localtime(&t) != gmtime(&t) bug

[Paul Eggert]

Ooops!  ado is right about standards conformance problems with my
proposed patch.  I looked at the C Standard and it turns out that even
tz-1994g has problems, because the standard requires that
localtime(&t) == gmtime(&t).  The relevant text is in section 4.12.3
of X3.159-1989:

	Except for the strftime function, these functions [asctime,
	ctime, gmtime, localtime] return values in one of two static
	objects: a broken-down time structure and an array of char.

So not only must ctime(&t) == asctime(&tm) (as tz-1994g provides),
localtime(&t) == gmtime(&t) is also required.  I assume that the
same is true of offtime.  Here's a patch.

===================================================================
RCS file: RCS/localtime.c,v
retrieving revision 1994.6.1.4
retrieving revision 1994.6.1.5
diff -c -r1994.6.1.4 -r1994.6.1.5
*** localtime.c	1994/05/11 21:32:24	1994.6.1.4
--- localtime.c	1994/05/11 21:50:59	1994.6.1.5
***************
*** 150,155 ****
--- 150,161 ----
  
  char *			tzname[2] = { wildabbr, wildabbr };
  
+ /*
+ ** The C Standard requires that localtime and gmtime
+ ** return values in the same object.
+ */
+ static struct tm tm;
+ 
  #ifdef USG_COMPAT
  time_t			timezone = 0;
  int			daylight = 0;
***************
*** 964,971 ****
  localtime(timep)
  const time_t * const	timep;
  {
- 	static struct tm	tm;
- 
  	localsub(timep, 0L, &tm);
  	return &tm;
  }
--- 970,975 ----
***************
*** 1014,1021 ****
  gmtime(timep)
  const time_t * const	timep;
  {
- 	static struct tm	tm;
- 
  	gmtsub(timep, 0L, &tm);
  	return &tm;
  }
--- 1018,1023 ----
***************
*** 1027,1034 ****
  const time_t * const	timep;
  const long		offset;
  {
- 	static struct tm	tm;
- 
  	gmtsub(timep, offset, &tm);
  	return &tm;
  }
--- 1029,1034 ----