Sammanfattning
The importance of software correctness has been accentuated as a growing number of safetycritical systems have been developed relying on software operating these systems. One of the more prominent methods targeting the construction of a correct program is formal verification. Formal verification identifies a correct program as a program that satisfies its specification and is free of defects. While in theory formal verification guarantees a correct implementation with respect to the specification, applying formal verification techniques in practice has shown to be difficult and expensive. In response to these challenges, various support methods and tools have been suggested for all phases from program specification to proving the derived verification conditions. This thesis concerns practical verification methods applied to diagrammatic modeling languages.
While diagrammatic languages are widely used in communicating system design (e.g., UML) and behavior (e.g., state charts), most formal verification platforms require the specification to be written in a textual specification language or in the mathematical language of an underlying logical framework. One exception is invariantbased programming, in which programs together with their specifications are drawn as invariant diagrams, a type of state transition diagram annotated with intermediate assertions (preconditions, postconditions, invariants). Even though the allowed program states—called situations—are described diagrammatically, the intermediate assertions defining a situation’s meaning in the domain of the program are still written in conventional textual form. To explore the use of diagrams in expressing the intermediate assertions of invariant diagrams, we designed a pictorial language for expressing array properties. We further developed this notation into a diagrammatic domainspecific language (DSL) and implemented it as an extension to the Why3 platform. The DSL supports expression of array properties. The language is based on Reynolds’s interval and partition diagrams and includes a construct for mapping array intervals to logic predicates.
Automated verification of a program is attained by generating the verification conditions and proving that they are true. In practice, full proof automation is not possible except for trivial programs and verifying even simple properties can require significant effort both in specification and proof stages. An animation tool which supports runtime evaluation of the program statements and intermediate assertions given any userdefined input can support this process. In particular, an execution trace leading up to a failed assertion constitutes a refutation of a verification condition that requires immediate attention. As an extension to Socos, a verificion tool for invariant diagrams built on top of the PVS proof system, we have developed an execution model where program statements and assertions can be evaluated in a given program state. A program is represented by an abstract datatype encoding the program state, together with a smallstep state transition function encoding the evaluation of a single statement. This allows the program’s runtime behavior to be formally inspected during verification. We also implement animation and interactive debugging support for Socos.
The thesis also explores visualization of system development in the context of model decomposition in EventB. Decomposing a software system becomes increasingly critical as the system grows larger, since the workload on the theorem provers must be distributed effectively. Decomposition techniques have been suggested in several verification platforms to split the models into smaller units, each having fewer verification conditions and therefore imposing a lighter load on automatic theorem provers. In this work, we have investigated a refinementbased decomposition technique that makes the development process more resilient to change in specification and allows parallel development of submodels by a team. As part of the research, we evaluated the technique on a small case study, a simplified version of a landing gear system verification presented by Boniol and Wiels, within the EventB specification language.
While diagrammatic languages are widely used in communicating system design (e.g., UML) and behavior (e.g., state charts), most formal verification platforms require the specification to be written in a textual specification language or in the mathematical language of an underlying logical framework. One exception is invariantbased programming, in which programs together with their specifications are drawn as invariant diagrams, a type of state transition diagram annotated with intermediate assertions (preconditions, postconditions, invariants). Even though the allowed program states—called situations—are described diagrammatically, the intermediate assertions defining a situation’s meaning in the domain of the program are still written in conventional textual form. To explore the use of diagrams in expressing the intermediate assertions of invariant diagrams, we designed a pictorial language for expressing array properties. We further developed this notation into a diagrammatic domainspecific language (DSL) and implemented it as an extension to the Why3 platform. The DSL supports expression of array properties. The language is based on Reynolds’s interval and partition diagrams and includes a construct for mapping array intervals to logic predicates.
Automated verification of a program is attained by generating the verification conditions and proving that they are true. In practice, full proof automation is not possible except for trivial programs and verifying even simple properties can require significant effort both in specification and proof stages. An animation tool which supports runtime evaluation of the program statements and intermediate assertions given any userdefined input can support this process. In particular, an execution trace leading up to a failed assertion constitutes a refutation of a verification condition that requires immediate attention. As an extension to Socos, a verificion tool for invariant diagrams built on top of the PVS proof system, we have developed an execution model where program statements and assertions can be evaluated in a given program state. A program is represented by an abstract datatype encoding the program state, together with a smallstep state transition function encoding the evaluation of a single statement. This allows the program’s runtime behavior to be formally inspected during verification. We also implement animation and interactive debugging support for Socos.
The thesis also explores visualization of system development in the context of model decomposition in EventB. Decomposing a software system becomes increasingly critical as the system grows larger, since the workload on the theorem provers must be distributed effectively. Decomposition techniques have been suggested in several verification platforms to split the models into smaller units, each having fewer verification conditions and therefore imposing a lighter load on automatic theorem provers. In this work, we have investigated a refinementbased decomposition technique that makes the development process more resilient to change in specification and allows parallel development of submodels by a team. As part of the research, we evaluated the technique on a small case study, a simplified version of a landing gear system verification presented by Boniol and Wiels, within the EventB specification language.
Originalspråk  Engelska 

Handledare 

Utgivningsort  Turku 
Förlag  
Tryckta ISBN  9789521242137 
Elektroniska ISBN  9789521242144 
Status  Publicerad  2022 
MoEpublikationstyp  G5 Doktorsavhandling (artikel) 