Deser: Rethinking Rust Serialization

summary

Armin Ronacher introduced Deser, a Rust serialization library designed to address structural limitations in Serde caused by its stack-based recursion and unified data model. By inverting the architecture so formats push events into heap-allocated sinks via a driver, Deser eliminates stack overflow risks, enables stream parsing, and cleanly handles complex structures like XML namespaces and flattened maps. Although it brings binary size tradeoffs and drops support for non-self-describing formats like protobuf, it provides a robust alternative for complex self-describing data. Performance benchmarks indicate mixed runtime overheads but faster derived-code compile times.

key points

1. Deser uses a heap-backed driver-sink model instead of Serde's recursive visitor pattern, preventing stack overflows and supporting stream parsing.
2. It directly resolves notorious Serde edge cases including arbitrary precision numbers, map key type conversions under flattening, and XML namespaces.
3. Runtime performance is mixed—averaging roughly 10% slower for JSON reading—while release compilation times for derived code are roughly 2.3x faster.

who should read it

Rust developers building systems that process complex, nested, or untrusted self-describing formats like JSON, TOML, or XML.

discussion positive