[Marble-devel] Review Request 120004: Solve copy problems by adding missing detach() calls. Other minor fixes.

Calin Cruceru calin at rosedu.org
Thu Sep 4 06:58:56 UTC 2014

This is an automatically generated e-mail. To reply, visit:

(Updated Sept. 4, 2014, 9:58 a.m.)

Review request for Marble.


Added a test for tours and another one to also highlight the geometry parent handling in GeoDataPlacemark.

Repository: marble


So I witnessed some time ago that the copy of GeoDataGeometry based classes did not work as expected and I figured out that the main cause were some missing detach() calls in functions returning either non-const references or pointer to some non-const member, which made possible their modification without creating a new *Private instance.

* This patch adds those missing detach() calls on all GeoDataGeometry and GeoDataFeature based classes (shout me if we are using this technique in other classes so I can take a look there as well).

* There is also one important change I thought of in GeoDataPlacemark. Before, in the copy constructor, we used to do p()->m_geometry->setParent( this ) which broke the invariant m_geometry->parent() == this for other. However, there was one more problem: when detach() was called for the first time, the newly created geometry (in GeoDataPlacemarkPrivate) did not have a parent set (it was zeroed) so after a GeoDataPlacemark copy was done, the other geometry's parent would point to the old placemark and the newly created (not immediately, but after the first detach() call) placemark wouldn't have its parent set at all. So, as a workaround, I added p()->m_geometry->setParent(this) after each detach() call in GeoDataPlacemark and removed the call from the copy constructor, so that the invariant is only broken for this, but only until the first detach() call and then each geometry correctly points to its parent.

There is one thing I did not find a solution for and may be a problem: the folderList(), placemarkList() and featureList() methods in GeoDataContainare are marked as const but return QVector<GeoDataFeature*> so it may be possible to modify some GeoDataFeature objects through those pointers immediately after a copy of GeoDataContainer, without calling detach() - so it would affect members of the copied object.

Diffs (updated)

  Solve-copy-problems-in-GeoDataFeature-and-GeoDataGeometry-based-classes-v4.diff PRE-CREATION 
  src/lib/marble/geodata/data/GeoDataPolygon.cpp b8e0f32 
  src/lib/marble/geodata/data/GeoDataPoint_p.h a7c19e0 
  src/lib/marble/geodata/data/GeoDataPoint.cpp 8c5d9c9 
  src/plugins/render/annotate/AnnotatePlugin.cpp b1c8072 
  src/lib/marble/geodata/data/GeoDataContainer.cpp f0fae00 
  src/lib/marble/geodata/data/GeoDataFeature.cpp 9b354a2 
  src/lib/marble/geodata/data/GeoDataGeometry.cpp 8dfdd52 
  src/lib/marble/geodata/data/GeoDataMultiGeometry.cpp 7c6d148 
  src/lib/marble/geodata/data/GeoDataMultiGeometry_p.h df96074 
  tests/TestGeometryDetach.cpp PRE-CREATION 
  tests/CMakeLists.txt 47b2c68 
  tests/TestFeatureDetach.cpp PRE-CREATION 
  src/lib/marble/geodata/data/GeoDataPoint.h 401b25f 
  src/lib/marble/geodata/data/GeoDataPlacemark.cpp b2d0304 
  src/lib/marble/geodata/data/GeoDataMultiTrack.cpp 9cc89bc 

Diff: https://git.reviewboard.kde.org/r/120004/diff/



Calin Cruceru

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/marble-devel/attachments/20140904/4928e6a1/attachment.html>

More information about the Marble-devel mailing list