NRE in realtime-session-with-stacks line 123

[desdesdes]

Hi Brian,

Thanks for this great example. I sometimes got an NullReferenceException then running the code on program.cs line 123. After I replaced line 122 to 124 with the code below and the exception was fixed.

Task.Factory.StartNew((o) =>
{
    var localCurrent = (TraceCallStack)o;
    ResolveSymbolsForModule(localCurrent.CodeAddress.CodeAddresses, localCurrent.CodeAddress.ModuleFile);
}, current);

Best regards,
Bart Vries

[brianrob]

Thanks for reporting this @desdesdes. I'm not able to repro this. How commonly do you see this, and are you still running against .NET 6 as the current repro does? Your change seems reasonable, but I would have expected localCurrent to be properly captured here.

[desdesdes]

Hi Brian,

It was indeed the case that i was running against .net8! I tested it and the problem is reproducible in this repo if i change the targetframework from net6.0 to net8.0 in realtime-session-with-stacks.csproj, but not in the net6 version.

Best regards,
Bart Vries

[brianrob]

Thanks @desdesdes. I just tried running this against .NET 8, but I can't seem to repro the failure. How often were you seeing the failure? I'm wondering if you might be able to share a process dump of the failure. I am happy to take your change assuming it's the right thing, but I want to make sure there isn't a hidden GC hole or something like that here.

[desdesdes]

Hi Brain,

The main issue is that the NRE is thrown on a async background task and because of the behavior in .net core the exception on async background task swallowed by default. This behavior is documented here and here under remarks.

So you will only see the exception when you use a debugger or when you handle TaskScheduler.UnobservedTaskException.

I have build a quick repro to run with a debugger attached here.

If you like I can create a dump, but because the exception is swallowed it will be a first chance exception dump.

Best regards,
Bart Vries