scoder

Let's wait with changes like this until that part of the C-API is stable. Python-dev is still discussing changes in this area.

> Therefore, split tuples into suitable and unsuitable, and use existing code generation for unsuitable. Don't we do that already with the `is_literal` and `is_partly_literal` flags?

Yes, you can have tuples with slices and slices with tuples. Difficult to map this to a table then.

This kind of breakage shouldn't happen in a point release, so it's an unintended regression. However, now that the regression is there, I wonder if we should really go back...

Don't worry. Being able to mix C, C++ and Python APIs freely is at the core of Cython's feature set. This is just a rather special case where C++ and...

I'm probably missing a lot of historic trial and error here, but at first sight, as it stands now, this feels hackish to me. Like you're passing the types through...

Thanks a lot for going through this, @nsait-linaro. > ` libxml used: (2, 9, 5)` While we're at it, we should upgrade to libxml2 2.9.10 and libxslt 1.1.34. That's what...

So, you said that you made a copy of libiconv. Could you maybe use a patch against the original sources instead? That would simplify upgrades (such as the imminent one).

> I can prepare a patch that > we can keep in this repository that can be applied to extend the > submodule libiconv for win/arm64. Is that what do...

In nogil CPython (which has already landed in CPython 3.13), you can probably just use `threading.RLock` instead – unless you're asking about fastrlock's C-API. The change that CPython needed seems...