[tobias-tengler]

The composite result is serialized through one JsonWriter call per token, even when the output is minimized. Every property name goes through an escape check, length validation and its own GetSpan/Advance pair, although response names never need escaping.

This PR writes minimized results (no indentation, no null omission, default or relaxed encoder) straight into the response buffer through a cached span. Property names come from a token cached on the selection that already holds the quotes and the colon, and source leaves and source-owned JSON scalars are copied as raw bytes. Indented output, null omission and custom encoders keep the existing JsonWriter path. The output is byte-identical, and new tests compare both paths on composed gateway results.

Method	Version	Mean	Ratio	Allocated	Alloc Ratio
Format 87 KB result	main	33.30 μs	1.00	96 B	1.00
Format 87 KB result	PR	21.91 μs	0.66	96 B	1.00

k6 constant scenario, A B B A per gateway (3 × 30 s runs per invocation; values are the mean of the per-invocation medians). The first composite schema invocation of the PR was discarded: in all three of its runs the gateway, subgraph and k6 cores all idled at about half their usual load while latency doubled, which points to an external stall, so the composite PR column is the second invocation only.

Gateway	Version	RPS	CPU per request	p95	p99
Composite schema (net10.0)	main	4,417	589 μs	19.48 ms	25.46 ms
Composite schema (net10.0)	PR	4,575	573 μs (−2.8%)	18.89 ms	24.85 ms
Apollo Federation (net11.0)	main	4,123	563 μs	19.82 ms	25.41 ms
Apollo Federation (net11.0)	PR	4,188	547 μs (−2.8%)	19.81 ms	25.70 ms

Gateway CPU per request drops by 2.8% on both gateways, well beyond the run-to-run noise measured on main (CPU per request CV about 0.5% for the composite schema gateway and 0.2% for the Apollo Federation gateway). The RPS gains (+158 and +65) are not significant on their own given that noise (σ about 65 RPS for the composite schema gateway, 40 RPS for the Apollo Federation gateway), and latency is unchanged within noise.

[github-actions]

Patch coverage

91.5% of changed lines covered (367/401)

File	Covered	Changed	Patch %
…/Fusion/src/Fusion.Execution/Text/Json/ThrowHelper.cs	0	1	0.0% 🔴
…/src/Fusion.Execution/Text/Json/MinimizedJsonEmitter.cs	107	129	82.9% 🟡
…/Text/Json/CompositeResultDocument.WriteTo.cs	137	144	95.1% 🟡
…/Fusion.Execution/Text/Json/SourceResultDocument.WriteTo.cs	111	115	96.5% 🟡
…/Fusion/src/Fusion.Execution/Execution/Nodes/Selection.cs	3	3	100.0% 🟢
…/Execution/Nodes/SourceResponseNameMapping.cs	2	2	100.0% 🟢
…/src/Fusion.Execution/Text/Json/CompositeResultDocument.cs	1	1	100.0% 🟢
…/Fusion/src/Fusion.Execution/Text/Utf8StringCache.cs	5	5	100.0% 🟢
src/HotChocolate/Json/src/Json/JsonWriter.cs	1	1	100.0% 🟢

Uncovered changed lines (JSON)

{
  "sha": "9b0ec888e8188e0e2c908c03d6c765248e17dc9c",
  "files": [
    { "path": "src/HotChocolate/Fusion/src/Fusion.Execution/Text/Json/ThrowHelper.cs", "ranges": [[45, 45]] },
    { "path": "src/HotChocolate/Fusion/src/Fusion.Execution/Text/Json/MinimizedJsonEmitter.cs", "ranges": [[116, 116], [120, 122], [124, 127], [129, 135], [143, 147], [182, 183]] },
    { "path": "src/HotChocolate/Fusion/src/Fusion.Execution/Text/Json/CompositeResultDocument.WriteTo.cs", "ranges": [[335, 337], [341, 341], [362, 364]] },
    { "path": "src/HotChocolate/Fusion/src/Fusion.Execution/Text/Json/SourceResultDocument.WriteTo.cs", "ranges": [[152, 152], [165, 166], [250, 250]] }
  ]
}

---

Project coverage: 59.2% (314259/530629 lines)

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.