Shad Storhaug commented on LUCENENET-627:

After updating the test framework to utilize the 
{{NUnit.Framework.TestContext.CurrentContext.Random}} property, I noticed that 
random test failures now repeat reliably until the next build. Whether or not 
there is a way to lock in that behavior from one build to the next, this is 
probably adequate enough to avoid porting the entire RandomizedTesting 

That said, it would probably make sense to create a new library with the random 
generators so they can be shared without a reference to our test framework and 
all of the dependencies that come with it. We should also migrate the extension 
methods for the {{Random}} class.

> Port RandomizedTesting Test Runner
> ----------------------------------
>                 Key: LUCENENET-627
>                 URL: https://issues.apache.org/jira/browse/LUCENENET-627
>             Project: Lucene.Net
>          Issue Type: Task
>          Components: Lucene.Net.TestFramework
>    Affects Versions: Lucene.Net 4.8.0
>            Reporter: Shad Storhaug
>            Priority: Major
>              Labels: up-for-grabs
>   Original Estimate: 150h
>  Remaining Estimate: 150h
> While the test framework can run with the standard NUnit test runner and we 
> have managed to make it work, it is missing the ability to reproduce a random 
> test failure, which is crucial for debugging some of the more complicated 
> random tests. Lucene uses a custom JUnit test runner called 
> [RandomizedTesting|https://github.com/randomizedtesting/randomizedtesting] to 
> accomplish this.
> It also includes some other nice features
> # Ensure when a test fails, the random seeds are included in error messages 
> and logs
> # Code analysis to ensure the tests are setup properly
> # Run tests in a random order to ensure they are not dependent upon each other
> Preliminary analysis shows that the API for NUnit allows building custom test 
> runners and it is a close enough match to implement the functionality. Most 
> likely, this will require a custom adapter as well, so the test runner can 
> integrate into Visual Studio and dotnet test/mstest.
> Do note that without doing some pretty heavy refactoring on the Codec, 
> Directory, and LuceneTestCase classes, it is not possible to run tests in 
> parallel within the same AppDomain because the codecs use a static variable 
> to turn codec impersonation on/off. For now, it would probably be best to run 
> tests serially.

This message was sent by Atlassian Jira

Reply via email to