FPU talk Friday 24 September, 12-1pm, 5.08 ICT Building

[Bernie Pope]

(Note the unusual time and place - this talk is also a Postgrad seminar).

Title: Static Aliasing Analysis for Efficient Destructive Update in Higher-Order Languages

Speaker: Matt Giuca

Abstract:

Declarative languages provide a powerful restriction which helps manage code complexity: no data structure may be modified once created. While good for robust software, this restriction can make code inefficient (especially array manipulations), and programmers are often forced to use alternative data structures, such as linked lists, to compensate.

I will present an abstract interpretation for declarative languages, developed with Peter Schachte, which can be used to automatically insert data structure re-use instructions into a declarative program, allowing a predictable portion of updates to be performed efficiently. The call-independent analysis uses sharing graphs to track aliasing between program nodes. It handles arbitrarily-nested data structures (with full precision) and recursive data structures, and as-yet informal work shows it can handle higher-order functions without closure-conversion or loss of precision.

The analysis will be demonstrated on programs written in the Mars programming language, which was created especially for this purpose.

We hope to see you there.