koffice; 2.0 use <command>cmake</command>. This tutorial
describes the <command>automake</command> way of creating plugins. 
<!-- If I have not updated this manual when we release KOffice 2.0, please
remind me to do so. -->
</para><para>
Plugins are &tde; modules and should be tagged as such in their
<filename>Makefile.am</filename>. Filters, tools, paintops, colorspaces and
import/export filters need <literal role="extension">.desktop</literal> files;
viewplugins need a <application>KXMLGui</application>
<filename>pluginname.rc</filename> file in addition. The easiest way to get
started is to checkout the chalk-plugins project from the &koffice; Subversion
repository and use it as the basis for your own project. We intend to prepare
a skeleton &chalk; plugin pack for KDevelop, but haven’t had the time to do
so yet.
</para>

<sect4 id="d-p-c-a-makefile">
<title><filename>Makefile.am</filename></title>

<para>
Let's look at the skeleton for a plugin module. First, the
<filename>Makefile.am</filename>. This is what &tde; uses to generate the
makefile that builds your plugin:

<programlisting>
kde_services_DATA = chalkLIBRARYNAME.desktop

INCLUDES = $(all_includes)

chalkLIBRARYNAME_la_SOURCES = sourcefile1.cpp sourcefile2.cpp

kde_module_LTLIBRARIES = chalkLIBRARYNAME.la
noinst_HEADERS = header1.h header2.h

chalkLIBRARYNAME_la_LDFLAGS = $(all_libraries) -module $(KDE_PLUGIN)
chalkLIBRARY_la_LIBADD = -lchalkcommon

chalkextensioncolorsfilters_la_METASOURCES = AUTO
</programlisting>

This is the makefile for a filter plugin. Replace
<replaceable>LIBRARYNAME</replaceable> with the name of your work, and you are
set.
</para><para>
If your plugin is a viewplugin, you will likely also install a <literal
role="extension">.rc</literal> file with entries for menubars and toolbars.
Likewise, you may need to install cursors and icons. That is all done through
the ordinary &tde; <filename>Makefile.am</filename> magic incantantions:

<programlisting>
chalkrcdir = $(kde_datadir)/chalk/chalkplugins
chalkrc_DATA = LIBRARYNAME.rc
EXTRA_DIST = $(chalkrc_DATA)

chalkpics_DATA = \
   bla.png \
   bla_cursor.png
chalkpicsdir = $(kde_datadir)/chalk/pics
</programlisting>

</para>
</sect4>

<sect4 id="d-p-c-a-desktop">
<title>Desktop files</title>

<para>
The <literal role="extension">.desktop</literal> file announces the type of plugin:

<programlisting>
[Desktop Entry]
Encoding=UTF-8
Icon=
Name=User-visible Name
ServiceTypes=Chalk/Filter
Type=Service
X-TDE-Library=chalkLIBRARYNAME
X-TDE-Version=2
</programlisting>
</para><para>
Possible ServiceTypes are:
</para>

<itemizedlist>
<listitem><para>Chalk/Filter</para></listitem>
<listitem><para>Chalk/Paintop</para></listitem>
<listitem><para>Chalk/ViewPlugin</para></listitem>
<listitem><para>Chalk/Tool</para></listitem>
<listitem><para>Chalk/ColorSpace</para></listitem>
</itemizedlist>

<para>
File import and export filters use the generic &koffice; filter framework and
need to be discussed separately.
</para>
</sect4>

<sect4 id="d-p-c-a-boilerplate">
<title>Boilerplate</title>

<para>
You also need a bit of boilerplate code that is called by the &tde; part
framework to instantiate the plugin &mdash; a header file and an implementation file.
</para><para>
A header file:
<programlisting>
#ifndef TOOL_STAR_H_
#define TOOL_STAR_H_

#include &lt;tdeparts/plugin.h&gt;

/**
* A module that provides a star tool.
*/
class ToolStar : public KParts::Plugin
{
   TQ_OBJECT
public:
   ToolStar(TQObject *parent, const char *name, const QStringList &amp;);
   virtual ~ToolStar();

};

#endif // TOOL_STAR_H_
</programlisting>
</para>

<para>
And an implementation file:
<programlisting>
#include &lt;kinstance.h&gt;
#include &lt;kgenericfactory.h&gt;

#include &lt;kis_tool_registry.h&gt;

#include "tool_star.h"
#include "kis_tool_star.h"


typedef KGenericFactory&lt;ToolStar&gt; ToolStarFactory;
K_EXPORT_COMPONENT_FACTORY( chalktoolstar, ToolStarFactory( "chalk" ) )


ToolStar::ToolStar(TQObject *parent, const char *name, const QStringList &amp;)
   : KParts::Plugin(parent, name)
{
   setInstance(ToolStarFactory::instance());
   if ( parent->inherits("KisToolRegistry") )
   {
       KisToolRegistry * r = dynamic_cast&lt;KisToolRegistry*&gt;( parent );
       r -> add(new KisToolStarFactory());
   }

}

ToolStar::~ToolStar()
{
}

#include "tool_star.moc"
</programlisting>
</para>
</sect4>

<sect4 id="d-p-c-a-registries">
<title>Registries</title>

<para>
Tools are loaded by the tool registry and register themselves with the tool
registry. Plugins like tools, filters and paintops are loaded only once: view
plugins are loaded for every view that is created. Note that we register
factories, generally speaking. For instance, with tools a new instance of a
tool is created for every pointer (mouse, stylus, eraser) for every few. And a
new paintop is created whenever a tool gets a mouse-down event.
</para>

<para>
Filters call the filter registry: 
<programlisting>
   if (parent->inherits("KisFilterRegistry")) {
       KisFilterRegistry * manager = dynamic_cast&lt;KisFilterRegistry *&gt;(parent);
       manager->add(new KisFilterInvert());
   }
</programlisting>
</para><para>
Paintops the paintop registry:
<programlisting>
   if ( parent->inherits("KisPaintOpRegistry") ) {
           KisPaintOpRegistry * r = dynamic_cast&lt;KisPaintOpRegistry*&gt;(parent);
           r -> add ( new KisSmearyOpFactory );
   }
</programlisting>
</para><para>
Colorspaces the colorspace registry (with some complications):
<programlisting>
   if ( parent->inherits("KisColorSpaceFactoryRegistry") ) {
	KisColorSpaceFactoryRegistry * f = dynamic_cast&lt;isColorSpaceFactoryRegistry*&gt;(parent);

       KisProfile *defProfile = new KisProfile(cmsCreate_sRGBProfile());
       f->addProfile(defProfile);

       KisColorSpaceFactory * csFactory = new KisRgbColorSpaceFactory();
       f->add(csFactory);
   
       KisColorSpace * colorSpaceRGBA = new KisRgbColorSpace(f, 0);
       KisHistogramProducerFactoryRegistry::instance() -> add(
               new KisBasicHistogramProducerFactory&lt;KisBasicU8HistogramProducer&gt;
               (KisID("RGB8HISTO", i18n("RGB8 Histogram")), colorSpaceRGBA) );
   }
</programlisting>
</para><para>
View plugins do not have to register themselves, and they get access to a
<classname>KisView</classname> object:
<programlisting>
   if ( parent->inherits("KisView") )
   {
       setInstance(ShearImageFactory::instance());
       setXMLFile(locate("data","chalkplugins/shearimage.rc"), true);

       (void) new TDEAction(i18n("&amp;Shear Image..."), 0, 0, this, TQ_SLOT(slotShearImage()), actionCollection(), "shearimage");
       (void) new TDEAction(i18n("&amp;Shear Layer..."), 0, 0, this, TQ_SLOT(slotShearLayer()), actionCollection(), "shearlayer");

       m_view = (KisView*) parent;
   }
</programlisting>
</para><para>
Remember that this means that a view plugin will be created for every view the
user creates: splitting a view means loading all view plugins again.
</para>
</sect4>

<sect4 id="d-p-c-a-versioning">
<title>Plugin versioning</title>

<para>
&chalk; 1.5 loads plugins with <literal>X-TDE-Version=2</literal> set in the
<literal role="extension">.desktop</literal> file. &chalk; 1.6 plugins will
probably be binary incompatible with 1.5 plugins and will need the version
number 3. &chalk; 2.0 plugins will need the version number 3. Yes, this is not
entirely logical.
</para>

</sect4>
</sect3>
</sect2>

<sect2 id="developers-plugins-colorspaces">
<title>Colorspaces</title>

<para>
Colorspaces implement the <classname>KisColorSpace</classname> pure virtual
class. There are two types of colorspaces: those that can use
<command>lcms</command> for transformations between colorspaces, and those
that are too weird for <command>lcms</command> to handle. Examples of the
first are cmyk, rgb, yuv. An example of the latter is watercolor or wet &amp;
sticky. Colorspaces that use <command>lcms</command> can be derived from
<classname>KisAbstractColorSpace</classname>, or of one of the base classes
that are specialized for a certain number of bits per channel.
</para><para>
Implementing a colorspace is pretty easy. The general principle is that
colorspaces work on a simple array of bytes. The interpretation of these bytes
is up to the colorspace. For instance, a pixel in 16-bit GrayA consists of
four bytes: two bytes for the gray value and two bytes for the alpha value.
You are free to use a struct to work with the memory layout of a pixel in your
colorspace implementation, but that representation is not exported. The only
way the rest of &chalk; can know what channels and types of channels your
colorspace pixels consist of is through the
<classname>KisChannelInfo</classname> class.
</para><para>
Filters and paintops make use of the rich set of methods offered by
<classname>KisColorSpace</classname> to do their work. In many cases, the
default implementation in <classname>KisAbstractColorSpace</classname> will
work, but more slowly than a custom implementation in your own colorspace
because <classname>KisAbstractColorSpace</classname> will convert all pixels
to 16-bit L*a*b and back.
</para>

<sect3 id="developers-plugins-colorspaces-kischannelinfo">
<title><classname>KisChannelInfo</classname></title>

<programlisting>
(http://websvn.kde.org/trunk/koffice/chalk/chalkcolor/kis_channelinfo.h)
</programlisting>
<para>
This class defines the channels that make up a single pixel in a particular
colorspace. A channel has the following important characteristics:
</para>
<itemizedlist>
<listitem><para>a name for display in the user interface</para></listitem>
<listitem><para>a position: the byte where the bytes representing this channel
start in the pixel.</para></listitem>
<listitem><para>a type: color, alpha, substance or substrate. Color is plain
color, alpha is see-throughishness, substance is a representation of amount of
pigment or things like that, substrate is the representation of the canvas.
(Note that this may be refactored at the drop of a hat.)</para></listitem>
<listitem><para>a valuetype: byte, short, integer, float — or
other.</para></listitem>
<listitem><para>size: the number of bytes this channel takes</para></listitem>
<listitem><para>color: a <classname>TQColor</classname> representation of this
channel for user interface visualization, for instance in
histograms.</para></listitem>
<listitem><para>an abbreviaton for use in the GUI when there’s not much
space</para></listitem>
</itemizedlist>
</sect3>

<sect3 id="developers-plugins-colorspaces-kiscompositeop">
<title><classname>KisCompositeOp</classname></title>

<para>
As per original Porter-Duff, there are many ways of combining pixels to get a
new color. The <classname>KisCompositeOp</classname> class defines most of
them: this set is not easily extensible except by hacking the chalkcolor
library.
</para><para>
A colorspace plugin can support any subset of these possible composition
operations, but the set must always include "OVER" (same as "NORMAL") and
"COPY". The rest are more or less optional, although more is better, of
course.
</para>
</sect3>

<sect3 id="developers-plugins-colorspaces-kiscolorspace">
<title><classname>KisColorSpace</classname></title>

<para>
The methods in the <classname>KisColorSpace</classname> pure virtual classs
can be divided into a number of groups: conversion, identification and
manipulation.
</para><para>
All classes must be able to convert a pixel from and to 8 bit RGB (i.e., a
<classname>TQColor</classname>), and preferably also to and from 16 bit L*a*b.
Additionally, there is a method to convert to any colorspace from the current
colorspace.
</para><para>
Colorspaces are described by the <classname>KisChannelInfo</classname> vector,
number of channels, number of bytes in a single pixel, whether it supports
High Dynamic Range images and more.
</para><para>
Manipulation is for instance the combining of two pixels in a new
pixel: bitBlt, darkening or convolving of pixels.
</para><para>
Please consult the API documentation for a full description of all methods you
need to implement in a colorspace.
</para><para>
<classname>KisAbstractColorSpace</classname> implements many of the virtual
methods of <classname>KisColorSpace</classname> using functions from the
<command>lcms</command> library. On top of
<classname>KisAbstractColorSpace</classname> there are base colorspace classes
for 8 and 16 bit integer and 16 and 32 bit float colorspaces that define
common operations to move between bit depths.
</para>

</sect3>
</sect2>

<sect2 id="developers-plugins-filters">
<title>Filters</title>

<para>
Filters are plugins that examine the pixels in a layer and them make changes
to them. Although &chalk; uses an efficient tiled memory backend to store
pixels, filter writers do not have to bother with that. When writing a filter
plugin for the &Java; imaging API, Photoshop or The Gimp, you need to take care
of tile edges and <quote>cobble</quote> tiles together: &chalk; hides that
implementation detail.
</para>
<note><para>Note that it is theoretically easy to replace the current tile
image data storage backend with another backend, but that backens are not true
plugins at the moment, for performance reasons.</para></note>
<para>
&chalk; uses iterators to read and write pixel values. Alternatively, you can
read a block of pixels into a memory buffer, mess with it and then write it
back as a block. But that is not necessarily more efficient, it may even be
slower than using the iterators; it may just be more convenient. See the API
documentation.
</para><para>
&chalk; images are composed of layers, of which there are currently four
kinds: paint layers, group layers, adjustment layers (that contain a filter
that is applied dynamically to layers below the adjustment layer) and part
layers. Filters always operate on paint layers. Paint layers contain paint
devices, of the class <classname>KisPaintDevice</classname>. A paint device in
its turn gives access to the actual pixels.
</para><para>
<classname>PaintDevice</classname>s are generally passed around wrapped in
shared pointers. A shared pointer keeps track of in how many places the paint
device is currently used and deletes the paint device when it is no longer
used anywhere. You recognize the shared pointer version of a paint device
through its <literal>SP</literal> suffix. Just remember that you never have to
explicitly delete a <classname>KisPaintDeviceSP</classname>.
</para><para>
Let's examine a very simple filter, one that inverts every pixel. The code for
this filter is in the <filename
class="directory">koffice/chalk/plugins/filters/example</filename> directory. 
The main method is 
<programlisting>
KisFilterInvert::process(KisPaintDeviceSP src, KisPaintDeviceSP dst,
                        KisFilterConfiguration* /*config*/, const QRect&amp; rect).
</programlisting>
The function gets passed two paint devices, a configuration object (which is
not used in this simple filter) and a <varname>rect</varname>. The
<varname>rect</varname> describes the area of the
paint device which the filter should act on. This area is described by
integers, which means no sub-pixel accuracy.
</para><para>
The <varname>src</varname> paint device is for reading from, the
<varname>dst</varname> paint device for writing to. These parameters may point
to the same actual paint device, or be two different paint devices. (Note:
this may change to only one paint device in the future.)
</para><para>
Now, let's look at the code line by line:
</para>
<programlisting>
void KisFilterInvert::process(KisPaintDeviceSP src, KisPaintDeviceSP dst,
                             KisFilterConfiguration* /*config*/, const QRect&amp; rect)
{
   Q_ASSERT(src != 0);
   Q_ASSERT(dst != 0);

   KisRectIteratorPixel srcIt = src->createRectIterator(rect.x(), rect.y(), rect.width(), rect.height(), false); <co id="invert1" />
   KisRectIteratorPixel dstIt = dst->createRectIterator(rect.x(), rect.y(), rect.width(), rect.height(), true ); <co id="invert2" />

   int pixelsProcessed = 0;
   setProgressTotalSteps(rect.width() * rect.height());

   KisColorSpace * cs = src->colorSpace();
   TQ_INT32 psize = cs->pixelSize();

   while( ! srcIt.isDone() )
   {
       if(srcIt.isSelected()) <co id="invert3" />
       {
           memcpy(dstIt.rawData(), srcIt.oldRawData(), psize); <co id="invert4" />

           cs->invertColor( dstIt.rawData(), 1); <co id="invert5" />
       }
       setProgress(++pixelsProcessed);
       ++srcIt;
       ++dstIt;
   }
   setProgressDone(); // Must be called even if you don't really support progression
}
</programlisting>

<calloutlist>
<callout arearefs="invert1">
<para>This creates an iterator to read the existing pixels. Chalk has three
types of iterators: horizontal, vertical and rectangular. The rect iterator
takes the most efficient path through the image data, but does not guarantee
anything about the location of the next pixel it returns. That means that you
cannot be sure that the pixel you will retrieve next will be adjacent to the
pixel you just got. The horizontal and vertical line iterators do guarantee
the location of the pixels they return.
</para></callout>
<callout arearefs="invert2"><para>
(2) We create the destination iterator with the <literal>write</literal>
setting to <literal>true</literal>. This means that if the destination paint
device is smaller than the rect we write, it will automatically be enlarged to
fit every pixel we iterate over. Note that we have got a potential bug here:
if <varname>dst</varname> and <varname>src</varname> are not the same device,
then it is quite possible that the pixels returned by the iterators do not
correspond. For every position in the iterator, <varname>src</varname> may be,
for example, at 165,200, while <varname>dst</varname> could be at 20,8 &mdash;
and therefore the copy we perform below may distort the image...
</para></callout>
<callout arearefs="invert3"><para>
Want to know if a pixel is selected? That is easy &mdash; use the
<methodname>isSelected</methodname> method. But selectedness is not a binary
property of a pixel, a pixel can be half selected, barely selected or almost
completely selected. That value you can also got from the iterator. Selections
are actually a mask paint device with a range between 0 and 255, where 0 is
completely unselected and 255 completely selected. The iterator has two
methods: <methodname>isSelected()</methodname> and
<methodname>selectedNess()</methodname>. The first returns true if a pixel is
selected to any extent (i.e., the mask value is greater than 1), the other
returns the maskvalue.
</para></callout>
<callout arearefs="invert4"><para>
As noted above, this <literal>memcpy</literal> is a big bad bug...
<methodname>rawData()</methodname> returns the array of bytes which is the
current state of the pixel; <methodname>oldRawData()</methodname> returns the
array of bytes as it was before we created the iterator. However, we may be
copying the wrong pixel here. In actual practice, that will not happen too
often, unless <varname>dst</varname> already exists and is not aligned with
<varname>src</varname>.
</para></callout>
<callout arearefs="invert5"><para>
But this is correct: instead of figuring out which byte represents which
channel, we use a function supplied by all colorspaces to invert the current
pixel. The colorspaces have a lot of pixel operations you can make use of.
</para></callout>
</calloutlist>

<para>
This is not all there is to creating a filter. Filters have two other
important components: a configuration object and a configuration widget. The
two interact closely. The configuration widget creates a configuration object,
but can also be filled from a pre-existing configuration object. Configuration
objects can represtent themselves as XML and can be created from XML. That is
what makes adjustment layers possible.
</para>

<sect3 id="developers-plugins-filters-iterators">
<title>Iterators</title>

<para>
There are three types of iterators: 
</para>

<itemizedlist>
<listitem><para>Horizontal lines</para></listitem>
<listitem><para>Vertical lines</para></listitem>
<listitem><para>Rectangular iterors</para></listitem>
</itemizedlist>

<para>
The horizontal and vertical line iterators have a method to move the iterator
to the next row or column: <methodname>nextRow()</methodname> and
<methodname>nextCol()</methodname>. Using these is much faster than creating a
new iterator for every line or column.
</para><para>
Iterators are thread-safe in &chalk;, so it is possible to divide the work
over multiple threads. However, future versions of &chalk; will use the
<methodname>supportsThreading()</methodname> method to determine whether your
filter can be applied to chunks of the image (&ie;, all pixels modified
independently, instead of changed by some value determined from an examination
of all pixels in the image) and automatically thread the execution your
filter.
</para>
</sect3>

<sect3 id="developers-plugins-filters-kisfilterconfiguration">
<title><classname>KisFilterConfiguration</classname></title>

<para>
<classname>KisFilterConfiguration</classname> is a structure that is used to
save filter settings to disk, for instance for adjustment layers. The
scripting plugin uses the property map that’s at the back of
<classname>KisFilterConfigaration</classname> to make it possible to script
filters. Filters can provide a custom widget that &chalk; will show in the
filters gallery, the filter preview dialog or the tool option tab of the
paint-with-filters tool.
</para>
<para>
An example, taken from the oilpaint effect filter:
</para>
<programlisting>
class KisOilPaintFilterConfiguration : public KisFilterConfiguration
{

public:

   KisOilPaintFilterConfiguration(TQ_UINT32 brushSize, TQ_UINT32 smooth)
       : KisFilterConfiguration( "oilpaint", 1 )
       {
           setProperty("brushSize", brushSize);
           setProperty("smooth", smooth);
       };
public:

   inline TQ_UINT32 brushSize() { return getInt("brushSize"); };
   inline TQ_UINT32 smooth() {return getInt("smooth"); };

};
</programlisting>
</sect3>

<sect3 id="developers-plugins-filters-kisfilterconfigurationwidget">
<title><classname>KisFilterConfigurationWidget</classname></title>

<para>
Most filters can be tweaked by the user. You can create a configuration widget
that Chalk will use where-ever your filter is used. An example:
</para>

<para>
<screenshot>
<screeninfo>The <guilabel>Oilpaint</guilabel> dialog</screeninfo>
<mediaobject>
<imageobject>
<imagedata fileref="dialogs-oilpaint.png" format="PNG" />
</imageobject>
<textobject>
<phrase>The <guilabel>Oilpaint</guilabel> dialog</phrase>
</textobject>
<caption><para>The <guilabel>Oilpaint</guilabel> dialog</para></caption>
</mediaobject>
</screenshot>
</para>

<para>
Note that only the left-hand side of this dialog is your responsibility:
&chalk; takes care of the rest. There are three ways of going about creating
an option widget:
</para>

<itemizedlist>
<listitem><para>Use &TQt; Designer to create a widget base, and subclass it for
your filter</para></listitem>
<listitem><para>Use one of the simple widgets that show a number of sliders
for lists of integers, doubles or bools. These are useful if, like the above
screenshot, your filter can be configured by a number of integers, doubles or
bools. See the API dox for <classname>KisMultiIntegerFilterWidget</classname>,
<classname>KisMultiDoubleFilterWidget</classname> and
<classname>KisMultiBoolFilterWidget</classname>.</para></listitem>
<listitem><para>Hand-code a widget. This is not recommended, and if you do so
and want your filter to become part of &chalk;’s official release, then I’ll ask
you to replate your hand-coded widget with a &TQt; Designer
widget.</para></listitem>
</itemizedlist>

<para>
The oilpaint filter uses the multi integer widget:
</para>

<programlisting>
KisFilterConfigWidget * KisOilPaintFilter::createConfigurationWidget(TQWidget* parent, KisPaintDeviceSP /*dev*/)
{
   vKisIntegerWidgetParam param;
   param.push_back( KisIntegerWidgetParam( 1, 5, 1, i18n("Brush size"), "brushSize" ) );
   param.push_back( KisIntegerWidgetParam( 10, 255, 30, i18n("Smooth"), "smooth" ) );
   return new KisMultiIntegerFilterWidget(parent, id().id().ascii(), id().id().ascii(), param );
}

KisFilterConfiguration* KisOilPaintFilter::configuration(TQWidget* nwidget)
{
   KisMultiIntegerFilterWidget* widget = (KisMultiIntegerFilterWidget*) nwidget;
   if( widget == 0 )
   {
       return new KisOilPaintFilterConfiguration( 1, 30);
   } else {
       return new KisOilPaintFilterConfiguration( widget->valueAt( 0 ), widget->valueAt( 1 ) );
   }
}

std::list&lt;KisFilterConfiguration*&gt; KisOilPaintFilter::listOfExamplesConfiguration(KisPaintDeviceSP )
{
   std::list&lt;KisFilterConfiguration*&gt; list;
   list.insert(list.begin(), new KisOilPaintFilterConfiguration( 1, 30));
   return list;
}
</programlisting>

<para>
You can see how it works: fill a vector with your integer parameters and
create the widget. The <methodname>configuration()</methodname> method
inspects the widget and creates the right filter configuration object, in this
case, of course, <classname>KisOilPaintFilterConfiguration</classname>. The
<methodname>listOfExamplesConfiguration</methodname> method (which should be
renamed to correct English...) returns a list with example configuration
objects for the filters gallery dialog.
</para>
</sect3>

<sect3 id="developers-plugins-filters-conclusion">
<title>Filters conclusion</title>

<para>
There’s more to coding interesting filters, of course, but with this
explanation, the API documentation and access to our source code, you should
be able to get started. Don’t hesitate to contact the &chalk; developers on
IRC or on the mailing list.
</para>
</sect3>

</sect2>

<sect2 id="developers-plugins-tools">
<title>Tools</title>

<para>
Tools appear in &chalk;’s toolbox. This means that there is limited space for
new tools &mdash; think carefully whether a paint operation isn’t enough for
your purposes. Tools can use the mouse/tablet and keyboard in complex ways,
which paint operations cannot. This is the reason that Duplicate is a tool,
but airbrush a paint operation.
</para><para>
Be careful with static data in your tool: a new instance of your tool is
created for every input device: mouse, stylus, eraser, airbrush &mdash; whatever.
Tools come divided into logical groups:
</para>
<itemizedlist>
<listitem><para>shape drawing tools (circle, rect)</para></listitem>
<listitem><para>freehand drawing tools (brush)</para></listitem>
<listitem><para>transform tools that mess up the geometry of a
layer</para></listitem>
<listitem><para>fill tools (like bucket fill or gradient)</para></listitem>
<listitem><para>view tools (that don’t change pixels, but alter the way you
view the canvas, such as zoom)</para></listitem>
<listitem><para>select tools (that change the selection
mask)</para></listitem>
</itemizedlist>

<para>
The tool interface is described in the API documentation for
<classname>KisTool</classname>. There are three subclasses:
<classname>KisToolPaint</classname>, <classname>KisToolNonPaint</classname>
and <classname>KisToolShape</classname> (which is actually a subclass of
<classname>KisToolPaint</classname>) that specialize
<classname>KisTool</classname> for painting tasks (i.e., changing pixels) ,
non-painting tasks and shape painting tasks.
</para><para>
A tool has an option widget, just like filters. Currently, the option widgets
are shown in a tab in a dock window. We may change that to a strip under the
main menu (which then replaces the toolbar) for &chalk; 2.0, but for now,
design your option widget to fit in a tab. As always, it’s best to use &TQt;
Designer for the design of the option widget.
</para><para>
A good example of a tool is the star tool:
</para>

<screen>
kis_tool_star.cpp       Makefile.am   tool_star_cursor.png  wdg_tool_star.ui
kis_tool_star.h        Makefile.in   tool_star.h           
chalktoolstar.desktop  tool_star.cpp  tool_star.png         
</screen>

<para>
As you see, you need two images: one for the cursor and one for the toolbox.
<filename>tool_star.cpp</filename> is just the plugin loader, similar to what
we have seen above. The real meat is in the implementation:
</para>

<programlisting>
KisToolStar::KisToolStar()
   : KisToolShape(i18n("Star")),
     m_dragging (false),
     m_currentImage (0)
{
   setName("tool_star");
   setCursor(KisCursor::load("tool_star_cursor.png", 6, 6));
   m_innerOuterRatio=40;
   m_vertices=5;
}
</programlisting>

<para>
The constructor sets the internal name &mdash; which is not translated
&mdash; and the call to the superclass sets the visible name. We also load the
cursor image and set a number of variables.
</para>

<programlisting>
void KisToolStar::update (KisCanvasSubject *subject)
{
   KisToolShape::update (subject);
   if (m_subject)
       m_currentImage = m_subject->currentImg();
}
</programlisting>

<para>
The <methodname>update()</methodname> method is called when the tool is
selected. This is not a <classname>KisTool</classname> method, but a
<classname>KisCanvasObserver</classname> method. Canvas observers are notified
whenever something changes in the view, which can be useful for tools.
</para><para>
The following methods (<methodname>buttonPress</methodname>,
<methodname>move</methodname> and <methodname>buttonRelease</methodname>) are
called by &chalk; when the input device (mouse, stylus, eraser etc.) is
pressed down, moved or released. Note that you also get move events if the
mouse button isn’t pressed. The events are not the regular &TQt; events, but
synthetic &chalk; events because we make use of low-level trickery to get
enough events to draw a smooth line. By default, toolkits like &TQt; (and GTK)
drop events if they are too busy to handle them, and we want them all.
</para>

<programlisting>
void KisToolStar::buttonPress(KisButtonPressEvent *event)
{
   if (m_currentImage &amp;&amp; event->button() == LeftButton) {
       m_dragging = true;
       m_dragStart = event->pos();
       m_dragEnd = event->pos();
       m_vertices = m_optWidget->verticesSpinBox->value();
       m_innerOuterRatio = m_optWidget->ratioSpinBox->value();
   }
}

void KisToolStar::move(KisMoveEvent *event)
{
   if (m_dragging) {
       // erase old lines on canvas
       draw(m_dragStart, m_dragEnd);
       // move (alt) or resize star
       if (event->state() &amp; TQt::AltButton) {
           KisPoint trans = event->pos() - m_dragEnd;
           m_dragStart += trans;
           m_dragEnd += trans;
       } else {
           m_dragEnd = event->pos();
       }
       // draw new lines on canvas
       draw(m_dragStart, m_dragEnd);
   }
}

void KisToolStar::buttonRelease(KisButtonReleaseEvent *event)
{
   if (!m_subject || !m_currentImage)
       return;

   if (m_dragging &amp;&amp; event->button() == LeftButton) {
       // erase old lines on canvas
       draw(m_dragStart, m_dragEnd);
       m_dragging = false;

       if (m_dragStart == m_dragEnd)
           return;

       if (!m_currentImage)
           return;

       if (!m_currentImage->activeDevice())
           return;
       
       KisPaintDeviceSP device = m_currentImage->activeDevice ();;
       KisPainter painter (device);
       if (m_currentImage->undo()) painter.beginTransaction (i18n("Star"));

       painter.setPaintColor(m_subject->fgColor());
       painter.setBackgroundColor(m_subject->bgColor());
       painter.setFillStyle(fillStyle());
       painter.setBrush(m_subject->currentBrush());
       painter.setPattern(m_subject->currentPattern());
       painter.setOpacity(m_opacity);
       painter.setCompositeOp(m_compositeOp);
       KisPaintOp * op =
	   KisPaintOpRegistry::instance()->paintOp(m_subject->currentPaintop(), m_subject->currentPaintopSettings(), &amp;painter);
       painter.setPaintOp(op); // Painter takes ownership

       vKisPoint coord = starCoordinates(m_vertices, m_dragStart.x(), m_dragStart.y(), m_dragEnd.x(), m_dragEnd.y());

       painter.paintPolygon(coord);
       
       device->setDirty( painter.dirtyRect() );
       notifyModified();

       if (m_currentImage->undo()) {
           m_currentImage->undoAdapter()->addCommand(painter.endTransaction());
       }
   }
}
</programlisting>

<para>
The <methodname>draw()</methodname> method is an internal method of
<classname>KisToolStar</classname> and draws the outline of the star. We call
this from the <methodname>move()</methodname> method to give the user feedback
of the size and shape of their star. Note that we use the
<varname>TQt::NotROP</varname> raster operation, which means that calling
<methodname>draw()</methodname> a second time with the same start and end
point the previously drawn star will be deleted.
</para>

<programlisting>
void KisToolStar::draw(const KisPoint&amp; start, const KisPoint&amp; end )
{
   if (!m_subject || !m_currentImage)
       return;

   KisCanvasController *controller = m_subject->canvasController();
   KisCanvas *canvas = controller->kiscanvas();
   KisCanvasPainter p (canvas);
   QPen pen(TQt::SolidLine);

   KisPoint startPos;
   KisPoint endPos;
   startPos = controller->windowToView(start);
   endPos = controller->windowToView(end);

   p.setRasterOp(TQt::NotROP);

   vKisPoint points = starCoordinates(m_vertices, startPos.x(), startPos.y(), endPos.x(), endPos.y());

   for (uint i = 0; i &lt; points.count() - 1; i++) {
       p.drawLine(points[i].floorQPoint(), points[i + 1].floorQPoint());
   }
   p.drawLine(points[points.count() - 1].floorQPoint(), points[0].floorQPoint());

   p.end ();
}
</programlisting>

<para>
The <methodname>setup()</methodname> method is essential: here we create the
action that will be plugged into the toolbox so users can actually select the
tool. We also assign a shortcut key. Note that there’s some hackery going on:
remember that we create an instance of the tool for every input device. This
also means that we call <methodname>setup()</methodname> for every input
device and that means that an action with the same name is added several times
to the action collection. However, everything seems to work, so why worry?
</para>

<programlisting>
void KisToolStar::setup(TDEActionCollection *collection)
{
   m_action = static_cast&lt;TDERadioAction *&gt;(collection->action(name()));

   if (m_action == 0) {
       TDEShortcut shortcut(TQt::Key_Plus);
       shortcut.append(TDEShortcut(TQt::Key_F9));
       m_action = new TDERadioAction(i18n("&amp;Star"),
                                   "tool_star",
                                   shortcut,
                                   this,
                                   TQ_SLOT(activate()),
                                   collection,
                                   name());
       TQ_CHECK_PTR(m_action);

       m_action->setToolTip(i18n("Draw a star"));
       m_action->setExclusiveGroup("tools");
       m_ownAction = true;
   }
}
</programlisting>

<para>
The <methodname>starCoordinates()</methodname> method contains some funky math
&mdash; but is not too interesting for the discussion of how to create a tool
plugins.
</para>

<programlisting>
KisPoint KisToolStar::starCoordinates(int N, double mx, double my, double x, double y)
{
   double R=0, r=0;
   TQ_INT32 n=0;
   double angle;

   vKisPoint starCoordinatesArray(2*N);

   // the radius of the outer edges
   R=sqrt((x-mx)*(x-mx)+(y-my)*(y-my));

   // the radius of the inner edges
   r=R*m_innerOuterRatio/100.0;

   // the angle
   angle=-atan2((x-mx),(y-my));

   //set outer edges
   for(n=0;n&lt;N;n++){
       starCoordinatesArray[2*n] = KisPoint(mx+R*cos(n * 2.0 * M_PI / N + angle),my+R*sin(n *2.0 * M_PI / N+angle));
   }

   //set inner edges
   for(n=0;n&lt;N;n++){
       starCoordinatesArray[2*n+1] = KisPoint(mx+r*cos((n + 0.5) * 2.0 * M_PI / N + angle),my+r*sin((n +0.5) * 2.0 * M_PI / N + angle));
   }

   return starCoordinatesArray;
}
</programlisting>

<para>
The <methodname>createOptionWidget()</methodname> method is called to create
the option widget that &chalk; will show in the tab. Since there is a tool per
input device per view, the state of a tool can be kept in the tool. This
method is only called once: the option widget is stored and retrieved the next
time the tool is activated.
</para>

<programlisting>
TQWidget* KisToolStar::createOptionWidget(TQWidget* parent)
{
   TQWidget *widget = KisToolShape::createOptionWidget(parent);

   m_optWidget = new WdgToolStar(widget);
   TQ_CHECK_PTR(m_optWidget);

   m_optWidget->ratioSpinBox->setValue(m_innerOuterRatio);

   QGridLayout *optionLayout = new QGridLayout(widget, 1, 1);
   super::addOptionWidgetLayout(optionLayout);

   optionLayout->addWidget(m_optWidget, 0, 0);

   return widget;
}
</programlisting>

<sect3 id="developers-plugins-tools-conclusions">
<title>Tool Conclusions</title>

<para>
Tools are relatively simple plugins to create. You need to combine the
<classname>KisTool</classname> and <classname>KisCanvasObserver</classname>
interfaces in order to effectively create a tool. 
</para>

</sect3>

</sect2>

<sect2 id="developers-plugins-paintoperations">
<title>Paint operations</title>

<para>
PaintOps are one of the more innovative types of plugins in Chalk (together
with pluggable colorspaces). A paint operation defines how tools change the
pixels they touch. Airbrush, aliased pencil or antialiased pixel brush: these
are all paint operations. But you could &mdash; with a lot of work &mdash;
create a paintop that reads Corel Painter XML brush definitions and uses those
to determine how painting is done.
</para><para>
Paint operations are instantiated when a paint tool receives a
<literal>mouseDown</literal> event and are deleted when the mouseUp event is
received by a paint tool. In between, the paintop can keep track of previous
positions and other data, such as pressure levels if the user uses a tablet.
</para><para>
The basic operation of a paint operation is to change pixels at the cursor
position of a paint tool. That can be done only once, or the paint op can
demand to be run at regular intervals, using a timer. The first would be
useful for a pencil-type paint op, the second, of course,  for an
airbrush-type paintop.
</para><para>
Paintops can have a small configuration widget which is placed in a toolbar.
Thus, paintop configuration widgets need to have a horizontal layout of
widgets that are not higher than a toolbar button. Otherwise, &chalk; will
look very funny.
</para><para>
Let’s look at a simple paintop plugin, one that shows a little bit of
programmatic intelligence. First, in the header file, there’s a factory
defined. This factory creates a paintop when the active tool needs one:
</para>

<programlisting>
public:
   KisSmearyOpFactory() {}
   virtual ~KisSmearyOpFactory() {}

   virtual KisPaintOp * createOp(const KisPaintOpSettings *settings, KisPainter * painter);
   virtual KisID id() { return KisID("paintSmeary", i18n("Smeary Brush")); }
   virtual bool userVisible(KisColorSpace * ) { return false; }
   virtual TQString pixmap() { return ""; }

};
</programlisting>

<para>
The factory also contains the <classname>KisID</classname> with the public and
private name for the paintop &mdash; make sure your paintop’s private name
does not clash with another paintop! &mdash; and may optionally return a
pixmap. &chalk; can then show the pixmap together with the name for visual
identifcation of your paintop. For instance, a painter’s knife paintop would
have the image of such an implement.
</para><para>
The implementation of a paintop is very straightforward:
</para>

<programlisting>
KisSmearyOp::KisSmearyOp(KisPainter * painter)
   : KisPaintOp(painter)
{
}

KisSmearyOp::~KisSmearyOp()
{
}
void KisSmearyOp::paintAt(const KisPoint &amp;pos, const KisPaintInformation&amp; info)
{
</programlisting>

<para>
The <methodname>paintAt()</methodname> method really is where it’s at, with
paintops. This method receives two parameters: the current position (which is
in floats, not in whole pixels) and a
<classname>KisPaintInformation</classname> object. which contains the
pressure, x and y tilt, and movement vector, and may in the future be extended
with other information.
</para>

<programlisting>
   if (!m_painter->device()) return;

   KisBrush *brush = m_painter->brush();
</programlisting>

<para>
A <classname>KisBrush</classname> is the representation of a Gimp brush file:
that is a mask, either a single mask or a series of masks. Actually, we don’t
use the brush here, except to determine the <quote>hotspot</quote> under the
cursor.
</para>

<programlisting>
   Q_ASSERT(brush);

   if (!brush) return;

   if (! brush->canPaintFor(info) )
       return;

   KisPaintDeviceSP device = m_painter->device();
   KisColorSpace * colorSpace = device->colorSpace();
   KisColor kc = m_painter->paintColor();
   kc.convertTo(colorSpace);

   KisPoint hotSpot = brush->hotSpot(info);
   KisPoint pt = pos - hotSpot;

   // Split the coordinates into integer plus fractional parts. The integer
   // is where the dab will be positioned and the fractional part determines
   // the sub-pixel positioning.
   TQ_INT32 x, y;
   double xFraction, yFraction;

   splitCoordinate(pt.x(), &amp;x, &amp;xFraction);
   splitCoordinate(pt.y(), &amp;y, &amp;yFraction);

   KisPaintDeviceSP dab = new KisPaintDevice(colorSpace, "smeary dab");
   TQ_CHECK_PTR(dab);
</programlisting>

<para>
We don’t change the pixels of a paint device directly: instead we create a
small paint device, a dab, and composite that onto the current paint device.
</para>

<programlisting>
   m_painter->setPressure(info.pressure);
</programlisting>

<para>
As the comments say, the next bit code does some programmatic work to create
the actual dab. In this case, we draw a number of lines. When I am done with
this paintop, the length, position and thickness of the lines will be
dependent on pressure and paint load, and we’ll have create a stiff, smeary
oilpaint brush. But I haven’t had time to finish this yet.
</para>

<programlisting>
   // Compute the position of the tufts. The tufts are arranged in a line
   // perpendicular to the motion of the brush, i.e, the straight line between
   // the current position and the previous position.
   // The tufts are spread out through the pressure

   KisPoint previousPoint = info.movement.toKisPoint();
   KisVector2D brushVector(-previousPoint.y(), previousPoint.x());
   KisVector2D currentPointVector = KisVector2D(pos);
   brushVector.normalize();

   KisVector2D vl, vr;

   for (int i = 0; i &lt; (NUMBER_OF_TUFTS / 2); ++i) {
       // Compute the positions on the new vector.
       vl = currentPointVector + i * brushVector;
       KisPoint pl = vl.toKisPoint();
       dab->setPixel(pl.roundX(), pl.roundY(), kc);

       vr = currentPointVector - i * brushVector;
       KisPoint pr = vr.toKisPoint();
       dab->setPixel(pr.roundX(), pr.roundY(), kc);
   }

   vr = vr - vl;
   vr.normalize();
</programlisting>

<para>
Finally we blt the dab onto the original paint device and tell the painter
that we’ve dirtied a small rectangle of the paint device.
</para>

<programlisting>
   if (m_source->hasSelection()) {
       m_painter->bltSelection(x - 32, y - 32, m_painter->compositeOp(), dab.data(),
                               m_source->selection(), m_painter->opacity(), x - 32, y -32, 64, 64);
   }
   else {
       m_painter->bitBlt(x - 32, y - 32, m_painter->compositeOp(), dab.data(), m_painter->opacity(), x - 32, y -32, 64, 64);
   }

   m_painter->addDirtyRect(QRect(x -32, y -32, 64, 64));
}


KisPaintOp * KisSmearyOpFactory::createOp(const KisPaintOpSettings */*settings*/, KisPainter * painter)
{
   KisPaintOp * op = new KisSmearyOp(painter);
   TQ_CHECK_PTR(op);
   return op;
}
</programlisting>

<para>
That’s all: paintops are easy and fun!
</para>

</sect2>

<sect2 id="developers-plugins-viewplugins">
<title>View plugins</title>

<para>
View plugins are the weirdest of the bunch: a view plugin is an ordinary KPart
that can provide a bit of user interface and some functionality. For instance,
the histogram tab is a view plugin, as is the rotate dialog.
</para>

</sect2>

<sect2 id="developers-plugins-importexport">
<title>Import/Export filters</title>

<para>
&chalk; works with the ordinary &koffice; file filter architecture. There is a
tutorial, a bit old, but still useful, at: <ulink
url="http://koffice.org/developer/filters/oldfaq.php" />. It is probably best
to cooperate with the &chalk; team when developing file filters and do the
development in the &koffice; filter tree. Note that you can test your filters
without running &chalk; using the <command>koconverter</command> utility.
</para><para>
Filters have two sides: importing and exporting. These are usually two
different plugins that may share some code.
</para><para>
The important <filename>Makefile.am</filename> entries are:
</para>

<programlisting>
service_DATA = chalk_XXX_import.desktop chalk_XXX_export.desktop
servicedir = $(kde_servicesdir)
kdelnk_DATA = chalk_XXX.desktop
kdelnkdir = $(kde_appsdir)/Office
libchalkXXXimport_la_SOURCES = XXXimport.cpp
libchalkXXXexport_la_SOURCES = XXXexport.cpp
METASOURCES = AUTO
</programlisting>

<para>
Whether you are building an import filter or an export filter, your work always
boils down to implementing the following function:
</para>

<programlisting>
virtual KoFilter::ConversionStatus convert(const QCString&amp; from, const QCString&amp; to);
</programlisting>

<para>
It is the settings in the <literal role="extension">.desktop</literal> files
that determine which way a filter converts:
</para><para>
Import:
</para>

<programlisting>
X-TDE-Export=application/x-chalk
X-TDE-Import=image/x-xcf-gimp
X-TDE-Weight=1
X-TDE-Library=libchalkXXXimport
ServiceTypes=KOfficeFilter
</programlisting>

<para>
Export:
</para>

<programlisting>
X-TDE-Export=image/x-xcf-gimp
X-TDE-Import=application/x-chalk
ServiceTypes=KOfficeFilter
Type=Service
X-TDE-Weight=1
X-TDE-Library=libchalkXXXexport
</programlisting>

<para>
And yes, the mimetype chosen for the example is a hint. Please, pretty please,
implement an xcf filter?
</para>

<sect3 id="plugins-developers-importexport-import">
<title>Import</title>

<para>
The big problem with import filters is of course your code to read the data on
disk. The boilerplate for calling that code is fairly simple:
</para>

<note><para>Note: we really, really should find a way to enable &chalk; to keep
a file open and only read data on a as-needed basis, instead of copying the
entire contents to the internal paint device representation. But that would
mean datamanager backends that know about tiff files and so on, and is not
currently implemented. It would be ideal if some file filters could implement
a class provisionally named <classname>KisFileDataManager</classname>, create
an object of that instance with the current file and pass that to KisDoc.  But
&chalk; handles storage per layer, not per document, so this would be a hard
refactor to do.</para></note>

<programlisting>
KoFilter::ConversionStatus XXXImport::convert(const QCString&amp;, const QCString&amp; to)
{
   if (to != "application/x-chalk") <co id="import1" />
       return KoFilter::BadMimeType;

   KisDoc * doc = dynamic_cast&lt;KisDoc*&gt;(m_chain -> outputDocument()); <co id="import2" />
   KisView * view = static_cast&lt;KisView*&gt;(doc -> views().getFirst()); <co id="import3" />
   
   TQString filename = m_chain -> inputFile(); <co id="import4" />
   
   if (!doc)
       return KoFilter::CreationError;

   doc -> prepareForImport(); <co id="import5" />

   if (!filename.isEmpty()) {
   
       KURL url(filename);

       if (url.isEmpty())
           return KoFilter::FileNotFound;
           
       KisImageXXXConverter ib(doc, doc -> undoAdapter()); <co id="import6" />

       if (view != 0)
           view -> canvasSubject() ->  progressDisplay() -> setSubject(&amp;ib, false, true);

       switch (ib.buildImage(url)) <co id="import7" /> {
           case KisImageBuilder_RESULT_UNSUPPORTED:
               return KoFilter::NotImplemented;
               break;
           case KisImageBuilder_RESULT_INVALID_ARG:
               return KoFilter::BadMimeType;
               break;
           case KisImageBuilder_RESULT_NO_URI:
           case KisImageBuilder_RESULT_NOT_LOCAL:
               return KoFilter::FileNotFound;
               break;
           case KisImageBuilder_RESULT_BAD_FETCH:
           case KisImageBuilder_RESULT_EMPTY:
               return KoFilter::ParsingError;                
               break;
           case KisImageBuilder_RESULT_FAILURE:
               return KoFilter::InternalError;
               break;
           case KisImageBuilder_RESULT_OK:
               doc -> setCurrentImage( ib.image()); <co id="import8" />
               return KoFilter::OK;
           default:
               break;
       }

   }
   return KoFilter::StorageCreationError;
}
</programlisting>

<calloutlist>
<callout arearefs="import1"><para>This is supposed to be an importfilter, so
if it is not called to convert to a &chalk; image, then something is
wrong.</para></callout>
<callout arearefs="import2"><para>The filter chain already has created an
output document for us. We need to cast it to <classname>KisDocM</classname>,
because &chalk; documents need special treatment. It would not, actually, be
all that bad an idea to check whether the result of the cast is not 0, because
if it is, importing will fail.</para></callout>
<callout arearefs="import3"><para>If we call this filter from the GUI, we try
to get the view. If there is a view, the conversion code can try to update the
progressbar.</para></callout>
<callout arearefs="import4"><para>The filter has the filename for our input
file for us.</para></callout>
<callout arearefs="import5"><para><classname>KisDoc</classname> needs to be
prepared for import. Certain settings are initialized and undo is disabled.
Otherwise you could undo the adding of layers performed by the import filter
and that is weird behaviour.</para></callout>
<callout arearefs="import6"><para>I have chosed to implement the actual
importing code in a separate class that I instantiate here. You can also put
all your code right in this method, but that would be a bit
messy.</para></callout>
<callout arearefs="import7"><para>My importer returns a statuscode that I
can then use to set the status of the import filter. &koffice; takes care of
showing error messages.</para></callout>
<callout arearefs="import8"><para>If creating the
<classname>KisImage</classname> has succeeded we set the document's current
image to our newly created image. Then we are done: <literal>return
KoFilter::OK;</literal>.</para></callout>
</calloutlist>

</sect3>

</sect2>

</sect1>