Demo Data from EDBT 2026 / reviewers in the wild / expert
Stefan Radziuk
dblp:396/4986
· DBLP profile ↗
1ranked-venue papers
0as first author
1since 2021 · last 2025
0009-0009-1220-9200ORCID · reported
Domains — the database's venue-derived domains; a paper can count in several
Software engineering, systems software and programming languages · 1 · 1 since 2021
Expertise — from the expertise taxonomy: the topics of the expert's papers under the CCF categories. A weight counts papers with recency: 1 for a paper about the topic, 0.3 when the topic is its context, halved every five years.
| Software engineering, system software, and programming languages
1 paper |
Programming languages and type systems · 93% Compilers and program optimization · 7% |
Topics — the 4 heaviest of 5, each with the papers that count most for it
| Topic | Weight | Papers | Last | Evidence papers |
|---|---|---|---|---|
Programming languages and type systems › language semantics › formal semantics
mechanized semantics |
0.9 | 1 | 2025 | Progressful Interpreters for Efficient WebAssembly Mechanisation · Proc. ACM Program. Lang. 2025 |
Programming languages and type systems › type systems › type soundness
progress and preservation |
0.9 | 1 | 2025 | Progressful Interpreters for Efficient WebAssembly Mechanisation · Proc. ACM Program. Lang. 2025 |
Programming languages and type systems › type systems
type soundness |
0.9 | 1 | 2025 | Progressful Interpreters for Efficient WebAssembly Mechanisation · Proc. ACM Program. Lang. 2025 |
Programming languages and type systems
webassembly |
0.9 | 1 | 2025 | Progressful Interpreters for Efficient WebAssembly Mechanisation · Proc. ACM Program. Lang. 2025 |
Methods — techniques the papers use, named apart from their topics
coq dependent types · 0.9constructive progress proof · 0.9
| Year | Publication | Venue | Position |
|---|---|---|---|
| 2025 | Progressful Interpreters for Efficient WebAssembly MechanisationabstractMechanisations of programming language specifications are now increasingly common, providing machine-checked modelling of the specification and verification of desired properties such as type safety. However it is challenging to maintain these mechanisations, particularly in the face of an evolving specification. Existing mechanisations of the W3C WebAssembly (Wasm) standard have so far been able to keep pace as the standard evolves, helped enormously by the W3C Wasm standard’s choice to state the language’s semantics in terms of a fully formal specification. However a substantial incoming extension to Wasm, the 2.0 feature set, motivates the investigation of strategies for more efficient production of the core verification artefacts currently associated with the WasmCert-Coq mechanisation of Wasm. In the classic formalisation of a typed operational semantics as followed by the W3C Wasm standard, both the type system and runtime operational semantics are defined as inductive relations, with associated type soundness properties (progress and preservation) and an independent sound interpreter. We investigate two more efficient strategies for producing these artefacts, which are currently all separately defined by WasmCert-Coq. First, the approach of Kokke, Siek, and Wadler for deriving a sound interpreter from a constructive progress proof — we show that this approach scales to the W3C Wasm 1.0 standard, but results in an inefficient interpreter in our setting. Second, inspired by results from intrinsically-typed languages, we define a progressful interpreter which uses Coq’s dependent types to certify not only its own soundness, but also the progress property. We show that this interpreter can implement several performance optimisations while maintaining these certifications, which are fully erasable when the interpreter is extracted from Coq. Using this approach, we extend the WasmCert-Coq mechanisation to the significantly larger Wasm 2.0 feature set, discovering and correcting several errors in the expanded specification’s type system. Xiaojia Rao, Stefan Radziuk, Conrad Watt, Philippa Gardner |
Proc. ACM Program. Lang. | 2 |