OK, so it *is* a 32-bit problem. This is awesome :):

- on my 32-bit system the highest I can go with an error-less format-time-string is 

(format-time-string "%a %b %d %H:%M:%S %Y GMT" (list 32767 65535 0 0))

That will give a total of 

32767*2**16+65535 = 2147483647 = 2**31-1

seconds from the Unix epoch, i.e. spot-on the year-2038 problem:

https://en.m.wikipedia.org/wiki/Year_2038_problem

- on the other hand, on my 64-bit systems the highest I can go is 

(format-time-string "%a %b %d %H:%M:%S %Y GMT" (list 1034058169428 61391 0 0))

This puzzled me a bit, but looking at the date, it's Wed Dec 31 23:59:59 GMT year (wait for it)

2147485547 = 1900 + 2**31-1

So it's again the usual issue, kicked up to higher timestamps in the 64-bit case:

http://www.unixguide.net/sun/faq/6.27.shtml

On October 29, 2019 10:06:28 AM EDT, "A.C." <achirvasub@gmail.com> wrote:
>
>
>On October 29, 2019 8:47:02 AM EDT, Lars Ingebrigtsen <larsi@gnus.org>
>wrote:
>>"A.C." <achirvasub@gmail.com> writes:
>>
>>> On the other hand, the Debian i686 system that doesn't load
>arxiv.org
>>has a
>>> largest valid n of
>>
>>I'm not quite sure I understood the bug report -- are you saying that
>>this is only an issue on 32-bit systems?
>>
>
>I don't have access to enough systems to tell for sure. I do know it's
>only happening on the one 32-bit system I have access to right now, and
>also that I hit the problem a while back on 32-bit ARM:
>
>https://github.com/termux/termux-packages/issues/2614
>
>(and have not see the issue on any 64-bit systems)
>
>>In any case, Emacs 27 has bignum support, so this should hopefully not
>>be a problem any more.  Can you try the test case with the current
>>Emacs
>>trunk and see whether it's still present?
>
>That will be tough. This 32-bit machine overheats something fierce; I
>doubt I can even clone the git repo without it dying, let alone
>compile.
>
>-- 
>Sent from my Android device with K-9 Mail. Please excuse my brevity.

-- 
Sent from my Android device with K-9 Mail. Please excuse my brevity.