feat: make MethodAssertionGenerator incremental

[TimothyMakkison]

Add incremental caching for MethodAssertionGenerator.

• Uses ImmutableEquatableArray for structural equality on collections
• I have created tests to verify this, as usual I can't run any projects with Sourcy so I created a new test project see Method assertion incre test #4565
• Plan is to do this for all generators to improve ide responsiveness, should help with repeated build times 👍

[thomhurst]

Summary

Adds incremental caching to MethodAssertionGenerator by introducing ImmutableEquatableArray<T> for structural equality on collections, improving IDE responsiveness and build performance.

Critical Issues

None found ✅

Suggestions

1. Missing snapshot test updates

Impact: Medium
The PR modifies source generator output by restructuring internal data models (e.g., AssertionMethodData now uses new record types like MethodData, TargetTypeData, and ParameterData). While the generated code output may be identical, source generator changes typically require snapshot test verification.

Action: Run dotnet test TUnit.Assertions.SourceGenerator.Tests to verify no snapshot differences. If .received.txt files are generated, review and commit .verified.txt files per CLAUDE.md instructions.

2. Unused field BuildAssertion

Location: MethodAssertionGenerator.cs:20

publicstaticstringBuildAssertion="MethodAssertionData";

This field is only used once in line 71 with .WithTrackingName(BuildAssertion). Consider:

• Making it const instead of a field (no runtime allocation)
• Or inlining the string literal directly if it's only used once

Minor optimization: Avoids unnecessary static field allocation for a constant value.

3. Record types for incremental caching

Observation: The new record types (ParameterData, MethodData, TargetTypeData, ContainingTypeData) all implement value equality, which is correct for incremental generation. However, they contain string fields that reference fully-qualified type names and expressions.

Question: Have you verified that these strings produce stable output across compilations? Type display strings can sometimes include metadata tokens or ordering differences that break incremental caching.

Test recommendation: Add a test that verifies incremental compilation works correctly by running the generator twice with identical input and confirming no regeneration occurs.

4. ImmutableEquatableArray source attribution

The comment at the top credits dotnet/runtime. Consider adding a license header or more explicit attribution if required by the source repository's license (MIT in this case allows it, but attribution is good practice).

Previous Review Status

No previous review comments.

Verdict

✅ APPROVE - No critical issues

This is a well-executed performance improvement. The refactoring to value-based records and ImmutableEquatableArray follows best practices for incremental source generators. The code is clean and the approach aligns with TUnit's performance-first principles (Rule 4).

Minor suggestions above are optional improvements, not blockers.

[TimothyMakkison]

1. Missing snapshot test updates

Shouldn't alter the output (I hope) 🤞

2. Unused field BuildAssertion

See separate PR for the incremental tests.

3. Record types for incremental caching

Should be fine, I rewrote Mapperly, Mediator and Refit with the same technique 👍

Should I add the tests to TUnit.Assertion.SourceGenerator.Tests in a different PR? Once it's added, running them will be a pain for me due to the sourcy issue.

[TimothyMakkison]

Okay, looking at TestMetadataGenerator I see a lot of dead or unused code, is it okay if I make a pr removing this, or are you keeping it for future plans?

I'm pretty sure InterfaceCache is creating a memory leak as ITypeSymbol is always different between runs, at the very least it keeps at least on stale ITypeSymbol permanently alive

[thomhurst]

Okay, looking at TestMetadataGenerator I see a lot of dead or unused code, is it okay if I make a pr removing this, or are you keeping it for future plans?

I'm pretty sure InterfaceCache is creating a memory leak as ITypeSymbol is always different between runs, at the very least it keeps at least on stale ITypeSymbol permanently alive

If it's dead yeah go for it! 🔥