[davnicwil]

I find it catches a lot of design mistakes for me which I would put in the non-trivial bugs category. Many bugs that result from design mistakes are less immediate, like hard to spot edge cases (a better design makes them impossible) or code that lends itself to becoming messier and more bug prone in the future as it's changed/extended (a better design makes this less likely).

[jshen]

I’ve had the opposite experience with complex type systems. They make it harder to evolve out of bad designs. This is particularly true when a lot of inheritance is used.

[davnicwil]

Interesting. What I've found is that when I come up with some pattern that is hard to type, or I have two parts of a system that need to work together but the interface can't be typed simply or doesn't quite 'fit', it is almost always the case that changing things around until the types do work straightforwardly uncovers a better design.

This is definitely sometimes not easy, and of course it'll get more difficult generally the more code there is or the more complex (usually meaning bad!) the design is, but I'd almost put it the other way round to your perspective - the types are like the signposts in this design evolution and discovery process. They don't make it easier to do the work of refactoring a design, but they do make it much easier to understand when I'm going in a good/bad direction and also when I have succeeded.

[jshen]

You are describing writing code alone. I’m thinking about other peoples code I’ve had to deal with where I didn’t create the types. Not saying I’m great at it, but that you can’t always control the quality of the code you have to work on.

[zarzavat]

That is true for nominal type systems such as Java’s.

However TypeScript’s structural typing is much more flexible, because it is designed to describe the types of all the existing JavaScript code out there (via the DefinitelyTyped project).

This gives the best of both worlds: the freedom to express the architecture you need, plus the freedom to actually refactor an existing code base into that new architecture without introducing many bugs. Dynamically typed languages give you the former but not the latter.