tdepim/kontact/Thoughts

* Note: Lines starting with a d are my comments - Daniel
* Note: Lines starting with a # are my comments - Cornelius
* Note: Lines starting with a "z" are my comments - Zack :)
* Note: Lines starting with a "s" are my comments - Simon
* Note: Lines starting with a "Don:" are my comments - Don
* Note: Lines starting with a "g" are my comments - Guenter
* Note: Lines starting with a "m" are my comments - Matthias Kretz
* Note: Lines starting with a "MiB:" are my comments - Michael
* Note: Lines starting with a "h" are my comments - Holger

Misc:
=====

Configuration Merge
-------------------

d Idea: The KOffice way of life: Offer a method that adds a given wiget of a
d predefined type as page in a KDialogBase or offer a pointer to a KDialogBase
d -> requires a Kontact part or an external lib per part

m I believe this is a more general problem. Please take a look at
m tdegraphics/kview/kpreferences{dialog,module}.{h,cpp}. I'd like to generalize
m these classes and include them into tdelibs (the same configuration merge is
m being done in Kate, Noatun, Kopete, KView and probably more).

# The problem is even more generic. We also have to merge about boxes, tips of
# the day and maybe more.


Merged Foldertree View
----------------------

d Idea: Let the part send a description of their folders and reaction to calls
d as XML, similar to XMLGUI

# Is a folder tree really the right tool to represent events, todos or
# contacts?

MiB: On the one hand, Notes can be hierarchic, so a folder tree would be the
MiB: nearest solution...

z I think so. Applications could send the root of their tree to
z Kontact so that the interface looks like

- Mail
  |    \
  |     - Local Folders
  |                    \
  |                     Inbox
  |                     |
  |                     Thrash
  |                     |
  |                     Sent
- Notes
  |    \
  |     Notes 1
  |     |
  |     Notes 2
  |
- Events
        \
         Event 1
         |
         Event 2

z which is not that bad. The question would be how to render the tree
z on the Kontact side while keeping the items on the parts side ( because
z e.g. KMails hold custom pixmaps for the folders which had to be
z displayed in the Kontact tree).

g I'm currently having 248 events. A tree is a very bad solution to visualize
g them. selecting "Events" in the tree should just only start the korganizer
g part.

MiB: ...OTOH... yes, /me agrees with g, a folder tree becomes complex quite fast.

Don: The folder tree makes sense for advanced users, but I think
Don: the simplicity of the current navigator widget has advantages for
Don: non power users.
Don:
Don: Actually instead of the navigator widget I think it makes sense
Don: to consider reusing the widget choosing widget in the latest
Don: version of the Qt designer, which in a sense can be
Don: considered a generalization of the navigator widget. And could
Don: make the folder tree in kmail unnecessary.
Don:
Don: I might investigate the Qt designer widget further but if someone
Don: else wants to look at a folder tree widget that's cool with me.

# I had a look at the Qt designer widget choosing widget. I think it has a
# severe usability problem, because the buttons (or kind of tabs) which are used
# to access the widget subgroups are not always at the same place but move
# around when you click on them. Dependening on which group is shown, the button
# is at the top or at the bottom of the widget. In my opinion this solution is
# unacceptable.

# But Daniel had a good idea how to improve that. It looks similar to the Qt
# designer widget, but it opens the current group always at the top of the
# widget and only highlights the current group in the list at the bottom, but
# doesn't move it. This seems to also be the way Outlook does it.

Don: Guenter, agree.
Don: Wouldn't the idea to be to show calendars in the tree or
Don: navigator widget, rather than individual events?

# Yes, that makes sense. Calendars are much more similar to mail folders than
# single events. You wouldn't integrate individual mails in the folder tree,
# would you?

d That raises an interesting point: The KNotes plugin would not need an own
d canvas in the WidgetStack then. It's sufficient to have the notes in the
d folder view, an RMB menu on them and a "New Note" action.
d So the new design must be able to catch that case (the current one does not).

# I think notes are on the same level as mails or events. They should be listed
# in the view. KNotes would probably just create a single entry in the folder
# tree.


KNotes integration
------------------

MiB: Which reminds me of my own concern about the 'how' of integrating KNotes:
MiB: * the current solution is to start KNotes extern, it is not embedded in Kontact
MiB:   at all. Thus opening a note that is on another desktop either leaves the Kontact
MiB:   window or moves the note. Either not perfect. Also, Kontact is likely to cover
MiB:   notes that reside on the desktop, easy working is impossible. Which is the reason
MiB:   I don't like the current approach too much.
MiB: * but there's always hope---my idea would be to show the notes in Kontact itself.
MiB:   Now I tend to say it's a bit intrusive to not allow starting KNotes and
MiB:   Kontact/KNotes at the same time which raises the following issues:
MiB:     - if KNotes and Kontact are running at the same time, changes to the notes have
MiB:       to be synchronized (not much of a problem). Changes to be synced are the
MiB:       text/contents itself, the text color/style..., the note color. Not sure about
MiB:       the note size. Not to be synced is the position.
MiB:     - so the position in Kontact has to be saved individually and independently
MiB:       of the real desktop position (realized by attaching two display config
MiB:       files, works in make_it_cool branch mostly). Kontact's size is generally
MiB:       smaller than the desktop.
MiB:     - normally notes are on a specific desktop, now they have to be displayed on one
MiB:       area---how to do this best?

MiB: what does M$ do? How do they manage the notes in their PIM app? (I don't know
MiB: it, never seen that thing)


Toolbar Items
-------------

d The KParts Technology only provides actions for the current part. It might be
d desireable to have common actions that are always available.

Don: I agree that it is desireable to have common actions always
Don: available (and parts too like the todo list)
Don:
Don: But are you sure Kparts is limited in this way? KOrganizer can load
Don: multiple plugins simultaneously. And all of these plugins are tdeparts
Don: (eg. birthday import), and tdeactions for all loaded plugins are
Don: created and made available simultaneously.
Don:
Don: Yeah, I'm quite positive you can load multiple parts simultaneously.

# Certainly. Actions like "New Mail", "New Contact", "New Event" should be
# available independently of a selected part.

Don: This is a very important issue, I think we need a library with three
Don: methods:
Don: KAddressBookIface loadKAddressBook()
Don: KMailIface loadKMail()
Don: KOrganizerIface loadKOrganizer()
MiB: And don't forget KNotesIface loadKNotes() :-)

h: That doesn't sound extendable ;)
h: So if I would like to add a 'New ShortMessage' part we would have to extend
h: that library... better use TDETrader and some sort of a common framework
h: and Mib's comments shows that problem!

d: That's what KDCOPServiceStarter is for :)

Don: Now if kontact is running then loadX will load the X part in kontact
Don: (if it is not already loaded) and return a dcop iface for that
Don: part.
Don:
Don: If kontact is not running but is the users preferred application
Don: then loadX will start kontact and then do the above.
Don:
Don: If kontact is not running and is not the users preferred application
Don: then a standalone version of X should be started, and an iface for
Don: that standalone app returned.
Don:
Don: I think this library should be in libtdepim ad all the tdepim apps
Don: should be moved into tdepim, so their iface files all be in one
Don: package. Or alternatively a new kdeinterfaces package be created
Don: and used as a general repository for interface files.
Don:
Don: Another important issue is invokeMailer and the fact that currently
Don: KDE just runs kmail with command line arguments by default. That has
Don: to be made smarter.
Don:
Don: I guess when kmail is run with command line arguments it could
Don: actually use loadKMail() and then use the resulting iface.
Don:
Don: And the same for all other loadX apps.


Status Bar
----------

d We need a more sophisticated handling (progressbar, etc)

Don: Definitely.

# We now have tdelibs/tdeparts/statusbarextension. This is intended to solve these
# problems, right?

d: Right. Simply add it as childobject in your part and use it's API. Works even
d: for other KPart hosts than Kontact


Kontact plugin unification
-------------------------

# Currently all Kontact plugins look quite similar. It would be nice, if we
# could provide infratructure to reduce duplicated code as far as possible.

d I thouht of a KontactPart, similar to a KOPart, if that makes sense. I don't think
d a normal KPart is sufficient for us.

Don: I've spent quite a bit of time in all pim *_part files and IIRC
Don: the amount of duplicated code, is pretty much negligible.
Don:
Don: But a KontactPart could make sense for when the parts want to communicate with
Don: the container. Eg. if the parts want to add folders to the container
Don: apps folder tree (or navigator)
Don:
Don: And maybe for communicating with the status bar.
	

Communication/Interaction:
==========================

d Invoking parts when they are needed for the first time takes too long,
d starting all takes too long on startup
d Idea: Mark complex parts as basic parts that get loaded anyway

# parts could be loaded in the background based on usage patterns. Kontact could
# remember which parts were used at the last session and load them in the
# background after loading the initial part to be shown at startup.

z This idea seems to be similar to Microsoft's
z hide-unused-item-in-the-menu strategy. But it probably mess up
z kaddressbook integration. Although not used during every session
z this part is needed and should be always loaded. This strategy
z would be great for could-to-come parts, like a summary part.
z Background loading of parts is OK. The idea is simple : load the
z last used part on startup. Make sure its loading finishes and then
z load the rest once the user can already interact with the last used
z loaded part.

g why do we always need the addressbook? Is libtdeabc not sufficient?

Don: I guess my machine is too fast, starting parts is pretty quick here :-)

d DCOP is too slow, internal communication should be handled via a dedicated
d interface, communication with external applications (i.e. knotes) should be
d done via wrapper parts that communicate with their respective IPC method to
d their application using the native protocol (DCOP, Corba, etc).

# Are you sure that DCOP is too slow for in-process communications? I thought it
# would handle this special case efficiently.

s It is only efficient in the sense that it won't do a roundtrip to the server but
s dispatch locally. What remains is the datastream marshalling. Not necessarily
s ueberfast. But I think the point is a different one: It is simply not as intuitive
s to use as C++. Yes, DCOPRef already helps a lot for simple calls, but talking to
s remote components still requires one to do error checking after each method call.
s in addition the stub objects one deals with (AddressBookIface_stub for example)
s are no real references. To the programmer they look like a reference to a
s remote addressbook component, but it really isn't. there is no state involved.
s like if between two method calls on the stub the addressbook process gets restarted,
s the state is lost and the programmer on the client side has no way to find out
s about that. you'll end up with really complex code on the caller side to handle things
s like that.

d Yes, but of course one should always prefer in-process IPC if possible. DCOP
d currently _works_ for Kontact, but that's all about it. It isn't exactly elegant.
d The only advantange of the current approach is that we can allow the user to
d run one of the parts standalone. I am not really sure we want that. I used to find
d it desireable, but I am not sure anymore.

MiB: But that's the whole idea behind Kontact---to be able to integrate apps
MiB: _and_ to have standalone versions. Just think about KNotes... impossible
MiB: to have it limited to only Kontact!

Don: I love being able to run the apps inside or outside of the
Don: container, it's just really cool being able to choose I think it's a
Don: great feature and users will really love having the
Don: choice. Especially when they are migrating.

MiB: Definitely.

Don: I think if we use the loadX methods defined above then we can still
Don: support this. I'm PRO DCOP. And this way we don't have to special
Don: case of the code depending on whether the application is running in
Don: a container app or not.
Don:
Don: I find difficult to imagine a function that DCOP is not fast enough
Don: to support. It supports all our current PIM IPC needs fine.

MiB: yes, not too much against DCOP. But for KNotes I thought about turning
MiB: a note into a plugin that can be loaded by Kontact and KNotes independently.
MiB: like this, there's no DCOP necessary anymore and makes it much more flexible.
MiB: e.g. usage of different display configs, a note embedded somewhere and having
MiB: a parent or standalone on the desktop.

# Communication with external applications is something which doesn't fit too
# well with the 'integrated' approach of Kontact. Is this really necessary?

d We won't get around it, think knotes, maybe sync tools, think abstact 3rd party
d projects (not sure the latter is really that important, but we should consider it.
d it barely plays a role anyway).

MiB: hm. true. But not too important, IMHO. Just add a Kontact-DCOP interface :-)

h: Pretty much to talk about...
h: 1. the speed of DCOP is not that important. I worry more about the integration
h:    of all parts. So how would I cross reference an 'Event' with a 3rd party
h:    Kaplan Part? A common base class for all PIM records comes into my mind - again -
h:    Now with normal C++ you can pass a pointer through the framework
h:    Doing it with DCOP we need to marshall and demarshall it. This part can get really
h:    ugly if we want more tight integration of all KaplanParts. We could add
h:    a pure virtual method to marshall to a QDataStream. So now marshalling is done.
h:    For demarshalling we need to get the type of the QDataStream content and then we need
h:    to ask someone - a factory - to get a object for the type and then call another pure
h:    virtual.....
h:    The question is if this is really necessary
h: 2. stand a lone apps
h:    The 'stand a lone' app can always run in the same address space but be a top level widget
h:    itself. WIth some DCOP magic clicking on the KMAIL icon code make Kaplan detach the part...
h: 3. Integration!
h:    The goal of Kaplan should not be to merge some XML files an give a common Toolbar for
h:    X applications in one shell. I want true integration. Yes KMAIL can use KABC to show
h:    all emails for one contact but a generic way to do such things would be more than nice.
h:    It would be nice if I could relate the PIM objects in a common way. So I create an Event and
h:    relate some todos to it. So for KDE4 I want a common base class for all PIM classes including mail
h:    see Opies OPimRecord for a bit too huge base class

Security
--------

d If we use the tdeparts (ktrader) approach to find a parts by looking
d for an application with the correct mime type this might raise security
d problems. (Martin's concern)

# Looking up Kontact parts isn't based on mime types but on services of type
# "Kontact/Plugin". This is just as save as starting a program statically linking
# its parts. I really don't see any security concerns here.

d Ok, if we limit stuff to Kontact/Plugin and Kontact/Part that might be safe enough
d indeed. I (and Martin, who raise this concern initially) was just afraid of
d allowing "any" part.

h: hmm If somebody can install a Service into the global kde dir or the user kde home
h: there is something else broken IMHO


Summary View
------------
h: How would one best integrate a summary view into kontact?
h: a) add a virtual QWidget *summary(const QDateTime&, QWidget* parent );
h:    to get a summary widget for a day?
h: b) use some sort of XML to UI to represent the summary informations
h: c) have a stand a lone part which opens the PIM data seperately? ( How
h:    to synchronize access? )