Review Request 125217: KDirWatch: fix thread safety

[David Faure]

-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://git.reviewboard.kde.org/r/125217/
-----------------------------------------------------------

(Updated Sept. 14, 2015, 7:06 a.m.)


Review request for KDE Frameworks, Albert Astals Cid and Michael Pyne.


Changes
-------

only call qAddPostRoutine once; and in the post routine, don't create KDirWatch::self just to clean it up (for the case where a process only created KDirWatch instances other than KDirWatch::self)


Repository: kcoreaddons


Description (updated)
-------

The use of KDirWatch from KSycoca exposed that KDirWatch was really not
thread-safe. The internal singleton design brings a ton of races between
KDirWatch instances. Easy solution: one "singleton" per thread, using
QThreadStorage.

The main impact is on memory management; the "singleton" is no longer refcounted
but simply deleted at the end of the thread, by QThreadStorage (including for
the main thread). Since this happens at ~QApp time, we need to zero the d pointer
of KDirWatch::self(), which is deleted by a global-static dtor later on.

REVIEW: 125217


Diffs (updated)
-----

  src/lib/io/kdirwatch_p.h b2ac9d419bb0b0939a4eaa235a55634482db1cc6 
  src/lib/io/kdirwatch.cpp 5e015c10cb99dbe4a21f944d4c0d0b364f2a7263 
  src/lib/io/kdirwatch.h f1593e73898c3610c2a560714ed608367120afa8 

Diff: https://git.reviewboard.kde.org/r/125217/diff/


Testing
-------

unittests in kcoreaddons still pass, but now kservice's ksycocathreadtest passes every time, unlike before.


Thanks,

David Faure

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/kde-frameworks-devel/attachments/20150914/0d1a5b60/attachment-0001.html>