Coverage causes hangs

Hi,

When adding coverage for .NET Framework, it’s causing hangs. It works fine on .NET Core: https://dev.azure.com/onovotny/GitBuilds/_build/results?buildId=2203&view=logs&j=12f1170f-54f2-53f3-20dd-22fc7dff55f9

This commit has the changes: https://github.com/onovotny/bc-csharp/commit/5ddb17530b4175ce31092ac671d2ed496c2ecd06

Here’s the command that fails, when on .NET Framework: https://github.com/onovotny/bc-csharp/commit/5ddb17530b4175ce31092ac671d2ed496c2ecd06

Seems like it’s happening with .NET Core too: https://dev.azure.com/onovotny/GitBuilds/_build/results?buildId=2204&view=logs&j=12f1170f-54f2-53f3-20dd-22fc7dff55f9&t=bfbec40a-1b5e-5690-b870-859627cad0c0&l=20

Should be able to repro with dotnet test -f netcoreapp2.1 ...

Repro steps:

Clone: https://github.com/onovotny/bc-csharp Checkout commit: 4a401ed22df13ef61b2bb7a306ee8a083167f1c6 Go to crypto\test Run dotnet test -f netcoreapp2.1 and see that it passes. About 1.5 min on my machine Run dotnet test -f netcoreapp2.1 -s ..\..\CodeCoverage.runsettings

Tests are taking much longer to run (never complete) and fail in some cases.

I think it’s struggling on this test: https://github.com/onovotny/bc-csharp/blob/netstandard/crypto/test/src/math/ec/test/ECPointPerformanceTest.cs#L176

Perhaps with some of the random number generators in use there?

[jkotas]

@jkotas do you think that aggressive inlining could help here(I think that tiering cannot help to remove the call)

This is small-enough method to be inlined by JIT without any hints. I do not expect that aggressive inlining would help anything here.

[MarcoRossignoli]

@onovotny I did a lot of test on 2 different machine yesterday evening 4/5 time x machine, and no luck…I cannot repro, my command was

C:\git\coverletissue\bc-csharp> dotnet test .\crypto\test\crypto.test.csproj -s .\CodeCoverage.runsettings 

runsettings

<?xml version="1.0" encoding="utf-8"?>
<!-- File name extension must be .runsettings -->
<RunSettings>
  <DataCollectionRunSettings>
    <DataCollectors>      
       <DataCollector friendlyName="XPlat code coverage">
        <Configuration>
          <Format>cobertura</Format>
          <Exclude>[xunit.*]*,[*Tests]*,[nunit.*]*,[NUnit3.*]*</Exclude> <!-- [Assembly-Filter]Type-Filter -->
          <ExcludeByAttribute>Obsolete,GeneratedCodeAttribute,CompilerGeneratedAttribute</ExcludeByAttribute>
          <SingleHit>true</SingleHit>
        </Configuration>
      </DataCollector>    
    </DataCollectors>
  </DataCollectionRunSettings>
</RunSettings>

Can you check that my config it’s the same as yours?

This is the instrumentation of Init() on RsaKeyPairGenerator

I’ve one suspect for https://github.com/tonerdo/coverlet/issues/511#issuecomment-519087222 …that for some reason tests are running in parallel and two test ask for static access to https://github.com/onovotny/bc-csharp/blob/netstandard/crypto/test/src/cms/test/CMSTestUtil.cs#L84-L91 and the first init kpg so null check above return non initilized instance for another thread that could lead to null ref. Maybe could happen because instrumented code are slower than non instrumented or is no more inlined so the randomic nature of failure. Can you try to force tests serialization?