[thomhurst]

Summary

The TUnit.Core.SourceGenerator.Tests snapshot harness parsed fixtures from tests/TUnit.TestProject with Roslyn 4.7's default parse options. Collection expressions did not bind there, so attribute arguments such as Arguments = [5] were dropped. The snapshots then recorded what the generator does with broken input, not what a real build produces.

• Before: [MethodDataSource(nameof(AsyncDataMethodWithArgs), Arguments = [5])] was snapshotted with no Arguments line and a factory calling AsyncDataMethodWithArgs() with no argument (code that would not even compile).
• After: the snapshot shows Arguments = new object[] { 5 } and AsyncDataMethodWithArgs(((int)arguments[0])), which is what a real build generates.

Cause

Two things combined:

1. CSharpSyntaxTree.ParseText(source) used the default language version, not Preview as TUnit.TestProject builds with (CS8652).
2. The test project compiles the fixtures with Roslyn 4.7, the generator's build-time minimum. Even in preview mode, Roslyn 4.7 rejects collection expressions in attribute arguments (CS0182) and params collections (CS0225). Switching to Preview alone changed no snapshot.

The harness checks only generator diagnostics, so none of this was reported.

Fix

• Parse fixtures, and the trees the generator adds (via the driver's parseOptions), with LanguageVersion.Preview.
• Host the generator on Roslyn 4.14 in this test project (VersionOverride on Microsoft.CodeAnalysis.Common/CSharp/Workspaces.Common). The generator is still built against 4.7. The Core generator has no ROSLYN4_14 conditionals, so this matches the Roslyn414 variant that a current SDK loads.
• New guard ThrowIfInputCannotBeParsed: a snapshot test now fails if its input has syntax errors, or if declaration binding reports CS0182/CS0225/CS8652, the errors a too-old compiler gives for valid newer syntax. Other compile errors stay allowed, because some fixtures reference packages the harness does not load (AutoFixture, Vogen, OneOf...). I checked that the guard works by temporarily removing the Roslyn override: 7 snapshot tests then fail with "The test input does not parse as a real build would".
• The harness's UnconditionalSuppressMessageAttribute stub had no constructor, so every [UnconditionalSuppressMessage("x", "y")] in a fixture failed with CS1729. It now has the real (category, checkId) shape. No snapshot changed because of this.
• CrossCompilationAttributeTests failed deterministically after the Roslyn upgrade, but only when it ran in parallel with a snapshot test. Roslyn caches assembly symbols on each reference's metadata. With the shared ReferencesHelper.References, the derived compilation could bind different cached symbols than the base compilation did. It then saw the base project through RetargetingAttributeData, which has no ApplicationSyntaxReference, so the IDE-shape precondition failed. The class now creates fresh references per test. That passed 6 of 6 runs in the parallel combination that failed every time before.

Snapshot changes, reviewed one by one

All the changes come from attribute arguments written as collection expressions, which used to be dropped.

1. AsyncMethodDataSourceDrivenTests.Test: Arguments = [5] is now emitted, and the factory calls AsyncDataMethodWithArgs(int) with the argument. Correct. The fixture has a single AsyncDataMethodWithArgs(int baseValue) overload. The old snapshot called it with no arguments, which a real build cannot do.
2. MethodDataSourceDrivenTests.Test and ConflictingNamespaceTests.MethodDataSource_WithConflictingNamespace (DotNet8_0/9_0/10_0/Net4_7, same fixture): three [MethodDataSource(nameof(SomeMethod), Arguments = [...])] attributes now carry their Arguments = new object[] { 5 } / { "Hello World!", 5, true } / { "Hello World!", 7, true } lines. The Arguments lines are correct. The new output also exposes an existing generator bug: the factory still calls SomeMethod(), the first overload, instead of SomeMethod(int) / SomeMethod(string, int, bool). The sibling attributes written as new object[] { ... } already showed this bug in the old snapshots, so it is not new. Open PR fix(sourcegen): infer generic method types from any MethodDataSource row shape #7143 fixes it ("the factory always called the first member with that name") and updates these same snapshots, so I did not file a duplicate issue. When fix(sourcegen): infer generic method types from any MethodDataSource row shape #7143 merges, these snapshots should show the overload-correct calls for the collection-expression attributes too. Whichever PR lands second needs a snapshot re-accept.

No other snapshot changed on any TFM.

Tests

• TUnit.Core.SourceGenerator.Tests on net8.0, net9.0, net10.0 and net472: all pass after accepting the snapshots above (192 passed, 1 pre-existing skip on net8/9/10; 191 passed, 1 skip on net472).
• /*/*/*AttributeTests/* (the parallel combination that exposed the retargeting race) passed 6 of 6 runs.
• Guard check: with the Roslyn 4.14 override removed, 7 tests fail with the new parse error, as intended.

The change only affects the test harness. The generator, the engine and the public API are unchanged.

Fixes #7162

Follow-up filed: #7190 (the same parse-options pattern in the Assertions/Should/Mocks generator harnesses).

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-20T05:35:04.981849Z	a76127b	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[coderabbitai]

Warning

Review limit reached

You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository.

Next included review available in 29 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used all 8 included reviews currently available.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 9d98cda0-11d1-4a25-b9ab-af07d46c4944

📥 Commits

Reviewing files that changed from the base of the PR and between c4b9977 and a76127b.

📒 Files selected for processing: 9

• tests/TUnit.Core.SourceGenerator.Tests/AsyncMethodDataSourceDrivenTests.Test.verified.txt
• tests/TUnit.Core.SourceGenerator.Tests/ConflictingNamespaceTests.MethodDataSource_WithConflictingNamespace.DotNet10_0.verified.txt
• tests/TUnit.Core.SourceGenerator.Tests/ConflictingNamespaceTests.MethodDataSource_WithConflictingNamespace.DotNet8_0.verified.txt
• tests/TUnit.Core.SourceGenerator.Tests/ConflictingNamespaceTests.MethodDataSource_WithConflictingNamespace.DotNet9_0.verified.txt
• tests/TUnit.Core.SourceGenerator.Tests/ConflictingNamespaceTests.MethodDataSource_WithConflictingNamespace.Net4_7.verified.txt
• tests/TUnit.Core.SourceGenerator.Tests/CrossCompilationAttributeTests.cs
• tests/TUnit.Core.SourceGenerator.Tests/MethodDataSourceDrivenTests.Test.verified.txt
• tests/TUnit.Core.SourceGenerator.Tests/TUnit.Core.SourceGenerator.Tests.csproj
• tests/TUnit.Core.SourceGenerator.Tests/TestsBase.cs

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[greptile-apps]

Confidence Score: 5/5

[Low impact] The PR appears safe to merge; no new actionable issues were found.

Summary

The snapshot harness now uses Roslyn 4.14 with preview parsing so newer fixture syntax binds correctly.

• Source-generator snapshots now keep arguments written with collection expressions.
• Each cross-compilation test builds from its own metadata references.

thomhurst explicitly acknowledges the existing SomeMethod() overload bug and defers its fix to PR #7143. It is not reported again here.

No new actionable issues were found. Tests were not run during this review.

_{Reviews (1) · Last reviewed commit: "test(sourcegen): parse snapshot fixtures..." · Reviewed by Greptile}

[github-actions]

Review

Solid test-harness fix. The root cause (default parse options plus Roslyn 4.7 silently dropping collection-expression attribute arguments) is diagnosed correctly. The ThrowIfInputCannotBeParsed guard stops the same problem recurring unnoticed. Because the guard fires on CS0182/CS0225/CS8652, it would also catch a future Roslyn downgrade.

Snapshot changes are all additions of Arguments = ... lines, plus the one factory fix for AsyncDataMethodWithArgs. They match what a real build produces. The remaining SomeMethod() overload issue is pre-existing and covered by #7143.

Notes (non-blocking):

• The VersionOverride of 4.14.0 is hard-coded in the csproj in three places. A single MSBuild property (e.g. RoslynHostVersion) would keep them in sync. Also check that central package management allows VersionOverride here (it evidently builds, so it does).
• The guard's ID list is a hand-maintained allowlist. If another feature-gating error appears (e.g. CS9xxx for a newer syntax), it will be silently tolerated again. Consider also failing on any error whose location is inside an attribute argument list, or at least documenting how to extend it.
• CreateIsolatedReferences reloads every reference from disk per test. That is fine for 2 tests. If more cross-compilation tests are added, cache one isolated set per test class instead.
• Follow-up Other source-generator test harnesses parse fixtures with Roslyn 4.7 default language version #7190 for the other generator harnesses is the right call.

Overall: LGTM.