[tz] NetBSD vs Darwin timezone API

[Paul Eggert]

On 11/10/2016 03:17 PM, Tom Lane wrote:
> I think that for backwards compatibility's sake, if nothing else, the
> default behavior should be no auto-update.

As Christos says, we're talking only about localtime_rz and mktime_z. 
Emacs uses these on NetBSD, and typically will want autoupdate (because 
it wants to show the current time in the mode line, for example) but 
sometimes will not (because it wants to make several related requests to 
infer time zone rules). The Emacs Lisp API currently does not provide a 
way to request or to suppress autoupdate, so that will have to be added 
to the Elisp API regardless of what the tzallocate default is and so 
this does not give us much guidance on what the tzallocate default 
should be.

localtime and localtime_r could go either way; that is, they can invoke 
tzallocate and specify the autoupdate flag, regardless of the default. 
There are reasonable arguments in both directions. Darwin localtime and 
localtime_r do autoupdate, and grab a lock which is a performance 
bottleneck in multithreaded applications. NetBSD localtime and 
localtime_r do not do autoupdate, which fails to track changes.

All in all I'm now inclined to say that the default for tzallocate 
autoupdate should be off, as that's the traditional behavior, but this 
is merely a mild preference.