Jus deslare the assert as "safe". The proper fix is coming.

CCBUG:447506

BUG:443621

BUG:439729

luz paz authored Jun 11, 2021 and Dmitry Kazakov committed Jun 14, 2021

Found via `codespell -q 3 -S *.xpm,./3rdparty -L aline,ba,fo,hsi,lod,pard,redy,uint,uptodate,vertexe,vertexes`

We shouldn't ask for projection() in group layers too often.

BUG:438384

after duplication; The canvas would get stuck w/ rendered
shaded selection mask overlay and not properly respond to
selection of other nodes.

BUG:430474

The ASL layer styles should not store the resources inside themselves.
Instead, they should upload the resources into the global resource server
and fetch them from there when needed. It fixes GUI elements to know
what resource is active in the layer, which actually has a cloned
layer style.

The patch introduces a special helper class KisLinkedResourceWrapper,
which remembers md5 and filename of the resource and then can be used
to fetch the resource from the server using this info.

Renamed some functions in KisKeyframeChannel

(with Eoin.)

We shouldn't use layer->extent() for the thumbnail generation, because
extent may also include indirect painting device, which may have a
different size, not like the one of projection (e.g. when selection is
active).

CCBUG:394439

Also correctly shows onion skins during isolation. Also fixes some edge
cases where drawing on isolated layers didn't work properly.

BUG:394199

We need to take layer's masks' change rect into account. We cannot
use normal changeRect() call, because it never shrinks the update rect
and adds projection device bounds (obviously, because it is supposed
to render stuff and not just shwo a decoration).

That is why the patch introduces KisLayer::tightUserVisibleBounds(), which
calculates tight bounding rect of the layer, taking all the masks into
account.

BUG:410163

Before this commit, KisAslStorage would always be empty.
This commit adds support for reading from asl files
and extracting both patterns and layer styles.

Overlay-like layer styles (overlay, inner shadow, inner glow and satin)
seem to be applied in a complicated way:

1) Firstly the copy of the layer-stack projection is created
2) Then a copy of original layer alpha channel is extracted into a
   selection
3) Alpha channel of the layer itself is maxed-out to 1.0
4) The layer and all the layer overlay-like styles are merged into
   the projection
5) The the copy of the original projection from step 1) is
   copied onto the projection again using selection from step 2).

This weird technique is used to ensure that color mixing happens
exactly in full opacity, as if the colors would have no alpha-channel.

BUG:412283

We cannot just skip rendering of layers, which has any clones attached,
becasue clones may fetch data from their projections.

BUG:408167
BUG:405536

Without any image present, the store will just drop the
projection without any reusing. It may cause small slowdowns
in the beginning of the stroke and (theoretically) memory
fragmentation on Windows.

I'm not sure if it actually the reason of the 408133, so we
need a confirmation from the reporter first.

CCBUG:408133

In Krita we have two [0] types of "dependent" nodes, that is,
whose contents depends on the result of a merge of other layers.

KisAdjustmentLayer depends on the result of composing all the
layer below it. KisGroupLayer depends on the result of composing
its children.

All these layers can have filter masks and/or layer styles. To
handle them, the patch introduces a term needRectForOriginal().
needRectForOriginal(rect) means "what area of layer's original
device should be prepared to correct result in 'rect' of the
'final device'". Under 'final device' it may be considered
projection for a layer or a final composed planes for a
projection plane.

What is the difference between needRectForOriginal() and needRect?

needRect() defines "what area of layers below should be prepared
to get correctly merged image". Obviously, group layer doesn't
depend on layers below (it depends on its children) and adjustment
layer's needRect also includes filter's needRect, which is not the
case for needRectForOriginal().

[1] --- actually, we have three. KisCloneLayer is also "dependent",
        but it is not covered by this patch.

BUG:366419
CCBUG:390299

We have implemented correct painting on masks, but we have never
implemented a projection for them. Indeed, for normal rendering
process they don't need it. But when a mask is selected in isolated
mode, it still needs a projection, so that the use could see the result
of the painting.

BUG:403576

When indirect painting is disabled (or a mask is removed) the layer
becomes projection-free, that is, projection device should be recycled.
The problem is that the projection may still be accessed by worker
threads, therefore the actual cleaning of the devices should happen in
an exclusive context.

The patch adds a special exclusive spontaneous job, that does all the
cleaning when all workers are guaranteed to be idle/paused.

Since the places where this method is called check whether 0 is
returned or not, it's wrong to assert here.

Not having a layer style projection plane is a
normal situation when the style is empty

BUG:398788

We should deep copy the layer style projections
when cloning the image. Otherwise shape layers and
other delayed update nodes might cause rerender of
the layers stack and lacking layers styles will become
visible in the saved image.

BUG:396224

We should deep copy the layer style projections
when cloning the image. Otherwise shape layers and
other delayed update nodes might cause rerender of
the layers stack and lacking layers styles will become
visible in the saved image.

BUG:396224

1) Selection masks now use KisProjectionLeaf hierarchy
   to embed themselves onto the top of the layers stack.
   That sounds a bit weird, but, yes, now the two
   hierarchies differ quite significantly.

2) KisImage::setOverlaySelectionMask() is used to mask one
   of the masks as an overlay.

3) KisNodeManager puts the mask into the overlay of the image
   as soon as the mask gets selected as a current layer.

4) KisProjectionLeaf now accesses the image using
   KisNodeGraphListener::graphOverlayNode(). That is not
   too nice, because it breaks the listener pattern, but
   that is the safest way to access the image from the
   internal nodes. We might want just rename this class :)

Some time ago we disabled updates after duplicating a layer. Therefore,
we should also copy the projection device. Otherwise, it will lazily
initialized with the content of the original device, which is not what
we usually want.

BUG:395142

Some time ago we disabled updates after duplicating a layer. Therefore,
we should also copy the projection device. Otherwise, it will lazily
initialized with the content of the original device, which is not what
we usually want.

BUG:395142

BUG:389503

It also fixes the tooltip used for e.g. the brush to show the brush in
the tooltip.
Additionally it is possible to show the tooltip by pressing the middle
button.

Closes T3499 Layer thumbnail on the timeline on hover

Summary:
Some actions should not be started before all the previous actions
are finished. That is especially true for the actions that work
with layers stack, like "Merge Down" and "Flatten".

Now KisDelayedSaveDialog is used not only for saving, but also for
waiting before doing usual actions. It also has a busy-loop timeout of
1 second before showing up, for not distracting painters from their
workflow when the action is postponed only a little bit.

BUG:372724
Fixes T4593

Test Plan:
Basically, the steps like in bug 372724.

1) Select multiple layers
2) Start a long stroke (1000px colorsmudge brush is a good example)
3) Press Ctrl+E ***multiple*** times quickly
4) After all the background work is finished, press Ctrl+Z

There should be no crash, and the result should be somewhat expected.

Repeat the same for:
1) Merge down
2) Flatten Image
3) Flatten Layer
4) Transform a layer using "Layers" menu(?)

Reviewers: #krita, timotheegiet, scottpetrovic

Maniphest Tasks: T4593

Differential Revision: https://phabricator.kde.org/D3449

Frederik Gladhorn authored Nov 15, 2016 and Scott Petrovic committed Nov 19, 2016

This is a simple search and replace for KisLayer which is a QObject.
qobject_cast doesn't require RTTI and is faster.