[tz] Troll throws zic for a loop

[Paul Eggert]

Marc Lehmann wrote:
> If the number of transitions in the table is ultimately bounded (which
> seems to be the case here), then not using malloc is safer, faster,
> simpler, more memory efficient, and cleaner for zic.

Having zic avoid malloc is less resilient in the face of future changes, 
such as the ones needed for Troll.  If zic had already been using 
malloc, we could ship the new Antarctica/Troll now, without worrying 
about old zic implementations rejecting it.  If zic starts using malloc 
now, future changes like this should be easier to deploy.

Plus, zic is already using malloc elsewhere, so using malloc for these 
cases doesn't introduce any fundamental new problems.

malloc is more problematic for localtime, not just in terms of runtime 
overhead, but in terms of safety and correctness: there are applications 
where it's not nice to display the time incorrectly (or not at all -- or 
worse, dump core) merely because malloc failed.  There are embedded 
environments where use of malloc is frowned upon entirely, for these 
reasons.  These considerations generally don't apply to zic, which makes 
it more reasonable to use malloc in zic than in localtime.