[Kde-pim] Review Request 119933: Optimize: Share set of \SEEN flag across items.

[Milian Wolff]

> On Aug. 26, 2014, 4:52 a.m., Kevin Krammer wrote:
> > akonadi/protocolhelper.cpp, line 546
> > <https://git.reviewboard.kde.org/r/119933/diff/1/?file=307604#file307604line546>
> >
> >     just checking: this is still fine if multiple threads execute it simultantiously?
> 
> Milian Wolff wrote:
>     According to the C++11 memory model, yes - this is threadsafe. Afaik it was threadsafe when using GCC for a long time already, not sure about MSVC though.
> 
> Dan Vrátil wrote:
>     Wouldn't declaring the variable in global scope be safe in all cases?
> 
> Milian Wolff wrote:
>     No, global statics are a bad idea and I've seen crashes due to them on shutdown - at least when using clang as the compiler. See e.g. https://projects.kde.org/projects/extragear/kdevelop/kdevplatform/repository/revisions/e059f07d79a58f4a755b338e89b52a34e6984472
>     or
>     https://projects.kde.org/projects/extragear/kdevelop/kdevelop/repository/revisions/03efc1faabd69368c7685ca3778bc8fd22528cb9
>     
>     it probably depends on the type you create globally but generally I'd try to avoid global statics. If you guys think the current patch too unsafe for some reason, I'll provide a different patch which will then init the seen flag in a function-static. By calling this function then at a defined place before threads start, we can ensure that no issues arise, even without c++11.
> 
> Christian Mollekopf wrote:
>     Unless a global variable x is used from within the destructor of a global object y that was initialized before (and hence destructed after) the global variable x, that shouldn't be a problem.
>     Regarding thread-safety, with C++11 the initialization is threadsafe (and apparently this has been available as a gcc specific feature before as well), but I'm not sure about the acess of the variable. If this is indeed safe I wouldn't see why we can't just enable C++11 and move forward.

Access of const Qt types that are implicitly shared is threadsafe. The only thing that gets changed internally is the refcounter which is atomic and threadsafe.


- Milian


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://git.reviewboard.kde.org/r/119933/#review65259
-----------------------------------------------------------


On Aug. 25, 2014, 6:29 p.m., Milian Wolff wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://git.reviewboard.kde.org/r/119933/
> -----------------------------------------------------------
> 
> (Updated Aug. 25, 2014, 6:29 p.m.)
> 
> 
> Review request for KDEPIM-Libraries.
> 
> 
> Repository: kdepimlibs
> 
> 
> Description
> -------
> 
> Most items just have a single \SEEN flag in their list of flags.
> Currently, we share the \SEEN flag, but not the list itself. By
> special-casing this extremely often occurring case, we can decrease
> the memory consumption easily. In my case it's a difference of ca.
> 7MB (171MB before vs. 164MB after).
> 
> 
> Diffs
> -----
> 
>   akonadi/protocolhelper.cpp 3d496667fa622cac2dda7d5a0ca4c92b5a59d2fe 
> 
> Diff: https://git.reviewboard.kde.org/r/119933/diff/
> 
> 
> Testing
> -------
> 
> ran it before and after this patch and compared the memory consumption. That went down noticeably. Also added a static counter to compare cache hits/misses temporarily, which showed this cache is easily hit ca. 40k times and only missed 4k times (so a 10:1 ratio, not bad at all :))
> 
> 
> Thanks,
> 
> Milian Wolff
> 
>

_______________________________________________
KDE PIM mailing list kde-pim at kde.org
https://mail.kde.org/mailman/listinfo/kde-pim
KDE PIM home page at http://pim.kde.org/