Support for composable/extensible test context

[bkoelman]

@DustinCampbell Here the pull request for my comment earlier. I left out the FluentAssertions parts. But all my other features are in here, because I'm using a quite different API. Putting all features in separate pull requests at this moment would case much duplication and painful merges later. Based on discussions around this pull request, I can update it and remove parts that may be added later in another form. At a later time, I'd like to add some extensibility points to plug a custom assertion framework. Consider this pull request an expression to the direction I'd like to take this.

A key difference is that testing a CodeFixProvider can only be done from the output of its associated analyzer. This makes it possible to transfer analyzer state through Diagnostic.Properties, along with its reported span(s) and automatically test-apply all suggested fixes. Personally, as an analyzer developer, I am more interested in testing a working combination of Analyzer + FixProvider, rather than unittesting scenarios for a standalone FixProvider.

So I have one base class, AnalysisTestFixture, from which unittests derive. To run a test, an immutable context is passed in. It has Withers to supply additional information, such as extra references, filename etc.

Lastly, I changed reporting of builtin compiler errors, which were discarded before. I found that valuable, as it revealed some broken tests I had written weeks ago. The assertion method is virtual, so can be overridden from unittest classes. Callback could also be moved into context, if desired.

As an example, I rewrote the template tests for a new Analyzer/FixProvider project to use my API:

using Microsoft.CodeAnalysis;
using Microsoft.CodeAnalysis.CodeFixes;
using Microsoft.CodeAnalysis.Diagnostics;
using NUnit.Framework;
using RoslynNUnitLight;

namespace UppercaseClass.Test
{
    [TestFixture]
    internal class UppercaseClassNameUnitTest: AnalysisTestFixture
    {
        protected override string DiagnosticId => UppercaseClassAnalyzer.DiagnosticId;

        [Test]
        public void TestMethod1()
        {
            // No diagnostics expected to show up

            const string source = @"";

            var context = new AnalyzerTestContext(source, LanguageNames.CSharp);
            AssertDiagnostics(context);
        }

        [Test]
        public void TestMethod2()
        {
            // Diagnostic and CodeFix both triggered and checked for

            const string source = @"
    using System;

    namespace ConsoleApplication1
    {
        class [|SomeTypeName|]
        {   
        }
    }";

            string expected = source.Replace("[|SomeTypeName|]", "SOMETYPENAME");

            var analyzerContext = new AnalyzerTestContext(source, LanguageNames.CSharp);
            var fixContext = new FixProviderTestContext(analyzerContext, new[] { expected }, reformatExpected: false);
            AssertDiagnosticsWithCodeFixes(fixContext);
        }

        protected override DiagnosticAnalyzer CreateAnalyzer()
        {
            return new UppercaseClassAnalyzer();
        }

        protected override CodeFixProvider CreateFixProvider()
        {
            return new UppercaseClassCodeFixProvider();
        }
    }
}

[cezarypiatek]

Hi @bkoelman

I know it's a very long time since this PR was reported. RoslynNUnitLight is in the process of deprecating and the recommended successor is https://github.com/cezarypiatek/RoslynTestKit - this is a newer version of RoslynNUnitLight. Your problem was solved in RoslynTestKit - if you need to test CodeFix that fixes an issue reported by given/provided DiagnosticAnalyzer you can do that by overriding CreateAdditionalAnalyzers() method. You can find an example here https://github.com/cezarypiatek/RoslynTestKit#example-test-code-fix-that-fixes-issue-reported-by-provided-diagnosticanalyzer