Localized Static Program Analysis for Optimization and Error Detection

Oskar Haarklou Veileborg

Research output: Book/anthology/dissertation/reportPh.D. thesis

Abstract

Software development processes continuously evolve, but remain imperfect.
In this thesis we identify two issues related to software development in the modern
programming languages, Java and Go.
The first issue concerns a type of software defect commonly known as concurrency bugs.
To realize the full potential of modern multi-core processors, programmers must write
concurrent, multi-threaded programs.
However, development of concurrent programs is complex, and it can be difficult to avoid
common shared-memory concurrency bugs, such as data races and deadlocks, in part because
scheduling non-determinism makes it hard to reason about behaviors of a multi-threaded
program. The Go programming language endorses the use of a channel communication model
as a means to avoid concurrency bugs, but recent work shows that Go developers who use
channels still introduce many concurrency bugs in their programs, some of which are
directly caused by misuse of channels.

The second issue involves abstraction overhead of using functional programming paradigm
features in mainstream languages, in particular Java.
The use of immutable data and pure functions is an important principle of functional
programming. It makes it easier to reason about potential behaviors of programs, which
in turn makes it easier to prevent the introduction of bugs.
Realizing the benefits of this programming style, mainstream languages have introduced
features that are commonly associated with functional programming, such as declarative
data-processing.
This feature was introduced in Java 8 through the stream API and has been widely
adopted. However, previous work shows that using this API instead of imperative
data-processing methods comes with a performance cost, referred to as abstraction
overhead, which dissuades some developers from using it.

In this thesis we present two novel static program analysis techniques based on abstract
interpretation. One enables ahead-of-time optimizations for Java programs that use
streams, while the other allows for detection of blocking errors involving channels in
Go programs. We built proof-of-concept implementations of these techniques and have
experimentally validated their efficacy.
The techniques scale to large real-world programs, which consist of both application
code and many libraries. Both techniques build on the idea of localized analysis,
a form of static program analysis that splits the analyzed program into smaller
fragments that can be analyzed individually with a high degree of precision.
Original languageEnglish
PublisherAarhus Universitet
Number of pages121
Publication statusPublished - Jul 2023

Fingerprint

Dive into the research topics of 'Localized Static Program Analysis for Optimization and Error Detection'. Together they form a unique fingerprint.

Cite this