Enable all net48 integration tests and stabilise tests

[scott-xu]

No description provided.

[Rob-Hague]

We can merge this one next assuming it passes. No need to create and resolve conflicts in the other branches

[scott-xu]

Deal 🥂

[scott-xu]

public static System.Security.Cryptography.ECDsa Create(System.Security.Cryptography.ECParameters parameters) applies to net47 and above.

See https://learn.microsoft.com/en-us/dotnet/api/system.security.cryptography.ecdsa.create?view=net-8.0#system-security-cryptography-ecdsa-create(system-security-cryptography-ecparameters)

[scott-xu]

Unfortunately, Mono doesn't implement ECDsa. See https://github.com/mono/mono/blob/main/mcs/class/referencesource/System.Core/System/Security/Cryptography/ECDsa.cs#L32

[WojciechNagorski]

So we can run them on Windows. What do you think?

[Rob-Hague]

Unfortunately, Mono doesn't implement ECDsa. See https://github.com/mono/mono/blob/main/mcs/class/referencesource/System.Core/System/Security/Cryptography/ECDsa.cs#L32

It is the reason behind #1406

So we can run them on Windows. What do you think?

I tried a while back. The problem is running a Linux container on a Windows host, which doesn't seem so easy on CI machines (the way to do it is WSL2 AFAIK but most Windows server images do not come with WSL2). But I'd welcome anyone to try because I'd very much like to see it happen.

[Rob-Hague]

It might be possible now that Windows Server 2022 includes WSL2. The problems I ran into were seemingly related to LCOW (#1220 (comment)). I wouldn't know if it is possible to run docker via WSL2 rather than LCOW in appveyor (I don't really know much about this stuff)

[Rob-Hague]

I think there's some work involved here to make sure we don't introduce more unstable tests

[scott-xu]

Unfortunately, Mono doesn't implement ECDsa. See https://github.com/mono/mono/blob/main/mcs/class/referencesource/System.Core/System/Security/Cryptography/ECDsa.cs#L32

It is the reason behind #1406

We can consider using BouncyCastle as a fallback when in Mono. (https://www.mono-project.com/docs/faq/technical/#how-can-i-detect-if-am-running-in-mono)

Then we can split the code into 3 implementations.

1. Bcl implementation (.NET Framework 4.7+ and not Mono)
2. Cng implementation (.NET Framework 4.6.2 and not Mono)
3. BouncyCastle implementation (.NET Framework and Mono)

I can have a try when I have time.

[scott-xu]

It seems not easy to split EcdsaKey into three implementations because it is public already with ECDsaCng Ecdsa and CngAlgorithm HashAlgorithm properties when target to .NET Framework.

[Rob-Hague]

I kind of expected mono to use a .NET Standard build, but I guess it can use a .NET Framework build as well. The error in #1406 comes from calling ECDsa.Create(ECParameters) which only occurs #if !NET_FRAMEWORK.

Anyway I think it's fine to tweak the public members at this level if needed, especially if it doesn't work in the first place.

I'll run the CI on this a couple more times to check for any more misbehaving tests

[scott-xu]

Any idea to stabilize this test? @Rob-Hague

  Failed Test_ExecuteAsync_CancellationToken [1 s]
  Error Message:
   Assert.AreEqual failed. Expected:<TERM>. Actual:<(null)>. 
  Stack Trace:
     at Renci.SshNet.IntegrationTests.OldIntegrationTests.SshCommandTest.Test_ExecuteAsync_CancellationToken() in /home/appveyor/projects/ssh-net/test/Renci.SshNet.IntegrationTests/OldIntegrationTests/SshCommandTest.cs:line 139

https://ci.appveyor.com/project/drieseng/ssh-net/builds/50322490/job/yo6pooiew5j86wcw

The 500ms wait may not enough in unit test.
See https://github.com/sshnet/SSH.NET/blob/develop/src/Renci.SshNet/SshCommand.cs#L446

[Rob-Hague]

Hmm, I would not expect the 500ms to be relevant here because we know OpenSSH implements signals. I think we can ignore that one and see if it happens more often

[Rob-Hague]

Thanks