Skipped tests still execute test constructor

[lizalc]

Maybe this is intentional (per #1728 (comment))? Tests that are marked as skipped will still run the test constructor.

Here is a simple repo test class (lizalc/tunit-constructor-runs-before-skipattribute):

namespace tunit_constructor_runs_before_skipattribute;

public class Tests : IAsyncDisposable
{
    public Tests()
    {
        Console.WriteLine("Constructor runs before any test, even if skipped");
    }

    [Test]
    [Skip("Skip")]
    public void Basic()
    {
        Console.WriteLine("This is a basic test");
    }

    public ValueTask DisposeAsync()
    {
        Console.WriteLine("Dispose is skipped despite constructor running");
        return ValueTask.CompletedTask;
    }
}

 dotnet run

████████╗██╗   ██╗███╗   ██╗██╗████████╗
╚══██╔══╝██║   ██║████╗  ██║██║╚══██╔══╝
   ██║   ██║   ██║██╔██╗ ██║██║   ██║
   ██║   ██║   ██║██║╚██╗██║██║   ██║
   ██║   ╚██████╔╝██║ ╚████║██║   ██║
   ╚═╝    ╚═════╝ ╚═╝  ╚═══╝╚═╝   ╚═╝

   TUnit v0.24.0.0 | 64-bit | Darwin 24.5.0 Darwin Kernel Version 24.5.0: Tue Apr 22 19:53:27 PDT 2025; root:xnu-11417.121.6~2/RELEASE_ARM64_T6041 | osx-arm64 | .NET 9.0.6 | Microsoft Testing Platform v1.7.1

   Engine Mode: SourceGenerated

Constructor runs before any test, even if skipped
skipped Basic (0ms)
  Skip

Test run summary: Zero tests ran - bin/Debug/net9.0/tunit-constructor-runs-before-skipattribute.dll (net9.0|arm64)
  total: 1
  failed: 0
  succeeded: 0
  skipped: 1
  duration: 64ms

[lizalc]

Noted this when porting tests from xUnit which used constructors for test setup. Should I instead convert every constructor to a method decorated with Before(Test)? Inconvienent but doable.

[thomhurst]

Hmm.. if you've got a basic test skip attribute on a class, I could probably put in special logic to check for this on test discovery.

For custom skip attributes (where you plug in your own logic to determine wether it skips) I think it would still have to behave as current, as that logic is only ran on test start. But I could check if it's a non derived skip attribute probably and skip it at discovery time.

[lizalc]

Note that the docs mention using constructors for test setup:

Most setup for a test can be performed in the constructor (think setting up mocks, assigning fields.)

[lizalc]

For my use case I'm providing custom skip attributes.

[thomhurst]

For my use case I'm providing custom skip attributes.

This definitely complicates things. But it might be fine. I just need to have a lil play about. I reckon we can sort this though.

[thomhurst]

I would say though, regardless of this issue, any constructor logic should be free from side effects. Because that would make this issue redundant.

I know that isn't always the case though.

But using IAsyncInitializer for any setup should only execute when a test using it starts. That means any process needing this or that only happens if needed. If you skip a class, then that never happens if you have it in the constructor.

Saying that, I'm going to leave this open for now and I can probably just dispose classes if they're skipped but still new'd up. It just means extra setup and teardown work that ends up being redundant, which is why I recommend the other pattern :)

[lizalc]

I'll probably move the test setup out of the constructors in my test classes but others porting from xUnit may run into this too and need to be aware.

[thomhurst]

Yep completely. xUnit are all about constructors so I'm in agreement this will affect others.

I just slightly disagree with their mindset. I know it doesn't matter because I can't force my viewpoint on everyone, but I like constructors just setting necessary members, and actual logic being outside of that