Standalone Objective-C tests for the macOS texture, built against
FlutterMacOS.framework of the active Flutter SDK by macos/test/run_tests.sh
(optionally with ASan or TSan). The plugin sources are compiled into the
test; the plugin registers through a fake plugin registrar and is driven
through handleMethodCall:result:. The fake engine follows
FlutterTextureRegistrar.mm and FlutterExternalTexture.mm of engine a18df97c:
the texture id is the texture's address, the registrar keeps registered
textures, and copyPixelBuffer is wrapped into a Metal texture whose
CVMetalTexture and pixel buffer are released at once while the Metal texture
stays drawn. Frames come through FlutterRgbaRendererPluginOnRgba on a thread
without an autorelease pool. Memory is the task's phys_footprint, which counts
IOSurfaces, read after the engine's CVMetalTextureCache has let go of what it
wrapped (it keeps that until it makes a texture a second later, measured);
it is not checked under ASan, whose shadow of each new IOSurface mapping
counts in it. No app, window or screen capture.
On 50efc18e ten of the thirteen tests fail (Mac mini M4, macOS 26.5.2):
every frame gets a new IOSurface (60 for 60 frames), the raster thread waits
for the copy (median 0.9 to 1.1 ms in three runs, up to 537 of 600 calls over
1 ms, with 4K frames), frames are announced off the main thread, closed
textures keep their last frame (634 MB more after 20 4K textures) and still
take and announce late frames, a texture closed with recycle is never freed,
and rows of another stride than the pixel buffer's are copied in one piece
(shifted rows for 1366, 1001, 1279 and 3838 pixels wide with stride_align 1,
and 1001 with 256), short frames included. The two tests on frames the engine
was handed pass, as no buffer is ever written twice yet.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Retain closed raw-pointer identities for late frames, release frame buffers
after engine completion, and serialize termination with decoded frame writes.
Add native regression controls for held raster buffers, inline completion,
failed raster posts, missing registrations, teardown and concurrent closure.
B252; Apache-2.0 dependency license and upstream history preserved.
Each session created a new TextureRgba and closeTexture never released it, so
the engine's FlPixelBufferTexture kept its GL texture (one frame of GPU or,
with llvmpipe, process memory per session). Releasing the object is not safe:
FlPixelBufferTexture deletes its GL texture in dispose without selecting a GL
context, so a release from another window deletes that window's texture of
the same name (upstream PR #2 turned the remaining window white).
closeTexture(key, recycle: true) now keeps the texture in a list of the
plugin instance, i.e. of the engine (window) that created it, and the next
createTexture of that engine registers it again. Its GL texture is only ever
touched by that engine's populate in its own context, nothing is released,
and the GL memory is bounded by the number of textures a window had open at
once. When the engine goes away the recycled textures stay referenced for the
same reason. Without recycle (old callers, or a tab that moves to another
window while its session still holds the pointer) a closed texture stays
terminated and is never handed out again.
With no frame waiting, copy_pixels hands out one transparent pixel instead
of no buffer, so a recycled texture cannot show the last frame of its previous
session before the first new frame.
Dart: closeTexture(int key, {bool recycle = false}) sends "recycle"; other
platforms ignore it. Tests: all ten Linux lifecycle tests pass (clang and
gcc, -Wall -Werror, also under ASan/UBSan); Dart 4/4.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
closeTexture marked the texture terminated but kept the frame last handed to
the engine and any frame still waiting, and TextureRgba had no finalize, so
each closed session kept one or two frame-sized buffers forever (2 x
width x height x 4 bytes measured per 1080p session).
terminate now frees both buffers under the texture mutex: the texture is
already unregistered and OnRgba stops copying once terminated is set. A
finalize frees them as well. GObject and GL lifetime are unchanged here.
Tests: close_frees_frame_buffers, closed_texture_ignores_late_frames,
finalize_frees_frame_buffers and frames_racing_close pass (also under ASan);
the four recycling tests still fail.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Standalone C++ tests for the Linux texture, built against the engine of the
active Flutter SDK by linux/test/run_tests.sh (optionally with ASan/UBSan).
The plugin source is compiled into the test and driven through a real
FlMethodChannel on fake engine objects (binary messenger, texture registrar
with the reference rules of engine a18df97c, plugin registrar); frame
buffers copied with new[] are counted.
On 42797e0f eight of the ten tests fail: closing a texture keeps both frame
buffers, the GObject has no finalize, and closed textures are never reused,
so every session leaves a frame buffer and an engine GL texture behind.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>