Minimal reproducer for breakage introduced by PR #3402.

[rwgk]

Description

WIP — Starting with a minimal reproducer.

Breakage introduced by PR #3402 (comment)

Suggested changelog entry:

[henryiii]

@jagerman, it looks like the new code doesn't handle self correctly. The old code counted from the end, so self was ignored (slightly incorrect. I helped write on the old code, but it's been quite some time). But the new code counts from the start, so it has to account for self.

[rwgk]

Hi @jagerman, could you please take a look? This is a really tiny reproducer. Looks like an edge case that's not handled correctly, maybe something simple like an off-by-one. I'm guessing a minor tweak will fix it, the trick is to know where. I'm hoping that you as the original author may have a clue straightaway.

[rwgk]

I expanded the new test in a trivially systematic way, placing the py::kw_only at different positions:

0: *, int i
1: int, *, int i
2: int, int, *, int i

Then defing (A) a simple function and (B) init.
Calling with i=1 works consistently for (A), fails for (B) 0, but works for (B) 1 & 2.

@jagerman does that give you any clues?

[jagerman]

I will look into this later today. I think you're right that there's some edge case being triggered here but will likely take some debugging through it to figure it out.

[rwgk]

Issue fixed by @jagerman's PR #3488 :-)