Pip Cet wrote:
>> So we now silently replace the argument with another, smaller value,
>> and then go ahead as if business as usual?  Is that a good,
>> user-friendly behavior?
> It's certainly incorrect for circular lists.

Yes, sorry, I forgot the circular case. Fixed by installing the attached patch.

This patch improves circular-list nthcdr performance for fixnums too. For 
example, on my Fedora 28 x86-64 platform (AMD Phenom II X4 910e, circa 2010) the 
third line of the following benchmark runs about 8 million times faster:

(setq bench-circular (list 1 2 3 4 5 6))
(setcdr (nthcdr 5 bench-circular) bench-circular)
(nthcdr 536870911 bench-circular)

Normally I wouldn't bother with this sort of performance improvement (I mean, 
how often to people write code that deliberately goes around in circles? :-), 
but nthcdr is used so often that it seemed worth doing. Plus it was fun to fix 
this mostly by using machine arithmetic rather than GMP.