My app is used a lot in the background, and I noticed high CPU usage in
the native macOS version. With this PR, CPU usage is reduced when window
is invisible by disabling drawing. This was already implemented for the
JVM, and that behaviour is now copied to native macOS. So the code is
similar to JVM parts in:
-
https://github.com/JetBrains/skiko/blob/master/skiko/src/awtMain/kotlin/org/jetbrains/skiko/redrawer/MetalRedrawer.kt
-
https://github.com/JetBrains/skiko/blob/master/skiko/src/awtMain/objectiveC/macos/MetalRedrawer.mm

The draw function is now suspend, which required moving some other
functions around. I tried to keep the implementation similar to the JVM
version.

# Testing
Tested using `./gradlew runNative` in `samples/SkiaMultiplatformSample`.
Now the CPU usage is reduced a lot when moving another (non transparent)
app over the window.

# Old behaviour


https://github.com/user-attachments/assets/60ad3398-3f44-4e2f-aca7-44a4607d0840

# New behaviour with 300ms timeout
See reduction in CPU usage when window is invisible


https://github.com/user-attachments/assets/106efdeb-0044-4411-9b26-5a6b8886ff04


# New behaviour without 300ms timeout



https://github.com/user-attachments/assets/b986548b-d7ab-497b-9ddb-5b46ca016f83

Fixes
https://youtrack.jetbrains.com/issue/CMP-7603/Fix-Maven-Central-publication
for Skiko

Just a quick fix that adds an updated library.

For a proper fix I created an issue
https://youtrack.jetbrains.com/issue/SKIKO-1003/Unify-Maven-publication-of-Skiko-with-Compose

Official Skia documentation suggests that it's [_highly
recommended_](https://skia.org/docs/user/build/#highly-recommended-build-with-clang-cl)
to build Skia with Clang-CL on Windows, and that this dramatically
improves Skia performance with Software rendering and in other areas.
This corresponds with my experience, so here it is.

Related Skia-pack PR: https://github.com/JetBrains/skia-pack/pull/64

Prerequisite: #1024

This PR is split-off from and is a prerequisite of Clang-cl PR: #1020

BreakIterator.clone() is broken on JVM and seemingly leads to a
use-after-free behaviour. Clang-cl makes this issue more visible, likely
due to address sanitisation.

Explanation why it is deleted in an internal channel:
https://jetbrains.slack.com/archives/C02DDNREC77/p1738678000232729
* `breakIteratorCloneTest` is failing with an access violation in Debug
mode with additional checks, so using it wasn't safe
* it is not used anywhere

This is a breaking change and will require bumping Skiko version to
0.9.0.

Required to fix https://youtrack.jetbrains.com/issue/CMP-7508

Test: covered by existing tests

It is a simplified adoption of Android's `RenderNode`.

```kt
/**
 * <p>RenderNode is used to build hardware accelerated rendering hierarchies. Each RenderNode
 * contains both a display list as well as a set of properties that affect the rendering of the
 * display list. RenderNodes are used internally for all Views by default and are not typically
 * used directly.</p>
 *
 * <p>RenderNodes are used to divide up the rendering content of a complex scene into smaller
 * pieces that can then be updated individually more cheaply. Updating part of the scene only needs
 * to update the display list or properties of a small number of RenderNode instead of redrawing
 * everything from scratch. A RenderNode only needs its display list re-recorded when its content
 * alone should be changed. RenderNodes can also be transformed without re-recording the display
 * list through the transform properties.</p>
 ```
 
 This is a more correct approach to make `GraphicsLayer` in Compose invalidation independently.
 - Moved drawing callback to C++ side to avoid extra interop costs
 - Switched from `SkPicture` placeholder to custom `SkDrawable` implementation

Needed to run benchmarks without vsync interference.

The code extracted from Direct3DSwingRedrawer with minimal changes.

- implementation for other API will be added when it is needed (Metal is
considered later)
- there are no plans to make commonMain abstraction in `skiko` as it is
not needed at the moment

## Testing
- manually in benchmarks with this code:
```
@OptIn(ExperimentalSkikoApi::class)
class DirectXGraphicsContext() : DesktopGraphicsContext {
    private val context = DirectXOffscreenContext()
    private var texture: DirectXOffscreenContext.Texture? = null

    override fun surface(width: Int, height: Int): Surface {
        texture?.close()
        texture = context.Texture(width, height)
        return Surface.makeFromBackendRenderTarget(
            context.directContext,
            texture!!.backendRenderTarget,
            SurfaceOrigin.TOP_LEFT,
            SurfaceColorFormat.BGRA_8888,
            ColorSpace.sRGB,
            SurfaceProps(pixelGeometry = PixelGeometry.UNKNOWN)
        ) ?: throw IllegalStateException("Can't create Surface")
    }

    override suspend fun awaitGPUCompletion() {
        texture?.waitForCompletion()
    }

    override fun close() {
        texture?.close()
        context.close()
    }
}

interface DesktopGraphicsContext : GraphicsContext, AutoCloseable
```
- new test for SkiaSwingLayer
- new test for DirectXOffscreenContext

Currently fails with
```
Error: This request has been automatically failed because it uses a deprecated version of `actions/upload-artifact: v3`. Learn more: https://github.blog/changelog/2024-04-16-deprecation-notice-v3-of-the-artifact-actions/
```

Fixes
https://youtrack.jetbrains.com/issue/SKIKO-994/Add-logging-for-rendering

That approach worked but had performance issues. Also, `SkPicture` is
immutable by design, so we shouldn't try to change that.

#1014 is a more correct and efficient way to implement it

New `PictureFilterCanvas` class that allows override picture rendering
during drawing

Required for
https://github.com/JetBrains/compose-multiplatform-core/pull/1766

Fixes
https://youtrack.jetbrains.com/issue/CMP-7352
https://youtrack.jetbrains.com/issue/CMP-7345

This is an alternative (better) fix for
https://youtrack.jetbrains.com/issue/CMP-6721

Similar change is skia is required too:
https://github.com/JetBrains/skia-pack/pull/60

Reason: macos-12 is deprecated and not available anymore
https://github.com/actions/runner-images/issues/10721
(Several previous CI runs were incomplete bacause of this:
https://github.com/JetBrains/skiko/actions, including a commit merged to
master -
https://github.com/JetBrains/skiko/commit/bccb17a47610b17ee5e297d3665b4fc0625da9e1)

In order to use two different versions of skiko in one app (using
multiple classloaders), it is required to use the unpacking strategy, as
the skiko.library.path is a system property, which is same for all
classloaders.

This change will allow to use presigned natives shipped together with
the app, while still being able to use multiple versions as the property
skiko.library.path will be loaded from current classloader resources,
with fallback to system property.
Co-authored-by: Jakub Senohrabek <jakub@senohrabek.net>

Co-authored-by: Oleksandr Karpovich <a.n.karpovich@gmail.com>

Fix for
https://youtrack.jetbrains.com/issue/CMP-6722/Excessive-garbage-generation-from-redrawing

Previously, the `draw` method would allocate a `BufferedImage` on every
redraw, which is wasteful. This fix changed `SwingOffscreenRenderer` to
reuse the same instance of `BufferedImage` and only clear it on redraws.
Only changing the draw size would allocate a new `BufferedImage`.

Co-authored-by: Artem Bobrov <artem.bobrov@jetbrains.com>

Fixes the following error when calling `URIHandler_openUri` on iOS 18:

```
The caller of UIApplication.openURL(_:) needs to migrate to the non-deprecated UIApplication.open(_:options:completionHandler:). Force returning false (NO).
```

Fixes: https://youtrack.jetbrains.com/issue/CMP-6699

Support configurable vertical centering for implementing Compose's
`LineHeightStyle.Alignment`.

https://youtrack.jetbrains.com/issue/CMP-2602

See also:
- https://github.com/JetBrains/skia/pull/5
- https://github.com/JetBrains/compose-multiplatform-core/pull/1569

Fixes
https://youtrack.jetbrains.com/issue/CMP-5856/Desktop-ComposePanel-size-breaks-with-.fillMax-modifiers#focus=Comments-27-10632441.0-0

Fixes
https://youtrack.jetbrains.com/issue/CMP-5968/Compose-content-is-rendered-in-the-wrong-place-in-IJ-when-using-AWT-compositing

Regression after
https://github.com/JetBrains/skiko/pull/661/files#diff-910a6e28fda20a00bc98c6a8a04f74ab701d79e841b9baddf146b810e610572fR363
(`setBounds` is called more often, but not enough as `ancestorMoved`)

When a panel changes its position without changing its size, `doLayout`
isn't called because the content itself wasn't changed. But we still
need to update the bounds of the underlying layer.

## Testing
-
https://youtrack.jetbrains.com/issue/CMP-5856/Desktop-ComposePanel-size-breaks-with-.fillMax-modifiers#focus=Comments-27-10632441.0-0
isn't reproducible after the fix
- there are no resize glitches

## Release Notes
### Fixes
- Fix bounds of ComposePanel in IntelliJ on macOs

Adding a new flag `-Pskiko.wasm.withProfiling=true` which adds
`--profiling` flag to emcc, so it preserves the function names - makes
it easier to use the browser profiler.

By default, it's disabled.

See https://github.com/JetBrains/skiko/issues/973

Skia-pack used in this skiko:
https://github.com/JetBrains/skia-pack/releases/tag/m126-1d69d9b-2

This is supposed to be fix for
https://github.com/JetBrains/skiko/issues/967

The idea is pass strings as (utf8) byte arrays directly through the
border just like we do it on native

The download directory path now depends on the buildType: Debug and
Release


___
It affected our buildServer CI pipeline where it builds Release first
and then Debug. And downloadTask was considered UP-TO-DATE when Debug
build was attempted. Therefore the build used "Release" version of skia
instead of "Debug. "

Added a Debug build to the githiub workflow too.

Add a flag to control number of buffers used by Metal. Add a flag to control Vsync on window resize (#968)

#### Number of buffers
MacOS
[doc](https://developer.apple.com/documentation/quartzcore/cametallayer/2938720-maximumdrawablecount?language=objc).
As far as I know AWT uses double buffering by default. So we would like
to experiment with this setting in Fleet. Also I expect that double
buffering should reduce user interaction latency.

#### Control VSync on window resize

I've noticed that on Linux and Windows we don't wait for VSync when
repainting synchronously, but only on macOS we do.
I suspect that locking EDT when we are waiting for VSync might lead to
problems with app responsivenes. So it would be nice to experiment with
this setting.

Usage of main-thread AppKit APIs in other thread is assumed to lead to
the crash.
Second attempt for https://github.com/JetBrains/skiko/pull/958.

Structure display link control reactively around Wake/Sleep and Screen
change notifications.

---------
Co-authored-by: Igor Demin <igordmn@users.noreply.github.com>

One still can use optional platfrom-specific skia dependencies using
exactly the property which was used before, however, by default all
build targets are using the same version of skia prebuilt artefacts.