dwenzek

In this paper, Daniel revisits the ideas developed in his book, "The essence of software" [2]. placing them in their historical context and opening a future for more modular software implementations.

He presents a convincing argumentation on how and why an object-oriented approach does not really contribute to reducing coupling. And his proposal is truly appealing: building an application as a set of external synchronization rules governing the interactions between modules (so called concepts). Each module comes with its own view on the entities of the application domain and provides the actions featuring the concept. Only the synchronization rules of the application relate the entities and actions of these modules.

That said, I'm left wanting more and I will need to take the time to put these ideas in code.

One main goal of Rel is to give users the ability to grow the language using libraries [2], say for linear algebra or graph processing. And I'm really impressed by the result. With ability to defines (possibly recursive) relations over relation variables and arbitrary-length tuples, Rel is a nice successor of Datalog.

I also like the oppiniated approach which is to enforce strong data modeling constraints on all relations (6th normal form and identifier uniqueness across entity types). This removes the need for nulls and beyond helps to remove complicated corner cases.

Since long, I love Alistair's books which are extraordinary insightful. As an example, [1] has drastically changed my way of writing specs and even thinking about user-software interactions.

Now, it's good to discover that Alistair is also a great speaker highlighting the key points that will help to understand his point.