SVGReader in conjunction with TransformSRT causes massive memory leak

bjoern · February 10, 2023, 2:53pm

Also the vvvv process sometimes keeps lingering when closing vvvv while the patch is running.

Tested with 5.0 preview-0588 and 2021.4.12.

To reproduce open attached patch and have a look at memory consumption in Taskmanager.
Removing SVGReader or TransformSRT will stop the leak.

MemLeak.7z (5.2 KB)

Elias · February 10, 2023, 5:13pm

Sounds very much like Memory leak in SKIA drawSVG - #10 by schlonzo which we never figured out.

bjoern · February 10, 2023, 5:23pm

Thanks, I guess we’ll have to use images then.

bjoern · March 18, 2023, 3:05pm

Just stumbled upon this issue which seems kind of related:

github.com/mono/SkiaSharp

[BUG][MemoryLeak] Animated SKPath causes memory leak. Example Project included

opened 05:50PM - 28 Jan 23 UTC

AquilaAbriel

**Description** I'm currently experimenting with animated SKPaths, but I stum…bled upon a memory leak issue I couldn't resolve. I'm not entirely sure but I suspect there is some unmanaged resource, which is not released correctly. (Either in SKPath, or in SKSurfce, maybe related to both?) I tried two different methods of creating an animated SKPath. Either by calling Reset() and rebuilding a new shape, or by calling Dispose() on the old path and creating a new one. Both options result in a memory leak. Some further observations: - The memory leak affects both RAM and VRAM. - Forcing a garbage collection run does not free up the leaked memory. - The SKPaint settings (Antialias / Stroke / Fill) don't affect the memory leak. (Apparently) - SKCanvas matrix changes don't affect the memory leak. (Translation and Scale get applied each frame) - Only rebuilding but not drawing the path doesn't cause a memory leak. - Only drawing the same path, without rebuilding it, doesn't cause a memory leak. - Using dispose on the SKPath, instead of resetting and reusing it, doesn't stop the memory leak. - Disposing + recreating or resetting + rebuilding the SKPath but skipping the "animation" (essentially re-adding the same points each frame) stops the memory leak?! (How is that even possible?) After digging though some of the Google Skias documentation about SKPath, I found the [isVolatile] flag. The description of which sounds suspiciously like what is going wrong here: _Returns true if the path is volatile; it will not be altered or discarded by the caller after it is drawn. [SkPath](https://api.skia.org/classSkPath.html) by default have volatile set false, allowing [SkSurface](https://api.skia.org/classSkSurface.html) to attach a cache of data which speeds repeated drawing. If true, [SkSurface](https://api.skia.org/classSkSurface.html) may not speed repeated drawing._ I don't know if the volatile setting was added to Skia in a later Version, than what SkiaSharp is build on, but both the Flag and the related setIsVolatile() function are absent in SkiaSharp. **Code** Alternative methods of creating the animated path (Running once each frame): ``` //-----------------------------------------------------------------------------------------------------// //Reset and reuse existing SKPath //-----------------------------------------------------------------------------------------------------// PathBuffer.Reset(); //Move to first point position PathBuffer.MoveTo(CurrentPoints[0].Position); //Connect bezier points for (int x = 1; x < CurrentPoints.Length; x++) { PathBuffer.CubicTo(CurrentPoints[x - 1].HandleNext, CurrentPoints[x].HandleLast, CurrentPoints[x].Position); } //Close loop SKPoint first = PathBuffer.GetPoint(0); PathBuffer.CubicTo(CurrentPoints.Last().HandleNext, CurrentPoints[0].HandleLast, first); PathBuffer.Close(); ``` ``` //-----------------------------------------------------------------------------------------------------// //Dispose old SKPath and create new SKPath //-----------------------------------------------------------------------------------------------------// PathBuffer.Reset(); PathBuffer.Dispose(); PathBuffer = new SKPath(); //Move to first point position PathBuffer.MoveTo(CurrentPoints[0].Position); //Connect bezier points for (int x = 1; x < CurrentPoints.Length; x++) { PathBuffer.CubicTo(CurrentPoints[x - 1].HandleNext, CurrentPoints[x].HandleLast, CurrentPoints[x].Position); } //Close loop SKPoint first = PathBuffer.GetPoint(0); PathBuffer.CubicTo(CurrentPoints.Last().HandleNext, CurrentPoints[0].HandleLast, first); PathBuffer.Close(); ``` Drawing the animated path to the screen: ``` //Clear rendertarget gameTarget.Clear(new SKColor(0, 0, 0, 0)); //Center canvas in the middle of the screen gameTarget.surface.Canvas.Translate(gameTarget.width / 2f, gameTarget.height / 2f); //Scale canvas (Path point coords are within -1 to 1 space) gameTarget.surface.Canvas.Scale(100f, 100f); using (SKPaint paint = new SKPaint()) { paint.Color = new SKColor(255, 255, 255); paint.IsAntialias = true; paint.IsStroke = true; paint.StrokeWidth = 0.02f; if (!DEBUG_SKIP_DRAW) { //Draw current path to canvas gameTarget.surface.Canvas.DrawPath(PathBuffer, paint); } } //Reset rendertarget matrix gameTarget.surface.Canvas.ResetMatrix(); ``` **Expected Behavior** Correctly releasing all unmanaged data (and caches?) related to SKPath and SKSurface if SKPath gets disposed. **Actual Behavior** I have no idea, what is going on "behind the curtains"... All I see is a 100mb of RAM leaking every 10 seconds. Even though I called Dispose() on everything I have access to. **Basic Information** - Version with issue: 2.88.3, GPU Renderer - Last known good version: Unknown - IDE: Visual Studio 2022 - Platform Target Frameworks: Windows10, .NET 7 **Screenshots** dotMemory screenshot: ![dotMemory](https://user-images.githubusercontent.com/7701826/215282730-e7c93cf4-bc03-441b-b0b3-5474115eb346.jpg) **Reproduction Link** Very simplified example project to showcase the memory leak issue. (Everything importand is in the DEBUG_SKPath.cs file) This project uses .NET 7 and requires the newest version of Visual Studio 2022 to compile: [ExampleProjectFiles.zip](https://github.com/mono/SkiaSharp/files/10527777/ExampleProjectFiles.zip) Memory profile produced by JetBrains dotMemory profiler (I have not enough experience to know, what I have to look for): [MemoryProfiling.zip](https://github.com/mono/SkiaSharp/files/10527894/MemoryProfiling.zip)

OT: Also, judging from the number of unaddressed issues on github I don’t think skiasharp is a safe bet for the future.