Add mypy type hints?

[Daenyth]

This would help people who'd like to use sortedcontainers in a project with mypy type checking.

https://github.com/python/mypy/wiki/Creating-Stubs-For-Python-Modules

[grantjenks]

I am open to it. Pull request welcome.

[Daenyth]

Fair enough. If I can get some time to devote to it I'll see what I can do.

[solomon-b]

If this is still desired, I would like to take this issue on.

[solomon-b]

I wrote a stubfile for SortedList and SortedListWithKeys. I've never written a type hints for someone else's project so I'm hoping to get some feedback before I continue with the remaining classes.

https://github.com/ssbothwell/sorted_containers/blob/master/sortedcontainers/sortedlist.pyi

Thanks!

[Daenyth]

Off the bat I think it's probably wrong to use the Orderable type that you've made there. As far as I know, it works for any type defining inequality protocols.

The class should probably extend Generic[_A] where _A = TypeVar('_A'). Unfortunately as far as I'm aware there's no way to restrict it to be orderable types, but the alternative is to say Iterable[Any] and throw out element safety.

[solomon-b]

Thanks for the feedback. The Orderable type alias seemed questionable to me but I went with it as an initial draft.

I'm assuming that in this type of situation the type hints are meant mainly for IDE support rather then static analysis. Given that, maybe Iterable[Any] is not a big problem. I could also alias Iterable[Any] to Orderable as a hint for developers that the method is actually taking an orderable object while still not breaking static analysis.

What do you think?

[solomon-b]

PEP484 has a solution using a Type Variable with an upper bound. Any type replacing the type variable must implement the __lt__ method.

This ensures only types with inequality are passed but not that they all can be compared to each other. It sounds like this is the best solution we are going to get for the time being. I updated my stub file to include this.

On another note, mypy throws some "incompatible with supertype" errors for methods in SortedListWithKey that overload those in SortedList. An example is key() which returns None in SortedList and returns the key function in SortedListWithKey. I think this is violating the covariance rule for return types.

I also get an error about MutableSequence (which SortedList inherits from) not being defined. Do I need to import the base class into the stubfile?

[grantjenks]

Sorry for the silence on this issue. I don't remember getting email updates.

I've been working on V2 which will adopt Python 3 semantics for some APIs while still being backwards-compatible with Python 2. I'm afraid I've changed some of the parameter names which will require updating in your pyi file.

I'm also not sure how to test your pyi file.

[bryanforbes]

If you're still open to this, I can work up a PR

[grantjenks]

I'm open to it.

Still prefer pyi over source changes. And I'm not sure how to test it.

It'll be a learning experience for me.

[Daenyth]

I'm not actively doing python at the moment but do feel free to ping me with any questions, I'd be happy to help you work through any issues you come across.

[Madoshakalaka]

@grantjenks Oh, I just saw this thread after I submitted #136 . What do you think of the pull request?

[nacitar]

I'd just like to express my interest in this feature. Has there been any more progress with the pull request?