Image logo Bob conference 2017.

Talk: 12:35-13:20 (English)

Performance and Safety: an Example of using Liquid Haskell in the Real World

No one should ever be forced to choose between performant and robust software. Yet sometimes, when we have to squeeze every bit of performance from our code, we are tempted to leave abstractions behind and go down to the level of bare pointer arithmetic. Thus, we reduce all overhead for the sake of improving speed, at the price of risking segmentation faults and worse.

Using the example of the Store serialisation library, we show how to get the best of both worlds, by using Liquid Haskell. Liquid Haskell is an extension of Haskell’s Type system that allows annotating types with refinement predicates. For example, this allows us to encode, in the type system, that some pointer points to a region in memory of a certain size. Thus, the absence of pointer overflows can be encoded in the types of our functions, and be proven by Liquid Haskell. This is superior to performing bounds checks in several ways:

  • There is no runtime overhead.
  • Since the correctness is statically proven, we do not need to handle errors at runtime.
  • Writing down the involved invariances in Liquid Haskell helps to get a precise understanding of our code and possible sources of errors.

The talk aims to be practical, showing in detail how to use Liquid Haskell to prove the correctness of assumptions about your code.

Philipp Kant

Philipp is a physicist and spent several years doing calculations in theoretical particle physics. In 2014, he turned to software development. He now works at FP Complete, where he is concerned with implementing high performance, parallelised software for scientific simulations.

Slides
kant.html
Video
http://youtu.be/8Y78eTaEkBU