SIGSEGV when using SKCanvas.DrawText() and then SKCanvas.Flush()

[mattleibow]

Description

The app crashes when drawing text and then flushing the canvas. Waiting on a response from Google to address this.

Basic Information

• Version with issue: 1.68.0
• Last known good version: 1.60.3
• IDE: N/A
• Platform Target Frameworks:
  • Android: 8.1
• Target Devices:
  • Pixel 2 and Razer Phone

Links

• skia bug: https://bugs.chromium.org/p/skia/issues/detail?id=8425
• "repro" case: https://github.com/mattleibow/NativeAndroidSkiaSharp

VS bug #736300

[mattleibow]

After doing some more research, I found out several things:

• it works fine in the emulator
• it works fine as a debug build

As a result, there appears to be something going wrong for ARM when the optimizations are at -Os. I had a try with -O3 and it appears to solve the issue.

However, this does incur a penalty with the size no longer being the main point. As a result, the size increases by about 800KB-1200KB depending on the architecture.

For this release, I will go with a larger binary that works over a smaller one that can't do GPU. I have reported the issue to Google along with all the information I could find. Hopefully, they can find out exactly what the issue is and provide a real fix.

[apaatsio]

I found this issue on google because I'm experiencing a similar bug on a Flutter app. See flutter/flutter#29751 and my comment

There seem to be many common denominators (Pixel 2, Android 8.1, drawing text, SIGSEGV during SKCanvas.Flush()) so I was wondering if these things could be related.

Unfortunately, I can't see the Skia issue 8425 that @mattleibow linked. It's giving me "Permission denied" so I can't dig deeper into this.

[blimkemann]

I am getting this error even in debug builds using 1.68.0 and above when targeting Android 10 as a build target. It happens every time and is repeatable. If I remove all calls to DrawText there is no crash. Even a single DrawText will cause a crash with a "Could not allocate vertices" message.

Is there any news on this?

Here is the managed stack trace:

Memory around native instruction pointer (0x73a030012cfe):0x73a030012cee  48 89 74 24 48 ff 50 18 48 8b 8c 24 e8 00 00 00  H.t$H.P.H..$....
0x73a030012cfe  ff 41 08 48 89 44 24 28 48 8b 6c 24 60 48 89 4c  .A.H.D$(H.l$`H.L
0x73a030012d0e  24 60 48 85 ed 74 35 48 8d 5d 08 ff 4d 08 75 0d  $`H..t5H.]..M.u.
0x73a030012d1e  48 8d 7b f8 e8 d9 8c bf ff 84 c0 74 1f 83 7d 0c  H.{........t..}.

======================================
===========================
	Managed Stacktrace:
=================================================================
	  at <unknown> <0xffffffff>
	  at SkiaSharp.SkiaApi:sk_canvas_flush <0x0013f>
	  at SkiaSharp.SKCanvas:Flush <0x00097>
	  at Redbook.Mobile.Document.Renderer:RenderPage <0x0051b>
	  at Redbook.Mobile.Document.Page:Render <0x00113>
	  at <>c__DisplayClass27_0:<OnPaintSurface>b__0 <0x0172b>
	  at RunnableImplementor:Run <0x000d1>
	  at Java.Lang.IRunnableInvoker:n_Run <0x000db>
	  at Android.Runtime.DynamicMethodNameCounter:32 <0x0011a>
	  at Android.Runtime.DynamicMethodNameCounter:32 <0x000fa>
=================================================================

[mattleibow]

@apaatsio, I had a look at that issue on the Google bug list, no feedback at all. The only comment a long while back was:

Well, it is dying while trying to upload glyphs to the font atlas.

I don't know why that would be occurring though.

@blimkemann could you create a test/repo project for your example. It was working when I reduced the optimization level a while back. Maybe this is a slightly different case that I can start testing on.

I am looking on getting a brand new skia engine out in #986, just trying to fix the last few bugs in the current version. Moving this issue to that release so it gets at least a test.

[blimkemann]

SkiaTextBug.zip
@mattleibow, here is a repro. Pretty simple really. Just start in debugger on Android and the drawing will crash as soon as Canvas.Flush is called. Please let me know if you have questions about this.

[mattleibow]

Thanks! I'll have a look.

[mattleibow]

This has been some time, I will try using the new v2.80 packages. Have you still been having this issue?

[blimkemann]

Yes, still having this issue.

[mattleibow]

Not sure how I missed it, but the issue with your code is that you are drawing outside of the main render loop.

This code:

MainThread.BeginInvokeOnMainThread(() => {
    // drawing code
});

The MainThread.BeginInvokeOnMainThread returns immediately and the surface/GL context is reset.

Just remove those blocks and draw directly.

[blimkemann]

Thanks. That fixed that issue. I am now always wrapping InvalidateSurface in MainThread.BeginInvokeOnMainThread. However, as I use some commands on the page and drawing happens, after about 4 calls to InvalidateSurface, the whole SKGLView appears as a black rectangle (only on Android) and then randomly comes back after a few more redraws. Am I still missing something?

[mattleibow]

Are there any errors in the log? Is there something you are doing just before it happens?

[blimkemann]

Each time, I call Canvas.Clear(), Canvas.Flush(), do my drawing, and Canvas.Flush() again. Other than that, exactly the same thing happens between drawings. The logs show only this: Time Device Name Type PID Tag Message 07-02 11:26:07.704 moto_g7 Info 24845 .redbook.mobil Explicit concurrent copying GC freed 10877(367KB) AllocSpace objects, 0(0B) LOS objects, 49% free, 3MB/6MB, paused 311us total 7.868ms 07-02 11:26:07.707 moto_g7 Debug 24845 Mono GC_TAR_BRIDGE bridges 7 objects 95968 opaque 280479 colors 7 colors-bridged 7 colors-visible 7 xref 0 cache-hit 0 cache-semihit 0 cache-miss 0 setup 0.02ms tarjan 16.60ms scc-setup 0.02ms gather-xref 0.01ms xref-setup 0.01ms cleanup 2.08ms 07-02 11:26:07.707 moto_g7 Debug 24845 Mono GC_BRIDGE: Complete, was running for 10.88ms 07-02 11:26:07.707 moto_g7 Debug 24845 Mono GC_MINOR: (Nursery full) time 20.11ms, stw 21.25ms promoted 676K major size: 55024K in use: 50942K los size: 4096K in use: 2560K From: Matthew Leibowitz <notifications@github.com> Sent: Wednesday, July 1, 2020 3:27 PM To: mono/SkiaSharp <SkiaSharp@noreply.github.com> Cc: Brian Limkemann <brian@decisivecomputing.com>; Mention <mention@noreply.github.com> Subject: Re: [mono/SkiaSharp] SIGSEGV when using SKCanvas.DrawText() and then SKCanvas.Flush() (#647) Are there any errors in the log? Is there something you are doing just before it happens? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<#647 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ACPIWVO4Q7FNZJ276AENGYLRZOEXRANCNFSM4FX2OECQ>.

[mattleibow]

@blimkemann Is this still happening? Have you tried the v2 of SkiaSharp?

[blimkemann]

It is not still happening. Turns out is was a threading issue From: Matthew Leibowitz <notifications@github.com> Sent: Tuesday, October 13, 2020 7:08 PM To: mono/SkiaSharp <SkiaSharp@noreply.github.com> Cc: Brian Limkemann <brian@decisivecomputing.com>; Mention <mention@noreply.github.com> Subject: Re: [mono/SkiaSharp] SIGSEGV when using SKCanvas.DrawText() and then SKCanvas.Flush() (#647) @blimkemann<https://github.com/blimkemann> Is this still happening? Have you tried the v2 of SKiaSharp? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<#647 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ACPIWVM2DFMHHOTPZ2AWDVDSKTMV7ANCNFSM4FX2OECQ>.

[mattleibow]

Thanks for the feedback. I'm closing it now. If things start happening, then feel free to open a new issue and link to this for context.