The result of calling getTimezone() on a Date object results in a 
DateTimeZone object with a reference to the dateobj-> time->tz_info 
object which may get later destroyed.

This can cause unexpected script behaviour or interpreter crashes, test 
case in attachment (1).

When I fixed this the obvious way, per attachment 3, by adding a clone, 
I wondered where the cloned tzinfo structures would get destroyed and I 
can't see anywhere.

After looking at this further - so far as I can tell, the duplication of 
the tzinfo structures throughout this code is not actually necessary; 
when the structures are created they are referenced from the global 
tzcache and will hence last "forever" anyway.  The structures are not 
changed anywhere either, again AFAICT; though they aren't treated as 
const so maybe I'm missing something there.

So simply copying pointers around would simplify the code, fix leaks and 
fix the bug as well.  That's attachment (2).

What do you think?

Regards, Joe