GitLab

In file included from /home/boud/dev/krita/libs/pigment/compositeops/KoCompositeOpCopy2.h:26,
                 from /home/boud/dev/krita/libs/pigment/colorspaces/KoAlphaColorSpace.cpp:35:
/home/boud/dev/krita/libs/pigment/compositeops/KoCompositeOpBase.h: In instantiation of ‘void KoCompositeOpBase<_CSTraits, _compositeOp>::genericComposite(const KoCompositeOp::ParameterInfo&, const QBitArray&) const [with bool useMask = true; bool alphaLocked = true; bool allChannelFlags = true; _CSTraits = KoColorSpaceTrait<half, 1, 0>; _compositeOp = AlphaColorSpaceMultiplyOp<KoColorSpaceTrait<half, 1, 0> >]’:
/home/boud/dev/krita/libs/pigment/compositeops/KoCompositeOpBase.h:67:72:   required from ‘void KoCompositeOpBase<_CSTraits, _compositeOp>::composite(const KoCompositeOp::ParameterInfo&) const [with _CSTraits = KoColorSpaceTrait<half, 1, 0>; _compositeOp = AlphaColorSpaceMultiplyOp<KoColorSpaceTrait<half, 1, 0> >]’
/home/boud/dev/krita/libs/pigment/compositeops/KoCompositeOpBase.h:58:10:   required from here
/home/boud/dev/krita/libs/pigment/compositeops/KoCompositeOpBase.h:110:27: warning: ‘void* memset(void*, int, size_t)’ clearing an object of type ‘KoCompositeOpBase<KoColorSpaceTrait<half, 1, 0>, AlphaColorSpaceMultiplyOp<KoColorSpaceTrait<half, 1, 0> > >::channels_type’ {aka ‘class half’} with no trivial copy-assignment; use assignment or value-initialization instead [-Wclass-memaccess]
                     memset(dst, 0, pixel_size);
                     ~~~~~~^~~~~~~~~~~~~~~~~~~~

We don't have the HDR profiles on Jenkins, so we need to check
which ones are available.

No need to assert if a certain profile cannot be found; we can
disregard that and handle not getting a colorspace back.

Apparently, on Jenkins we cannot create all of these colorspaces,
so don't fail by dereferencing a null pointer:

********* Start testing of TestLcmsRGBP2020PQColorSpace *********
Config: Using QtTest library 5.12.4, Qt 5.12.4 (x86_64-little_endian-lp64 shared (dynamic) release build; by GCC 9.1.1 20190611 [gcc-9-branch revision 272147])
PASS   : TestLcmsRGBP2020PQColorSpace::initTestCase()
AddressSanitizer:DEADLYSIGNAL
=================================================================
==6181==ERROR: AddressSanitizer: SEGV on unknown address 0x000000000008 (pc 0x7f3f02031167 bp 0x7fff6f5c1290 sp 0x7fff6f5c1280 T0)
==6181==The signal is caused by a READ memory access.
==6181==Hint: address points to the zero page.
    #0 0x7f3f02031166 in KoColorSpace::id() const /home/jenkins/workspace/Extragear/krita/kf5-qt5 SUSEQt5.12/libs/pigment/KoColorSpace.cpp:106
    #1 0x41535a in testRoundTrip(KoColorSpace const*, KoColorSpace const*, SourceType) /home/jenkins/workspace/Extragear/krita/kf5-qt5 SUSEQt5.12/plugins/color/lcms2engine/tests/TestLcmsRGBP2020PQColorSpace.cpp:44
    #2 0x418726 in testRoundTrip(KoID const&, KoID const&, SourceType) /home/jenkins/workspace/Extragear/krita/kf5-qt5 SUSEQt5.12/plugins/color/lcms2engine/tests/TestLcmsRGBP2020PQColorSpace.cpp:118
    #3 0x4192b2 in TestLcmsRGBP2020PQColorSpace::test() /home/jenkins/workspace/Extragear/krita/kf5-qt5 SUSEQt5.12/plugins/color/lcms2engine/tests/TestLcmsRGBP2020PQColorSpace.cpp:143
    #4 0x40aa2c in TestLcmsRGBP2020PQColorSpace::qt_static_metacall(QObject*, QMetaObject::Call, int, void**) /home/jenkins/workspace/Extragear/krita/kf5-qt5 SUSEQt5.12/build/plugins/color/lcms2engine/tests/TestLcmsRGBP2020PQColorSpace_autogen/EWIEGA46WW/moc_TestLcmsRGBP2020PQColorSpace.cpp:78
    #5 0x7f3efcc6889a in QMetaMethod::invoke(QObject*, Qt::ConnectionType, QGenericReturnArgument, QGenericArgument, QGenericArgument, QGenericArgument, QGenericArgument, QGenericArgument, QGenericArgument, QGenericArgument, QGenericArgument, QGenericArgument, QGenericArgument) const (/usr/lib64/libQt5Core.so.5+0x29689a)
    #6 0x7f3f050a0962  (/usr/lib64/libQt5Test.so.5+0x19962)
    #7 0x7f3f050a1352  (/usr/lib64/libQt5Test.so.5+0x1a352)
    #8 0x7f3f050a1910  (/usr/lib64/libQt5Test.so.5+0x1a910)
    #9 0x7f3f050a1cda in QTest::qRun() (/usr/lib64/libQt5Test.so.5+0x1acda)
    #10 0x7f3f050a1edb in QTest::qExec(QObject*, int, char**) (/usr/lib64/libQt5Test.so.5+0x1aedb)
    #11 0x41b9d7 in main /home/jenkins/workspace/Extragear/krita/kf5-qt5 SUSEQt5.12/plugins/color/lcms2engine/tests/TestLcmsRGBP2020PQColorSpace.cpp:185
    #12 0x7f3efc4f2bca in __libc_start_main (/lib64/libc.so.6+0x26bca)
    #13 0x40a8f9 in _start (/home/jenkins/workspace/Extragear/krita/kf5-qt5 SUSEQt5.12/build/plugins/color/lcms2engine/tests/TestLcmsRGBP2020PQColorSpace+0x40a8f9)

AddressSanitizer can not provide additional info.
SUMMARY: AddressSanitizer: SEGV /home/jenkins/workspace/Extragear/krita/kf5-qt5 SUSEQt5.12/libs/pigment/KoColorSpace.cpp:106 in KoColorSpace::id() const
==6181==ABORTING

Either an update to Qt broke testing unregistered types, or
this test has always been broken, or some other change to
the KisImageSignalRouter class made it necessary that the
arguments were registered with qRegisterMetaType...

But KoColorProfile and KoColorSpace cannot be registered
with qRegisterMetaType because they are abstract classes,
so we cannot check those signals in this test.

This does not work if source was closed inbetween operations or
source is in another krita instance.

BUG:408421
BUG:404595

This makes the polygon tool behave as the manual states, if a modifier is
pressed it will be retained even it is let go mid selection.

BUG:376007

We now return a tuple for every action, this allows to place the action
in different contexts. We only use multiple contexts for reload action
for now.

Differential revision: https://phabricator.kde.org/D10839

Before this crash, Overview docker would use its canvas in case
of rotation without checking if the canvas exist. This patch
fixes it by adding additional checks and returning from the function
in case of empty canvas.

Supposedly "it happens because when the canvas is deleted,
canvasOffsetXChanged is emitted".

Before this commit, that would result in crash if you drop
the layer or the group with the layer on the canvas of another file
(in the subwindows mode). This commit contains a fix so now it works
as expected.

BUG:408019

This reverts commit 6b0ebe2d.

This is actually wrong. EXRConverter::Private::makeLayerNamesUnique()
actually can cause exr layer names to be different from krita layer
names.

When zooming/rotating, because of subtle displacement
events sent by qt, KisInputManager started painting or
to be precise, created dots on canvas. This patch
discards such displacements, so, no dots.

Maniphest: T10784

This facilitates both rotation and zooming actions of the
canvas

Maniphest: T10784

Maniphest: T10784

Because otherwise krita will just seem to crash.

This is not a fatal error, and we should be able to load files we
saved in a broken way.

On loading an exr, we check whether the exr layername and the
krita layername are the same with a safe assert, so we probably
should make sure of that when saving as well.

BUG:396158

In case of conflict in i18n, keep the version of the branch "ours"
To resolve a particular conflict, "git checkout --ours path/to/file.desktop"

Before this commit this action would mirror them vertically.
(It's under Layer -> Transform All Layers -> menu entry)

OpenGL 2.1 systems are supported, but Small Colro Selector is not
available for them, because it doesn't have legacy shaders (needs
to be implemented).

CCBUG:408872

Before this commit if you first selected the color model,
then color profile, then color depth, the color profile would switch to
default for this color depth no matter that the user just set it
manually. This is mostly the problem if someone is for example using
Rec2020 profile for HDR images and switches between 32 and 16 bit
color depths.

Behaviour now:
- by default color profile changes to the default one that suits
the color model and color depth
- if the user sets the profile manually, it is remembered
and if it changes color depth, the profile doesn't change
(keep in mind that "(Default)" suffix is still attached
to the default profile, so it's easy to find in case the user wants it
- if the user changes the color model, the default behaviour
(switching profile automatically) is restored until the user
sets up the profile maually again.

BUG:406700

They cause artifacts and we don't have a fix yet :(

CCBUG:404133

Note from the committer:
This commit was initially made on krita/4.2 branch.
However between 4.2.1 and 4.2.2 the stable branch was
reconstructed, which caused all commits that were exclusively
on the previous krita/4.2 (and not on master) to be missing.

The previous commit hash: d3e4c3aa299cf10a3ef167213f50799baba9c782

Also regarding the artifacts: it never worked, optimization were
enabled during the creamy flow implementation, but they never should
be enabled until someone fix them properly.

ShapeGroupContainerModel::remove() will call KoShapeGroup::invalidateSizeCache(),
even when the shape group is going to be destroyed. This method is called after
KoShapeGroup's destructor, so at this time we no longer have access to members of
KoShapeGroup, i.e. its d-pointer. So this call will cause access violation as the
instance `QScopedPointer<Private> d` in KoShapeGroup has already been destroyed.

This is a result of getting rid of derived d-pointers, because if we use a derived
d-pointer, it only gets destroyed in the ancestor class (KoShape). We temporarily
fix this problem by manually call `model()->deleteOwnedShapes()` in KoShapeGroup's
destructor.

CCMAIL:tusooa@vista.aero

This fixes a crash when pressing cancel when you're trying to
create a RGB8 linear gamma image, which shouldn't be possible.

The class KisSharedDescendent used to be employed to wrap up
derived d-pointers, but since we have already refactored them
out, there is no reason to keep them now.

KoImageData was explicitly shared, and the implicit sharing
breaks TestImageCollection