Symbology brush transform #2666
Conversation
…oint if rendering a map tile. Therefore it is assured, that the image pattern matches for adjacent tiles of the same scale
|
@mhugent that sounds exciting. Do you have any screenshots at hand demonstrating the positive impact of this PR? |
|
@nirvn: there are two test images in this pull request, each composed of four part images. Without the PR, there would be discontinuities at the borders of the part images. |
| //transform brush to upper left corner of geometry bbox | ||
| QPointF leftCorner = points.boundingRect().topLeft(); | ||
| QTransform leftCornerTransform = QTransform::fromTranslate( leftCorner.x(), leftCorner.y() ); | ||
| mBrush.setTransform( leftCornerTransform ); |
There was a problem hiding this comment.
Looks like in some places a transform is already set on the brush. Might be safer to take the existing brush transform and then translate it (similar to how rotation is done in line 1593)
|
Looks good to me! just the few minor things noted above + failing tests to be addressed. |
| @@ -50,7 +50,7 @@ class QgsVectorLayerDiagramProvider; | |||
| * @note added in 2.4 | |||
| * @note not available in Python bindings | |||
| */ | |||
| class QgsVectorLayerRenderer : public QgsMapLayerRenderer | |||
| class CORE_EXPORT QgsVectorLayerRenderer : public QgsMapLayerRenderer | |||
There was a problem hiding this comment.
Why to export the class in qgis_core library? it was not meant for that... I see the class is used in the unit test but it is an implementation detail that QgsVectorLayerRenderer exists and unit tests should not use it....
…es and restore transformation after rendering
|
@mhugent - to get the test to pass under OSX you'll need to use the Eg - find failure result on dash (http://dash.orfeo-toolbox.org/testDetails.php?test=36433187&build=214498), then run (where http://dash.orfeo-toolbox.org/displayImage.php?imgid=71581 is the url for the rendered image) This will create a new "_mask.png" file alongside the test reference image which will give the test tolerance respecting that mask. Not sure why the sip build is failing though - maybe it's the missing enum values? |
| @@ -276,6 +276,9 @@ class QgsMapCanvas : QGraphicsView | |||
| //! true if antialising is enabled | |||
| bool antiAliasingEnabled() const; | |||
|
|
|||
| //! sets map tile rendering flag | |||
| void enableMapTileRendering( bool theFlag ); | |||
There was a problem hiding this comment.
should this be 'setMapTileRendering' for consistency?
There was a problem hiding this comment.
I don't mind if 'set' or 'enable'. Just thought enableMapTileRendering would be consistent with enableAntiAliasing.
|
@mhugent I was going to say - if you want to get this in before freeze maybe #ifdef out the tests for now, and we'll sort them out later |
|
Now it succeeds on Mac but not on Linux. The error message: [ 95%] Building CXX object python/CMakeFiles/python_module_qgis__gui.dir/gui/sip_guipart0.cpp.o clang: error: unable to execute command: Killed is not really revealing. Is that a timeout or something? |
|
+1 from me to see this merged even though the freeze window for 2.14 is closed! Thanks for doing the @mhugent - tiling with current QGIS releases is a pain because of the issues you mentioned. |
|
This PR can also be seen as some sort of bug fix, so feature freeze might not be a problem. |
|
Compiles fine for me on Fedora linux and I can't see where the source of the problem is. I'd say revert all sip file changes and re-enable them one-by-one to find the culprit? |
|
@m-kuhn , @nyalldawson : looks like the linux build now already fails when creating the build environment. |
|
Same sip problems even if no sipfile was changed |
|
@mhugent can you rebase, readd the sip bindings, and force push? Hopefully it's fixed now. |
This reverts commit f728465.
|
Worked. Thanks, @nyalldawson ! |
Implements the improved mouse/key modifier behaviour discussed in: http://osgeo-org.1560.x6.nabble.com/Key-modifiers-with-selection-tc5239653.html Specifically, For click-and-drag selections: - holding shift = add to selection - holding ctrl = substract from selection - holding ctrl+shift = intersect with current selection - holding alt (can be used with shift/ctrl too) = change from "intersects" to "fully contains" selection mode For single-click selections: - holding shift or ctrl = toggle whether feature is selected (ie either add to current selection or remove from current selection) This brings the canvas behaviour into line with other design apps and also with the composer behaviour. (fix qgis#2666)
This pull request adds a new render flag 'RenderMapTile'. The idea is that if setting this in a project, QGIS knows that a map tile is generated for use in a tile cache. E.g. with the current pull request, QGIS makes sure that image fills between adjacent tiles match if the flag is set (it uses a brush transformation in QgsImageFillSymbolLayer). In future, there are other things which could be considered for tile image rendering. E.g. some wms servers don't place labels into the border zone if images for tile caches are rendered.
A unit test is provided too. The test assembles four adjacent tiles to one image and compares against a reference image.