…0a47d407

feature(scripting): serialize implemented methods (#12670) 8957983a44
feature(scripting): serialize implemented methods, detected by executing the generator in the editor

Co-authored-by: Gordon <pggordonhayes@gmail.com>

…25aa1

* fix(flutter): improve render texture flicker on resize

* chore: review comments and additional improvements

* fix: perform block on main thread when swapping textures

* chore: clang format

* fix: laggy graphic on macOS during window resize

* chore: update CHANGELOG

* chore: additional texture cleanup hardening

* chore: revert native changes

* docs: update CHANGELOG

gl: route feathers to atlas on Mali-G52 to avoid PLS corruption (#12753) f60e429709
On Mali-G52 (Panfrost driver, e.g. MediaTek MT8169), shapes rendered with a
feather come out with corrupted/fringed edges, and the corruption can spread to
other geometry rendered in the same pixel-local-storage (PLS) pass.

When a complex feather is rasterized directly into pixel local storage (the
`CoverageType::pixelLocalStorage` path selected in `PathDraw::SelectCoverageType`),
the Mali-G52 driver corrupts the PLS planes. This is the same class of failure
already documented and worked around for PowerVR in `RenderContextGLImpl`
(`alwaysFeatherToAtlas`). Small feathers are unaffected because they already
route to the atlas via the `atlasScaleFactor <= 0.5` check; only larger feathers
fall through to the direct-to-PLS path.

Extend the existing `alwaysFeatherToAtlas` workaround (previously PowerVR-only) to
Mali-G52, forcing feathers through the offscreen atlas instead of PLS.

Reproduced on a Mali-G52 device. The corruption is deterministic and appears only
on frames that render a feather; feather-free frames are unaffected. Setting
`alwaysFeatherToAtlas = true` eliminates the corruption, confirmed with an
on-device build.

feat(renderer): Add scissor rectangle support for clipping (#12672) ae1042e910
This adds scissor rectangle support for clipped paths (currently only enabled in Vulkan), which currently is a fill rate improvement, but can be further leveraged to add even more optimizations in the future.

chore(vulkan, android): Renderer now errors on Androids older than 10.0 (#12685) 1453e46bc1
We decided on a minimum version of Android 10.0 to allow Vulkan to run, but there was no enforcement in the Vulkan code. Now there is!

chore(scripting): allow scripts to read image properties sourced from… (#12710) 57bfa223c1
chore(scripting): allow scripts to read image properties sourced from property values and not assets

chore: prevent critical failures on future incompatible features (#12732) 4e27795b8a

feat(cmdq): Command Server Canvas Support (#12717) 47b201bfae
* made CommandServer create ScriptedVMs.

* wrapped scripting context in correct macro

* make the ScriptedVM an rcp param

* moved params around to support android

* avoid exposing ScriptVM

* remove defaults

* added no op render context impl

* to many errors revert rcp para,m

* remove more setOreContext locations

* update guard tests

* changed to static cast

* build ore_gpu_guard_test no-rtti to fix typeinfo link error

* fix detection VM ore context lifetime (retain render context)

* updated buffer tests and also factored out mock context

* fix doc

* removed uneeded function

Bezier render prototype (#12727) 4b2501f748
* Create .gitignore

* Files

* Remove loop blinn

* Move to premake

* clang format

* Fix winding loop stuff

* Remove loop blinn stuff

* Update main.cpp

* Comments

* Moved files

* clang

* Update main.cpp

test: bump RSTB version assertions to v4 (#12724) db24f50ecb

ore: GPUBuffer descriptor-table constructor, expose colorBufferHalfFloat (#12718) 0bfcc5fff7

feature(ore): resolve shader entry points by name (RSTB v4) (#12715) b553e7dcb1
feat(ore): RSTB v4 entry-point containers, resolve @vertex/@Fragment by name

ore: gate float render targets and baseVertex on backend features (#12712) 10e90773d5
* ore: gate float render targets and baseVertex on backend features

* ore: test float renderTarget and baseVertex feature guards

fix(ore): d3d12 upload stages caller region, not full subresource (#12702) 0cc614761c

fix(ore): ContextVulkan TextureVulkan upload defer to next recording CB (#12700) cccee5e480
* fix ContextVulkan TextureVulkan upload defer to next recording CB

* ore vulkan upload: batch flush barriers, fix array depth default, 64-bit region bounds

* ore vulkan upload: 64-bit bytesPerRow fallback before uint32_t guard

Deferred GPUCanvas reports speculative format (#12701) 557c18ea05
* feat(scripting): deferred GPUCanvas reports speculative format, drop preferredCanvasFormat

* docs(scripting): GPUCanvas.format is always rgba8unorm, including deferred

fix(ore): shader lookup folderPath and ContextGL uniform buffer save (#12698) 61e10a350d
fix shader lookup folderPath and ContextGL uniform buffer save

Nnnnn scripting and gamepad fixes (#12696) cb60c2b893
* fix(scripting): ensure stack is sized before pushing values

* fix(gamepad editor): fix value types on ffi

Bind native GPU texture as Rive RenderImage (#12606) f8c8d7cf8b
* feat(unity): bind native GPU texture as RenderImage for video

* feat(unity): Vulkan + D3D12 video binding backends

* harden: sRGB hint, source-format gates, tighter comments

* drop unused markChanged + RequestRedraw

refactor(ore): Render context own ore (#12661) af004065e9
* started moving to external command buffer only

* started converting all ore context to use only external cmd list

* updated vulkan and d3d12 to use gpuresources

* started moving to gpuresource

* use an upload buffer now so that it can be frame tracked

* updated android testing window

* update metal testing window

* update metal context

* fixed lua isue

* update to new api

* fix

* more

* no command buffer for native

* metlQueue requirement

* update testing window android vulkan

* fix

* framebuffer now properly gets destroyed

* updated wagyu testing window

* begin / end frame for metal texture

* updated vulkan texture and testing window gpu

* fix compile error

* fix(ore): refcount persistent VkDescriptorPool so it outlives bind groups

* ore vulkan: generational descriptor pool fix (#12678)

* ore vulkan: generational descriptor pool, no per-set free

* ore vulkan: pool rollover gm

* ci: run PR workflows on any base branch, not just master

* chore: adding baselines for ore_pool_rollover

* ore wgpu: keep CommandEncoder alive past Ore beginFrame move

* ci: switch java distribution from zulu to temurin

* fix: missing goldens

* updated no_op_ore_context for new function

* scripting: NoOpOreContext default ctor passes nullptr manager (#12692)

* scripting: link gpu_resource.cpp into documentation_generator (#12693)

feat(editor): Expose ScrollConstraint velocity and scrollActive computed properties (#12671) d797171a0b
Customer asked whether its possible to hide scrollbars when a scroll constraint is inactive. Rather than implement that behavior, we expose computed velocity values in both x and y directions which will be non-zero when the scroll constraint is being dragged or swiped and released. In addition, we expose a scrollActive computed property that indicates if the scroll constraint is active in any way (dragging, swiped and released until being settled, being dragged by a scrollbar).

Co-authored-by: Gordon <pggordonhayes@gmail.com>

feat(js): poll Rive focus state to conditionally drive focus back to the canvas from elsewhere in the DOM (#12713) 6169c50d9f
* feat(js): poll Rive focus state to conditionally drive focus back to the canvas from elsewhere in the DOM

* comment clarifications

* Early return on polling logic before looking for active SM to work with

* make fake sm stub in focusManagement tests more complete

* Formatting fix

* sp. and cleanup

* ran simplify on changes

* call focusNext/Previous when browser moves focus onto canvas automatically for keyboard interactions

feat(Android): Vulkan (#12653) 528ab110eb
Adds Vulkan support to the new Android runtime.

Includes fixes to resize behavior.

fix(runtime): assign view model instances to the right view models (#12756) 018a0724d9

Co-authored-by: Gordon <pggordonhayes@gmail.com>

…vents (#12785) c7e9a02fd8

* feat(flutter): immediately advance state machine on pointer up/down events

* chore: gate advance call on non none HitResult

* refactor: scope advanceOnPointerUpDown on RiveWidgetController (highest level)

* chore: remove extra schedule repaint call

* refactor: remove opt-out

* refactore: more removals

* chore: more docs cleanup

feat(Android/Cmdq): Advance after pointer down and up (#12786) ba9b94960c
There is possibility that if pointer down and up are received on the same frame, the down event would be missed. This addresses that by advancing by 0 in between down and up, allowing events/VMI bindings to propagate.

Nnnnn export bindable object with different (#12779) a680ab80e5
* chore(runtime): remove unused list methods
* fix(editor): export data bind if primary is not found

Add Factory::ore() and drop the redundant ore context cache in ScriptingContext (#12769) 76eeeed5d9
* add Factory::ore() and drop the redundant ore context cache in ScriptingContext

* store the command server render context as a Factory to avoid renderer header in core

* move Factory::ore() to the end of the virtual section to keep vtable slots stable

* drop the redundant command server render context arg and use the factory directly

Co-authored-by: Gordon <pggordonhayes@gmail.com>

…2794) d2e7e96c9a

* chore(flutter): remove experimental tag from shared texture usage

* chore: review comment cleanup

C++ Runtime Readme revamp (#12537) c8bb275d99

Co-authored-by: Gordon <pggordonhayes@gmail.com>

…95e2a4b2f4

* fix(flutter): treat touch/stylus pointer up as pointer exit

* chore: cleanupg

Add ASTC and ETC2 compression modes (#12518) 0ed7a0a13d
* Add files

* Update files

* Clang format

* Update image_asset.dart

* remove old rtex references

* Add block size stuff

* Update render_context.cpp

* Update premake5_pls_renderer.lua

* Clang format

* Fix build break after merge

* Fix break due to conflict

* Fix build break

* Fix break

* Fix conflict

* Update export_rive_test.dart

* Update export_rive_test.dart

* Update premake5_pls_renderer.lua

* Update theme_rename_audit_test.dart

* Update Texture_compressor.cpp

* Update encode_etc2.cpp

* Update image_asset.json

* Update image_asset.dart

* Update comment

* Potential fix for pull request finding

* Potential fix for pull request finding

* Fix astcBlockSize field-id collision with master's dragMultiplier

master added ScrollConstraint.dragMultiplier at field key 1029, colliding
with ImageAsset.astcBlockSize. Bump astcBlockSize to the next free key 1030
in both def trees and the generated Dart constant (rive_core_context.dart
references it symbolically). Minimal change — no full regen / dart-fix churn.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Update testing_window_android_vulkan.cpp (#12898) 4bf87f97a6
* Update testing_window_android_vulkan.cpp

* Update testing_window_android_vulkan.cpp

Add include instead of fwd declare (#12897) c5b59b956e
* Add include instead of fwd declare

* Update player.cpp

chore(runtime): notify property changes (#12683) 32f2243f54
* chore(runtime): add property notifier to avoid polling values on every frame
* chore: improve memory usage

feat(wgpu,webgl): Get WebGPU and WebGL deploying to a real browser (#12849) 684824c253
Also add CI testing for WebGL and WebGPU in Chrome Canary.

fix(renderer): Get an external target building again (#12884) 3a6064b95e
This change fixes one of the external platforms so that it builds again.

There is no ore support for it at the moment, but the gms/goldens run on device

feat: Scroll constraint drag multiplier (#12880) 2551181ab2
New dragMultiplier property (double, default 1.0) that scales pointer movement → scroll movement. 0.5 halves it, 2.0 doubles it. Applied before both the scroll-offset update and the physics accumulator, so drag feel and inertial release stay in sync. Inspector row added under Threshold.

Velocity reporting
A few related fixes while in this code:

velocityX/Y now decays smoothly through fling and settle (previously froze at the last drag value at release).
Velocity resets to 0 when a drag holds still; acceleration is preserved so release-into-fling still works.
Scrollbar drag also updates velocityX/Y; scrollbar release clears it immediately.

text vertical trim (#12861) a3fb919b91
* feature: add support for vertical trim

* add passthrough properties for vertical trimming

fix(Android): Unblock Android release flow (#12868) 77c82242bb
Commit `c8bb275d99` introduced a new Git attribute for an existing file without re-normalizing that file. This fixes the normalization issue while also making the Android release process more resilient to it recurring in the future.

When the commit changed the expected git attributes for .bat files, the existing .bat file was left stored with line endings that did not match Git's normalized index expectations. This caused the runtime submodule to appear dirty, which then blocked Android releases because release-it requires a clean working tree. The release workflow is now more resilient on top of fixing the dirty status.

feat(command-queue): add file asset listing via FileListener (#12662) 4a612198f1

fix(runtime): invalidate all scripted VM property caches on instance swap (#12863) ca0054c077

fix: Snap enabled scrolling not respecting last item padding (#12856) f11274b5d9
With a scroll constraint inside a parent layout that has trailing padding, snapping past the last item settled flush to the viewport edge instead of leaving padding visible. Non-snap scrolling already respected padding; only snap was broken. Pulling past the end and releasing also snapped back without padding.

chore(editor): Generate Windows pdbs for symbolication (#12720) 2fce4282b6
PR to get crash reports from Windows desktop editor.

Build sentry with crashpad on Windows
Generate debug symbols for Windows during release and upload to Sentry.
Tested locally and verified Windows crashes are being sent to Sentry.

fix(renderer/d3d11): guard MicroProfile GPU init against multi-context re-entry (#12766) 13fba5c3d5

chore: Stateful Component handling of source artboard databinds (#12776) e9483b78c4
Adds improved handling of binding source artboards on stateful components. Now the expected behavior is:

The edit time source artboard set on the component is the one that gets a stateful viewmodel instance created at edit time
At runtime, if that source artboard property is databound, we check if the selected artboard has the same viewmodel as the edit time artboard. If so, just reuse the stateful viewmodel instance. If the viewmodel is different, we instance a viewmodel at runtime and bind that to the components artboard.
Relevant changes to VM instance ownership:

NestedArtboard now holds a pointer to its stateful VM instance (not an rcp) which allows Artboard to unref it, NestedArtboard doesn't need to be concerned with that
In the case where isStateful is true and the source artboard is databound, if the artboard's VM is different than the stateful VM, we do get an rcp to that instance, which we (NestedArtboard) own and need to unref
Artboard's destructor still needs to go through the same process, the only thing is that it no longer needs to check if a VM instance's parent is nullptr because it now owns the stateful VM instances as well.

fix(build): locate Visual Studio via vswhere as a fallback (#12840) c5d462f9d9
* fix(build): locate Visual Studio via vswhere as a fallback

setup_windows_dev.ps1 only checked three hardcoded VS2022 edition paths,
so it failed to find newer installs (e.g. VS2026) and aborted. Add a
vswhere-based fallback (only on the path that previously errored) that
finds any VS install with the C++ toolset, regardless of version or
edition. Existing runners are unaffected: they short-circuit on
fxc-on-PATH or the hardcoded 2022 paths before reaching the fallback.

Needed by the self-hosted console build runners (Xbox/PS5), which run
VS 2026.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* fix(build): require VS 2022+ in the vswhere fallback

Per review: -latest alone would happily return VS2019 (or older) if it's
the newest installed, which can't build Rive. Add -version "[17.0,)" so the
vswhere fallback only matches VS2022 (v17) or newer (e.g. VS2026 = v18).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

feat(wgpu): Implement InterlockMode::atomics (#12730) 72ac79b91f
PLS planes (color, clip, coverage) are emulated as storage buffer<u32>
arrays on RenderTargetWebGPU, indexed by a 32x32-tile swizzle for cache
locality. glsl.glsl gains a new PLS_IMPL_STORAGE_BUFFER path alongside
STORAGE_TEXTURE.

renderPassInitialize becomes a real draw that seeds the color buffer on
LoadAction::clear or LoadAction::preserveRenderTarget, and clears the
coverage/clip buffers. The existing STORE_COLOR_CLEAR #ifdef and the new
LOAD_COLOR_FROM_DST_TEXTURE branch are gated by specialization constants
so one compiled shader covers all combinations. atomic_resolve_coalesced
is a new shader variant for non-FFCO resolve that writes directly to the
framebuffer attachment from the PLS color buffer.

The per-flush bind group + pipeline layout move into a
DrawPipelineLayout class keyed off InterlockMode, since atomic mode
needs Fragment visibility on PATH/PAINT/PAINT_AUX storage buffers while
other modes only access them from vertex.

The Dawn fiddle context is pinned to D3D11 for now; other backends show
sync issues in atomic mode that look like a Dawn bug.

chore(runtime): add computed root values test (#12830) d17e462c1a

fix(metal): exempt prepareToFlush from thread-safety analysis (#12734) 3a7bd9bfdf
* fix(metal): exempt prepareToFlush from thread-safety analysis

RenderContextMetalImpl::prepareToFlush acquires a buffer-ring lock that
is released asymmetrically in a GPU completion handler scheduled by
postFlush. Clang's -Wthread-safety analysis cannot follow a lock handed
off into that block and flags the unbalanced lock(). The newer Clang in
Xcode 26.4.1 turns this into a hard error for downstream builds that
compile with -Wthread-safety -Werror.

Add a portable RIVE_NO_THREAD_SAFETY_ANALYSIS macro and apply it to the
prepareToFlush definition to exempt this intentional, correct locking
pattern from the analysis.

feature: fit text content by font size (#12792) c623003932

chore(tests): add test to catch missing BrowserStack gm images (#12806) 65383d26e5
Nothing in our CI was previously testing to make sure that the browserstack_gms images contained all of the images needed (when a new gm was introduced), now there is a test to do so.

fix(ore/vk): MSAA resolve corruption on Xclipse 920 (resolve targets missed the post-pass layout hand-off) (#12810) bd3b3f3b75
* fix(ore/vk): hand off MSAA resolve target layout to Rive after the render pass

* ci: un-skip ore_msaa_resolve on galaxy-s22 vk jobs, TEMP gate CI to s22 only

* clang-format

* rebaseline galaxy-s22 vk ore_msaa_resolve goldens

* ci: restore full CI, revert temporary s22-only gating

* review: ResolveTarget struct instead of parallel arrays, std::exchange in move ctor, drop stale glmsaa comment

Co-authored-by: Gordon <pggordonhayes@gmail.com>

* chore(flutter): release 0.14.9

* chore: update changelog

Rev Luau to rive_0_726 (upstream 0.726) (#12953) 61822bb247
Rev Luau to rive_0_726 and adapt scripting workspace to the 0.726 API

chore(editor): add view model uses to dependencies panel (#12948) 63b8450361

fix(renderer): Update dead macro names (#12933) 09c54e8737
DISABLE_CLIP_RECT_FOR_VULKAN_MSAA was renamed to
DISABLE_CLIP_DISTANCE_FOR_UBERSHADERS in #11040 (generalized from
Vulkan-only to all MSAA ubershaders), which moved the #ifndef check
sites to the new name. However, three #define sites were still on the
old name.

  - draw_msaa_stencil.main
  - clear_clockwise_atomic_clip.main
  - init_clockwise_atomic_workaround.frag

The latter two were added in #12120, five months after the rename,
copy-pasting the already-dead define, so they had never once taken
effect.

fix(runtime): bidirectional data bind with source preference does not… (#12927) adb0834474
fix(runtime): bidirectional data bind with source preference does not overwrite target

feat(renderer): Add scissor support to more backends (#12778) 424d2323f9
This adds support for using scissor rects during clips to OpenGL, D3D 11 and 12, and Metal.

WebGPU and the external platforms are not yet done

This also adds a few new gms, and updates the rebaseline script to work with a new directory (which does not have a subdirectory)

Nnnnn focus management fixes (#12303) b2438309fa
* sort focus nodes hierarchically

* chore(focus): correctly handle focus with lists

* fix virtualizableChanged lifecycle

Feat: Updates to recorder to enable thumbnail generation in prod (#12878) fcbebb25e4
* Add ability to run recorder as a service

* Fix includes

* Further optimizations

* ADd profiling

* include profiler include

* Make png compression level configurable

* Add webp support

* Fix webp build

* Make webp lossless

* productionise code

* Fix writing

* back to lossless

* Fix up main

* Clang format

* PR feedback fixes

* Split libweb enc

* Fix PR feedback

Co-authored-by: Gordon <pggordonhayes@gmail.com>

Adds riveAdvanceProfilingEnabled flag and Stopwatch-based
instrumentation to RiveWidgetController.advance(). When enabled,
emits Timeline.instantSync events with artboard name, advance
duration (us), and whether the state machine changed.

(cherry picked from commit 0ccd246)

Apps that override MediaQuery.devicePixelRatio (device-preview shells,
resolution scalers) diverge from the real view dpr the shared texture is
sized/flushed at, mispositioning offset painters. Read the real,
non-overridable view dpr for the value while keeping a MediaQuery dpr
dependency so runtime dpr changes still rebuild.

Upstream 0.14.9 already carries the full-affine transform (Mat2D.fromMat4)
and native shared-texture settling, so only the dpr source remains as an
Ello delta.