JavaScript is a widely used dynamic programming language. What started out
as a client-side scripting language for browsers, is now used for large applications
in many different settings. As for other dynamic languages, JavaScript makes
it easy to write programs quickly without being constrained by the language,
and programmers exploit that power to write highly dynamic programs.
Automated tools for helping programmers and optimizing programs are used
successfully for many programming languages. Unfortunately, the automated
tools for JavaScript are not as good as for other programming languages. The
program analyses, that the automated tools are built upon, are poorly suited
to deal with the highly dynamic nature of JavaScript programs. The lack
of language restrictions on the programmer are detrimental to the quality of
program analyses for JavaScript.
The aim of this dissertation is to address the highly dynamic nature of
JavaScript programs in order to improve program analysis for JavaScript.
This is done for two kinds of program analysis: dataflow analysis and type
systems. We design, implement and evaluate new abstractions for dataflow
analysis. These abstractions improve on state-of-the-art dataflow analyses, and
indicate that good, automated tools for highly dynamic JavaScript programs
are attainable.
To aid the design of abstractions for both dataflow analyses and type
systems, we have developed several methodologies. The methodologies aid
the analysis designer in identifying the aspects of highly dynamic JavaScript
programs that needs new abstractions.