- Plats: Room 2446, Institutionen för informationsteknologi, Polacksbacken, Lägerhyddsvägen 2, Uppsala
- Doktorand: Brandauer, Stephan
- Om avhandlingen
- Arrangör: Avdelningen för datalogi
- Kontaktperson: Brandauer, Stephan
References are a programming language construct that lets a programmer access a datum invariant of its location.
References permit aliasing -- several references to the same object, effectively making a single object accessible through different names (or paths). Aliasing, especially of mutable data, is both a blessing and a curse: when used correctly, it can make a programmer's life easier; when used incorrectly, for example through accidental aliases that the programmer is unaware of, aliasing can lead to hard to find bugs, and hard to verify programs.
Aliases allow us to build efficient data structures by connecting objects together, making them immediately reachable. Aliases are at the heart of many useful programming idioms. But with great power comes great responsibility: unless a programmer carefully manages aliases in a program, aliases propagate changes and make parts of a program's memory change seemingly for no reason. Additionally, such bugs are very easy to make but very hard to track down.
This thesis presents an overview of techniques for controlling how, when and if data can be aliased, as well as how and if data can be mutated. Additionally, it presents three different projects aimed at conserving the blessings, but reducing the curses. The first project is disjointness domains, a type system for expressing intended aliasing in a fine-grained manner so that aliasing will not be unexpected; the second project is Spencer, a tool to flexibly and precisely analyse the use of aliasing in programs to improve our understanding of how aliasing of mutable data is used in practise; and the third project is c flat, an approach for implementing high-level collection data structures using a richer reference construct that reduces aliasing problems but still retains many of aliasing's benefits.