Type Checking and Whole-program Inference for Value Range Analysis
MetadataShow full item record
Value range analysis is important in many software domains for ensuring the safety and reliability of a program and is a crucial facet in software development. The resulting information can be used in optimizations such as redundancy elimination, dead code elimination, instruction selection, and improve the safety of programs. This thesis explores the use of static analysis with type systems for value range analysis. Properly formalized type systems can provide mathematical guarantees for the correctness of a program at compile time. This thesis presents (1) a novel type system, the Narrowing and Widening Checker, (2) a whole-program type inference, the Value Inference for Integral Values, (3) a units-of-measurement type system, PUnits, and (4) an improved algorithm to statically analyze the data-flow of programs. The Narrowing and Widening Checker is a type system that prevents loss of information during narrowing conversion of primitive integral data types and automatically distinguishes the signedness of variables to eliminate the ambiguity of a widening conversion from type \!byte! and \!short! to type \!int!. This additional type system ensures soundness in programs by restricting operations that violate the defined type rules. While type checking verifies whether the given type declarations are consistent with their use, type inference automatically finds the properties at each location in the program, and reduces the annotation burden of the developer. The Value Inference for Integral Values is a constraint-based whole-program type inference for integral analysis. It supports the relevant type qualifiers used by the Narrowing and Widening type system, and reduces the annotation burden when using the Narrowing and Widening Checker. Value Inference can infer types in two modes: (1) ensure a valid integral typing exists, and (2) annotate a program with precise and relevant types. Annotation mode allows human inspection and is essential since having a valid typing does not guarantee that the inferred specification expresses design intent. PUnits is a type system for expressive units of measurement types and a precise, whole-program inference approach for these types. This thesis presents a new type qualifier for this type system to handle cases where the method return and method parameter type are context-sensitive to the method receiver type. This thesis also discusses the related work and the benefits and trade-offs of using PUnits versus existing Java unit libraries, and demonstrates how PUnits can enable Java developers to reap the performance benefits of using primitive types instead of abstract data types for unit-wise consistent scientific computations in real-world projects. The Dataflow Framework is a data-flow analysis for Java used to evaluate the values at each program location. Data-flow analysis is considered a terminating, imprecise abstract interpretation of a program and many false-positives are issued by the Narrowing and Widening Checker due to its imprecision. Three improvements to the algorithm in the framework are presented to increase the precision of the analysis: (1) implementing a dead-branch analysis, (2) proposing a path-sensitive analysis, and (3) discussing how loop precision can be improved. The Narrowing and Widening Checker is evaluated on 22 of the Apache Commons projects with a total of 224k lines of code. Out of these projects, 18 projects failed with 717 errors. The Value Inference for Integral Values is evaluated on these 18 Apache Commons projects. Out of these projects, 5 projects are successfully evaluated to SAT and the Value Inference inferred 10639 annotations. The 13 projects that are evaluated to UNSAT are manually examined and all of them contain a real narrowing error. Manual annotations are added to 5 of these projects to resolve the reported errors. In these 5 projects, the Narrowing and Widening Checker detects 69 real errors and 26 false-positives, with a false-positive rate of 37.7\%. The type system performs adequately with a compilation time overhead of 5.188x for the Narrowing and Widening Checker and 24.43x for the Value Inference. These projects are then evaluated with the addition of dead-branch analysis to the framework; the additional evaluation time is negligible. Its performance is suitable for use in a real-world software development environment. All the presented type systems build on techniques from type qualifier systems and constraint-based type inference. Our implementation and evaluation of these type systems show that these techniques are necessary and are effective in ensuring the correctness of real-world programs.
Cite this version of the work
Tongtong Xiang (2020). Type Checking and Whole-program Inference for Value Range Analysis. UWSpace. http://hdl.handle.net/10012/16445
Showing items related by title, author, creator and subject.
Testing the spatial affordance hypothesis: Evidence from factor analysis, mathematical models, and behavioural analysis Barton, Kevin Richard (University of Waterloo, 2015-09-29)This thesis attempts to explain the apparent link between how we navigate the world around us and the physical properties that define that world. Despite a number of works indicating the substantial effect of the layout ...
Data Envelopment Analysis may Obfuscate Corporate Financial Data: Using Support Vector Machine and Data Envelopment Analysis to Predict Corporate Failure for Nonmanufacturing Firms Yang, Xiaopeng; Dimitrov, Stanko (Taylor & Francis, 2017)Corporate failure prediction has drawn numerous scholars’ attention because of its usefulness in corporate risk management, as well as in regulating corporate operational status. Most research on this topic focuses on ...
Yu, Miao; Olkowicz, Mariola; Pawliszyn, Janusz (Elsevier, 2019-03-07)In LC-MS based untargeted analysis, data is collected at the peak or ion level, although the investigated biochemistry processes occur at the compound or reaction level. To this end, the presence of redundancy peaks such ...