Use this URL to cite or link to this record in EThOS:
Title: Inferring useful static types for duck typed languages
Author: Lamaison, Alexander
ISNI:       0000 0004 2752 7996
Awarding Body: Imperial College London
Current Institution: Imperial College London
Date of Award: 2013
Availability of Full Text:
Access from EThOS:
Access from Institution:
Complete and precise identification of types is essential to the effectiveness of programming aids such as refactoring or code completion. Existing approaches that target dynamically typed languages infer types using flow analysis, but flow analysis does not cope well with heavily used features such as heterogeneous containers and implicit interfaces. Our solution makes the assumption that programs that are known to work do not encounter run-time type errors which allows us to derive extra type information from the way values are used, rather than simply where those values originate. This is in keeping with the “duck typing” philosophy of many dynamically typed languages. The information we derive must be conservative, so we describe and formalise a technique to ‘freeze’ the duck type of a variable using the features, such as named methods, that are provably present on any run of the program. Development environments can use these sets of features to provide code-completion suggestions and API documentation, amongst other things. We show that these sets of features can be used to refine imprecise flow analysis results by using the frozen duck type to perform a structural type-cast. We first formalise this for an idealised duck-typed language semantics and then show to what extent the technique would work for a real-world language, Python. We demonstrate its effectiveness by performing an analysis of several real-world Python programs which shows that we can infer the types of method-call receivers more precisely than can flow analysis alone.
Supervisor: Wolf, Alexander Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral