Contribute all good stuff to mypy.stubgenc and freeze this repo

[sizmailov]

At the point when mypy.stubgen (or other) will be able to generate full-featured stubs for pybind11 there would be no need in this tool.

Things to ship:

• function signatures
• default argument parsing
• attribute types
• attribute values
• nested classes
• functions with positional-only arguments
• pybind11-specific type transformations (e.g. iterator -> typing.Iterator)

[virtuald]

FWIW, I like the smallness of this dependency. Just copy a single file to a repo and we're good to go. :)

[sizmailov]

Finally I've managed to make some progress here (see https://github.com/sizmailov/mypy/tree/enhance-stubgenc-pybind-compatibility).

Hopefully I'll be able to split mentioned branch into a couple of PRs and make it's way to official mypy repo.

I'll keep this issue update along the progress with PRs. Comments/reviews as always are very welcome.

[sizmailov]

First PR is merged 🎉 python/mypy#9877
A bit more is on the way.
With those two stubgen will be dangerously really close to pybind11-stubgen output :)

[hiaselhans]

thanks for the good work @sizmailov !! :)

[allanleal]

Great work. Please keep it on!

[varunsh-xilinx]

@sizmailov are you also planning on contributing your docstring parsing? Currently, the stubs from mypy don't have docstrings and there's an open issue suggesting it could be an added option.

[sizmailov]

Sorry, I don't have time/plans to contribute to mypy in foreseeable future.
I'm personally unsatisfied with my docstring parsing implementation to contribute it anywhere, especially to a mature project such as mypy.
At the same time, I don't mind if you reuse parts (or even all) of my code to create a PR to mypy or any other project. Feel free to do so without asking, and don't worry about my credits.

[cunnane]

Please don't give up :) I'm using your codebase to create documentation stubs: I want to publish docs on readthedocs but don't want to build the pybind binary there (in fact I can't because it's MSVC-only...). Just tweaking a few lines was all that was necessary for this use case

[sizmailov]

I didn't return to the problem for quite some time and I'm not planning to in foreseeable future. So closing as not planned.

[bluenote10]

Out of curiosity: What exactly is the relationship between this implementation and mypy.stubgen? Is one of them using the other, or did they start as a fork of each other or so? The original issue sounded like mypy.stubgen is the better place to have this functionality. So just curious what made you change your mind -- the amount of work involved, or are there simply too many pybind11 specific features involved that would making an integration into mypy difficult?

[sizmailov]

@bluenote10 Initially, I was unsatisfied with how mypy.stubgen performed on my pet project, and I decided to make a tool to get things done quickly.

At some point, I thought it would be nice to avoid the responsibility of solely maintaining a package with long-standing bugs/feature requests, so I've contributed a few features to mypy.stubgen. It was too slow and time-consuming and I was carried away by other things.

I think there is a need for stubgen as:

1. an opinionated CLI tool with a handful of options, for the most common use cases
2. a flexible library, to obtain fine-tuned results

I believe mypy.stubgen could satisfy the first but not the second. So, pybind11-stubgen and mypy.stubgen would co-exist for quite a while.