Národní úložiště šedé literatury Nalezeno 2 záznamů.  Hledání trvalo 0.01 vteřin. 
Určení spolehlivosti výsledků statické analýzy pomocí strojového učení
Beránek, Tomáš ; Fiedor, Jan (oponent) ; Vojnar, Tomáš (vedoucí práce)
The Meta Infer static analyzer is a tool for detecting various types of errors in source code. However, its results contain more than 95 % of false alarms. This thesis proposes a solution that ranks Infer’s reports using Graph Neural Networks (GNNs) based on the likelihood of being a real error, thus mitigating the issue with false alarms. The system consists of a training pipeline, which converts the D2A dataset – a set of labeled reports from Meta Infer – into Extended Code Property Graphs (ECPGs) and GNN models trained on these ECPGs. Experimental results indicate that the developed GNN models can match, and in some cases even surpass, existing models developed by strong industrial teams. Moreover, these existing solutions are closed source, making the solution developed in this thesis a promising open-source alternative.
Pokročilá statická analýza výkonnosti v nástroji Meta Infer
Pavela, Ondřej ; Rogalewicz, Adam (oponent) ; Vojnar, Tomáš (vedoucí práce)
Looper is a static complexity analysis tool for inference of tight upper bounds on the exe- cution cost of programs. It is based on the previously existing Loopus tool which used abstract program model of difference constraints (inequalities of the form + ), which allows for natural abstraction of common loop counter updates = + + and = + 0. Looper was initially proposed and implemented in author’s bachelor’s thesis as a checker for the Meta Infer framework but the tool failed to meet the expectations when tested on real-world code. This master’s thesis proposes a new improved version of Looper that aims at solving the main limitations of the original tool, namely through introduction of interprocedural analysis. Additionally, various extensions target- ing improved precision of the intraprocedural analysis, such as new abstraction algorithm, handling of compound loop conditions and more, were implemented. Moreover, logging, issue reporting and collection of results has been significantly improved. Finally, through extensive experiments with the new Looper version, the ability to analyze real-world code in a more general, scalable and precise way was shown.

Chcete být upozorněni, pokud se objeví nové záznamy odpovídající tomuto dotazu?
Přihlásit se k odběru RSS.