University of Cambridge > Talks.cam > Microsoft Research Cambridge, public talks > Tunable Static Inference for Generic Universe Types

Tunable Static Inference for Generic Universe Types

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Microsoft Research Cambridge Talks Admins.

This event may be recorded and made available internally or externally via http://research.microsoft.com. Microsoft will own the copyright of any recordings made. If you do not wish to have your image/voice recorded please consider this before attending

Object ownership is useful for many applications, including program verification, thread synchronization, and memory management. However, the annotation overhead of ownership type systems hampers their widespread application. This talk addresses this issue by presenting a tunable static type inference for Generic Universe Types. In contrast to classical type systems, ownership types have no single most general typing. Our inference chooses among the legal typings via heuristics. Our inference is tunable: users can indicate a preference for certain typings by adjusting the heuristics or by supplying partial annotations for the program. We present how the constraints of Generic Universe Types can be encoded as a boolean satisfiability (SAT) problem and how a weighted Max-SAT solver finds a correct Universe typing that optimizes the weights. We implemented the static inference tool, applied our inference tool to four real-world applications, and inferred interesting ownership structures.

This talk is part of the Microsoft Research Cambridge, public talks series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.

 

© 2006-2024 Talks.cam, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity