[PATCH 1/2] Merge mktime, timegm from upstream Gnulib

[Paul Eggert]

Florian Weimer wrote:
> Is this the expected new behavior on 32-bit architectures?

No, tst-mktime2.c should still work on 32-bit architectures, and it does work 
for me with glibc master on i686 with GCC 7.3. So I'm a bit puzzled why it 
doesn't work for armv7l-unknown-linux-gnueabihf with GCC 7.3. Almost surely this 
is a bug I introduced, though it is possible it's a bug in localtime that the 
new mktime unearthed.

With a bit of luck the patches I just proposed in 
<https://www.sourceware.org/ml/libc-alpha/2018-11/msg00239.html> will fix the 
problem you reported, as there are indeed some obscure bugs in the current 
master near the Y2038 boundary.