Differential Testing of Pushdown Reachability with a Formally Verified Oracle

Publikation: Bidrag til bog/antologi/rapportKonferencebidrag i proceedingsForskningfagfællebedømt

Pushdown automata are an essential model of recursive computation. In model checking and static analysis, numerous problems can be reduced to reachability questions about pushdown automata and several efficient libraries implement automata-theoretic algorithms for answering these questions. These libraries are often used as core components in other tools, and therefore it is instrumental that the used algorithms and their implementations are correct. We present a method that significantly increases the trust in the answers provided by the libraries for pushdown reachability by (i) formally verifying the correctness of the used algorithms using the Isabelle/HOL proof assistant, (ii) extracting executable programs from the formalization, (iii) implementing a framework for the differential testing of library implementations with the verified extracted algorithms as oracles, and (iv) automatically minimizing counter-examples from the differential testing based on the delta-debugging methodology. We instantiate our method to the concrete case of PDAAAL, a state-of-the-art library for pushdown reachability. Thereby, we discover and resolve several nontrivial errors in PDAAAL.

OriginalsprogEngelsk
TitelProceedings of the 22nd Conference on Formal Methods in Computer-Aided Design, FMCAD 2022
RedaktørerAlberto Griggio, Neha Rungta
ForlagTU Wien Academic Press
Publikationsdato2022
Sider369-379
ISBN (Elektronisk)9783854480532
DOI
StatusUdgivet - 2022
Begivenhed22nd International Conference on Formal Methods in Computer-Aided Design, FMCAD 2022 - Trento, Italien
Varighed: 17 okt. 202221 okt. 2022

Konference

Konference22nd International Conference on Formal Methods in Computer-Aided Design, FMCAD 2022
LandItalien
ByTrento
Periode17/10/202221/10/2022

Bibliografisk note

Funding Information:
This research was supported by the Independent Research Fund Denmark (DFF project QASNET) and by Novo Nordisk Fonden (NNF20OC0063462).

Publisher Copyright:
© 2022 FMCAD Association and authors.

ID: 339333121