unsetenv() patch for TZ

[Yaakov Selkowitz]

On Mon, 2015-03-30 at 11:38 -0400, Craig Howland wrote:
>       OK.  While I disagree that this is the best approach, it is your 
> prerogative and attached is a patch for the 
> ctime()/localtime()/mktime()/strftime() approach.  It does have the benefit of 
> also fixing the problem identified by Yaakov in "[PATCH] strftime: use tzname if 
> TM_ZONE is NULL" (https://sourceware.org/ml/newlib/2015/msg00321.html).

My patch would still be needed as posted, this just takes care of the
additional questions posted therein.

>       Since ctime() calls localtime(), it is not directly patched (i.e. the 
> patch only directly adds to 3 of the 4 functions mentioned).  The strftime.c 
> patch also gets the wcsftime() function due to the shared source.
>       I made a minor size/efficiency call in strftime(), and put in two tzset() 
> calls, one for %z and one for %Z, so that tzset() will only be called if it is 
> needed, at the expense of one more function call in the overall size.  (Which is 
> what Corinna says GLIBC does.0

Corinna also mentioned that it was guarded from multiple invocations by
what sounded like a local variable.

-- 
Yaakov Selkowitz
Associate Software Engineer, ARM
Red Hat, Inc.