Review Request 113591: Reduce UDSEntry memory usage by sharing the contained QStrings if possible

[Frank Reininghaus]

-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
http://git.reviewboard.kde.org/r/113591/
-----------------------------------------------------------

Review request for kdelibs and David Faure.


Repository: kdelibs


Description
-------

This patch is a subset of https://git.reviewboard.kde.org/r/113355/ (I'll continue working on the other part of that request, which can be dealt with separately, at some later point).

It adds a unit test to makes sure that saving UDSEntries to a QDataStream and re-loading them works as expected, and makes use of implicit sharing of QStrings in UDSEntryPrivate::load(QDataStream &s, UDSEntry &a) to reduce the memory usage of this class (which is the major consumer of memory in Dolphin and other applications that list the contents of large directory contents with KIO).

It caches the most recently loaded QString for each UDS field in a simple QVector<QString>. This works because sharable strings like, e.g., the user and the group, usually appear at the same position in the QDataStream when retrieving a large number of UDSEntries that have been stored by a kioslave.

Note that I had made an earlier attempt to achieve the same thing using a QHash<uint, QString> to look up the cached strings ( http://pastebin.kde.org/p52a24b49 ), but the QVector<QString>-based solution turns out to be faster.


Diffs
-----

  kio/tests/udsentrytest.h PRE-CREATION 
  kio/tests/udsentrytest.cpp PRE-CREATION 
  kio/tests/CMakeLists.txt 5a1f9b5 
  kio/kio/udsentry.cpp 1e1f503 

Diff: http://git.reviewboard.kde.org/r/113591/diff/


Testing
-------

kdelibs unit tests still pass. The memory usage of both Dolphin and a simple test program that uses KIO::listDir to list the contents of a large directory (see r4 of https://git.reviewboard.kde.org/r/113355/) is reduced by ~128 bytes per item according to my tests.

A simple benchmark that simulates how 100,000 UDSEntries stored by kio_file are loaded (see r3 of https://git.reviewboard.kde.org/r/113355/) runs in 234 ms instead of 266 ms on my machine - it seems that growing the heap to provide space for the non-shared QStrings is more expensive than comparing all loaded QStrings with the cached values.


Thanks,

Frank Reininghaus

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/kde-core-devel/attachments/20131103/e959e495/attachment.htm>