This file is meant to be a place to track of medium to large-scale API changes that we would like to make for the next binary incompatible release. - Check for forked classes in kde pim and other modules - There is no reason why KConfigBase should inherit from QObject, get rid of that. - Change all FooPrivate *d; -> Private * const d; and place initialization in the constructor (for classes that would benefit from this). To help catch silly mistakes since d should never change. Also consider changing to use KStaticDeleter to help prevent mistakes where developers forget to delete the pointer. Maybe make use of Qt4 helper macros? - Move all utility functions away from KApplication. TBD: Make KApplication a very thin wrapper around QApplication. Ideally, KApplication should go away completely. The kapp pointer references everywhere must die. Find a way to make that possible - Remove all KDE_NO_COMPAT from the KDE2 -> KDE3 port before adding new ones for KDE4. - Remove the smart-to-dumb-pointer conversion operator from KSharedPtr and replace it with a T* Get() member function. An implict conversion is far too dangerous. See: http://groups.google.com/groups?selm=4ognb0%243cd%40linda.teleport.com&oe=UTF-8&output=gplain Also free the client from the requirement to inherit from KShared. - Remove all KSocket, KServerSocket, KSocks and KExtendedSocket references throughout KDE in favor of KNetwork's socket implementations. Also, revise the namespace usage. thiago.macieira@kdemail.net - Remove all methods marked 'deprecated' in KDE 3.x. - Clean up the uses of virtual_hook, use normal virtuals instead (e.g. KZip/KArchive) - Make it possible to create a KSystemTray without a KMainWindow and instead lazily create the main window on demand. - Make KSystemTray's Quit option in the context menu call kapp->quit() instead of closing all windows, so apps that save their window state on exit don't need workarounds to detect a window close caused by the tray rather than the user. - Make KSystemTray add its own reference to KApp, so tray apps that adhere to the style guide can safely close their main window without having to worry about the app quitting instead. - Make some long-lasting operations threaded. There are some operations in KDE that take too long on at least ~500MHz hardware to be done using event loops. Examples are the use of KLibLoader to dlopen plugins and KRun to fire up other processes. Konqueror and KMail tend to stall occasionally too, but that's harder to refactor because the bulk of the work there is X11-related. - Make it possible to use SSL-ed sockets outside KIO. KSSL isn't very suitable for reuse, and the certificate dialogs are not even available in public API at all, leading e.g. to Kopete IRC duplicating a lot of stuff and all KDE Jabber clients based on libpsi (PSI, Light Hawk and Kopete) using homebrewn solutions. [How does this have to do with changing KIO? It's just a request to have that code in a different form....] - Manually, properly clean up the includes in all of the headers. Remove unnecessary ones and change pointers to use class foo; decorations and move the headers into the cpp files. When finished fixkdeincludes shouldn't find anything (but don't rely on fixkdeinclude, it should be done manually first!) - Use of the C++ keyword "explicit". Make all ctors explicit which can be used with one parameter to minimize wrong use of the class (i.e. have a look at Dirk's work on KURL). Another candidate: KService( KDesktopFile *config ). KGuiItem has one of these dangerous ctors, too. KUser has a whole bunch of them. - Solve need for casting when using bitfield enums (e.g. KProcess::Communication), if Simon's solution works. - Merge kdelibs/kdecore/kuser.h and kdeadmin/kuser/kuser.h. - Simplify memory management with KParts. The part and the widget delete each other, which in fact makes it more complex to handle. The part owns the widget so it should delete it, but not the other way round. - KParts: remove plugin loading from setInstance, add an explicit loadPlugins() call in all parts that should load plugins (at the end of the derived-part ctor). - KParts: come up with better names for ReadOnlyPart and ReadWritePart, ReadOnly is a minimum requirement, but this does not become clear until you dig deeper into the class. With QTextEdit and QTextBrowser for instance, the inheritance is the other way around. E.g.: ReaderPart? ViewerPart? - Other ideas in this thread: http://lists.kde.org/?l=kde-core-devel&m=107608067302571&w=2 (Contributors to that thread, please add specific items here.) - KConfigBase: For all the read*Entry methods that take a QFont *, QPoint *, QSize *, etc. for the default value, add overloads that use a const reference, so that one can write resize( config->readSizeEntry( "Bleh", size() ); for example. Constructing a null QFont/QPoint/QSize is cheap enough to justify a consistent and usable API (IMHO) . Keep source compatibility. - Sort out reference counting of applications and KMainWindows: Add a setRefCountEnabled() function to KApplication; deref()ing the last reference should only quit the application if reference counts are enabled, and the reference count should be disabled by default so apps that don't use a KMainWindow mostly continue to work. Make the KMainWindow constructor call setRefCountEnabled(true), and make each KMainWindow hold its own reference to the application. The KMainWindow should drop its app reference on hide, and regain it on show. Make KSysTray also hold a reference to the app. Then apps that use more than one main window and the tray (like KSIrc and Kopete) will quit at the right time without nasty hacks. - Change classes that store bool data using strings ("true"/"false") to use bool calls (readBoolEntry()). Although not technicaly a binary incompatiblity... just something to bring up for possible change (for the better) Use KConfigXT instead! /FransE - Is it really necessary to bother people using the KCM APIs with withFallback? I have never heard of it failing and when would you not like to have that fallback? I think a default of 'true' would be fine(and remove the argument in the APIs). - Do we need to have factory arguments in the KCM API? (who would use it anyway?) - KCModuleInfo should use d-pointer. Perhaps there is more classes. - The config options in kdeglobals needs to be cleaned up a little. Perhaps I'm being picky.. * They should be named properly, for example "enableIcons" instead of "EnableIcons". Or atleast settle on one thing. * AFAICT it would be the right thing if KWin had its own rc file instead of having its "WM" section in kdeglobals. * Shouldn't KSpell have its own rc file instead of kdeglobals? * Should we allow spaces in option names? * Some option names are too short for being understandable. fixed->fixedFont; font->systemWideFont * Perhaps an own font section? - kpassdlg.h needs to be renamed to kpassworddialog.h - consistent with other classes(it won't break many apps). The class needs a facelift - see the comments in kpassdlg.{h,cpp} - KProcess needs a major facelift. Not all of these items need to be BIC, but not needing to worry about legacy will certainly allow an overall cleaner solution. - the internals need to be encapsulated properly. many of the virtuals shouldn't be virtual; lots of protected members should be accessible only via accessors. the current model is theoretically super-flexible, but in fact it is only super-fragile. - clear up pipe() vs. socketpair() usage. - move the draining loop from commClose() to wait() - add additional "channels" beyond std*, with the built-in capability of passing the fd on the command line - redirections, including piping multiple kprocesses together. this should make most cases of reimplementing commSetupDone{P,C} superflous. - possibly create KForkProcess that covers most/all of the remaining commSetupDoneC reimplementations. - do something about the NoRead comm mode - add setComm(); remove comm arg from start(). as a side effect, this will allow for using writeStdin() in Block mode. - merge KProcIO into KProcess; add setBuffered() for separate channels. - use QByteArray for writeStdin(). better than the buf+len arg, as it is ref-counted. - To discuss: Migrate the about dialog to use qhtml or khtml rather then the current setup of all of the widgets and sub widgets. Might be a lot simpler, require a lot less code and be much more flexible for additions/enhancments. Sidenote: currently the about information seems to be duplicated. - Make API use US English, (eg. KCatalogue -> KCatalog) - It would be nice if some insan^H^H^H brave soul had a look at the KAccel/KShortcut/KKeyWhatever classes and cleaned it up. - KKeyDialog needs to be improved. It should simplify using KKeyChooser by not having to instantiate KDialogBase without restricting the possibilities of KKeyChooser (for instance, being able to configure global and application accels in one dialog). If there is only little time, simply expose the KKeyChooser object instead of making it private and inaccessible. - Fix the APIs taking Matthias Ettrich's hints into account, most importantly fix the massive amount of bool-usage in the CTORs (see KDialogBase, for instance) - Get rid of KXMLClient::conserveMemory() and the conserveMemory argument in createGUI. It created too many problems in the past. Not worth the trouble. - Split KProtocolInfo into KProtocolInfoBase in kdecore (with almost everything) and KProtocolInfo in kio (with the methods taking a KURL, which need KProtocolManager) - Rename KDirNotify_stub and update API (no need for '*' arguments anymore) - KWallet::Wallet::* functions - remove the default = 0 argument for mainwindow for the dialog - Make libkwalletclient part of libkio and get rid of libkwalletclient. - Get rid of libkdesu dependency in libkio, get rid of SessionData::AuthData*, get rid of SlaveInterface::authorizationKey and SlaveInterface::delAuthorization - Make functions in KHelpMenu virtual so they can be overridden (ex: to provide an alternative help->contents action) - Replace KPasswordEdit with KLineEdit and adjust KPasswordDialog accordingly. - Merge KPIM::ClickLineEdit with KLineEdit - Merge all K*Label classes into one that offers all the features such as ellipsis (this time for center, middle and right in order to allow for sqeezing the text), links and selectable text with a consistant API. - Make sure all GUI classes adopt well to palette changes. - Rework KPanelExtension and KPanelApplet to have a sane set of global position enums, make the geometry handling less spaghetti like and extend the API so requests for things like re-reading the config of a KPanelExtension can be done by its parent. - Fix KURLRequester API to use KURL for urls instead of QString to make clear that we work with URLs and not with paths. - Dump KPixmapIO class. QPixmap with qt-copy patches #0005 and #0007 can perform just as well, KPixmapIO can't dither, and it generally doesn't make much sense to have this class. - KListView: merge setDropHighlighter and setDropVisualizer, add a setDnDMode to choose between "dnd to move items" (e.g. keditbookmarks) and "dnd onto items" (e.g. kmail). Unless Qt4 does it all :) - Move KIO::findDeviceMountPoint, findPathMoundPoint, probably_slow_mounted, and testFileSystemFlag to KMountPoint, to merge that code. - Refactor the kio/bookmarks code so we don't need to use the static d-pointer trick anymore. - Rename the parameters to KDEDesktopMimeType::pixmap so that they use meaningful names rather than just a, b, c, and d. Probably not something that has to wait for KDE 4, but seems like a good idea to have here for later doing. - Move KRichTextLabel into kdeui if still required with Qt 4. - Add the concept of a session to KIO, in particular for KHTML so that it can have all of its jobs associated in some way (a unique key of sorts). Will make SSL much easier to implement and allow removal of many hacks, mostly involving metadata. - Make KLibLoader default to RTLD_GLOBAL and make necessary changes to make KDE namespace clean.