[KDE/Mac] Review Request 120931: [OS X] improvements to KWindowSystem

René J.V. Bertin rjvbertin at gmail.com
Mon Nov 17 09:14:28 UTC 2014



> On Nov. 15, 2014, 3:43 p.m., Thomas Lübking wrote:
> > kdeui/windowmanagement/kwindowsystem_mac.cpp, line 556
> > <https://git.reviewboard.kde.org/r/120931/diff/2/?file=328516#file328516line556>
> >
> >     Does this *really* cut it on OSX?
> >     The function is not supposed to be an extra superfluous wrapper around QWidget, but typically used to control windows IN ANOTHER PROCESS.
> >     
> >     This raises the question whether that's possible on OSX at all.
> >     If not, testing for an in-process window (search toplevels only?) is ok, but the failure should cause a big fat warning to the developer that this code isn't portable.
> 
> René J.V. Bertin wrote:
>     It works for in-process windows, obviously, but no it won't work for windows from another process. I highly doubt that one could meddle with those, and as I must have written in the comments somewhere, you cannot convert the WId to a pointer to an actual window object if it's not owned by ourselves.
>     
>     What exactly are you proposing concerning that big fat warning?
>     Do you know of code that uses this kind of functionality cross-process, apart from kwin and maybe a couple of goodies that aren't relevent outside of a Plasma workspace?
> 
> Thomas Lübking wrote:
>     Well, how does the OSX docker etc. raise/activate a window?
>     Does this imply that activating won't work either for other PID windows? (The comments actually seem to suggest so)
>     In case: why mess with the cocoa API itfp? You could just as well wrap around Qt (w/ a //TODO or similar)
>     
>     > What exactly are you proposing concerning that big fat warning?
>     qWarning("BlahFooBar does not work on OSX, please fix your stuff");
>     if (m_DebugClient)
>        abort();
>        
>     m_DebugClient could be set in a header inline, depending on whether QT_DEBUG is defined (or QT_NO_DEBUG is not defined)
> 
> René J.V. Bertin wrote:
>     The OS X Dock is a case apart. It's system software, and probably interacts with the window manager.
>     But I should rephrase, maybe: "it won't work" means there's no documented way to achieve raising and the like across process boundaries. As long as you don't want to or cannot use AppleScript, and in this case we cannot because we cannot use the WId.
>     
>     I do think that the Cocoa API gives a bit more functionality than wrapping Qt calls would, but I can have another look at that.
> 
> Thomas Lübking wrote:
>     Are you or another OS_X hacker maybe in touch with more regular Cocoa API devs (forum, mailing list personal contact)?
>     I find it hard to believe that there's no access to windows on Cocoa (but on apple script) - wouldn't eg growl allow you to activate the sender window?
> 
> René J.V. Bertin wrote:
>     Growl is a notification framework, and indeed it doesn't seem it can do anything else but sending messages and displaying them.
>     
>     But I've spoken a little bit too soon. There *is* actually a potential way to tell another application to raise a specific window. In ObjC, invoking a class instance's member function is called sending a message to that instance ... and those messages *can* be used for IPC. See http://stackoverflow.com/questions/7448068/in-cocoa-or-generally-in-objective-c-is-there-a-way-to-send-a-message-to-objec . It's a feature I've never used so I tend to forget about it, and the question remains if we can actually use this without a substantial rewrite. And the big question remains: how to glean the required information from a WId. The documentation suggests that a WId is actually an `NSWindow*` on OS X, but from what I've seen this is no (no longer) correct.
> 
> René J.V. Bertin wrote:
>     I've had a look at using distributed objects in ObjC. That could be the solution we're looking for, given certain conditions:
>     
>     - WId values must be unique across processes during a session and not potentially identical in multiple processes at the same time like they could be if they are pointers (like the documentation suggests). 
>     - Because the KWindowSystem API only provides a WId to work on, we can only check the system for published distributed objects registered under, for instance, the hex. representation of that WId (which is why they must be unique)
>     - As a consequence, each time a window is opened we must create a dedicated NSConnection object registered with the WId's representation, so KDE's windowing layer must be adapted to do that. I don't know yet whether this is actually possible, nor how large an overhead this would introduce.
>     
>     A more elegant (?) solution would create a single NSConnection object for some kind of application interface that can respond to queries like [give me the NSWindow* for this WId] and register that through kded. And of course kded would need to have a reverse interface to a dictionary mapping WIds to registered applications because of point 2 above.
>     
>     Sounds like a fun little project to get right, but somehow I doubt it stands much chance to be accepted for KDE4.
>     
>     And my main question remains: just how many applications try to do things with windows that are not their own?
> 
> Martin Gräßlin wrote:
>     > And my main question remains: just how many applications try to do things with windows that are not their own?
>     
>     That is the main purpose of KWindowSystem. For doing things with your own window one wouldn't need KWindowSystem, but could just QWidget/QWindow.
> 
> Ian Wadham wrote:
>     @René: Do the results of http://lxr.kde.org/ident?_i=KWindowSystem&_remember=1 mean anything to you?
> 
> René J.V. Bertin wrote:
>     Hi Ian,
>     
>     Interesting, I wasn't aware of that site. Yes, the result mean something to me - you have probably heard of digiKam too for instance. What this doesn't answer without going over all those hits is to what extent the KWindowSystem calls are actually used to control windows from other applications. I'm pretty sure digiKam doesn't, for instance.
>     I think it's relatively safe to say that we'd have noticed where window level control was missing - we sure did for DrKonqi and KWalletd ...
> 
> Thomas Lübking wrote:
>     Looking across some files:
>     largely virtual desktop assignment, setting window types not (been) supported by Qt, forcing to be the active window, obtain KWindowInfo on oneself.
>     Amarok at least once checks whether it is the active window (winId() != KWindowSystem::activeWindow())
>     
>     Indeed there's a lot of "abuse" here, where kwindowsystem is invoked w/o any need.
>     
>     @Martin, do you think it would make sense to add an (preproc controlled) mechanism where one does "export AM_I_ABUSING_KWINDOWSYSTEM=1" what would lead to an abort if one operates on a local window and to control sth. that QWindow just provides as well?
>     
>     Could be a macro:
>     WARN_ON_ABUSE("Use QWindow::requestActivate() or QWidget::activateWindow() instead")
> 
> Ian Wadham wrote:
>     I see too that KMessageBox and KFileDialog have "*WId" variants of their main methods --- with warnings to use them only when "necessary". FWIW KMessageBox implements its "normal" methods (based on parent-widgets) by finding the parent's WId and then calling the *WId variant "to avoid code duplication" (see line 444 of kmessagebox.cpp).
>     
>     Also there is a strong warning in http://api.kde.org/4.x-api/kdelibs-apidocs/kdeui/html/classKWindowSystem.html#a7a04fd62d97ed3104e02bfa3ffa19ad5, "Detailed Description" section, about using this class on OS X. Otherwise I am baffled by all this low-level access to the window system. Even KWordQuiz in KDE Edu has an instance of it.
>     
>     Maybe René should just ban "applications doing things with windows that are not their own" on OS X and see if anything breaks...

I missed that doc bit about the WIds returned by KWindowSystem. I'm not sure why that asymmetry would be there, and wonder when the warning was written. It could be true though: where Qt's documentation states that a WId is just a cast `NSWindow*`, in reality it's either a `QCocoaWindow*` (which inherits NSWindow) or a `QCocoaView*` (which inherits NSView, a superclass containing an NSWindow member). Obviously it won't work to give Qt a "NSWindow WId" for a widget that uses a "NSView WId".
Something to look into for sure.

In any case, this limitation applies to all windows (or should I say, widgets...), including those owned by the application.


- René J.V.


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://git.reviewboard.kde.org/r/120931/#review70399
-----------------------------------------------------------


On Nov. 15, 2014, 12:04 a.m., René J.V. Bertin wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://git.reviewboard.kde.org/r/120931/
> -----------------------------------------------------------
> 
> (Updated Nov. 15, 2014, 12:04 a.m.)
> 
> 
> Review request for KDE Software on Mac OS X and kdelibs.
> 
> 
> Repository: kdelibs
> 
> 
> Description
> -------
> 
> This is an attempt to improve the Mac-specific implementation of the `KWindowSystem` class.
> For convenience and future-proofness (and also because I like the language) I converted `kwindowsystem_mac.cpp` to ObjC++, i.e. `kwindowsystem_mac.mm`, and added the AppKit framework in the CMakeFile.
> 
> Much of the code in this file is hardly better than gentle hacking, but that probably concerns the functions that are of least interest on a platform where KDE doesn't do session management.
> 
> I should probably update the "not yet implemented" debug statements (to "unsupported"), and I might have another look at kwindowinfo_mac.cpp too.
> 
> 
> Diffs
> -----
> 
>   kdeui/CMakeLists.txt 1454790 
>   kdeui/tests/kwindowtest.cpp b4012d7 
>   kdeui/windowmanagement/kwindowsystem_mac.cpp 4200237 
>   kdeui/windowmanagement/kwindowsystem_mac_p.h PRE-CREATION 
>   kdeui/windowmanagement/kwindowsystem_macobjc.mm PRE-CREATION 
> 
> Diff: https://git.reviewboard.kde.org/r/120931/diff/
> 
> 
> Testing
> -------
> 
> On OS X 10.6.8, mostly with the updated kwindowtest utility (which calls KWindowSystem functions when clicking the Open button in its toolbar).
> Also tested on Mac OS X 10.9.4 rebuilding kdelibs from scratch.
> 
> 
> Thanks,
> 
> René J.V. Bertin
> 
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/kde-mac/attachments/20141117/b75bcdeb/attachment-0001.html>


More information about the kde-mac mailing list