Review Request 127437: Fix many threading issues in KUrlCompletion

[David Faure]

> On March 21, 2016, 6:18 p.m., David Edmundson wrote:
> > src/widgets/kurlcompletion.cpp, line 1445
> > <https://git.reviewboard.kde.org/r/127437/diff/1/?file=453655#file453655line1445>
> >
> >     this looks wrong, shouldn't it be insertItems (or ::addMatches which does the same thing)
> >     
> >     we'll get this method called twice, once for each thread that takes longer than the initial timeout, setItems will discard matches from the first thread.

Only *one* of the threads is running. We are *either* running user completion ("~") or directory completion.
Note how the previous code was calling stop+clear, as well.

I could have kept clear+addMatches, it would make no difference, it's the same as setItems (AFAICS).

I just don't call stop() anymore, because now stop() is "asking the thread to finish (abort)", while this slot is called once the thread has finished (or just before it really finishes, but at least it's about to, it doesn't need to be requested to stop).

Maybe an "else" between the last two if()s in the slot would make it clearer?

Thanks for the review, I'm open for more suggestions :-)


- David


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://git.reviewboard.kde.org/r/127437/#review93829
-----------------------------------------------------------


On March 20, 2016, 2:46 p.m., David Faure wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://git.reviewboard.kde.org/r/127437/
> -----------------------------------------------------------
> 
> (Updated March 20, 2016, 2:46 p.m.)
> 
> 
> Review request for KDE Frameworks.
> 
> 
> Repository: kio
> 
> 
> Description
> -------
> 
> * Major race with other threads due to using QDir::setCurrent()
> * Race conditions on m_terminationRequested and m_matches
> * Race with previous completion thread when its posted event arrives after cancelling
> * Cancellation code spread out in many methods and never done fully correctly
> * isRunning() was missing one of the two threads, making unit test fail in valgrind
> * Fix the rarely-hit code path where the thread isn't finished after 200ms
>    - the current search string was lost because finished() wasn't called
>    - the matches were not used, in case of user-completion (AFAICS)
>    - changing the search string while the thread was running could lead to the old search
>      string still being used for completion
>      (the misnamed finished() wasn't called, so KCompletion didn't get the new string)
>   => added a variant of the unittest which doesn't wait for the thread initially
> 
> + Simplify code using signal/slots rather than a custom event
> + Simplify code using enum rather than casting to/from int
> 
> Most of these bugs are from 2004 (e.g. commit ec83c408619e3) ;)
> 
> REVIEW: 127437
> 
> 
> Diffs
> -----
> 
>   autotests/CMakeLists.txt 4dff0a24d31897a9641a70017bd87e33415cef14 
>   autotests/kurlcompletiontest.cpp eef39ff17940fadcb9492e5e7092070c976310d4 
>   src/widgets/CMakeLists.txt 87dac50b377f6ea4c3cd39e9afa37a4680aecf31 
>   src/widgets/kurlcompletion.h 14fda22a0e08bcf2da30e19fed577c8bda64a4ca 
>   src/widgets/kurlcompletion.cpp 1dc8f4898fb78d6f49e687462446007a10305f98 
> 
> Diff: https://git.reviewboard.kde.org/r/127437/diff/
> 
> 
> Testing
> -------
> 
> Hit a crash in DirectoryListThread when playing with kopenwithtest.
> 
> kurlcompletiontest extended, kurlcompletion-nowait added, both pass, also in valgrind (different timings).
> 
> 
> Thanks,
> 
> David Faure
> 
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/kde-frameworks-devel/attachments/20160322/a36d1d92/attachment.html>