[SKIKO-1112](https://youtrack.jetbrains.com/issue/SKIKO-1112) Missed
`PathBuilder.offset` and `PathBuilder.transform` in `0.144.0`

Missed part from #1154
JNI/KN C++ bindings are already in place

Fixes [SKIKO-1094](https://youtrack.jetbrains.com/issue/SKIKO-1094)
Update skia to m144

They are out of the scope of the skiko library.

[SKIKO-1020](https://youtrack.jetbrains.com/issue/SKIKO-1020) Deprecate
and remove URIManager
[SKIKO-1021](https://youtrack.jetbrains.com/issue/SKIKO-1021) Deprecate
and remove ClipboardManager
[SKIKO-1022](https://youtrack.jetbrains.com/issue/SKIKO-1022) Deprecate
and remove CursorManager

All public links are updated to
https://redirector.kotlinlang.org/maven/compose-dev
Internal buildSrc links to maven directly to avoid unnecessary redirects

Fixes SKIKO-1108

Previously:
- The bash script required specifying `SKIA_VERSION` separately from
`gradle.properties`, leading to version mismatches
- Script hardcoded `SKIA_TARGET` default to `iosSim`, requiring manual
override on non-macOS systems

This change:
- Added `skiaVersionFromEnvOrProperties` property and `printSkiaVersion`
task. The bash script now queries Gradle for the version instead of
maintaining its own default.
- Target platform now defaults to current OS (`hostOs.id`) instead of
hardcoded `iosSim`.
- Introduced `BuildLocalSkiaTask` that handles Python build script
invocation, architecture detection, and validation.
- Added `SkiaTarget` enum centralizing platform-specific Gradle flags
and architecture mappings.
- Simplified bash script: Now focuses solely on git operations
(clone/checkout skia-pack), delegating all build logic to Gradle.

### Usage
```sh
# Version from gradle.properties, target auto-detected from OS
./build-with-local-skia.sh

# Override either via environment variables
SKIA_VERSION=m138-80d088a-2 SKIA_TARGET=iosSim ./build-with-local-skia.sh

# Or use Gradle tasks directly
./gradlew prepareLocalSkiaBuild -Pskia.target=linux
```

Fixes [SKIKO-1103](https://youtrack.jetbrains.com/issue/SKIKO-1103)
Properly handle Dockerfile changes in CI

### 1. Handle Dockerfile Changes in CI
When a PR modifies both the `Dockerfile` and C++ source, CI must build
the library against the **new** container logic before merging to catch
compatibility issues.
Workflows now detect `Dockerfile` changes and automatically build images
locally before running tests. When no `Dockerfile` changes are detected,
published images from `ghcr.io` are used.

### 2. Avoid GLIBC Mismatches: Run GitHub Actions Outside Containers
Running GitHub Actions native steps (like `actions/checkout`) inside
custom containers causes GLIBC version mismatches since GitHub's
Node.js-based actions require newer GLIBC than Amazon Linux 2 provides.
This change introduced new composite action `docker-skiko-run` that runs
GitHub actions outside the container on `ubuntu-24.04` runner, and only
executes build/test commands inside the Docker container via `docker
run`.

### 3. Environment Alignment: Use `linux-compat` for GitHub Actions
Builds
GitHub Actions used Ubuntu 20.04 images with GLIBC 2.31, while TeamCity
publishing used Amazon Linux 2 with GLIBC 2.26. This mismatch could mask
GLIBC compatibility issues during PR validation.
Most GitHub Actions workflows now use `linux-compat` (Amazon Linux 2).
- Web builds are out of the scope here because emsdk requires newer
GLIBC.
- Cross-compilation is out of the scope because there is no simply way
to get arm shared libraries to x64 image on AL2

Note: TeamCity publishing should be updated to use `linux-compat` too

### 4. Introduce Orchestrator Workflows
New orchestrator workflows compose existing test/build/docs workflows:
- **`pull-request.yml`** - Runs on every PR: detects Docker changes,
builds images if needed (dry-run), runs tests + publish dry run + docs
validation
- **`post-merge.yml`** - Runs on push to master/release: detects Docker
changes, publishes images if changed, runs tests + publish dry run +
docs publication

So, we should have fewer "Run CI" temporary PRs now

### 5. Documentation as Pre-Merge Check
Documentation builds now run inside the same `linux-compat` Docker
environment used for library builds, and are validated as part of PR
checks (previously only ran post-merge).


### 6. Docker Tags Use Branch Names
Published Docker images are tagged with the branch name (e.g., `master`,
`release/0.9.46`), so the release branches might publish its own version
of the image. This way changes in `master` shouldn't prevent making a
patch for a previous version if it's required

Fixes
[SKIKO-1059](https://youtrack.jetbrains.com/issue/SKIKO-1059/Output-the-argument-size-in-ColorMatrix-in-case-of-error)
Converts to value class for inlining and no class allocation and fixes
the requirement message

This function will copy the pixels bytes directly from JS ArrayBuffer to
skiko linear memory. The allocated memory will be freed when SkBitmap is
removed.

Fixes [SKIKO-1101](https://youtrack.jetbrains.com/issue/SKIKO-1101)
Prevent possible version mismatch between skiko Kotlin and JNI parts

Fixes CMP-1460

Requires skia-pack version update before merging:
https://github.com/JetBrains/skia-pack/pull/68

Required for my linuxArm64 device which only support EGL.

https://youtrack.jetbrains.com/issue/SKIKO-918

This is part 2 of [this
pr](https://github.com/JetBrains/skiko/pull/1137)

**What has changed?**
1. Added more extension functions that prevent intermediate object
allocations from a lot of the Skiko api surface
2. Migrated EnumClass.values() to entries in a lot of classes (maybe
some have been missed)

**Next**
Try (important keyword here :), cause if not wrong there are Skia and
Compose diffs on how they're treated) doing less copies of Matrix

## Release Notes
Added API that can accept primitive values instead of `Rect`, `RRect`, `Offset` to avoid additional object allocation

Fixes [SKIKO-1081](https://youtrack.jetbrains.com/issue/SKIKO-1081) Make
docker images for building skiko publicly available

The Skiko project previously used a private Docker registry
(`registry.jetbrains.team/p/ui/skiko-docker`) for build containers,
which required secrets for authentication. This prevented CI
verification from fork PRs even after manual approval.

The change migrates to GitHub Container Registry (GHCR), making
containers publicly accessible and enabling CI to run on fork PRs.

Containers built:
https://github.com/orgs/JetBrains/packages?repo_name=skiko

Triggers:
- Push to master: Automatically builds and publishes all Docker images
- Pull requests: Builds images in dry-run mode (no publishing) for
verification
- Manual workflow dispatch: Option to publish or just build
- File changes: Triggers on changes to `skiko/docker/**` or the workflow
itself

This PR fixes https://youtrack.jetbrains.com/issue/SKIKO-996

The PixelRef bindings have two mistakes in them:

1. `rowBytes` is defined as a `NativePointer`, whereas it is actually an
integer with the number of bytes per row. This is the same as
`Bitmap::rowBytes`.
2. It is missing the `pixels` property that actually returns a pointer
to the pixel data.

See: https://api.skia.org/classSkPixelRef.html

On January 12th `18.4` was removed from GitHub Actions runners
See https://github.com/actions/runner-images/issues/13392

Fixes [SKIKO-950](https://youtrack.jetbrains.com/issue/SKIKO-950)

Fixes
https://youtrack.jetbrains.com/projects/SKIKO/issues/SKIKO-1040/Web-targets-not-optimal-use-of-FinalizationRegistry-API

It improves some microbenhcmarks according to
https://github.com/WebKit/JetStream/pull/84#discussion_r2254866128

Added some functions to Skia canvas to be able to avoid creating
intermediate objects (SkRect, SkRRect, etc) on some hot paths.
See [this
conversation](https://github.com/JetBrains/compose-multiplatform-core/pull/2543#discussion_r2483008621)
for the context of the PR.
This is the first step to allow for Compose Ui SkiaBackedCanvas to avoid
unnecessary intermediate allocations:
1.
https://github.com/JetBrains/compose-multiplatform-core/blob/47af63a3986292608982f421bb345eb0a2032a98/compose/ui/ui-graphics/src/skikoMain/kotlin/androidx/compose/ui/graphics/SkiaBackedCanvas.skiko.kt#L189
2.
https://github.com/JetBrains/compose-multiplatform-core/blob/47af63a3986292608982f421bb345eb0a2032a98/compose/ui/ui-graphics/src/skikoMain/kotlin/androidx/compose/ui/graphics/SkiaBackedCanvas.skiko.kt#L141
3.
https://github.com/JetBrains/compose-multiplatform-core/blob/47af63a3986292608982f421bb345eb0a2032a98/compose/ui/ui-graphics/src/skikoMain/kotlin/androidx/compose/ui/graphics/SkiaBackedCanvas.skiko.kt#L128
4.
https://github.com/JetBrains/compose-multiplatform-core/blob/47af63a3986292608982f421bb345eb0a2032a98/compose/ui/ui-graphics/src/skikoMain/kotlin/androidx/compose/ui/graphics/SkiaBackedCanvas.skiko.kt#L189
5.
https://github.com/JetBrains/compose-multiplatform-core/blob/47af63a3986292608982f421bb345eb0a2032a98/compose/ui/ui-graphics/src/skikoMain/kotlin/androidx/compose/ui/graphics/SkiaBackedCanvas.skiko.kt#L211
among others.
**Next steps**
-> open pr in Compose Ui to use these more optimized functions
-> More optimizations on the Skiko side, Matrixes is a good example, CMP
and Skiko are processing Matrixes 2 times, toSkiaMatrix, which basically
is the same as a wrapper around FloatArray and then in the toInterop
method which copies the array (unavoidable). A good optimization would
be to just pass the Compose Matrix underlying `values` property which is
a FloatArray that toInterop accepts and so a lot of intermediate objects
and transformations could be avoided altogether

Part of [CMP-81](https://youtrack.jetbrains.com/issue/CMP-81) Check
Compose repositories for following JetBrains rules
and [CMP-82](https://youtrack.jetbrains.com/issue/CMP-82) Make a
contribution guide for Compose repositories

Partially reverts JetBrains/skiko#1117 (except excluding "Intel(R) HD
Graphics 4600")

Fixes
https://youtrack.jetbrains.com/issue/CMP-9352/Compose-doesnt-work-in-Parallels-Windows11-arm64

## Release notes
### Features - Desktop
- Cleanup old unpacked binaries from `~/.skiko`. Use
`skiko.data.cleanup.days` to configure the retention period in days

[SKIKO-1069](https://youtrack.jetbrains.com/issue/SKIKO-1069)
[libstdc++.so.6+0xdf9b4] std::__codecvt_utf8_utf16_base::do_unshift

It introduces symbol references to `GLIBCXX_3.4.22` (GCC 6.1.0 per
[doc](https://gcc.gnu.org/onlinedocs/libstdc++/manual/abi.html)), which
should be fine

Binary size of binary is reduced by ~3MB

Reverts JetBrains/skiko#1114

Fixes https://youtrack.jetbrains.com/issue/SKIKO-1042/Support-ANGLE

The fix is the same as https://github.com/JetBrains/skiko/pull/1082,
except:
- the code is rearranged in new commits after resolving merge conflicts
(the merge is solved by applying their/ours)
- fixed "Task
':publishSkikoJvmRuntimeAngleWindowsX64PublicationToComposeRepoRepository'
uses this output of task ':signSkikoJvmRuntimeWindowsX64Publication'
without declaring an explicit or implicit dependency" (as in
https://github.com/JetBrains/skiko/pull/1127)

WIP: fix CI failure

## Testing
1. `./gradlew publishToMavenLocal`
2. In SkiaAwtSample:
```
./gradlew runWithAngleEnabled
```
3. [CI is
successful](https://buildserver.labs.intellij.net/buildConfiguration/Skiko_PublishSnapshot?branch=revert-1126-revert-1125-igor.demin%2Ffix-skiko-build-awtRuntimeElements&buildTypeTab=overview&mode=builds)

Reverts JetBrains/skiko#1126 with additional fixes

Fixes:
```
org.gradle.internal.execution.WorkValidationException: A problem was found with the configuration of task ':signKotlinMultiplatformPublication' (type 'Sign').
  - Gradle detected a problem with the following location: '/mnt/agent/work/381d14fe560aa354/skiko/build/libs/skiko-0.9.33-javadoc.jar.asc'.

    Reason: Task ':publishAwtRuntimeElementsPublicationToComposeRepoRepository' uses this output of task ':signKotlinMultiplatformPublication' without declaring an explicit or implicit dependency. This can lead to incorrect results being produced, depending on what order the tasks are executed.
```

Part of [SKIKO-530](https://youtrack.jetbrains.com/issue/SKIKO-530)
support for all platforms of awt-runtime in one jar

## Testing
- doesn't fail on configuration stage:
```
./gradlew publishAwtRuntimeElementsPublicationToMavenLocal -Pmaven.central.sign=true -Pmaven.central.sign.key=TEST -Pmaven.central.sign.password=TEST
```
- [CI is
successful](https://buildserver.labs.intellij.net/buildConfiguration/Skiko_PublishSnapshot/827040952)

In 0.9.36, there was a missing symbol that caused
```
libskiko-linux-x64.so: undefined symbol: _ZNSt19_Sp_make_shared_tag5_S_eqERKSt9type_info
```

It's because the build attempts to statically link the C++ standard
library, but:
- Skia's object files reference newer libstdc++ symbols
- GCC 7.3's `libstdc++` doesn't contain these symbols
- The static linking fails to resolve these symbols

GCC was downgraded from 9 to 7 in #1130 during the switch to AL2 to
provide better GLIBC compatibility.
Since AL2
[provides](https://docs.aws.amazon.com/linux/al2/ug/c-cplusplus.html)
only GCC 7.3 by default and GCC 10, installing GCC 9 is problematic
there, so it's bumped to 10

However, GCC 10 introduces outline atomics as default for new ARM chips,
the build might fail due to missing symbols like
```
org_jetbrains_skia_FontMgrWithFallback__1nDefaultWithFallbackFontProvider: error: undefined reference to '__aarch64_ldadd4_relax'
org_jetbrains_skia_FontMgrWithFallback__1nDefaultWithFallbackFontProvider: error: undefined reference to '__aarch64_ldadd4_acq_rel'
```
so, to keep things compatible with the previous GCC version,
`-mno-outline-atomics` option was added to disable this new behavior.

[SKIKO-1066](https://youtrack.jetbrains.com/issue/SKIKO-1066) Support
glibc 2.27
[CMP-9307](https://youtrack.jetbrains.com/issue/CMP-9307) Support glibc
2.27

- Added `linux-compat` docker image based on #1123. It works on both x64
and arm64 architectures
- `multistrap` tool was replaced by installing required packages inside
`linux-compat` and `linux-amd64` docker images
- `linux-arm64` uses GCC instead of clang (aligning between
distributions)
- GitHub checks now use precompiled docker images to reuse the same
environment
- `ci.yml` and `web.yml` were combined into `tests.yml`
- `publishToMavenLocal` checks are moved to `publish-dry-run.yml`
- Add building an Android sample as part of CI checks

---------
Co-authored-by: Vladislav.Ertel <vladislav.ertel@jetbrains.com>

In the compose hot reload project, we have seen races from multiple
tests trying to unpack skiko into the .skiko directory. Currently, the
library loading is only synchronized within the project. This MR also
adds a .lock file, which has to be locked when modifying the dataDir.

Note: This MR also changed the Library.load method from always entering
a monitor, to resolving an atomic reference, best case.

## Release Notes
### Fixes - Desktop
- Fixed a race condition that occurred when multiple processes attempted
to unpack Skiko binary files at startup

See
[CMP-9080](https://youtrack.jetbrains.com/issue/CMP-9080/Native-Crash-in-Skiko-on-desktop#focus=Comments-27-13023060.0-0)
for details

Reverts JetBrains/skiko#1125

Fails on CI

Regression after https://github.com/JetBrains/skiko/pull/1119 was added
on CI

```
org.gradle.internal.execution.WorkValidationException: A problem was found with the configuration of task ':signKotlinMultiplatformPublication' (type 'Sign').
  - Gradle detected a problem with the following location: '/mnt/agent/work/381d14fe560aa354/skiko/build/libs/skiko-0.9.33-javadoc.jar.asc'.

    Reason: Task ':publishAwtRuntimeElementsPublicationToComposeRepoRepository' uses this output of task ':signKotlinMultiplatformPublication' without declaring an explicit or implicit dependency. This can lead to incorrect results being produced, depending on what order the tasks are executed.

    Possible solutions:
      1. Declare task ':signKotlinMultiplatformPublication' as an input of ':publishAwtRuntimeElementsPublicationToComposeRepoRepository'.
      2. Declare an explicit dependency on ':signKotlinMultiplatformPublication' from ':publishAwtRuntimeElementsPublicationToComposeRepoRepository' using Task#dependsOn.
      3. Declare an explicit dependency on ':signKotlinMultiplatformPublication' from ':publishAwtRuntimeElementsPublicationToComposeRepoRepository' using Task#mustRunAfter.
```

The fix is blind as we can only test it in master