Dialyzer as it exists today is a net negative value add for me. I only use it if forced. In principle it checks out, but in practice it always costs more than it saves.

I use it everywhere and all the time. After 6 years of full time Elixir I think it’s one of the most underrated tools available to us. I think dialyzer is only as good as you use it. More (and better) the types, the better it works. I think when it’s sprinkled in a code base and the type definitions aren’t very clear (like using any, term, keyword, and map types) most places the tool doesn’t really help. Being really clear in the typing and use of type docs also provides a benefit to the documentation of the code and even that provides longer lasting benefits to maintenance over time. It’s one of those tools where you get out of it what you put in IMO. It’s not perfect in developer experience but it has provided me massive value over the years. Drop me into a Elixir project with very little to no types I’d probably struggle, I am not smart enough to not use types in my code base.

As someone who is working a lot with Typespecs, for instance while working on the library TypeCheck, here’s some of the struggles I have with using Dialyzer:

• Adding it into any project of reasonable size is slow, sometimes unsustainably so.
• When it reports a problem, it is often very unclear why it is caused, and even moreso how it might be fixed. For example, help with this long-standing ‘function call with opaqueness mitchmatch’ problem would be greatly appreciated.
• Elixir’s standard library itself does not follow Dialyzer’s opaqueness rules correctly, and fixing this would be a difficult and backwards-incompatible change. Essentially this means that pattern-matching on “pattern matching on the struct name allowed” structs like MapSets will make Dialyzer complain about usage which is recommended by Elixir’s official documentation. (Source: elixir/#8463 and elixir/#8480)

However, I still use it regularly, especially on smaller and simpler projects. I like the ‘belt ánd suspenders’ kind of approach w.r.t. detecting potential bugs early.
And as @eksperimental already noted: Gradient/Gradualizer is a wonderful new project, and very much worth keeping check on!