[tz] [libc-coord] thread-safe localtime() for an arbitrary timezone

[Guy Harris]

On Jun 20, 2023, at 7:16 PM, Paul Gilmartin via tz <tz at iana.org> wrote:

> ( the Archives of this thread seem incomplete.)

tz mailing list archives, or lib-coord:

	https://www.openwall.com/lists/libc-coord/2020/01/30/1

archives?  The former are incomplete because the thread didn't start on the tz mailing list; the first message, to libc-coord, was linked from Paul Eggert's message to tz in this thread:

	https://www.openwall.com/lists/libc-coord/2023/06/14/1

> On 6/20/23 15:00:29, enh via tz wrote:
>>    ...
>>    We could rename 'struct state', but I'm a bit lost as to what the
>>    problem is, as I don't see mention of 'state' or 'timezone_t' in
>>    openbsd/lib/libc/time/wcsftime.c. Although there is a 'struct state' in
>>    openbsd/lib/libc/time/localtime.c I assume you're replacing that with
>>    tzcode's localtime.c somehow.
> 
> (Replying mostly to the Subject:)
> A thread-safe localtime() seems impractical. localtime() depends
> on the environment variable TZ.  If concurrent threads set
> different values of TZ the results will be unpredictable.  The
> same applies to localtime_r().

If concurrent thread set different values of TZ, perhaps the concurrent threads should use a reentrant localtime_z() routine, possibly with a different name, as has been proposed in a number of places, including in this list back in 2001:

	http://mm.icann.org/pipermail/tz/2001-June/011636.html

That requires making such a routine common on multiple platforms, of course....

And, in fact, in the first message in the thread, the author said

	is anyone aware of any standardization work in this area? i know netbsd
	has tzalloc()/localtime_rz()/tzfree() (and tzcode has implementations), but
	they're the only ones who've shipped anything, right? and there's no
	in-progress work on any alternative?

So the proposal there is for localtime_rz(), taking a "time zone" structure, or pointer to same, provided by tzalloc() and freed by jtzfree(), rather than something that looks at TZ.  (Perhaps the title was a bit misleading, and should have said "lcoaltime() equivalent".)