 
Summary: Refined Type Inference for ML \Lambda
(Abstract)
Manuel F˜ ahndrich y and Alexander Aiken y
University of California, Berkeley z
1 Introduction
Inclusion constraints over setexpressions [1, 4] provide
a general formalism to express a large class of program
analyses. Over the past two years, we have experimented
with inclusion constraints to model dataflow in typebased
analyses. One of our research goals is to determine how to
structure and implement precise constraintbased analyses
such that they scale to large programs. Program analyses
with O(n 3 ) complexity bounds often exhibit their worst
case complexity in practice and consequently do not scale
beyond programs of a few thousand lines. As a result,
coarser but faster analyses are usually used [2, 10].
Scaling behavior and precision are intimately con
nected and in an ideal formalism, one can be traded
for the other. Unfortunately, inclusion constraints over
setexpressions do not provide enough control over this
