wrong direction of label direction symbol

[strk]

As you can see in the attached image, the "direction symbol" added by the label disagrees with the "direction symbol" of the line decoration.

I know the line decoration (the endpoint arrow) is correct. Note that line decoration was fixed by r15076 (see #3434) so a similar approach could be just copied over

The bug occurs with r11539 on a x84-64

[mhugent]

Is this only for vertical lines or for others too? Could you attach a small testdataset to reproduce the problem?

[strk]

Seems to be vertical lines only. I attach a shapefile. Turn on arrows decoration on symbology-ng and orientation symbol on labels. Arrow decorations are correct.

[mhugent]

Fixed in r15591

[strk]

Indeed r15591 fixed the attached edge.zip case, but it's still wrong for another dataset :/ I'm attaching it..

[strk]

In particular, edges with id 5 and 7 of qgisbug3543_1.zip are disagreed upon between label and line decoration. They are both going down, so line decoration is correct and label orientation symbol is wrong.

[strk]

I'd add that the original testcase only contained lines oriented from bottom up, which may explain why this case wasn't cought.

[strk]

The attached patch (bug3643.diff) fixes it for me, for both datasets. Might not be extremely elegant, but should be also more robust..

[strk]

A slightly more elegant version:

diff --git a/src/core/pal/feature.cpp b/src/core/pal/feature.cpp
index 6f657b1..5501434 100644
--- a/src/core/pal/feature.cpp
+++ b/src/core/pal/feature.cpp
@@ -596,7 +596,7 @@ namespace pal
       {
         //std::cout << alpha*180/M_PI << std::endl;
         if ( flags & FLAG_MAP_ORIENTATION )
-          reversed = ( alpha > M_PI / 2 || alpha < -M_PI / 2 );
+          reversed = ( (ex == bx && ey < by ) || ( alpha > M_PI / 2 || alpha < -M_PI / 2 ) );
 
         if (( !reversed && ( flags & FLAG_ABOVE_LINE ) ) || ( reversed && ( flags & FLAG_BELOW_LINE ) ) )
           positions->push_back( new LabelPosition( i, bx + cos( beta ) *distlabel , by + sin( beta ) *distlabel, xrm, yrm, alpha, cost, this, reversed ) ); // Line

[strk]

Even smaller is just moving re-adding the equal sign but to the right part:

-          reversed = ( alpha > M_PI / 2 || alpha < -M_PI / 2 );
+          reversed = ( alpha > M_PI / 2 || alpha <= -M_PI / 2 );

Works fine in my case, altought the equality comparison between the return from atan2 and M_PI/2 looks scary (miracles!)

This final version seems to be the closest to the original code, where the equal sign was in the first rather than second side of

.

[mhugent]

Applied the second version in r15637. Thanks!

[strk]

Great. I confirm the fix. You can close the grave now :)