I opened this PR without having made up my mind if we really should merge it or not. Having a left-recursive rule, that's called so frequently, open with an invalid alternative seems like it will have some performance implications. Should we maybe revisit backporting #22111 to 3.9 before merging this one?

CC @pablogsal @gvanrossum

I opened this PR without having made up my mind if we really should merge it or not. Having a left-recursive rule, that's called so frequently, open with an invalid alternative seems like it will have some performance implications. Should we maybe revisit backporting #22111 to 3.9 before merging this one?

CC @pablogsal @gvanrossum

I would say that is wise to have an idea of the impact of this before doing anything, definitely.

The stdlib benchmark went from 7.530s (mean from running it 4 times) in the 3.9 branch to 7.546s here.

@lysnikolaou: Please replace # with GH- in the commit message next time. Thanks!