[Qgis-developer] Bug in PAL labelling over the dateline

[Marco Hugentobler]

Hi Jeremy

Applied in r14352, thank you for the good problem description and the 
testdata.

> Quick question: Is it ok to change the PAL library like this, or is it
> better to put an exception handler around the PAL calls. I'm just thinking
> about future merges of the PAL library into the QGIS source.

There are already other PAL changes in the QGIS source tree that need to be 
merged back into PAL, e.g. Martins curved labeling or the possibility to force 
certain label positions for data defined labeling. So agreed, we should do that 
soon to stay in sync with PAL.

Regards,
Marco

Am Mittwoch, 6. Oktober 2010, um 22.45:29 schrieb Jeremy Palmer:
> Hi Marco,
> 
> Yes you are right, it's definitely easier and it doesn't mean changing the
> interfaces or too much code. So if you think this is the path of least
> resistant I'd be keen to fly with it.
> 
> Quick question: Is it ok to change the PAL library like this, or is it
> better to put an exception handler around the PAL calls. I'm just thinking
> about future merges of the PAL library into the QGIS source.
> 
> Cheers,
> Jeremy
> 
> -----Original Message-----
> From: Marco Hugentobler [mailto:marco.hugentobler at sourcepole.ch]
> Sent: Thursday, October 07, 2010 12:51 AM
> To: qgis-developer at lists.osgeo.org
> Subject: Re: [Qgis-developer] Bug in PAL labelling over the dateline
> 
> Hi Jeremy
> 
> Wouldn't it be easiest to, in pal.cpp, just ignore layers that are already
> there instead of throwing an exception? And the same could be done for
> features crossing the dateline in pal/layer.cpp:239
> 
> If the label initialisation would be done in QgsMapRenderer, it would be
> necessary to change the layer interface to pass additional attribute
> indices to fetch (e.g. necessary in case of data defined label settings).
> 
> Regards,
> Marco
> 
> >Am Mittwoch, 6. Oktober 2010, um 08.03:58 schrieb Jeremy Palmer:
> >Hi All,
> >
> >I've discovered a label rendering bug for geographic vectors layer when
> >the mapcanvas bound spans the dateline. To replicate this do the
> >following:
> >
> >1) Add a vector layer with a geographic coordinate system, preferably one
> 
> that already spans the dateline. I have attached a simple Shapefile.
> 
> >2) Enable PAL labelling on this layer.
> >3) Turn on OTF projections. Make sure you use a coordinate system that
> >will work for data over the dateline. I used EPSG:2193 for the attached
> >Shapefile. 4) Zoom to a projected extent that will ensure that layer will
> >span +/- 180 dateline.
> >
> >This should raise a generic exception that is caught by QGIS, which leaves
> >the application in state that is not usable. This exception is actually a
> >re- throw of PalException::LayerExists.
> >
> >At this stage I have debugged the application and determined this:
> >
> >* When otf transformed are enabled, layer is has a geographic coordinate
> >system and the canvas extents span the dateline two separate render
> >operation will be initiated by QgsMapRenderer. See lines 475-419 in
> >qgsmaprenderer.cpp
> >
> >* Within the QgsMapRenderer::Render method the labelling engine is
> >initialised, calls are made to the layer to draw the data, the labels are
> >rendered and finally the labelling engine is cleaned up.
> >
> >* Within the layer draw method the labelling engine prepares the layer for
> >labelling i.e rendererContext.labelingEngine()->prepareLayer(...). In the
> >context of the PAL label engine this sets the layer up for label rendering
> >and adds metadata about the layer to the PAL object. i.e.
> >Pal::addLayer(...)
> >
> >* You can not add a duplicate layer to the PAL object otherwise it will
> >throw a PalException::LayerExists exception. See line 186 of pal.cpp.
> >
> >From a simplistic point of view it seems to me that preparation of the
> >label engine should occur within the QgsMapRenderer::Render method and
> >not deeper down in the layer draw methods. I see that in the layer draw
> >methods call the labelingEngine::prepareLayer() to determine if labelling
> >is required and which fields are required for the data select. It seems
> >that this information could be associated with the QgsRenderContext and
> >set within
> >QgsMapRenderer::Render() then and accessed within the layer draw.
> >
> >Can someone please comment on this?
> >
> >Thanks heaps,
> >Jeremy


-- 
Dr. Marco Hugentobler
Sourcepole -  Linux & Open Source Solutions
Webereistrasse 66, 8134 Adliswil, Switzerland
marco.hugentobler at sourcepole.ch http://www.sourcepole.ch
Technical Advisor QGIS Project Steering Committee