/**
\mainpage Okular, the unified document viewer
\section okular_overview Overview
- \ref okular_history
- \ref okular_design
- \ref okular_generators
- Website
\authors Tobias König
\licenses \lgpl
\page okular_history Historical background
Okular is the successor of kpdf, the PDF viewer in KDE 3.
kpdf was refactored and extended in a Google Summer of Code project to support not only
viewing PDF but also other types of document, e.g. PostScript files, images and many more.
\page okular_design The Design of Okular
To support a wide range of document formats, Okular was designed in a modular way, so you
have the following components:
\li \ref Shell
\li \ref Part
\li \ref Okular::Document Class
\li \ref Okular::Generator
The shell is the application which is started by the user as standalone application and
which embedds the part. The part contains all GUI elements of Okular, for example the
content list, the bookmark manager, menus and the graphical view of the document class.
The document class is an abstract presentation of the document content. It contains information
about every page of the document, its size, orientation etc.
But somehow the document class must retrieve these information from the various types of documents.
This is the task of the Generators. Generators are plugins which are loaded at runtime and which
have the knowledge about the internal structure of the different document types.
They extract the needed information from the documents, convert the data into a common format and
pass them to the document class.
Currently Generators for the following document types are available:
\li Portable Document Format (PDF)
\li PostScript
\li Device Independent Format (DVI)
\li DeJaVu Format
\li Comic Books
\li Images (JPEG, PNG, GIF, and many more)
\li TIFF Image Format
\li FictionBook Format
\li Plucker Format
\li OpenDocument Text Format
\li Microsofts CHM Format
\li Microsofts XML Document Format
Now the questions is how can these various formats be represented in a unified way?
Okular provides features like rotation, text search and extraction, zooming and many more, so how
does it match with the different capabilities of the formats?
\section okular_design_basics Basics of Generators
Lets start with the smallest commonness of all document formats:
\li they have pages (one ore more) of a given size
\li pages can be represented as pictures
So the first thing every Generator must support is to return the number of pages of a document.
Furthermore it must be able to return the picture of a page at a requested size.
For vector based document formats (e.g. PDF or PostScript) the Generators can render the page for
the requested size, for static documents formats (e.g. images), the Generator must scale the
content according to the requested size, so when you zoom a page in Okular, the Generators are
just asked to return the page for the zoomed size.
When the document class has retrieved the page pictures from the Generators, it can do further
image manipulation on it, for example rotating them or applying fancy effects.
\section okular_design_text_support Generators with Text support
Some document formats however support more functionality than just representing a page as an image.
PDF, PostScript, DVI and DeJaVu for example contains a machine readable representation of the
included text. For those document formats Okular provides additional features like text search,
text extraction and text selection.
How is that supported by the Generators?
To access the text from the documents the generators must extract it somehow and make it available
to the document class. However for the text selection feature the document class must also know where
the extracted text is located on the page. For a zoom factor of 100% the absolute position of
the text in the document can be used, however for larger or smaller zoom factors the position
must be recalculated. To make this calculation as easy as possible, the Generators return an
abstract represtentation (\ref Okular::TextPage) of the text which includes every character together
with its normalized position. Normalized means that the width and height of the page is
in the range of 0 to 1, so a character in the middle of the page is at x=0.5 and y=0.5.
So when you want to know where this character is located on the page which is zoomed at 300%, you just
multiply the position by 3 * page width (and page height) and get the absolute position for this zoom level.
This abstract text representation also allows an easy rotation of the coordinates, so that text selection
is available on rotated pages as well.
\section okular_design_meta_information Meta Information
Most documents have additional meta information:
\li Name of the author
\li Date of creation
\li Version number
\li Table of Content
\li Bookmarks
\li Annotations
These information can be retrieved by the generator as well and will be shown by Okular.
\page okular_generators How to implement a Generator
The power of Okular is its extensibility by Generator plugins. This section will describe how to
implement your own plugin for a new document type.
\li \ref okular_generators_basic
\li \ref okular_generators_with_text
\li \ref okular_generators_threaded
\li \ref okular_generators_extended
\section okular_generators_basic A Basic Generator
To provide a short overview and don't reimplementing an existing generator we'll work on a Generator
for the Magic document format, a non existing, pure virtual format :)
Lets assume we have some helper class (MagicDocument) which provides the following functionality for this
document format:
\li Loading a document
\li Retrieving number of pages
\li Returning a fixed size picture representation of a page
The class API looks like this
\code
class MagicDocument
{
public:
MagicDocument();
~MagicDocument();
bool loadDocument( const QString &fileName );
int numberOfPages() const;
QSize pageSize( int pageNumber ) const;
QImage pictureOfPage( int pageNumber ) const;
private:
...
};
\endcode
The methods should be self explaining, loadDocument() loads a document file and returns false on error,
numberOfPages() returns the number of pages, pageSize() returns the size of the page and pictureOfPage()
returns the picture representation of the page.
Our first version of our Generator is a basic one which just provides page pictures to the document class.
The API of the Generator looks like the following:
\code
#include "magicdocument.h"
#include
class MagicGenerator : public Okular::Generator
{
public:
MagicGenerator();
~MagicGenerator();
bool loadDocument( const QString &fileName, QVector &pages );
bool closeDocument();
bool canGeneratePixmap() const;
void generatePixmap( Okular::PixmapRequest *request );
private:
MagicDocument mMagicDocument;
};
\endcode
The implementation of the Generator looks like this:
\code
#include
#include "magicgenerator.h"
OKULAR_EXPORT_PLUGIN(MagicGenerator)
MagicGenerator::MagicGenerator()
: Okular::Generator()
{
}
MagicGenerator::~MagicGenerator()
{
}
bool MagicGenerator::loadDocument( const QString &fileName, QVector &pages )
{
if ( !mMagicDocument.loadDocument( fileName ) ) {
emit error( i18n( "Unable to load document" ), -1 );
return false;
}
pagesVector.resize( mMagicDocument.numberOfPages() );
for ( int i = 0; i < mMagicDocument.numberOfPages(); ++i ) {
const QSize size = mMagicDocument.pageSize( i );
Okular::Page * page = new Okular::Page( i, size.width(), size.height(), Okular::Rotation0 );
pages[ i ] = page;
}
return true;
}
bool MagicGenerator::closeDocument()
{
return true;
}
bool MagicGenerator::canGeneratePixmap() const
{
return true;
}
void MagicGenerator::generatePixmap( Okular::PixmapRequest *request )
{
QImage image = mMagicDocument.pictureOfPage( request->pageNumber() );
image = image.scaled( request->width(), request->height(), Qt::IgnoreAspectRatio, Qt::SmoothTransformation );
request->page()->setPixmap( request->id(), new QPixmap( QPixmap::fromImage( image ) ) );
signalPixmapRequestDone( request );
}
\endcode
As you can see implementing a basic Generator is quite easy. The loadDocument() method opens the document file
and extracts the number of pages. For every page in the document it adds an Okular::Page object to the pages vector
which is passed in as method argument. Each page is initialized with its page number, width, height and initial rotation.
These page objects will be stored in the document object and act as a container for the picture representation
of the pages. This code is the same for nearly every Generator. On an failure the error() signal can be emitted
to inform the user about the issue. This code is the same for nearly every Generator.
In the closeDocument() method you should close the document and free all resources you have allocated in openDocument().
Now we come to the picture creation methods. The canGeneratorPixmap() method returns whether the Generator is currently
able to handle a new pixmap generation request. For a simple Generator like our one that's always the case as it works
linear, however a multithreaded Generator might return false here if it is still waiting for one of its working
threads to finish. In this case the document class will try to request the pixmap later again.
The generatePixmap() method does the actual fetching of the picture for a page. The page number, requested width and
height of the page is encapsulated in the passed Okular::PixmapRequest object.
So the task of the Generator is to create a pixmap of the requested page in the requested size and then store this
pixmap in the Okular::Page object which is associated with the page request.
When this task is finished, the Generator has to call signalPixmapRequestDone() with the page request object
as argument. This extra call is needed to allow the Generator to use signals and slots internally and create the
pixmap asynchronously.
So now you have the code of a working Okular Generator, the next step is to tell Okular about the new plugin.
Like in other places in KDE that is done by .desktop files, which are installed to the services directory.
Every Generator needs 3 .desktop files:
\li libokularGenerator_<name>.desktop
\li okularApplication_<name>.desktop
\li okular<name>.desktop
where <name> should be the name of the document format. So for our Magic Document Generator we
create the following 3 files:
\li libokularGenerator_magic.desktop
\li okularApplication_magic.desktop
\li okularMagic.desktop
with the following content:
\verbatim
[Desktop Entry]
Encoding=UTF-8
Type=Service
Name=Magic Document
Name[x-test]=xxMagic Documentxx
Comment=Magic Document backend for okular
Comment[x-test]=xxMagic Document backend for okularxx
ServiceTypes=okular/Generator
MimeType=application/x-magic
X-KDE-Library=libokularGenerator_magic
X-KDE-Priority=1
X-KDE-okularAPIVersion=1
X-KDE-okularHasInternalSettings=false
\endverbatim
The first 6 fields are standard .desktop entries, the fields afterwards have a special meaning to Okular
\li ServiceType Must be 'okular/Generator' for all Okular Generator Plugins
\li MimeType The mimetype or list of mimetypes of the supported document format(s)
\li X-KDE-Library The name of the plugin library
\li X-KDE-Priority When multiple Generators for the same mimetype exists, the one with the highest priority is used
\li X-KDE-okularAPIVersion The version of the Generator Plugin API ('1' currently)
\li X-KDE-okularHasInternalSettings Is 'true' when the Generator provides configuration dialogs
The second .desktop file has the following content:
\verbatim
[Desktop Entry]
Encoding=UTF-8
MimeType=application/x-magic;
Terminal=false
Name=okular
Name[x-test]=xxokularxx
GenericName=Document Viewer
GenericName[x-test]=xxDocument Viewerxx
Exec=okular %U %i -caption "%c"
Icon=okular
Type=Application
DocPath=okular/index.html
InitialPreference=7
Categories=Qt;KDE;Graphics;Viewer;
NoDisplay=true
\endverbatim
You can use the file as it is, you just have to adapt the mimetype. This file is needed to allow Okular
to handle multiple mimetypes.
The third .desktop file looks like this:
\verbatim
[Desktop Entry]
Encoding=UTF-8
Icon=okular
Name=okular
Name[x-test]=xxokularxx
ServiceTypes=KParts/ReadOnlyPart
X-KDE-Library=libokularpart
Type=Service
MimeType=application/x-magic
\endverbatim
You can use the file as it is as well, you just have to adapt the mimetype. This file is needed to allow
the Okular part to handle multiple mimetypes.
The last piece you need for a complete Generator is a CMakeLists.txt which compiles and installs the
Generator. Our CMakeLists.txt looks like the following:
\verbatim
macro_optional_find_package(Okular)
include_directories( ${OKULAR_INCLUDE_DIR} ${KDE4_INCLUDE_DIR} ${QT_INCLUDES} )
########### next target ###############
set( okularGenerator_magic_SRCS generator_magic.cpp )
kde4_automoc( ${okularGenerator_magic_SRCS} )
kde4_add_plugin( okularGenerator_magic WITH_PREFIX ${okularGenerator_magic_SRCS} )
target_link_libraries( okularGenerator_magic ${OKULAR_LIBRARIES} ${KDE4_KDEUI_LIBS} )
install( TARGETS okularGenerator_magic DESTINATION ${PLUGIN_INSTALL_DIR} )
########### install files ###############
install( FILES libokularGenerator_magic.desktop okularMagic.desktop DESTINATION ${SERVICES_INSTALL_DIR} )
install( FILES okularApplication_magic.desktop DESTINATION ${XDG_APPS_DIR} )
\endverbatim
The macro_optional_find_package(Okular) call is required to make the ${OKULAR_INCLUDE_DIR} and ${OKULAR_LIBRARIES}
variables available.
Now you can compile the Generator plugin and install it. After a restart of Okular the new plugin is available
and you can open Magic documents.
\section okular_generators_with_text A Generator with TextPage support
In this section we want to extend our Generator to support text search, text extraction and selection
as well. As mentioned in \ref okular_design_text_support, the Generator must provide an Okular::TextPage
object for every page which contains readable text.
Since we use the helper class MagicDocument to read the data from the document we have to extend it first,
so the new API looks as the following:
\code
class MagicDocument
{
public:
MagicDocument();
~MagicDocument();
bool loadDocument( const QString &fileName );
int numberOfPages() const;
QSize pageSize( int pageNumber ) const;
QImage pictureOfPage( int pageNumber ) const;
class TextInfo
{
public:
typedef QList List;
QChar character;
qreal xPos;
qreal yPos;
qreal width;
qreal height;
};
TextInfo::List textOfPage( int pageNumber );
private:
...
};
\endcode
MagicDocument has the new internal class TextInfo now, which contains a character and
its absolute position on a page. Furthermore MagicDocument provides a method textOfPage()
which returns a list of all TextInfo objects for a page.
That's really an optimistic API, in reality it is sometimes quite hard to find out
the position of single characters in a document format.
With the extension of our helper class we can continue on extending our Generator now:
\code
#include "magicdocument.h"
#include
class MagicGenerator : public Okular::Generator
{
public:
MagicGenerator();
~MagicGenerator();
bool loadDocument( const QString &fileName, QVector &pages );
bool closeDocument();
bool canGeneratePixmap() const;
void generatePixmap( Okular::PixmapRequest *request );
virtual bool canGenerateTextPage() const;
virtual void generateTextPage( Okular::Page *page, enum Okular::GenerationType type = Okular::Synchronous );
private:
MagicDocument mMagicDocument;
};
\endcode
We have extended the MagicGenerator class by two methods canGenerateTextPage() and generateTextPage().
The first method is equal to canGeneratePixmap(), it returns whether the Generator is currently able to
handle a new text page generation request. For linear Generators that should be always the case, however
when the generation is done in a separated worker thread, this method might return false.
In this case the document class will try to request the text page later again.
The second method will generate the Okular::TextPage object for the passed page. Depending on the capabilities
of the Generator and the passed type parameter that is done synchronously or asynchronously.
Let us take a look at the implementation of these methods in our MagicGenerator:
\code
#include
...
MagicGenerator::MagicGenerator()
{
setFeature( TextExtraction );
}
bool MagicGenerator::canGenerateTextPage() const
{
return true;
}
void MagicGenerator::generateTextPage( Okular::Page *page, enum Okular::GenerationType )
{
MagicDocument::TextInfo::List characters = mMagicDocument.textOfPage( page->number() );
if ( characters.isEmpty() )
return;
Okular::TextPage *textPage = new Okular::TextPage;
for ( int i = 0; i < characters.count(); ++i ) {
qreal left = characters[ i ].xPos / page->width();
qreal top = characters[ i ].yPos / page->height();
qreal right = (characters[ i ].xPos + characters[ i ].width) / page->width();
qreal bottom = (characters[ i ].yPos + characters[ i ].height) / page->height();
textPage->append( characters[ i ].character,
new Okular::NormalizedRect( left, top, right, bottom ) );
}
page->setTextPage( textPage );
}
\endcode
As you can see the generateTextPage method just iterates over the list of characters returned
by our MagicDocument helper class and adds the character and its normalized bounding rect to
the Okular::TextPage object. At the end the text page is assigned to the page. We don't pay
attention to the GenerationType parameter here, if your Generator want to use threads, it should
check here whether the request shall be done asynchronously or synchronously and start the generation
according to that. Additionally we have to tell the Okular::Generator base class that we support
text handling by setting this flag in the constructor.
In this state we can now search, select and extract text from Magic documents.
\section okular_generators_threaded A Generator with Thread support
Sometimes it makes sense to do the generation of page pictures or text pages asynchronously to
improve performance and don't blocking the user interface. This can be done in two ways, either
by using signals and slots or by using threads. Both have there pros and cons:
- Signals and Slots
- Pro: Can be used with backend libraries which are not thread safe
- Con: Sometime difficult to implement
- Threads
- Pro: Easy to implement as you can make synchronous calls to the backend libraries
- Con: Backend libraries must be thread safe and you must prevent race conditions by using mutexes
The signal and slots approach can be achieved with a normal Generator by calling Okular::Generator::signalPixmapRequestDone()
from a slot after pixmap generation has been finished.
When using threads you should use a slightly different API, which hides most of the thread usage, to make
implementing as easy as possible.
Let's assume the pictureOfPage() and textOfPage methods in our MagicDocument helper class are thread safe,
so we can use them in a multithreaded environment.
So nothing prevents us from changing the MagicGenerator to use threads for better performance.
The new MagicGenerator API looks like the following:
\code
#include "magicdocument.h"
#include
class MagicGenerator : public Okular::Generator
{
public:
MagicGenerator();
~MagicGenerator();
bool loadDocument( const QString &fileName, QVector &pages );
bool closeDocument();
proteced:
virtual QImage image( Okular::PixmapRequest *request );
virtual Okular::TextPage* textPage( Okular::Page *page );
private:
MagicDocument mMagicDocument;
};
\endcode
As you can see the canGeneratePixmap() generatePixmap(), canGenerateTextPage() and generateTextPage() methods have
been removed and replaced by the image() and textPage() methods.
Before explaining why, we'll take a look at the implementation:
\code
MagicGenerator::MagicGenerator()
{
setFeature( TextExtraction );
setFeature( Threaded );
}
QImage MagicGenerator::image( Okular::PixmapRequest *request )
{
QImage image = mMagicDocument.pictureOfPage( request->pageNumber() );
return image.scaled( request->width(), request->height(), Qt::IgnoreAspectRatio, Qt::SmoothTransformation );
}
Okular::TextPage* textPage( Okular::Page *page )
{
MagicDocument::TextInfo::List characters = mMagicDocument.textOfPage( page->number() );
if ( characters.isEmpty() )
return 0;
Okular::TextPage *textPage = new Okular::TextPage;
for ( int i = 0; i < characters.count(); ++i ) {
qreal left = characters[ i ].xPos / page->width();
qreal top = characters[ i ].yPos / page->height();
qreal right = (characters[ i ].xPos + characters[ i ].width) / page->width();
qreal bottom = (characters[ i ].yPos + characters[ i ].height) / page->height();
textPage->append( characters[ i ].character,
new Okular::NormalizedRect( left, top, right, bottom ) );
}
return textPage;
}
\endcode
So the first obviously thing is that both methods return a value instead of modifying the page directly.
The reason for this is that both methods are executed in its own thread, so the code executed in them can
block as long as it wants, it won't block the GUI anyway. Additionally we have to tell the Okular::Generator
base class that we can handle threads by setting the flag in the constructor.
With only a small change we made our MagicGenerator multithreaded now!
\section okular_generators_extended An Extended Generator
Now we want to create a new generator with some additional functionality:
\li Support for document information (author, creation date etc.)
\li Support for a table of content
\li Support for printing the document
\li Support for exporting the document as text
The new Generator shall be able to handle HTML documents. We choose this format as example, because
we can use QTextDocument to load, render and print a HTML page, so a lot of code can be reused.
The API of our HTMLGenerator looks like the following:
\code
#include
#include
class HTMLGenerator : public Okular::Generator
{
public:
HTMLGenerator();
~HTMLGenerator();
bool loadDocument( const QString &fileName, QVector &pages );
bool closeDocument();
bool canGeneratePixmap() const;
void generatePixmap( Okular::PixmapRequest *request );
virtual const Okular::DocumentInfo* generateDocumentInfo();
virtual const Okular::DocumentSynopsis* generateDocumentSynopsis();
virtual bool print( KPrinter &printer );
virtual Okular::ExportFormat::List exportFormats() const;
virtual bool exportTo( const QString &fileName, const Okular::ExportFormat &format );
private:
QTextDocument *mTextDocument;
Okular::DocumentInfo mDocumentInfo;
Okular::DocumentSynopsis mDocumentSynopsis;
};
\endcode
The Generator doesn't support text search and selection, as the code would be quite complex, we'll show
how to do it in the next chapter \ref okular_generators_textdocument anyway.
As you can see we have 5 new methods in the class:
\li generateDocumentInfo() Creates an Okular::DocumentInfo (which is infact a QDomDocument)
which contains document information like author, creation time etc.
\li generateDocumentSynopsis() Creates an Okular::DocumentSynopsis (which is a QDomDocument as well)
which contains the table of content.
\li print() Prints the document to the passed printer.
\li exportFormats() Returns the supported export formats.
\li exportTo() Exports the document to the given file in the given format.
Now that you know what the methods are supposed to do, let's take a look at the implementation:
\code
#include
#include
#include
#include
#include
#include "htmlgenerator.h"
OKULAR_EXPORT_PLUGIN(HTMLGenerator)
HTMLGenerator::HTMLGenerator()
: Okular::Generator(),
mTextDocument( 0 )
{
}
HTMLGenerator::~HTMLGenerator()
{
delete mTextDocument;
}
bool HTMLGenerator::loadDocument( const QString &fileName, QVector &pages )
{
QFile file( fileName );
if ( !file.open( QIODevice::ReadOnly ) ) {
emit error( i18n( "Unable to open file" ), -1 );
return false;
}
const QString data = QString::fromUtf8( file.readAll() );
file.close();
mTextDocument = new QTextDocument;
mTextDocument->setHtml( data );
mTextDocument->setPageSize( QSizeF( 600, 800 ) );
pages.resize( mTextDocument->pageCount() );
for ( int i = 0; i < mTextDocument->pageCount(); ++i ) {
Okular::Page * page = new Okular::Page( i, 600, 800, Okular::Rotation0 );
pages[ i ] = page;
}
mDocumentInfo.set( "author", "Tobias Koenig", i18n( "Author" ) );
mDocumentInfo.set( "title", "The Art of Okular Plugin Development", i18n( "Title" ) );
Okular::DocumentViewport viewport = ... // get the viewport of the chapter
QDomElement item = mDocumentSynopsis.createElement( "Chapter 1" );
item.setAttribute( "Viewport", viewport.toString() );
mDocumentSynopsis.appendChild( item );
viewport = ... // get the viewport of the subchapter
QDomElement childItem = mDocumentSynopsis.createElement( "SubChapter 1.1" );
childItem.setAttribute( "Viewport", viewport.toString() );
item.appendChild( childItem );
return true;
}
bool HTMLGenerator::closeDocument()
{
delete mTextDocument;
mTextDocument = 0;
return true;
}
bool HTMLGenerator::canGeneratePixmap() const
{
return true;
}
void HTMLGenerator::generatePixmap( Okular::PixmapRequest *request )
{
QPixmap *pixmap = new QPixmap( request->width(), request->height() );
pixmap->fill( Qt::white );
QPainter p;
p.begin( pixmap );
qreal width = request->width();
qreal height = request->height();
p.scale( width / 600, height / 800 );
const QRect rect( 0, request->pageNumber() * 800, 600, 800 );
p.translate( QPoint( 0, request->pageNumber() * -800 ) );
d->mDocument->drawContents( &p, rect );
p.end();
request->page()->setPixmap( request->id(), pixmap );
signalPixmapRequestDone( request );
}
const Okular::DocumentInfo* HTMLGenerator::generateDocumentInfo()
{
return &mDocumentInfo;
}
const Okular::DocumentSynopsis* HTMLGenerator::generateDocumentSynopsis()
{
if ( !mDocumentSynopsis.hasChildNodes() )
return 0;
else
return &mDocumentSynopsis;
}
bool HTMLGenerator::print( KPrinter &printer )
{
QPainter p( &printer );
for ( int i = 0; i < mTextDocument->pageCount(); ++i ) {
if ( i != 0 )
printer.newPage();
QRect rect( 0, i * 800, 600, 800 );
p.translate( QPoint( 0, i * -800 ) );
mTextDocument->drawContents( &p, rect );
}
}
Okular::ExportFormat::List HTMLGenerator::exportFormats() const
{
return Okular::ExportFormat::standardFormat( Okular::ExportFormat::PlainText );
}
bool HTMLGenerator::exportTo( const QString &fileName, const Okular::ExportFormat &format )
{
QFile file( fileName );
if ( !fileName.open( QIODevice::WriteOnly ) ) {
emit error( i18n( "Unable to open file" ), -1 );
return false;
}
if ( format.mimeType()->name() == QLatin1String( "text/plain" ) )
file.writeBlock( mTextDocument->toPlainText().toUtf8() );
file.close();
return true;
}
\endcode
Let's take a closer look at the single methods. In the loadDocument() method we try to open the
passed file name and read all the content into the QTextDocument object. By calling
QTextDocument::setPageSize(), the whole document is divided into pages of the given size.
In the next step we create Okular::Page objects for every page in the QTextDocument and fill
the pages vector with them.
Afterwards we fill our Okular::DocumentInfo object with data. Since extracting the HTML meta data
would need a lot of code we work with static data here. [to be continued]
*/
// DOXYGEN_EXCLUDE = conf generators shell ui