Design for a possible reimplementation of the KDE help center
=============================================================

Preludium
---------
This document presents an alternative design for a 'help center' applicaiton
in KDE. Lines which start with a # are supposed to be thoughts I had while
writing this, much like the stuff you write on the side of a page when reading
a book.

Lines starting with ## were added by my as further comments - Cornelius

And I'll have the ### lines - Lauri

General
-------
- main() instantiates a KHC::Application
- KHC::Application() deals with parsing the commandline parameters and
  instantiates a KHC::MainWindow
- KHC::MainWindow creates the main UI, setting up actions, using a QSplitter
  as it's mainwidget to separate a KHC::Navigator at the left from a KHC::View
  at the right

That's the simple part. ;-)

## Apparently already done ;-)

KHC::Navigator
--------------

KHC::Navigator inherits QTabWidget and provides, on two tabs, a
KHC::ContentsTab object and a KHC::SearchTab object.

## KHC::Navigator shouldn't inherit from QTabWidget. This limits flexibility.
## It can create a QTabWidget instance as aggregate just as well.

# I fear premature generalization ("We could need that bit of flexibility one
# day), aggregation adds a level of indirection through a pointer variable as
# well. I would prefer not making the system more complex as long as we cannot
# predict changes which justify doing so.

1.) KHC::ContentsTab provides the following entires:
	- Welcome to TDE
	- KDE user's manual
	- KDE FAQ
	- Contact information
	- Supporting KDE

# Should we create an extra item for these five and put them in there?
# Something like "General KDE" or so? OTOH that makes them less visible, and
# these are really ought to be seen. - Frerich

## The items are ok, in principle, but we should have a look at the content of
## the documents they point at. This document could benefit from some attention.

### Yes, they would.  Also, there are license issues with one of them.
### I'd personally like to do an entire rewrite of the User Manual,
### without GPL encumbrance and sans the content that hasn't changed since
### KDE 1.x days.  The odds of me getting this done before KDE 3.1, slim to fair.


	- Application manuals
	- Tutorials
	- UNIX man pages
	- UNIX info pages
	- Glossary

# Do we really need this "Tutorials" item at all? right now it holds only two
# items, perhaps we can get rid of it. - Frerich

## Yes, please.

### There should be a "General" area, where documentation that isn't
### attached directly to an application can go.  Tutorials might not be
### the best name for it I agree, but there is now some further content to
### add (the DCOP tutorial, for example, or any of the numerous tutorials
### on the websites, documenting things that aren't in the handbooks.q

# Alright, after some talk on IRC this structure evolved:
#
# - Tasks - contains short, three to four paragraph documents about how to
#	        solve an everyday task, examples:
#		Browsing the web
#		Send and receive email
#		How to view images
#		Playing sound files
#		Installing new KDE themes
#		How to configure KDE fonts
#		Getting in touch with KDE contributors
#		Supporting the KDE team
#
# - Guides - slightly longer, Mini-HOWTO style guides (about three to four
#            pages long, perhaps) which talk about tackling jobs which don't
#			 occur very often, examples:
### I don't know about limiting the length.  Some of these topics can stand
### a much longer document, but one of the things that differentiates them 
### from the references is that they are not specific to a single application, 
### nor are they complete references in the manner of the "KDE User Guide"
### Specificaly, the dcop tutorial we have is about 15 pages already, but if 
### the user is interested in the topic, that isn't over much, and it's full of
### examples
#		How to debug KDE programs
#		Sending useful KDE bug reports
#		Extending KDE's service menus
#		Taking advantage of KDE's DCOP facilities
#		Creating panel applets
#		Phrasing questions most effectively
#
# - References - references. :-)
#		KDE API reference
#		KDE application manuals
#		Info pages
#		Man pages
#		FAQ
#		User's manual
#
# - Glossary - same as always.
#		- By topic
#		- Alphabetically
#
# My primary argument for such a structure is that it resembles a
# task-oriented interface much more closely than the simple list of
# application manuals. Imagine a user new to KDE who has a fairly precise
# description of what he's trying to do in mind (think "I want to view an
# image file") but no idea what tool to use for the job. The current list of
# application manuals requires the user to browse all the manuals which seem
# relevant, searching for the information he seeks. A task-oriented list
# solves that issue.
# This effectively enables people new to KDE in less time to become productive
# (a task-oriented list isn't so useful for peoplew ho are familiar with KDE's
# applications, of course).
# Implementation-wise, we should perhaps stop using a TDEListView and use a
# KOffice-style component selection widget like koshell has at the left?

The first five items are generated by KHC::Navigator itself and are direct
links to KDE documentations. The work of generating each of the last four
items is (with one exception) delegated to four helper classes, which inherit a
'KHC::TreeBuilder' class which has the following interface:

class KHC::TreeBuilder
	virtual void build( TDEListViewItem *parent ) = 0;

## What about the trees generated as children of the contents list view?

# Oops, that's a typo, what you mean is what I originally intented: a
# TreeBuilder should take a 'TDEListView' as it's parent, subclasses can then
# overload that method (such as the KHC::TOCBuilder which will want to provide
# a build( TDEListViewItem *parent ) method).

# This concept of using a TreeBuilder baseclass might make it possible to turn
# all the classes which use that interface into plugins. That way we could
# e.g. have a ScrollKeeper plugin. - Frerich

## What exactly do you mean by plugin? A shared library loaded at run time or
## the desktop file based insertion of documents into the help center?

# The former.

The classes which inherit this interface are:
	- KHC::ManualTreeBuilder: responsible for generating the tree below the
	  "Application manuals" item
	- KHC::TOCBuilder: responsible for generating a TOC tree below each of the
	  manual trees items, so that you can choose Application
	  Manuals->Editors->KWrite->Using KWrite->Menu bar transparently. This is
	  the only builder which is not instantiated by KHC::ContentsTab but
	  instead instantiated by KHC::ManualTreeBuilder
	- KHC::TutorialTreeBuilder: responsible for generating the tree below the
	  "Tutorials" item
	- KHC::ManTreeBuilder: responsible for building the tree below the "UNIX
	  man pages" item
	- KHC::InfoTreeBuilder: responsible for building the tree below the "UNIX
	  info pages" item
	- KHC::GlossaryTreeBuilder: guess what

## - KHC::ScrollkeeperTreeBuilder

## It's certainly a good idea to move stuff like the info and man pages and
## scrollkeeper support to its own classes. What  I consider as important is
## that the concept of representing the documentation by desktop meta files is
## used as far as possible. This makes the system very flexible and extandable.

2.) KHC::SearchTab provides a widget which lets the search through all
available help repositories, also defining some flags such as 'Search by
regexp' or 'Search case sensitive'.

# I think this means that we have to create a 'DataCollection' class which
# gets inherited by all classes which are "searchable". DataCollections should
# also be able to contains multiple child DataCollection, so that we have e.g.
# one DataCollection per application manual, and one "Manuals" collection
# which contains all the application manual collections.
# We'd probably also need a DataCollection for the info pages and man pages.
# And later, in the far future, we might extent this concept to web searches,
# so that e.g. Google represents a DataCollection which we can query.
# I'm not yet decided how to do that properly, perhaps using multiple
# inheritance, so that each TOCBuilder is a DataCollection - naw, we'd rather
# have a "TableOfContents" class which contains a TOCBuilder, and is a
# datacollection? Hm, not sure.
# In any case DataCollections should some sort of plugins, so that we can add
# e.g. new web search interfaces lateron.
# - Frerich

## What you call a DataCollection is currently represented by the DocEntry
## objects. Each DocEntry object represents a document or a collection of
## documents. It has information about the name and description of the
## document, the location and how it can be searched.
##
## Currently this information is based on URLs or file names and is optimized
## to be used by scripts, e.g. CGI scripts. A little exception from this is
## the htdig support where just a keyword "SearchMethod=htdig" is put in the
## desktop file and the help center figures out how to perform that search by
## using a special class. This could be extended to cover other search methods
## like web searches or special search methods optimized for certain kind of
## documents.

# I just thought about it - isn't that a bit overkill for the web search
# stuff? I just thought about it - all we need to do is to copy the .desktop
# files (at least some of them, like the ones for google, yahoo and excite)
# from the enhanced browsing thing and treat those as plugin .desktop files.
# We could show them in a listview on the Search tab, each found search engine
# being represented by a checkable listview item. So, we just let the user
# enter a term, replace the \{@} placeholder in the URIs specified in the
# selected .desktop files with that term, send out a request via TDEIO and show
# the results in our TDEHTMLPart (after all KHC::View is a TDEHTMLPart already). A
# problem with this: How to display the multiple HTML pages returned by the
# selected search engines? Using a QSplitter to split multiple TDEHTMLParts?
# Hmm... just wondered... perhaps we can work around that by not showing the
# returned HTML data at all but rather use a XSLT script (that is, one XSLT
# script per web search) which transforms the returned search results into a
# common format - that way, we could also filter out duplicates and then
# transform that filtered output into a nice, uniform HTML page. How about
# that?

# I like this idea very much, I just thought it and noticed you wrote this
# down already. What I thought of was having a .desktop/.xslt file pair per
# search engine: each .desktop file holds at least the name of the engine (for
# the listview) and a search URI with a placeholder, just like in your scenario.
# In additionl there could be a X-KHelpCenter-XSLT key which defines which .xslt
# stylesheet to use for that particular search engine. We then query that search
# engine by replacing the placeholder in the URI with whatever the user entered
# and hand it to TDEIO. All the HTML returned by the various search engines gets
# then transformed into a custom, intermediate, XML dialect, using the XSLT
# stylesheets define in the .desktop files. Using that intermediate step we
# can nicely drop duplicate hits, for example, or create a list of hits in the
# sidebar (much like http://www.copernic.com does). After that, we can use
# another XSLT stylesheet to transform that cleaned XML tree into HTML which
# we then feed to our TDEHTMLView. Since we then have one unified output, we don't
# need to worry about having multiple TDEHTMLParts, and it's also nice because
# the user doesn't see which search engine returned which hit.

# A problem with this would be that we cannot tell how a particular search
# engine treats boolean expressions (e.g. some search engines use 'foo AND bar',
# others use '+foo +bar', a third variation is '"foo bar"'). We thus cannot
# replace the placeholder in the URI but first have to translate the syntax
# entered by the user into a syntax which is appropriate for each single news
# engine. Right now I don't know how we could do this with just a .desktop/.xslt
# pair. We could always use fullblown C++ plugins which hold code which is able
# to do that translation, but I would really prefer to stick with .desktop files
# now since they're much easier to create.

# Another thing which would speak in favor of C++ plugins: different search
# engines support different features (like, google can search more than just the
# web, and you can sometimes tell a search engine to list only results in a
# certain language, or with a certain encoding), so it would be nice if we could
# let the user access those features: through a dialog which has to be tailored
# to the possibilities of the respective search engine. I wonder whether we
# could have some sort of XML tree which defines how an UI should look like, and
# then let KHelpCenter create a dialog using that XML markup, but that idea is
# very vague right now.

# Hmm, I just tried it and the XSLT idea didn't really take off: the problem
# is that many HTML pages returned by Google, Yahoo & co. don't seem to be 
# valid XML, which is why tools such as meinproc or xsltproc refuse to process
# themm. :-/

KHC::View
---------
KHC::View inherits TDEHTMLPart and does the actual job of showing some sort of
document. Most importantly, it has a slot which passes it a KURL pointing to a
document to show. KHC::View will invoke tdeio_help if necessary (if the URL's
protocol == "help") by itself and otherwise use the plain URL.

# TODO: Things I didn't really think about yet: the interface between the
# navigator and the view. I think this has to be a bidirectional association
# since the navigator can change the view (e.g. by clicking on a manual which
# shows it in the view), but the view can also change the navigator (think of
# clicking on a 'See also' link in the glossary which should also scroll to
# the corresponding entry in the navigator).

## That's a very important aspect. We should have one central place where all
## document requests are processed and the necessary actions (like updating
## the navigator, loading a new page, caching the search results, etc.) are
## done.
##
## The TreeBuilder might need some interface to tell, if a certain URL exist
## in their tree, to make it possible to select content entries which aren't
## created yet, because they are only created on demand (like the application
## manuals).

# Very good idea. Perhaps I think iterating over a list of TreeBuilder
# instances and doing something like 'if ((*it)->canHandle(url))
# (*it)->selectItem(url)' which checks whether a TreeBuilder provides an item
# which corresponds to an URL (hmm, this makes me think, TreeBuilder is a bad
# name. Perhaps just 'Tree'?) and selects it (using
# QListView::ensureItemVisible() or so) if requested. This probably implies.
# that a TreeBuilder needs an internal QMap<KURL, QListViewItem *>.

# Also, the whole search engine needs more thought, that DataCollection idea
# seems promising to me but I'm not yet decided on how to do it properly.

## See above. We already have something which isn't too bad, I think.

# I just thought about this a bit, I think KHC::MainWindow should act as the
# interface between KHC::Navigator and KHC::View.

## I would prefer to have an extra class which does no GUI stuff, but passes
## URL requests around, does the needed processing and stores data, if needed
## (e.g. caching search results).

# Agreed.

## One very important aspect of the help center is that it has to be fast. It's
## not acceptable to wait several seconds after clicking on the Help menu of an
## application. We should think about that. Perhaps we can do some tricks like
## showing the main window before creating the other widgets and processing data
## or something similar. We could also think about creating more stuff only on
## demand.

# My perception is that filling the Navigator's listview takes a significant
# amount of time, just like setting up the TDEHTML view (loading the stylesheet,
# showing the welcome page). We could easily do taht in the background - show
# the mainwindow, then tell the TreeBuilders to start populating (using a
# QTimer with a timeout of 0, for a snappy GUI). Since they're collapsed at
# the start, the users won't even notice (and we can "fake" that they're
# already populated by calling setExpandable(true) for all of them (or letting
# them do that themselves) at the start.

## Finally a crazy idea: Wouldn't it be cool, if we would make the manuals more
## interactive. So when you read about a certain menu or a certain dialog of an
## application you can click on a link in the manual and the menu or dialog gets
## opened in the real application, or some widgets get highlghted in the real
## application. Such a feature could also be used to create interactive
## tutorials, where you have a small helpcenter window and the application next
## to each other on the screen and you can go through the tutorial step by step
## and practice with the real application while reading the instructions.
## With the help of DCOP it shouldn't be too hard to implement such an
## interactive help system. Maybe it's even possible to do it in a general way
## in the libs, so that application authors don't have to think about that
## feature.

# Hmm, that's an interesting idea. That takes KHelpCenter way beyond what it's
# currently doing. I can imagine this: we introduce a virtual "dcop" protocol,
# so that e.g. <ulink url="dcop:/kfortune/KFortuneIface/nextFortune"/>
# represents the DCOP call 'dcop kfortune KFortuneIface nextfortune'.
# KHelpCenter catches that protocol (oh dear, a lot of special cases with
# gloss, info etc. already - guess another one won't hurt). That looks like a
# good way for encapsulating DCOP calls.
# Now, the problem is - the application has to provide a dedicated
# "documentation" DCOP interface for this, with lots of calls for highlighting
# the various widgets (hm, this probably means taht we can skip the first two
# parts in our 'dcop' URL syntax, the application is known anyway, and the
# interface is hardcoded in KHelpCenter).
# So, what could happen is this: We have a piece of HTML in the documentation
# for our SuperApp application which goes like 'The
# <a href="dcop:highlightConnectButton">button labelled Connect</a> makes
# SuperApp establish a connection.' - the user clicks on that link,
# KHelpCenter catches a dcop: URL, checks whether SuperApp has already been
# started. If not, it starts a SuperApp process and does the dcop call 'dcop
# SuperApp DocIface highlightConnectButton' and SuperApp starts highlighting
# that connect button. The thing is that this requires a lot of work on the
# application side. The idea is very cool, but we'd have to think about
# outsourceing parts of that functionality, either to KHelpCenter, or to
# tdelibs.

## And another idea: The WhatsThis help texts describe all widgets of an
## application (provided that the texts are set by the developers). Currently
## they aren't accessible very easily. You have to go to a special mode and
## can then click on one widget after another to get the help, if there is one.
## There is no visual indication which widgets have help and which not. But the
## application knows about the WhatsThis helps. Perhaps it's possible to use
## the Qt object inspection stuff to extract all the texts and put them on an
## automatically generated screenshot of the corresponding dialog and put this
## graphic into the docs. Maybe it's even possible to do this at run-time and
## decorate dialogs with all WhatsThis helps at once, if the user triggers this
## mode.

# Hmm yes, that should be possible. Take the toplevel widget, use
# TQObject::children() and iterate over all children, use QToolTip::textFor() to
# check whether the given qwidget has a tooltip and if so, use QToolTip::tip()
# to show the tooltip.
# One could probably add a standard dcop call to TDEMainWindow, like
# "showAllToolTips". KSnapShot could get a QCheckBox "Show all tooltips", and
# if that box is checked it tells the selected window to show all it's
# tooltips via that DCOP call right before it does the snapshot. The thing is
# - is it possible to map the WinID of the window the user clicked on to
# the process name we should send your DCOP call to?

## One thing we should also keep in mind is that it might be useful to provide
## the help center as a component. FOr example KDevelop has a very similar
## thing. It would be much nicer, if it could reuse the KHelpcenter code. This
## would probbaly also mean to at a DoxygenTreeBuilder or something similar.

# That probably implies that instead of a QSplitter which holds the Navigator
# and the View, we'd have a KHC::MainWidget KPart which in turn aggregates the
# splitter. The DoxygenTreeBuilder sounds like a reason to make TreeBuilders
# real plugins, with dynamically loaded libraries, so that KDevelop or other
# "IDE"-like applications (perhaps a KOffice help system?) can have their
# customized tree builders.

Font Configuration
------------------

### Many bug reports on KHelpCenter not honouring TDEHTML font settings,
### which is odd, because the stylesheet is intentionally loose,
### specifying only "sans-serif" as the font face.

### Ideas to fix: 

### Help pages already make heavy use of the cascading feature of CSS, we
### ought to be able to leverage that by writing to perhaps the
### kde-localized.css file or a copy of it in $TDEHOME.  There is already
### code in KControl to create a user CSS stylesheet, and we probably only
### need to configure the size and the face for KHC.

### Or, fix whatever is the reason KHC doesn't follow the rules.  It could
### be encoding related, the help pages specify utf-8 as the encoding, and
### previous incarnations of the TDEHTML settings allowed fonts set on a 
### per-encoding basis (at which time, this was apparently working, the bug 
### reports dropped off, and only returned post KDE 3.0

# FWIW I added a simple font configuration facility a while back, which should
# IMHO be sufficient for the vast majority of users.