The paper is focused around two main use cases for linear types. The one that really catches my imagination is encoding the usage protocol of a resource (e.g., you can’t read from a file before it is opened) in types. “Here, linearity does not affect efficiency, but rather eliminates many bugs.” We’ll return to some examples of this later in this post. It seems like there’s a very nice and natural fit to protocols in distributed systems (a bit like we looked at with Disel yesterday).

The second use case is safe update in place, and the reason we’re interested in that is efficiency – if we know it’s safe to update a value in place we don’t need to worry about making copies, or mutual exclusion mechanisms.