KisDabShape produces dab of shape 10x5. From my reading of the code it's the test that's wrong. However, if KisDabShape code is wrong, please fix it. Dmitry, please check this, I think you know this part best.

Jenkins has a 2 min time limit. Trying to avoid it.

Reducing the number of iterations to avoid Jenkins timeout.

Dmitry, approxChangeRect didn't match "true value". I can't tell which one is right. Please, check that this is expected value.

Dmitry, I see that several new signals are sent by setResource. I added them to the test. Please, check that that's expected behavior.

alarm has to be cmsMAXCHANNELS long with everything set to 0, except alarm values.

Wolthera, I'm assuming the test was incorrect. The input value is out of gamut, so we should get "alarm" value. So we should assert equality to alarm value. Please, check.

Your fix looks correct :)

I would recommend adding a diff like that:

diff --git a/libs/image/tests/kis_convolution_painter_test.cpp b/libs/image/tests/kis_convolution_painter_test.cpp
index 755ace9..9a7c045 100644
--- a/libs/image/tests/kis_convolution_painter_test.cpp
+++ b/libs/image/tests/kis_convolution_painter_test.cpp
@@ -300,7 +300,7 @@ void KisConvolutionPainterTest::benchmarkConvolution()
 
         dbgKrita << "Diameter:" << diameter << "time:" << timer.elapsed();
 
-        if(diameter < 10) {
+        if(diameter < 1) {
             diameter += 2;
         } else {
             diameter += 8;

We use different algorithms for diameter < 5 and diameter > 5: spatial vs fftw, so it is useful to have the results for both ranges.

yep :)

That should be ok, because the rect is generated on a basis of a set of points sampled from the transform. The test might have broken because the sampling algorithm has changed.

Looks good!

yes, lcms docs confirms that :)

Please use memset instead. bzero is a deprecated function.