I'm going to shill Gradualizer[1] in here as well.
1. I think you have to see typing as helpful information. Like:
-spec add(integer(), integer()) -> integer().
add(A, B) -> A + B.
Technically, nothing prevents add from accepting number() or float(). It's too much specification.
2. I was surprised how much there actually is but also disappointed that a lot of errors were in until OTP 23. Gradualizer also overrides some definitions [2]
3. Haven't encountered a problem with this yet with gradualizer, would need to test.
Yeah, I had thought that development of Gradualizer had halted, glad to see it's not. Part of why I wanted to do Mavis is because I wanted to make it easy for developers to make their own custom overrides and ship it with their libraries, especially for Elixir protocols[0]. still haven't figured out how to architect that, but I have an idea.
I think you have to see typing as helpful information
Indeed. Selectrix will allow you to chain your inference with fallbacks, so, maybe the first level is a fast cache, second level is checking files for user overrides, third level is dialyzer specs, fourth level is inference.
I think it should also be possible to create an addon module with Mavis that lets you type inbound messages for GenServers, given a set of constraints (all possible calls and casts are present in the module itself as an API)
[0] it's a big problem that in elixir, Enumerable protocol destroys typing information, so I want to make an typing override for Enumerable an "example" typing override system.
I think it did partially stopped. I'm the only one committing right now but some of the original devs will review PRs and are EXTREMELY helpful. Great bunch.
Comments
I'm going to shill Gradualizer[1] in here as well.
1. I think you have to see typing as helpful information. Like:
-spec add(integer(), integer()) -> integer(). add(A, B) -> A + B.
Technically, nothing prevents add from accepting number() or float(). It's too much specification.
2. I was surprised how much there actually is but also disappointed that a lot of errors were in until OTP 23. Gradualizer also overrides some definitions [2]
3. Haven't encountered a problem with this yet with gradualizer, would need to test.
[1] https://github.com/josefs/Gradualizer [2] https://github.com/josefs/Gradualizer/blob/master/priv/prelu...
Yeah, I had thought that development of Gradualizer had halted, glad to see it's not. Part of why I wanted to do Mavis is because I wanted to make it easy for developers to make their own custom overrides and ship it with their libraries, especially for Elixir protocols[0]. still haven't figured out how to architect that, but I have an idea.
Indeed. Selectrix will allow you to chain your inference with fallbacks, so, maybe the first level is a fast cache, second level is checking files for user overrides, third level is dialyzer specs, fourth level is inference.
I think it should also be possible to create an addon module with Mavis that lets you type inbound messages for GenServers, given a set of constraints (all possible calls and casts are present in the module itself as an API)
[0] it's a big problem that in elixir, Enumerable protocol destroys typing information, so I want to make an typing override for Enumerable an "example" typing override system.
I think it did partially stopped. I'm the only one committing right now but some of the original devs will review PRs and are EXTREMELY helpful. Great bunch.