GeezerVison

cc @ericsnowcurrently

@willingc Changes committed.

Let's give @ericsnowcurrently a chance to review too. Thanks. Please ping me if this is still open later this week and I will merge.

A Python core developer has requested some changes be made to your pull request before we can consider merging it. If you could please address their requests along with any other requests in other reviews from core developers that would be appreciated.

Once you have made the requested changes, please leave a comment on this pull request containing the phrase I have made the requested changes; please review again. I will then notify any core developers who have left a review that you're ready for them to take another look at this pull request.

@vstinner, I made the function internal instead. PTAL

I have made the requested changes; please review again .

Thanks for making the requested changes!

@vstinner: please review the changes made to this pull request.

Thanks @nanjekyejoannah! It now looks better: remove the function rather than documenting it :-) I prefer to have a smaller C API!

This breaks pybind 11. Opened an issue with them (pybind/pybind11#1932). Where this should be fixed is a bit over my head at the moment.

This breaks pybind 11. Opened an issue with them (pybind/pybind11#1932). Where this should be fixed is a bit over my head at the moment.

Your gil_scoped_acquire class implementation is scary! Would you mind to open a new issue at bugs.python.org to describe your use case and ask to revert the change which removed PyThreadState_DeleteCurrent?

Pybind11 supports some advanced GIL-related tricks that require access this function. For instance, pybind11 can intercept a Python function call on the main() thread and delete the associated thread state, launching a new thread and continuing Python execution there (with a newly created thread state).

Kind of crazy, but why is this useful? UI libraries. On some platforms, it is not legal to poll UI events on any thread other than the main thread. This means that it's impossible to implement a proper UI event loop because Python is hogging the main thread. With the functionality in pybind11's gil_scoped_acquire, I can launch an event polling loop on the main thread, continue running an interactive Python session on another thread, and even swap them back e.g. when the user interface is no longer needed.

@vstinner: I provided some feedback on the Python bugtracker and here.

See https://bugs.python.org/issue38266.

I'm confused by the attached issue; it looks like the decision was to simply revert the change with the function moved, and not add the underscore; and maybe even make it slightly easier to use. Now that 3.9b1 is out, PyBind11-based libraries cannot build, and the only fix would be to start using a now internal function in CPython. SciPy is an example of a PyBind11-using library.

PyThreadState_DeleteCurrent() is declared by the public C API, but it's now excluded from the limited C API. Does pybind11 uses the limited C API? What is your issue? Please comment https://bugs.python.org/issue37878 instead of this merged PR (if you have a bugs.python.org account).

I'm sorry, I'll add a bugs.python.org account or see if I had one in the past if I need to comment there. I realize, from looking through the code, that an underscore was not added, the function was moved intact (which was not clear from the PR), exactly as it was supposed to be in the issue. The only remaining problem is that the release notes here are wrong: https://docs.python.org/3.9/whatsnew/3.9.html#removed (which is a much better problem).

I'd love it if PyBind11 could use the limited API, but it can't.

Thank you, @vstinner !

The only remaining problem is that the release notes here are wrong: https://docs.python.org/3.9/whatsnew/3.9.html#removed (which is a much better problem).

I created #20489 to update the What's New in Python 3.9 document. Would you mind to review my PR?