Fix #3217: resolve enum contracts against the underlying type under source generation

[RaphaelFakhri]

Fixes #3217.

The bug as filed no longer reproduces. The mechanism identified in the thread, OpenApiAnyFactory.CreateFromJson deserializing to a JsonElement and failing without reflection, went away when that path was replaced by JsonModelFactory using JsonNode.Parse, which needs no metadata. With a strict source-generated TypeInfoResolver, master emits enum values correctly today, both integer and string forms. Nothing guards that, which is part of why the issue stayed open.

What is still live is the nullable case, and it is fatal rather than null:

System.NotSupportedException : JsonTypeInfo metadata for type 'System.Nullable`1[TimeRange]'
was not provided by TypeInfoResolver of type 'SourceGenerationContext'.
   at JsonSerializerDataContractResolver.JsonConverterFunc(Object value, Type type)
   at JsonSerializerDataContractResolver.GetDataContractForType(Type type)

The enum branch of GetDataContractForType unwraps Nullable<T> into effectiveType for every purpose except the two calls that serialize a sample value, which still receive the original type. Under a reflection-based resolver that is invisible, because System.Text.Json synthesizes Nullable<T> metadata on demand. Under a source-generated context the metadata exists only if some registered type happens to have a TEnum? member, and an optional enum route or query parameter, which is the shape in the issue, gives it no reason to. So document generation throws.

The fix is to pass effectiveType to both.

Four tests in a new JsonSourceGenerationSchemaGeneratorTests, driven by a real JsonSerializerContext set as the sole TypeInfoResolver: an int-backed enum, a nullable int-backed enum, a string enum through JsonStringEnumConverter<T>, and an enum reached as a property. Against master's source three pass and the nullable one throws with the exception above. With the change, Swashbuckle.AspNetCore.SwaggerGen.Test is 764 passed, 0 failed on net8.0.

One thing to say before it gets asked: the identical mismatch exists in the adjacent primitive branch, so int? and Guid? under source generation have the same latent crash. I tried the same swap there and it broke twelve GenerateSchema_SetsDefault_IfPropertyHasDefaultValueAttribute cases, which serialize a null default through the nullable type on purpose. That needs the null-default path handled separately, so this PR is deliberately scoped to the enum branch.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 95.20%. Comparing base (d0b1e26) to head (4c18b15).
⚠️ Report is 16 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #4131      +/-   ##
==========================================
- Coverage   95.24%   95.20%   -0.04%     
==========================================
  Files         111      111              
  Lines        4142     4174      +32     
  Branches      848      853       +5     
==========================================
+ Hits         3945     3974      +29     
- Misses        197      200       +3     

Flag	Coverage Δ
Linux	95.20% <100.00%> (-0.04%)	⬇️
Windows	95.20% <100.00%> (-0.04%)	⬇️
macOS	95.20% <100.00%> (-0.04%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[Copilot]

🟡 Changes recommended

The fix bypasses nullable-enum-specific converters even when nullable metadata is available.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

Fixes nullable enum schema generation with strict source-generated JSON metadata.

Changes:

• Resolves enum serialization against the underlying enum type.
• Adds source-generation coverage for integer, nullable, string, and property enums.

File summaries

File	Description
JsonSerializerDataContractResolver.cs	Uses the effective enum type during serialization.
JsonSourceGenerationSchemaGeneratorTests.cs	Adds source-generated enum regression tests.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 1
• Review effort level: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[RaphaelFakhri]

The review point is valid and is addressed. The enum contract now serializes against Nullable<T> when the serializer has metadata for it, and falls back to the underlying enum type only when it does not (for example a source-generated context that registers only the enum). A converter registered for Nullable<T> is therefore still used.

A new test registers a JsonConverter<TEnum?> together with a context that includes TEnum?, and asserts that the schema values come from the converter. It fails against the previous version of the change and passes now. The original source-generation tests still pass on net8.0, net9.0 and net10.0.

[Copilot]

Copilot review overview

🟡 Changes recommended

The metadata probe bypasses nullable enum converters under normal reflection-based serializer options.

Review effort: Balanced
Findings: 1

Open (1)

• Preserve nullable enum converters with default serializer options · New

Resolved since last review (1)

• Using effectiveType unconditionally changes existing serializer semantics for nullable enums:…

[Copilot]

Copilot review overview

🟢 Approval recommended

The targeted fix preserves converter behavior and is covered across the relevant source-generation scenarios.

Review effort: Balanced
Findings: None

Resolved since last review (1)

• Preserve nullable enum converters with default serializer options

[github-actions]

Thanks for your contribution @RaphaelFakhri - the changes from this pull request have been published as part of version 10.3.0 📦, which is now available from NuGet.org 🚀