Commit by dkazakov :: r1199447 krita/trunk/koffice/krita/image/ (6 files in 2 dirs): (link)

Fixed a full refresh for Clone layers

Now the Async Merger knows about one more type of layer position:
N_EXTRA. Such layer will be counted in a need rect chain and will be
updated (original+projection), but it will not be merged into the
final projection of the image. This is done for preliminary updates of
the sources of Clone layers.

Btw, i would suggest everyone to perform 'make clean' in
./krita/image/ folder, because many parts of walkers are
inlined. There is no need to do a full clean, just in ./image/
directory.

CCMAIL:kimageshop at kde dot org
BUG:257532

Commit by dkazakov :: r1199348 krita/trunk/koffice/krita/ui/ (3 files in 2 dirs): (link)

Removed deprecated stuff

• repaintOriginal() was deprecated by the beginning of February
• steDirty()'s are useless because the image is locked while loading

Commit by dkazakov :: r1199261 krita/trunk/koffice/krita/image/kis_transformation_mask.cpp: (link)

Make proper default values for KisTransformationMask

KisKraSaverTest uses default constructed value of the mask for the
testing so change the default scale to a sane value. Nevertheless we
don't use these masks anywhere yet.

Commit by dkazakov :: r1199240 krita/trunk/koffice/krita/ (4 files in 2 dirs): (link)

Fix Clone Layer loading

The link to the source layer should be initialized.

Commit by dkazakov :: r1198898 krita/trunk/koffice/krita/ui/ (kis_doc2.cc kis_view2.cpp): (link)

The full image refresh should be done on document loading only

BUG:257264

Commit by dkazakov :: r1198799 krita/trunk/koffice/krita/ui/canvas/ (2 files): (link)

Made KisToolPerspectiveGrid work in interactive mode

There was a bug that lead to an infinite loop. It happened due to a
(NaN,NaN) intersection points those appeared after calculation of
intersection of null-length lines %). Now we just check for the length
of the line.

Commit by dkazakov :: r1198207 krita/trunk/koffice/krita/ui/canvas/ (2 files): (link)

Fixed several bugs in KisPerspectiveGridManager

What is done:
1) Now the manager doesn't even try to paint the lines inside the
polygon that is not convex. It used to lead to a really weird
picture of a hedgehog on a user's screen :P
2) Fixed an "infinity"-bug. Eigen::Hyperplane::intersection() returns
a random point in case the lines are parallel. Now an additional
check returns a point somewhere around infinity to workaround this
behavior.
3) Changed the algorithm that chooses a line to chop up into
"subdivisions". Now the farthest line from the focus is used for
the division.

Still to be done:

• Make KisPerspectiveGridTool interactive. It will be a really short fix. I just need to overcome a small infinite loop somewhere ;) BUG:251731

Commit by dkazakov :: r1196934 krita/trunk/koffice/krita/ui/canvas/kis_grid_painter_configuration.cpp: (link)

Fix usual Grid Tool

The pen should be cosmetic to be painted nicely.
CCBUG:251731

Commit by dkazakov :: r1196930 krita/trunk/koffice/krita/ui/tool/kis_tool_freehand.cc: (link)

KisToolFreehand should accept all the events in painting mode

BUG:256415

Commit by dkazakov :: r1196807 krita/trunk/koffice/krita/image/ (3 files in 2 dirs): (link)

Added a benchmark for KisConvolutionPainter

Adjusted the workers as well. Now the type of the worker depends not
the available engine only, but on the size of the kernel as well. Even
from theoretical perspective, KisConvolutionWorker is faster when the
size of the kernel is less than 5x5px.

Some conclusions from benchmarking the FFTW convolution worker:
1) FFT and IFFT of the image take 30% of the time each.
2) FFT of the kernel (even though the kernel has the same size as the
   image, but it is very sparse) takes 7.5% only.
3) Creation of a plan for FFT and IFFT take 4% each.

4) There is a possibility to make the worker faster. That is to use
   the "wisdom" (FFTW_MEASURE flag) recomended by FFTW3
   documentation. In case of activation of this flag the convolution
   becomes about 20% faster. But the use of this flag will make us
   change our system a bit: we will have to use descrete sizes of
   the convolution areas and we will need to save these "words of
   wisdom" somewhere.

CCMAIL:kimageshop at kde dot org

Commit by dkazakov :: r1196674 krita/trunk/koffice/krita/image/kis_painter.cc: (link)

Fixed KisPainter::setAlphaLocked() not to break setChannelFlags()

Commit by dkazakov :: r1196595 krita/trunk/koffice/krita/image/ (3 files in 2 dirs): (link)

Made KisConvolutionWorkerSpatial do convolution instead of correlation

Commit by dkazakov :: r1196381 krita/trunk/koffice/krita/image/tests/ (5 files in 2 dirs): (link)

Made new tests for convolution painter

They are not relative now. They check absolute values.

Commit by dkazakov :: r1194751 krita/trunk/koffice/krita/image/ (3 files in 2 dirs): (link)

Fix spatial convolution filter

Rounding for integers should have been done on a level of
KisMathToolbox.

Commit by dkazakov :: r1194562 krita/trunk/koffice/krita/image/tests/ (2 files): (link)

Added a real unittest for KisConvolutionPainter

This test checks the correctness of the convolution algorithm by its
(convolution) definition.

Results:
KisConvolutionWorkerFFT works fine
KisConvolutionWorkerSpatial does not work fine =)

Commit by dkazakov :: r1194510 krita/trunk/koffice/krita/image/tests/kis_layer_test.cpp: (link)

Make a test in KisLayerTest faster

90s for a single test is too much for regular unit-testing. When
something happens you can always test this case more thoroughtly.

Commit by dkazakov :: r1193814 krita/trunk/koffice/krita/image/kis_image.h: (link)

Deprecate KisImage::mergedImage() function

It is intended to do a very strange thing (it doesn't guarantee
anything, because it can't even lock the image) and its functionality is
fully covered by other methods of KisImage. See a comment in the
header to get more info.

Commit by dkazakov :: r1193812 krita/trunk/koffice/krita/ui/ (7 files in 2 dirs): (link)

Fix the border effects in OpenGL canvas

This patch makes the textures representing the image overlap. The size
of the textures is still 256x256 pixels, But effectively they carry
254x254 pixels of information. The overlapping 1px border makes linear
interpolation algorithm happy, so scaling is done in the right way.

The patch fixes "image->bounds()" border effect as well. How does it
solve it? It just repeats the first and the last pixels of every row
and column. So the interpolation runs successfully.

BUG:251113

Commit by dkazakov :: r1193811 krita/trunk/koffice/krita/ui/opengl/ (5 files): (link)

A major clean-up for KisOpenGLImageTextures

o Encapsulated all the texture tiles into a separate objects
  (KisTextureTile). This will allow us to add borders to the tiles.
o Removed deprecated/unused methods.

Commit by dkazakov :: r1193259 krita/trunk/koffice/ (4 files in 2 dirs): (link)

A proper fix for bug 255682

The problem was in KoShapePaste, because it performed setParent() and
controller->addShape() calls separately. There is no need to change
Krita's part of shape layers for that (that is why this patch reverts
commit 1192386).

The change to KoShapePaste was approved by Thomas on a mailing list
yesterday.

CCBUG:255682