Review Request 120149: [OS X] improved menubar experience: protected Preferences menu and cleaner "system tray"

Thomas Lübking thomas.luebking-Re5JQEeQqe8AvxtiuMwx3w at public.gmane.org
Mon Sep 15 22:57:01 BST 2014



> On Sept. 15, 2014, 12:19 nachm., Thomas Lübking wrote:
> > kdeui/actions/kaction.cpp, line 149
> > <https://git.reviewboard.kde.org/r/120149/diff/2/?file=312029#file312029line149>
> >
> >     what if
> >     KAction *foo = new KAction(this);
> >     foo->setText("Foo");
> >     
> >     -> you rather want to monitor the "changed()" signal?
> 
> René J.V. Bertin wrote:
>     Yes, that would probably be more elegant. I'm not exactly sure how one would monitor that signal, but it seems that it might be a bit complicated in the context of something that's bound to disappear?

You'd add a private slot to KAction and "connect(this, SIGNAL(changed()), SLOT(fixMenuRole()))"
Before altering the text in ::fixMenuRole(), don't forget to block (and later unblock) signals to not enter a recursion.

Since the changed() signal is not only called for altering text (but also icons, the role and some other stuff) you also might want to keep a string member to check whether the text actually changed before updating the role.


> On Sept. 15, 2014, 12:19 nachm., Thomas Lübking wrote:
> > kdeui/actions/kaction.cpp, line 164
> > <https://git.reviewboard.kde.org/r/120149/diff/2/?file=312029#file312029line164>
> >
> >     would this eg. work with kwrite ("Configure Editor...")? - or other kpart driven things?
> 
> René J.V. Bertin wrote:
>     The question is, should it work? KDevelop for instance has a Configure Editor item in its Settings menu, added *before* "Configure KDevelop...", and in that case it's not the editor's configure dialog that should get linked to the Preferences menu item.
>     If kwrite is nothing but a wrapper around an editor kpart (kate?) then I guess the end result will depend on how that kpart creates its configure action... If it uses KStandardActions and the wrapper (kwrite) doesn't (or only does after the kpart did), the Preferences menu ought to link to "Configure Editor...".
>     
>     In short, it's neigh impossible to cater for all possible cases. In a sense it wouldn't disturb me at all if KDE applications simply used their own menu organisation, but it's Qt that obliges us to take action because otherwise it's there that the organisation can get messed up.
> 
> René J.V. Bertin wrote:
>     This is where Qt initialises `menuRole` with `TextHeuristicRole`, a value only used on Mac:
>     
>     Qt 4.8.6 :
>     
>     	QActionPrivate::QActionPrivate() : group(0), enabled(1), forceDisabled(0),
>     	                                   visible(1), forceInvisible(0), checkable(0), checked(0), separator(0), fontSet(false),
>     	                                   forceEnabledInSoftkeys(false), menuActionSoftkeys(false),
>     	                                   iconVisibleInMenu(-1),
>     	                                   menuRole(QAction::TextHeuristicRole), softKeyRole(QAction::NoSoftKey),
>     	                                   priority(QAction::NormalPriority)
>     
>     Qt 5.3.1:
>     
>     	QActionPrivate::QActionPrivate() : group(0), enabled(1), forceDisabled(0),
>     	                                   visible(1), forceInvisible(0), checkable(0), checked(0), separator(0), fontSet(false),
>     	                                   iconVisibleInMenu(-1),
>     	                                   menuRole(QAction::TextHeuristicRole),
>     	                                   priority(QAction::NormalPriority)
>     
>     
>     Overriding that one way or another will make KDE menu(bars) behave exactly as they do on Linux, with only a few OS-specific items in the Application menu (Hide, Hide others, Show All). The added benefit would be for KDE applications like `konsole` that do not use the global menubar.
>     
>     Does KDE have a particular/special status making it easier to propose changes to the Qt API?

> Does KDE have a particular/special status making it easier to propose changes to the Qt API?

dfaure, I assume ;-)

No - not really. None that i'm aware of.
Would it help to alter KStandardAction::preferences() to explcitly set the PreferencesRole for this one (ie. would an explicit PreferencesRole trump TextHeuristicRole hitting for "configure shortcuts" or similar?


> On Sept. 15, 2014, 12:19 nachm., Thomas Lübking wrote:
> > kdeui/actions/kaction.cpp, line 188
> > <https://git.reviewboard.kde.org/r/120149/diff/2/?file=312029#file312029line188>
> >
> >     given KAction, KMenu and KMenuBar are all deprecated in KF5, is this actually of any upstream relevance?
> 
> René J.V. Bertin wrote:
>     I wouldn't know. It's been mentioned often enough that KF5 is still quite a while away from being release-ready on OS X, and in the meantime the best way to maintain momentum in getting it that far is to improve the user experience of what *is* available on the platform. Meaning KDE4 ...
>     
>     As to upstream relevance ... it clearly is unfortunate that KAction, KMenu and KMenuBar have been deprecated, because Qt's baked-in heuristics haven't changed in Qt5. From what I see that makes it neigh impossible to address the issues I'm trying to address in KF5 .

Hint: KStandardAction is *not* deprecated so if setting the explicit role in KStandardAction::preferences() is sufficient, you won.


> On Sept. 15, 2014, 12:19 nachm., Thomas Lübking wrote:
> > kdeui/actions/kstandardaction.cpp, line 173
> > <https://git.reviewboard.kde.org/r/120149/diff/2/?file=312030#file312030line173>
> >
> >     Will it?
> >     I don't see such code in QAction, setText is not virtual and why setTextWithCorrectRole itfp then?
> 
> René J.V. Bertin wrote:
>     True.
>     
>     Does C++ offer any mechanism by which one could overload a class's member function (or extend a class) without creating a child class? That might offer a solution on KF5.
>     
>     If no, the setText call can go back to where it came from (personally I do find it more logical to set a role after having set the text, though).

you can simply override a protected or public function by just adding the exact same signature to the derived class, BUT:

KAction *ka = new KAction(this);
QAction *qa = ka;
ka->setText("foo"); // calls your reimplementation
qa->setText("foo"); // calls the QAction function, despite qa == ka, ie. is actually KAction

to have the above example work, the function needed to be virtual (and that means virtual in the base class, ie. QAction)
Q_SLOTS are usually not virtual, but QMetaObject::invokeMethod("slotFunction") will resolve the best matching ::slotFunction() for the object (though by runtime resolution)


> On Sept. 15, 2014, 12:19 nachm., Thomas Lübking wrote:
> > kdeui/widgets/kmenu.cpp, line 174
> > <https://git.reviewboard.kde.org/r/120149/diff/2/?file=312033#file312033line174>
> >
> >     this is *utterly* wrong - you're manipulating a QAction reference just because it (at this very time!) hints it won't show it's icon in menus?
> >     
> >     The only sane approach is to fix the menu to honor this value and if that's not possible, clone the action (w/o icon), watch the original one for changed() and destroyed() for aligning updates and forward the clones signals to the original one.
> 
> René J.V. Bertin wrote:
>     I don't understand at all what you're trying to say.
>     
>     Do you mean that the user could change `isIconVisibleInMenu()` while the app is running, and that the change should be apparent in the menu without relaunching the application, or recreating the menu?
>     If so, is that even possible on OS X? (I could check but I have no idea what kind of application might do such a thing...)
>     And if that's indeed what you mean, what do you think is worse - Qt ignoring `isIconVisibleInMenu()` altogether, or me not supporting potential changes to the flag and/or icon?
>     
>     I do agree that my code changes the action object the user passes in. The thing is that it's probably not possible simply to clone the action and add the clone, because then everything the user does to his/her copy after adding it will have no effect ...
> 
> René J.V. Bertin wrote:
>     I've played around a bit with Qt's systray example. Here are my findings:
>     
>     - In Qt4, `isIconVisibleInMenu` is ignored completely for system tray menus, be it when adding a QAction or when changing the property later.
>     - Icon changes are respected.
>     - In Qt 5.3, `isIconVisibleInMenu` is respected when adding a QAction to the system tray menu, but subsequent changes to it have no effect.
>     - Icon changes are respected, unless `isIconVisibleinMenu` has been turned off, in which case the previously show icon remains visible.
>     
>     Based on these findings, I think someone could file an upstream bug report for Qt5, but for KDE/Qt4 it seems that we can stick with my proposal. With the current state of things, the user will have to restart the application if s/he wants to change icon visibility in the systray menu, whether my patch is applied or not (actually, without my patch visibility will never change...)

You are changing a clients code QAction from with a library, what is completely inacceptable.
Eg. a very common constellation is to use the very same QAction for a toolbutton and a menu entry. When it passed your code, puff, gone is the icon from the toolbutton.

It is not *simple* to clone the action, but it is possible. You'll have to track it's changes and forward them to your clone as you've to forward your clones triggering to the original action (ie. what Qt should be doing for this ominous systray menus - is it dbusmenu-qt?)


> On Sept. 15, 2014, 12:19 nachm., Thomas Lübking wrote:
> > kdeui/widgets/kmenu.cpp, line 182
> > <https://git.reviewboard.kde.org/r/120149/diff/2/?file=312033#file312033line182>
> >
> >     why do you reimplement to other variants?
> 
> René J.V. Bertin wrote:
>     When I didn't, the `addAction(QAction*)` change wasn't picked up, or at least so it seemed. I admit I was a bit surprised by that, could it be related to the compiler I'm using (mostly clang 3.0, the fastest on my dev. platform)?

I suspect "or at least so it seemed" and would suggest to retry. If the reimplementations would call your KMenu reimplementation, it would make sense (because addAction() isn't virtual), but this way they make no sense (to me)


- Thomas


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://git.reviewboard.kde.org/r/120149/#review66546
-----------------------------------------------------------


On Sept. 15, 2014, 11:03 vorm., René J.V. Bertin wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://git.reviewboard.kde.org/r/120149/
> -----------------------------------------------------------
> 
> (Updated Sept. 15, 2014, 11:03 vorm.)
> 
> 
> Review request for KDE Software on Mac OS X, kdelibs, KDEPIM, Marco Martin, and Olivier Goffart.
> 
> 
> Repository: kdelibs
> 
> 
> Description
> -------
> 
> This review is for 2 sets of changes; an initial one to the way "system tray" are rendered, and a newer set that protects the Preferences menu from getting linked to any action with an appropriate title.
> 
> -- the system tray:
> Until now, "system tray" menus had some rendering issues on Mac OS X:
> 
> - The menu title, the 1st menu item that on Linux shows the application name, remained empty
> - Menu items that can (in principle, potentially) show an icon always showed the icon
> 
> Point 1 was resolved by emulating the Linux addTitle/setTitle action in `KStatusNotifierItemPrivate::init()` : the menu title is implemented as a deactive standard menuitem followed by a separator. This makes the item stand out on a GUI that doesn't support the kind of formatting in menus as used in the Linux implementation.
> 
> Point 2 was identified as a Qt issue: `isIconVisibleInMenu` is ignored for systray menus. It was resolved by adding `KMenu::addAction` methods that overload the ones from QMenu that were hitherto inherited unchanged by KMenu. The only different method is `addAction(QAction*)` which removes the icon from the `QAction` if `isIconVisibleInMenu()` is false. The other `addAction` methods are "overloaded with themselves" with `using QMenu::addAction;` in the header file.
> 
> -- the Preferences menu item
> This is a menu item living in the Application menu, a menu that sits in the menubar between the Apple (?) menu and the File menu. This menu also contains the Quit command.
> KDE and Qt applications typically do not set up their menus in this fashion, so Qt provides an automatic way to put relevant menu items (actions) in the Application menu, using Apple's naming. The algorithm is described under QMenuBar in the Qt documentation: for the Preferences action, it will consider any action that has a text containing `config`, `options`, `settings` or `preferences`, and put it under the Preferences label if its menu role is set to `heuristic` (which appears to be the default).
> In practice, many applications provide a series of menu actions with names that trigger this method, and they do not always create their own preferences/settings/configuration menu first. Yet it is the first menu action that matches that will be installed under the Preferences menu, with the Command-, shortcut. A good example is KDevelop: it will have a Preferences menu that activates the `Configure Selection` action - which does not open a settings dialog but launches the configure or cmake procedure for the selected project ...
> 
> My proposed solution overrides this Qt behaviour. On OS X, the `KAction(const QString &text, QObject *parent)` constructor calls a modified (static) function `setTextWithCorrectMenuRole` which checks the text against the patterns Qt will consider for `PreferencesRole`. If it finds a match, it will force the role to `NoRole`, unless it is a perfect match with the standard KDE configuration action for the application (`"&Configuration appName..."`) in which case it sets the role to `PreferencesRole`. This latter consideration allows kdelibs to "catch" the configuration menu for applications like KMail, which appear not to be created using KStandardActions.
> This approach can be extended to other menu actions that end up incorrectly in the OS X Application menu.
> 
> Applications that create menu actions using QAction or a different KAction constructor will see no change (and should use `setMenuRole` selectively on OS X).
> 
> 
> Diffs
> -----
> 
>   kdeui/actions/kaction.cpp 9e8f7fb 
>   kdeui/actions/kstandardaction.cpp 7de0c6f 
>   kdeui/notifications/kstatusnotifieritem.cpp 1b15d40 
>   kdeui/widgets/kmenu.h f96e263 
>   kdeui/widgets/kmenu.cpp 7dab149 
> 
> Diff: https://git.reviewboard.kde.org/r/120149/diff/
> 
> 
> Testing
> -------
> 
> Testing was done with kdelibs git/master and KDE/MacPorts on OS X 10.6.8 . The modified code is in compile-time conditional blocks used only on OS X, so no regressions are to be expected on other platforms.
> 
> KF5 is not production ready on OS X, so I am not currently able to port these modifications beyond KDE4. However, I did see that Qt5 has a new approach to adding titles to menus, which can be described as a "labelled separator". Backporting that function from the Qt5 source to kdelibs gave menu items that had the separator but not the text (title) label. It is thus likely that some kind of emulation will also be required with KF5, on OS X.
> 
> I considered doing the addTitle/setTitle emulation in kmenu.cpp, but decided against that for now. Menu titles are rendered as under Linux in menus that are not attached to the OS X toplevel menubar (say in context menus). Without knowing how to distinguish the kind of menu in KMenu methods the emulation will have to remain in the client code.
> 
> The Preferences menu protection should carry over easily to KF5, supposing Qt5 uses the same heuristics to place relevant menu actions under the OS X application menu, and supposing `KAction` has made the transition to KF5.
> 
> 
> Thanks,
> 
> René J.V. Bertin
> 
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/kde-core-devel/attachments/20140915/fcfcbfa2/attachment.htm>


More information about the kde-core-devel mailing list