New backward-compatible NamedTuple by ilevkivskyi · Pull Request #282 · python/typing
But it's just a bug with a confusing diagnostic, right? If it gets less confusing, fine, but hardly a priority.
Indeed, I find the current behavior (binding yield inside comprehension auxiliary function scope) confusing. But I understand that this is personal (this is just something that I wanted to work on for a long time, so that I want to do this while I have some time for it).
But more important, in your opinion what are the high priority things to do in core CPython now? Maybe I could also try to do some of those things. I have got very interested in core CPython while implementing PEP 526 (to be honest I was impressed by CPython codebase in terms of quality and ease of work with it).
I can't actually follow that thread -- isn't there a bug for it?
Unfortunately, there is no separate issue for this yet, it was discovered in http://bugs.python.org/issue24379 , but again I don't think it is something high priority in general, it is rather interesting for me.
OK -- but isn't #207 also intending to fix that?
Yes, but I would like to propose an alternative way, by making most thing like Union etc. not actual classes, just objects that record they parameters. This is how Mark Shannon proposed to do this some time ago, I tend to agree with him, this way will solve __subclasscheck__ problem, but also some other problems and will give a significant speed-up (not for Generic[T] although, these must be classes). But, I propose to postpone this discussion until I will make a detailed plan (or even a PR) for this.