[Fusion] Look up input object fields by UTF-8 name during variable coercion

[tobias-tengler]

Variable coercion allocated several objects for every field of an input object variable: JsonVariableCoercion read JsonProperty.Name up to three times per field (field lookup, error path segment and the coerced ObjectFieldNode), and each read decodes a new UTF-16 string. Every coerced field also got a new NameNode.

This PR looks input fields up by their raw UTF-8 property name. FusionInputFieldDefinitionCollection gets a TryGetField(ReadOnlySpan<byte>) overload, backed by the same sorted UTF-8 name index FusionEnumValueCollection already used (now shared as Utf8NameIndex). The coerced ObjectFieldNode reuses a NameNode cached on FusionInputFieldDefinition, and the error path uses the schema's field name. JSON allows escaping any character of a property name ({"\u0069d": 1} is a valid way to send id), so a property name containing an escape sequence falls back to the unescaped lookup, only when the UTF-8 lookup misses.

Benchmarks

Small input (2 objects, 5 fields)

Method	Version	Mean	Ratio	Allocated	Alloc Ratio
Variable coercion	main	792.3 ns	1.00	1.58 KB	1.00
Variable coercion	PR	590.2 ns	0.74	0.95 KB	0.60

Large input with a list of 20 input objects (22 objects, 66 fields)

Method	Version	Mean	Ratio	Allocated	Alloc Ratio
Variable coercion	main	8.146 μs	1.00	17.09 KB	1.00
Variable coercion	PR	5.074 μs	0.62	7.91 KB	0.46

Variable coercion allocates 40% less for the small input and 54% less for the large input, and is 26% and 38% faster, because each input field no longer allocates its name strings and NameNode.

[Copilot]

🟡 Changes recommended

The enum refactor introduces an avoidable array allocation for every enum collection.

1 open finding

• Avoid unnecessary parallel array allocation in Utf8NameIndex.Create · New

What changed in this PR

Optimizes Fusion input-object variable coercion by matching raw UTF-8 property names and reusing cached syntax nodes.

Changes:

• Adds a shared UTF-8 name index.
• Uses UTF-8 lookup with escaped-name fallback.
• Adds lookup and coercion coverage.

File	Description
VariableCoercionHelperTests.cs	Tests escaped JSON field names.
FusionInputFieldDefinitionCollectionTests.cs	Tests UTF-8 field lookup.
JsonVariableCoercion.cs	Uses UTF-8 lookup and cached names.
FusionInputFieldDefinition.cs	Caches each field’s NameNode.
Utf8NameIndex.cs	Implements shared UTF-8 indexing.
FusionInputFieldDefinitionCollection.cs	Adds span-based field lookup.
FusionEnumValueCollection.cs	Adopts the shared index.

🧠 Review effort: Balanced

---

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

[github-actions]

Patch coverage

100.0% of changed lines covered (86/86)

File	Covered	Changed	Patch %
…/Collections/FusionEnumValueCollection.cs	2	2	100.0% 🟢
…/Collections/FusionInputFieldDefinitionCollection.cs	10	10	100.0% 🟢
…/src/Fusion.Execution.Types/Collections/Utf8NameIndex.cs	58	58	100.0% 🟢
…/src/Fusion.Execution.Types/FusionInputFieldDefinition.cs	2	2	100.0% 🟢
…/src/Fusion.Execution/Execution/JsonVariableCoercion.cs	14	14	100.0% 🟢

---

Project coverage: 58.6% (314230/536486 lines)