A safe language on a safe OS

Have you heard of “managed” code? Generally it refers to code that has no direct access to memory and instead has to access everything through a protected interface of sorts. The main advantages are that a program can’t go poking memory that it shouldn’t and useful rules can be enforced like type safety.

The most prevalent example of managed code is nearly everything running under .NET/Mono. Admittedly you can mark parts as “unsafe” letting you use pointers and stopping the garbage collector arbitrarily moving your data around but most of the usefulness comes from avoiding this where possible.

The problem is, you can’t always avoid it. The main reason for this is you have to access existing non managed systems. Rewriting everything in managed code is not feasible and although you can lessen any problems by writing wrappers so there is only one point of contact between managed and unmanaged code, problems can still occur - the sort of problems the managed code was supposed to prevent.

Microsoft are investigating a solution. After reading that last paragraph, the form of the solution should be obvious but for the most part it’s unworkable in the real world - eliminate all unmanaged code.

A lot of people claim managed code, or specifically .NET is slow and inefficient. Well it is. But it can be made faster. Most inefficiency is caused by a lot of run time checks to make sure everything is as it should be. If the code lives in an entirely managed world however most of these checks can be removed since (barring random hardware failure) the program can be guaranteed to satisfy the run time checks at compile time.

For more details about Singularity, Microsoft’s research operating system written in C#, check out this article by James Larus, Galen Hunt, and David Tarditi over at MSDN.