Review Request 120202: [OS X] improvements to the kwallet/OSX keychain integration
Thomas Lübking
thomas.luebking at gmail.com
Wed Sep 24 15:22:23 BST 2014
> On Sept. 21, 2014, 6:11 nachm., Thomas Lübking wrote:
> > kdeui/util/qosxkeychain.h, line 99
> > <https://git.reviewboard.kde.org/r/120202/diff/2/?file=314175#file314175line99>
> >
> > If OSXKaychain is an exported class (i don't know), this is an ABI incompatible change.
> >
> > It's also massively invasive and adds quite some overhead.
> >
> > Why did you not just remove the #ifdef from the slot declaration in Wallet (former patch) and #ifdef the implementation body instead?
> > (There are several such internal slots present, you don't have to "fix" the Wallet architecture with this patch ;-)
> >
> >
> > If there's absolutely no other solution and you do not want to add the slot unconditionally, you can still reimplement protected ::timerEvent() and do the timer "the hard way", ie. "myTimer = startTimer(timeout); ... if (te->timerId() == myTimer()) { blahFoo; } else BaseClass::timerEvent(te);"
>
> René J.V. Bertin wrote:
> I would never have done things this way if QOSXKeychain was an exported class... but I'm sensible to the overhead argument. I also realise I could probably have declared it `protected QObject`.
> I considered your suggestion, but decided against it because it would require patching kwallet.cpp too. Or at least I think it would, I presume one has to provide an implementation for every member function that's declared in the header? Unless one can make the declaration virtual and only provide an implementation in kwallet_mac.cpp (the sort of detail I just cannot seem to memorise :-/ )?
>
> Thomas Lübking wrote:
> Ah, I now see: The various backends do not inherit a common base but are just split by architecture (on the same header)
>
> As long as it's not referenced, a function does not have to be implemented.
> BUT: moc will referece all slots, so NO. YOU MUST ADD IMPLEMENTATIONS EVERYWHERE.
>
> (What you must not do as well in this regard, is to introduce pure virtual functions, ie "virtual void foo() = 0;" - the class becomes abstract by this and cannot be instatiated)
>
> So the remaining option would be "int myTimer = QObject::startTimer(timeout);"
>
> René J.V. Bertin wrote:
> Is it a really a big issue to introduce an empty and unused slot in the other Wallet implementation? (After all, "my" Wallet implementation has a number of slots too that are there only to satisfy the needs of the other ;) )
>
> Thomas Lübking wrote:
> Not from my POV - it's far less invasive than altering the private baseclass.
> You "decided against it" (although, timerEvent() would have to be implemented everywhere just as well)
>
>
> Suggestion: isolate it by adding a QObject inheriting member to OSXKeyChain (to call a slot there) for this will no longer be a problem with Qt5 (YEAH for lambda slots! =)
>
> René J.V. Bertin wrote:
> Well, I "decided against it" until I realised it wasn't really a big deal :)
>
> Just how is "adding a QObject inheriting member to OSXKeyChain" any different in terms of additional overhead than just letting the class itself inherit QObject?
>
> What I might do is define an additional class in QOSXKeychain.h, inheriting QTimer (which inherits QObject), give it a reference to the wallet instance, and then replace the QTimer* member in WalletPrivate with a pointer to that new class. That way the QObject overhead only occurs when I'm actually going to use the timer. Sounds like a plan to me (but that may just be the time of night ^^)
> how is "adding a QObject inheriting member to OSXKeyChain" any different in terms of additional overhead
Isn't at all.
It's just easier to revert since in Qt5 you can just use a lambda slot (what means you can write the functionality inline into the connect - no need for a slot function anywhere)
You can use the QTimer Object for this, yes (it'd be a bit wonky for some public interface, but as a semi-private class, there's no problem with this)
- Thomas
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://git.reviewboard.kde.org/r/120202/#review67155
-----------------------------------------------------------
On Sept. 21, 2014, 4:40 nachm., René J.V. Bertin wrote:
>
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://git.reviewboard.kde.org/r/120202/
> -----------------------------------------------------------
>
> (Updated Sept. 21, 2014, 4:40 nachm.)
>
>
> Review request for KDE Software on Mac OS X and kdelibs.
>
>
> Repository: kdelibs
>
>
> Description
> -------
>
> I'm still working on (the KDE4-based version of) my OS X keychain backend for kwallet. I'm at a point where I think I can present a work-in-progress in an RR because at least one feature has been improved enough to be of interest for everyone, and also because I could use feedback on how to proceed.
> I'm currently focussing on 2 settings that are configured in the kwallet KCM (SystemSettings), and for which I'm working on an implementation not requiring kwalletd and/or DBus.
>
> - idle time closing of wallets. This feature was not supported in the commited version presented in https://git.reviewboard.kde.org/r/119838/ The present patch adds an idleTimer and a shared lastAccessTime member. The idleTimer is reset each time a client performs one of a series of actions that I count as wallet accesses, and before resetting I update the idle timeout value from KConfig. When the timer fires, the elapsed time is compared to the shared last access time, and if it is >= the timeout, the wallet is closed. This applies only to "KDE keychains", so keychains used by OS X applications should not be affected.
>
> - "close when last application exits". This requires maintaining a "user list" which keeps track of what application has what wallet open. I've implemented an "internal" version of such a registry, mapping wallet name to application names and the list of wallets they have open (a list of wallet reference, pid per application name). The registry is functional, but I have not yet decided (read: figured out) how to make a distributed representation of it.
>
> So the work-in-progress concerns the distributed user registry. The idea would be to maintain the registry in shared memory, meaning it'd be reset (= disappear) when the last application exits, contrary to a file which can go stale. This would be simple if QSharedMemory objects could be resized, but apparently they cannot, so I'll have to look at other solutions possibly involving OS X frameworks (NSData and it's non-objectiveC version CFDataRef or CFMutableDataRef might be candidates). Suggestions welcome.
>
> Other work in progress concerns a less wheel-reinventing approach that builds on kwalletd and DBus. I don't see why the code used in `kwallet.cpp` wouldn't work, but I must still misunderstand its finer details. The present patch contains outcommented code that does indeed cause kwalletd to be launched and slots and signals to become visible e.g. in `qdbusviewer`. But they don't work, which in turn makes the whole kwallet layer dysfunctional. Here too feedback is welcome on how what I'm missing and/or how to get this to work.
> Once kwalletd works, wallet idle timeout closing and closing when the last client exits should work out-of-the-box, or at least I suppose.
>
>
> Diffs
> -----
>
> kdeui/util/kwallet.h d7f703f
> kdeui/util/kwallet_mac.cpp 8344ebb
> kdeui/util/qosxkeychain.h d0934e6
> kdeui/util/qosxkeychain.cpp 7cb9a22
>
> Diff: https://git.reviewboard.kde.org/r/120202/diff/
>
>
> Testing
> -------
>
> OS X 10.6.8, kdelibs 4.14.1 git/master, KDE/MacPorts 4.12.5 .
> Once finalised, all changes should port easily to KF5's kwallet_mac.cpp .
>
>
> Thanks,
>
> René J.V. Bertin
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/kde-core-devel/attachments/20140924/27cdbcbf/attachment.htm>
More information about the kde-core-devel
mailing list