I will like to expand on this a little bit. What we have tried about 3 years ago was an idea that I knew it would fail: perform type inference of an Elixir codebase using a well defined typing-discipline (HM with intersection types). The goal was mostly to assess how much we would have to change the language to provide proper inference. The answer was a lot.
Since then our focus is to use the typing information we already have in our Elixir codebases (which is already considerable, given pattern matching, guards, protocols, etc) and emit warnings based on that. The recent data constructors warnings in v1.11 are built on top of this foundation and we will continue adding more. We don’t want to add any new syntax nor explicit types for this to work.
Also, as a personal parenthesis, I think adding types via type annotations (like we have today) is a missed opportunity since types allow us to express some things a bit more cleanly (like enums, which we don’t have). I did wish some languages that added types after the fact provided a more integrated experience. This is not something we are exploring, but more of a general thought that I have on the topic.
