News

The content has been bookmarked!

There was an error bookmarking this content! Please retry.

Anders Hejlsberg and Guy Steele on Concurrency and Language Design

Posted by Jonathan Allen on Oct 22, 2008

Sections: Development,; Architecture & Design
Topics: Language Design ,; .NET ,; Performance & Scalability
Tags: Functional Programming ,; Parallel Programming

An interview recorded at JAOO, Anders Hejlsberg and Guy Steele talk about concurrency in language design. Anders Hejlsberg is the chief architect of C#, creator of Turbo Pascal, and former chief architect of Delphi. Guy Steele is the creator of Scheme and is currently working on Sun's Fortress, a programming language that focuses on parallelism and mathematics.

According to Anders, the change in emphasis is that the focus on parallelism is moving from coarse task and process based concurrency to a fine-grained concurrency where "you want to break a single logical task into multiple workloads".

One of the areas Anders is working on is finding a way for the compiler to know if an object has been "revealed to the rest of the world" and using that information to determine if it is safe to mutate. The challenge is that programmers have been taught to write in ways that are incompatible with where they need to go.

Guy follows with defining two types of concurrency, those that are fundamental to the application such as transactional processing, and those that are the result of not having a single CPU that is fast enough. For the latter type he claims that developers would be happy not to use multi-threading if it were a viable option. He does not care if the language is functional or not, as long as you can tell the compiler where it's OK to use parallelism.

Anders counters that developers do not necessarily know if their code, or the code it uses, are actually safe. This is where the idea of making pure functions, functions without side-effects, able to call only other pure functions.

When the topic turns to functional programming languages they both mentioned there is a lot of interest in it, but acknowledge the problems with retraining developers who have already heavily invested in imperative programming skills. Anders also brings up the issue of efficiency in functional languages; specifically algorithms like QuickSort perform poorly in functional languages. He prefers a hybrid approach where there are "islands of purity" in an otherwise imperative language with a clear delineation between them.