zonefile changes and long running processes.

[Paul Eggert]

Ondřej Bílka wrote:
> its often better to crash than return date 2.3.1972

Isn't the general glibc preference to use whatever's simplest and 
cheapest when the behavior isn't specified?  Even if supporting 
atomicity is not that expensive -- which remains to be seen -- most 
likely it will cost *something*.  As I understand it, a design goal of 
POSIX's localtime_r spec was to not require interlocking with tzset, so 
as implementers why not exploit this to improve localtime_r's performance?

> patch will not write itself

Absolutely!